diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2016-10-14 20:43:08 (GMT) |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2016-10-14 20:43:08 (GMT) |
commit | ac9ef8cd075a1efee13036908a99a4bd5f9d4b8c (patch) | |
tree | 4375b2aa77cd7fcc52f2e8ca2de599924536ff2d /drivers/infiniband | |
parent | b292fb80bb44726ac1055d443d951a3058fc8263 (diff) | |
parent | 993d1b52615e1a549e55875c3b74308391672d9f (diff) | |
download | linux-ac9ef8cd075a1efee13036908a99a4bd5f9d4b8c.tar.xz |
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Pull rdma qedr RoCE driver from Doug Ledford:
"Early on in the merge window I mentioned I had a backlog of new
drivers waiting to be reviewed and that, in addition to the hns-roce
driver, I wanted to get possible a couple more reviewed. I ended up
only having the time to complete one of the additional drivers.
During Dave Miller's pull request this go around, there were a series
of 9 patches to the QLogic qed net driver that add basic support for a
paired RoCE driver. That support is currently not functional because
it is missing the matching RoCE driver in the RDMA subsystem. I
managed to finish that review. However, because it goes against part
of Dave's net pull, and a part that was accepted a day or two after
the merge window opened, to apply cleanly it has to be applied to
either the tip of Dave's net branch, or as I did in this case, I just
applied it to your master after you had taken Dave's pull request."
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma:
qedr: Add events support and register IB device
qedr: Add GSI support
qedr: Add LL2 RoCE interface
qedr: Add support for data path
qedr: Add support for memory registeration verbs
qedr: Add support for QP verbs
qedr: Add support for PD,PKEY and CQ verbs
qedr: Add support for user context verbs
qedr: Add support for RoCE HW init
qedr: Add RoCE driver framework
Diffstat (limited to 'drivers/infiniband')
-rw-r--r-- | drivers/infiniband/Kconfig | 2 | ||||
-rw-r--r-- | drivers/infiniband/hw/Makefile | 1 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/Kconfig | 7 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/Makefile | 3 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/main.c | 914 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/qedr.h | 495 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/qedr_cm.c | 622 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/qedr_cm.h | 61 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/qedr_hsi.h | 56 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 748 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/verbs.c | 3547 | ||||
-rw-r--r-- | drivers/infiniband/hw/qedr/verbs.h | 101 |
12 files changed, 6557 insertions, 0 deletions
diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig index 19a418a..fb3fb89 100644 --- a/drivers/infiniband/Kconfig +++ b/drivers/infiniband/Kconfig @@ -89,4 +89,6 @@ source "drivers/infiniband/sw/rxe/Kconfig" source "drivers/infiniband/hw/hfi1/Kconfig" +source "drivers/infiniband/hw/qedr/Kconfig" + endif # INFINIBAND diff --git a/drivers/infiniband/hw/Makefile b/drivers/infiniband/hw/Makefile index 21fe401..e7a5ed9 100644 --- a/drivers/infiniband/hw/Makefile +++ b/drivers/infiniband/hw/Makefile @@ -10,3 +10,4 @@ obj-$(CONFIG_INFINIBAND_OCRDMA) += ocrdma/ obj-$(CONFIG_INFINIBAND_USNIC) += usnic/ obj-$(CONFIG_INFINIBAND_HFI1) += hfi1/ obj-$(CONFIG_INFINIBAND_HNS) += hns/ +obj-$(CONFIG_INFINIBAND_QEDR) += qedr/ diff --git a/drivers/infiniband/hw/qedr/Kconfig b/drivers/infiniband/hw/qedr/Kconfig new file mode 100644 index 0000000..7c06d85 --- /dev/null +++ b/drivers/infiniband/hw/qedr/Kconfig @@ -0,0 +1,7 @@ +config INFINIBAND_QEDR + tristate "QLogic RoCE driver" + depends on 64BIT && QEDE + select QED_LL2 + ---help--- + This driver provides low-level InfiniBand over Ethernet + support for QLogic QED host channel adapters (HCAs). diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile new file mode 100644 index 0000000..ba7067c --- /dev/null +++ b/drivers/infiniband/hw/qedr/Makefile @@ -0,0 +1,3 @@ +obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o + +qedr-y := main.o verbs.o qedr_cm.o diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c new file mode 100644 index 0000000..7b74d09 --- /dev/null +++ b/drivers/infiniband/hw/qedr/main.c @@ -0,0 +1,914 @@ +/* QLogic qedr NIC Driver + * Copyright (c) 2015-2016 QLogic Corporation + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and /or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include <linux/module.h> +#include <rdma/ib_verbs.h> +#include <rdma/ib_addr.h> +#include <rdma/ib_user_verbs.h> +#include <linux/netdevice.h> +#include <linux/iommu.h> +#include <net/addrconf.h> +#include <linux/qed/qede_roce.h> +#include <linux/qed/qed_chain.h> +#include <linux/qed/qed_if.h> +#include "qedr.h" +#include "verbs.h" +#include <rdma/qedr-abi.h> + +MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver"); +MODULE_AUTHOR("QLogic Corporation"); +MODULE_LICENSE("Dual BSD/GPL"); +MODULE_VERSION(QEDR_MODULE_VERSION); + +#define QEDR_WQ_MULTIPLIER_DFT (3) + +void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num, + enum ib_event_type type) +{ + struct ib_event ibev; + + ibev.device = &dev->ibdev; + ibev.element.port_num = port_num; + ibev.event = type; + + ib_dispatch_event(&ibev); +} + +static enum rdma_link_layer qedr_link_layer(struct ib_device *device, + u8 port_num) +{ + return IB_LINK_LAYER_ETHERNET; +} + +static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str, + size_t str_len) +{ + struct qedr_dev *qedr = get_qedr_dev(ibdev); + u32 fw_ver = (u32)qedr->attr.fw_ver; + + snprintf(str, str_len, "%d. %d. %d. %d", + (fw_ver >> 24) & 0xFF, (fw_ver >> 16) & 0xFF, + (fw_ver >> 8) & 0xFF, fw_ver & 0xFF); +} + +static struct net_device *qedr_get_netdev(struct ib_device *dev, u8 port_num) +{ + struct qedr_dev *qdev; + + qdev = get_qedr_dev(dev); + dev_hold(qdev->ndev); + + /* The HW vendor's device driver must guarantee + * that this function returns NULL before the net device reaches + * NETDEV_UNREGISTER_FINAL state. + */ + return qdev->ndev; +} + +static int qedr_register_device(struct qedr_dev *dev) +{ + strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX); + + dev->ibdev.node_guid = dev->attr.node_guid; + memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC)); + dev->ibdev.owner = THIS_MODULE; + dev->ibdev.uverbs_abi_ver = QEDR_ABI_VERSION; + + dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) | + QEDR_UVERBS(QUERY_DEVICE) | + QEDR_UVERBS(QUERY_PORT) | + QEDR_UVERBS(ALLOC_PD) | + QEDR_UVERBS(DEALLOC_PD) | + QEDR_UVERBS(CREATE_COMP_CHANNEL) | + QEDR_UVERBS(CREATE_CQ) | + QEDR_UVERBS(RESIZE_CQ) | + QEDR_UVERBS(DESTROY_CQ) | + QEDR_UVERBS(REQ_NOTIFY_CQ) | + QEDR_UVERBS(CREATE_QP) | + QEDR_UVERBS(MODIFY_QP) | + QEDR_UVERBS(QUERY_QP) | + QEDR_UVERBS(DESTROY_QP) | + QEDR_UVERBS(REG_MR) | + QEDR_UVERBS(DEREG_MR) | + QEDR_UVERBS(POLL_CQ) | + QEDR_UVERBS(POST_SEND) | + QEDR_UVERBS(POST_RECV); + + dev->ibdev.phys_port_cnt = 1; + dev->ibdev.num_comp_vectors = dev->num_cnq; + dev->ibdev.node_type = RDMA_NODE_IB_CA; + + dev->ibdev.query_device = qedr_query_device; + dev->ibdev.query_port = qedr_query_port; + dev->ibdev.modify_port = qedr_modify_port; + + dev->ibdev.query_gid = qedr_query_gid; + dev->ibdev.add_gid = qedr_add_gid; + dev->ibdev.del_gid = qedr_del_gid; + + dev->ibdev.alloc_ucontext = qedr_alloc_ucontext; + dev->ibdev.dealloc_ucontext = qedr_dealloc_ucontext; + dev->ibdev.mmap = qedr_mmap; + + dev->ibdev.alloc_pd = qedr_alloc_pd; + dev->ibdev.dealloc_pd = qedr_dealloc_pd; + + dev->ibdev.create_cq = qedr_create_cq; + dev->ibdev.destroy_cq = qedr_destroy_cq; + dev->ibdev.resize_cq = qedr_resize_cq; + dev->ibdev.req_notify_cq = qedr_arm_cq; + + dev->ibdev.create_qp = qedr_create_qp; + dev->ibdev.modify_qp = qedr_modify_qp; + dev->ibdev.query_qp = qedr_query_qp; + dev->ibdev.destroy_qp = qedr_destroy_qp; + + dev->ibdev.query_pkey = qedr_query_pkey; + + dev->ibdev.create_ah = qedr_create_ah; + dev->ibdev.destroy_ah = qedr_destroy_ah; + + dev->ibdev.get_dma_mr = qedr_get_dma_mr; + dev->ibdev.dereg_mr = qedr_dereg_mr; + dev->ibdev.reg_user_mr = qedr_reg_user_mr; + dev->ibdev.alloc_mr = qedr_alloc_mr; + dev->ibdev.map_mr_sg = qedr_map_mr_sg; + + dev->ibdev.poll_cq = qedr_poll_cq; + dev->ibdev.post_send = qedr_post_send; + dev->ibdev.post_recv = qedr_post_recv; + + dev->ibdev.process_mad = qedr_process_mad; + dev->ibdev.get_port_immutable = qedr_port_immutable; + dev->ibdev.get_netdev = qedr_get_netdev; + + dev->ibdev.dma_device = &dev->pdev->dev; + + dev->ibdev.get_link_layer = qedr_link_layer; + dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str; + + return ib_register_device(&dev->ibdev, NULL); +} + +/* This function allocates fast-path status block memory */ +static int qedr_alloc_mem_sb(struct qedr_dev *dev, + struct qed_sb_info *sb_info, u16 sb_id) +{ + struct status_block *sb_virt; + dma_addr_t sb_phys; + int rc; + + sb_virt = dma_alloc_coherent(&dev->pdev->dev, + sizeof(*sb_virt), &sb_phys, GFP_KERNEL); + if (!sb_virt) + return -ENOMEM; + + rc = dev->ops->common->sb_init(dev->cdev, sb_info, + sb_virt, sb_phys, sb_id, + QED_SB_TYPE_CNQ); + if (rc) { + pr_err("Status block initialization failed\n"); + dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt), + sb_virt, sb_phys); + return rc; + } + + return 0; +} + +static void qedr_free_mem_sb(struct qedr_dev *dev, + struct qed_sb_info *sb_info, int sb_id) +{ + if (sb_info->sb_virt) { + dev->ops->common->sb_release(dev->cdev, sb_info, sb_id); + dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt), + (void *)sb_info->sb_virt, sb_info->sb_phys); + } +} + +static void qedr_free_resources(struct qedr_dev *dev) +{ + int i; + + for (i = 0; i < dev->num_cnq; i++) { + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i); + dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl); + } + + kfree(dev->cnq_array); + kfree(dev->sb_array); + kfree(dev->sgid_tbl); +} + +static int qedr_alloc_resources(struct qedr_dev *dev) +{ + struct qedr_cnq *cnq; + __le16 *cons_pi; + u16 n_entries; + int i, rc; + + dev->sgid_tbl = kzalloc(sizeof(union ib_gid) * + QEDR_MAX_SGID, GFP_KERNEL); + if (!dev->sgid_tbl) + return -ENOMEM; + + spin_lock_init(&dev->sgid_lock); + + /* Allocate Status blocks for CNQ */ + dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array), + GFP_KERNEL); + if (!dev->sb_array) { + rc = -ENOMEM; + goto err1; + } + + dev->cnq_array = kcalloc(dev->num_cnq, + sizeof(*dev->cnq_array), GFP_KERNEL); + if (!dev->cnq_array) { + rc = -ENOMEM; + goto err2; + } + + dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev); + + /* Allocate CNQ PBLs */ + n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE); + for (i = 0; i < dev->num_cnq; i++) { + cnq = &dev->cnq_array[i]; + + rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i], + dev->sb_start + i); + if (rc) + goto err3; + + rc = dev->ops->common->chain_alloc(dev->cdev, + QED_CHAIN_USE_TO_CONSUME, + QED_CHAIN_MODE_PBL, + QED_CHAIN_CNT_TYPE_U16, + n_entries, + sizeof(struct regpair *), + &cnq->pbl); + if (rc) + goto err4; + + cnq->dev = dev; + cnq->sb = &dev->sb_array[i]; + cons_pi = dev->sb_array[i].sb_virt->pi_array; + cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX]; + cnq->index = i; + sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev)); + + DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n", + i, qed_chain_get_cons_idx(&cnq->pbl)); + } + + return 0; +err4: + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i); +err3: + for (--i; i >= 0; i--) { + dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl); + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i); + } + kfree(dev->cnq_array); +err2: + kfree(dev->sb_array); +err1: + kfree(dev->sgid_tbl); + return rc; +} + +/* QEDR sysfs interface */ +static ssize_t show_rev(struct device *device, struct device_attribute *attr, + char *buf) +{ + struct qedr_dev *dev = dev_get_drvdata(device); + + return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor); +} + +static ssize_t show_hca_type(struct device *device, + struct device_attribute *attr, char *buf) +{ + return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET"); +} + +static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL); +static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL); + +static struct device_attribute *qedr_attributes[] = { + &dev_attr_hw_rev, + &dev_attr_hca_type +}; + +static void qedr_remove_sysfiles(struct qedr_dev *dev) +{ + int i; + + for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++) + device_remove_file(&dev->ibdev.dev, qedr_attributes[i]); +} + +static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev) +{ + struct pci_dev *bridge; + u32 val; + + dev->atomic_cap = IB_ATOMIC_NONE; + + bridge = pdev->bus->self; + if (!bridge) + return; + + /* Check whether we are connected directly or via a switch */ + while (bridge && bridge->bus->parent) { + DP_DEBUG(dev, QEDR_MSG_INIT, + "Device is not connected directly to root. bridge->bus->number=%d primary=%d\n", + bridge->bus->number, bridge->bus->primary); + /* Need to check Atomic Op Routing Supported all the way to + * root complex. + */ + pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val); + if (!(val & PCI_EXP_DEVCAP2_ATOMIC_ROUTE)) { + pcie_capability_clear_word(pdev, + PCI_EXP_DEVCTL2, + PCI_EXP_DEVCTL2_ATOMIC_REQ); + return; + } + bridge = bridge->bus->parent->self; + } + bridge = pdev->bus->self; + + /* according to bridge capability */ + pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val); + if (val & PCI_EXP_DEVCAP2_ATOMIC_COMP64) { + pcie_capability_set_word(pdev, PCI_EXP_DEVCTL2, + PCI_EXP_DEVCTL2_ATOMIC_REQ); + dev->atomic_cap = IB_ATOMIC_GLOB; + } else { + pcie_capability_clear_word(pdev, PCI_EXP_DEVCTL2, + PCI_EXP_DEVCTL2_ATOMIC_REQ); + } +} + +static const struct qed_rdma_ops *qed_ops; + +#define HILO_U64(hi, lo) ((((u64)(hi)) << 32) + (lo)) + +static irqreturn_t qedr_irq_handler(int irq, void *handle) +{ + u16 hw_comp_cons, sw_comp_cons; + struct qedr_cnq *cnq = handle; + struct regpair *cq_handle; + struct qedr_cq *cq; + + qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0); + + qed_sb_update_sb_idx(cnq->sb); + + hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr); + sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl); + + /* Align protocol-index and chain reads */ + rmb(); + + while (sw_comp_cons != hw_comp_cons) { + cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl); + cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi, + cq_handle->lo); + + if (cq == NULL) { + DP_ERR(cnq->dev, + "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n", + cq_handle->hi, cq_handle->lo, sw_comp_cons, + hw_comp_cons); + + break; + } + + if (cq->sig != QEDR_CQ_MAGIC_NUMBER) { + DP_ERR(cnq->dev, + "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n", + cq_handle->hi, cq_handle->lo, cq); + break; + } + + cq->arm_flags = 0; + + if (cq->ibcq.comp_handler) + (*cq->ibcq.comp_handler) + (&cq->ibcq, cq->ibcq.cq_context); + + sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl); + + cnq->n_comp++; + + } + + qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index, + sw_comp_cons); + + qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1); + + return IRQ_HANDLED; +} + +static void qedr_sync_free_irqs(struct qedr_dev *dev) +{ + u32 vector; + int i; + + for (i = 0; i < dev->int_info.used_cnt; i++) { + if (dev->int_info.msix_cnt) { + vector = dev->int_info.msix[i * dev->num_hwfns].vector; + synchronize_irq(vector); + free_irq(vector, &dev->cnq_array[i]); + } + } + + dev->int_info.used_cnt = 0; +} + +static int qedr_req_msix_irqs(struct qedr_dev *dev) +{ + int i, rc = 0; + + if (dev->num_cnq > dev->int_info.msix_cnt) { + DP_ERR(dev, + "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n", + dev->num_cnq, dev->int_info.msix_cnt); + return -EINVAL; + } + + for (i = 0; i < dev->num_cnq; i++) { + rc = request_irq(dev->int_info.msix[i * dev->num_hwfns].vector, + qedr_irq_handler, 0, dev->cnq_array[i].name, + &dev->cnq_array[i]); + if (rc) { + DP_ERR(dev, "Request cnq %d irq failed\n", i); + qedr_sync_free_irqs(dev); + } else { + DP_DEBUG(dev, QEDR_MSG_INIT, + "Requested cnq irq for %s [entry %d]. Cookie is at %p\n", + dev->cnq_array[i].name, i, + &dev->cnq_array[i]); + dev->int_info.used_cnt++; + } + } + + return rc; +} + +static int qedr_setup_irqs(struct qedr_dev *dev) +{ + int rc; + + DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n"); + + /* Learn Interrupt configuration */ + rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq); + if (rc < 0) + return rc; + + rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info); + if (rc) { + DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n"); + return rc; + } + + if (dev->int_info.msix_cnt) { + DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n", + dev->int_info.msix_cnt); + rc = qedr_req_msix_irqs(dev); + if (rc) + return rc; + } + + DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n"); + + return 0; +} + +static int qedr_set_device_attr(struct qedr_dev *dev) +{ + struct qed_rdma_device *qed_attr; + struct qedr_device_attr *attr; + u32 page_size; + + /* Part 1 - query core capabilities */ + qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx); + + /* Part 2 - check capabilities */ + page_size = ~dev->attr.page_size_caps + 1; + if (page_size > PAGE_SIZE) { + DP_ERR(dev, + "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n", + PAGE_SIZE, page_size); + return -ENODEV; + } + + /* Part 3 - copy and update capabilities */ + attr = &dev->attr; + attr->vendor_id = qed_attr->vendor_id; + attr->vendor_part_id = qed_attr->vendor_part_id; + attr->hw_ver = qed_attr->hw_ver; + attr->fw_ver = qed_attr->fw_ver; + attr->node_guid = qed_attr->node_guid; + attr->sys_image_guid = qed_attr->sys_image_guid; + attr->max_cnq = qed_attr->max_cnq; + attr->max_sge = qed_attr->max_sge; + attr->max_inline = qed_attr->max_inline; + attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE); + attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE); + attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc; + attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc; + attr->max_dev_resp_rd_atomic_resc = + qed_attr->max_dev_resp_rd_atomic_resc; + attr->max_cq = qed_attr->max_cq; + attr->max_qp = qed_attr->max_qp; + attr->max_mr = qed_attr->max_mr; + attr->max_mr_size = qed_attr->max_mr_size; + attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES); + attr->max_mw = qed_attr->max_mw; + attr->max_fmr = qed_attr->max_fmr; + attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl; + attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size; + attr->max_pd = qed_attr->max_pd; + attr->max_ah = qed_attr->max_ah; + attr->max_pkey = qed_attr->max_pkey; + attr->max_srq = qed_attr->max_srq; + attr->max_srq_wr = qed_attr->max_srq_wr; + attr->dev_caps = qed_attr->dev_caps; + attr->page_size_caps = qed_attr->page_size_caps; + attr->dev_ack_delay = qed_attr->dev_ack_delay; + attr->reserved_lkey = qed_attr->reserved_lkey; + attr->bad_pkey_counter = qed_attr->bad_pkey_counter; + attr->max_stats_queues = qed_attr->max_stats_queues; + + return 0; +} + +void qedr_unaffiliated_event(void *context, + u8 event_code) +{ + pr_err("unaffiliated event not implemented yet\n"); +} + +void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle) +{ +#define EVENT_TYPE_NOT_DEFINED 0 +#define EVENT_TYPE_CQ 1 +#define EVENT_TYPE_QP 2 + struct qedr_dev *dev = (struct qedr_dev *)context; + union event_ring_data *data = fw_handle; + u64 roce_handle64 = ((u64)data->roce_handle.hi << 32) + + data->roce_handle.lo; + u8 event_type = EVENT_TYPE_NOT_DEFINED; + struct ib_event event; + struct ib_cq *ibcq; + struct ib_qp *ibqp; + struct qedr_cq *cq; + struct qedr_qp *qp; + + switch (e_code) { + case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR: + event.event = IB_EVENT_CQ_ERR; + event_type = EVENT_TYPE_CQ; + break; + case ROCE_ASYNC_EVENT_SQ_DRAINED: + event.event = IB_EVENT_SQ_DRAINED; + event_type = EVENT_TYPE_QP; + break; + case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR: + event.event = IB_EVENT_QP_FATAL; + event_type = EVENT_TYPE_QP; + break; + case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR: + event.event = IB_EVENT_QP_REQ_ERR; + event_type = EVENT_TYPE_QP; + break; + case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR: + event.event = IB_EVENT_QP_ACCESS_ERR; + event_type = EVENT_TYPE_QP; + break; + default: + DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code, + roce_handle64); + } + + switch (event_type) { + case EVENT_TYPE_CQ: + cq = (struct qedr_cq *)(uintptr_t)roce_handle64; + if (cq) { + ibcq = &cq->ibcq; + if (ibcq->event_handler) { + event.device = ibcq->device; + event.element.cq = ibcq; + ibcq->event_handler(&event, ibcq->cq_context); + } + } else { + WARN(1, + "Error: CQ event with NULL pointer ibcq. Handle=%llx\n", + roce_handle64); + } + DP_ERR(dev, "CQ event %d on hanlde %p\n", e_code, cq); + break; + case EVENT_TYPE_QP: + qp = (struct qedr_qp *)(uintptr_t)roce_handle64; + if (qp) { + ibqp = &qp->ibqp; + if (ibqp->event_handler) { + event.device = ibqp->device; + event.element.qp = ibqp; + ibqp->event_handler(&event, ibqp->qp_context); + } + } else { + WARN(1, + "Error: QP event with NULL pointer ibqp. Handle=%llx\n", + roce_handle64); + } + DP_ERR(dev, "QP event %d on hanlde %p\n", e_code, qp); + break; + default: + break; + } +} + +static int qedr_init_hw(struct qedr_dev *dev) +{ + struct qed_rdma_add_user_out_params out_params; + struct qed_rdma_start_in_params *in_params; + struct qed_rdma_cnq_params *cur_pbl; + struct qed_rdma_events events; + dma_addr_t p_phys_table; + u32 page_cnt; + int rc = 0; + int i; + + in_params = kzalloc(sizeof(*in_params), GFP_KERNEL); + if (!in_params) { + rc = -ENOMEM; + goto out; + } + + in_params->desired_cnq = dev->num_cnq; + for (i = 0; i < dev->num_cnq; i++) { + cur_pbl = &in_params->cnq_pbl_list[i]; + + page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl); + cur_pbl->num_pbl_pages = page_cnt; + + p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl); + cur_pbl->pbl_ptr = (u64)p_phys_table; + } + + events.affiliated_event = qedr_affiliated_event; + events.unaffiliated_event = qedr_unaffiliated_event; + events.context = dev; + + in_params->events = &events; + in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS; + in_params->max_mtu = dev->ndev->mtu; + ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr); + + rc = dev->ops->rdma_init(dev->cdev, in_params); + if (rc) + goto out; + + rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params); + if (rc) + goto out; + + dev->db_addr = (void *)(uintptr_t)out_params.dpi_addr; + dev->db_phys_addr = out_params.dpi_phys_addr; + dev->db_size = out_params.dpi_size; + dev->dpi = out_params.dpi; + + rc = qedr_set_device_attr(dev); +out: + kfree(in_params); + if (rc) + DP_ERR(dev, "Init HW Failed rc = %d\n", rc); + + return rc; +} + +void qedr_stop_hw(struct qedr_dev *dev) +{ + dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi); + dev->ops->rdma_stop(dev->rdma_ctx); +} + +static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev, + struct net_device *ndev) +{ + struct qed_dev_rdma_info dev_info; + struct qedr_dev *dev; + int rc = 0, i; + + dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev)); + if (!dev) { + pr_err("Unable to allocate ib device\n"); + return NULL; + } + + DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n"); + + dev->pdev = pdev; + dev->ndev = ndev; + dev->cdev = cdev; + + qed_ops = qed_get_rdma_ops(); + if (!qed_ops) { + DP_ERR(dev, "Failed to get qed roce operations\n"); + goto init_err; + } + + dev->ops = qed_ops; + rc = qed_ops->fill_dev_info(cdev, &dev_info); + if (rc) + goto init_err; + + dev->num_hwfns = dev_info.common.num_hwfns; + dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev); + + dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev); + if (!dev->num_cnq) { + DP_ERR(dev, "not enough CNQ resources.\n"); + goto init_err; + } + + dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT; + + qedr_pci_set_atomic(dev, pdev); + + rc = qedr_alloc_resources(dev); + if (rc) + goto init_err; + + rc = qedr_init_hw(dev); + if (rc) + goto alloc_err; + + rc = qedr_setup_irqs(dev); + if (rc) + goto irq_err; + + rc = qedr_register_device(dev); + if (rc) { + DP_ERR(dev, "Unable to allocate register device\n"); + goto reg_err; + } + + for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++) + if (device_create_file(&dev->ibdev.dev, qedr_attributes[i])) + goto sysfs_err; + + DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n"); + return dev; + +sysfs_err: + ib_unregister_device(&dev->ibdev); +reg_err: + qedr_sync_free_irqs(dev); +irq_err: + qedr_stop_hw(dev); +alloc_err: + qedr_free_resources(dev); +init_err: + ib_dealloc_device(&dev->ibdev); + DP_ERR(dev, "qedr driver load failed rc=%d\n", rc); + + return NULL; +} + +static void qedr_remove(struct qedr_dev *dev) +{ + /* First unregister with stack to stop all the active traffic + * of the registered clients. + */ + qedr_remove_sysfiles(dev); + ib_unregister_device(&dev->ibdev); + + qedr_stop_hw(dev); + qedr_sync_free_irqs(dev); + qedr_free_resources(dev); + ib_dealloc_device(&dev->ibdev); +} + +static int qedr_close(struct qedr_dev *dev) +{ + qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ERR); + + return 0; +} + +static void qedr_shutdown(struct qedr_dev *dev) +{ + qedr_close(dev); + qedr_remove(dev); +} + +static void qedr_mac_address_change(struct qedr_dev *dev) +{ + union ib_gid *sgid = &dev->sgid_tbl[0]; + u8 guid[8], mac_addr[6]; + int rc; + + /* Update SGID */ + ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr); + guid[0] = mac_addr[0] ^ 2; + guid[1] = mac_addr[1]; + guid[2] = mac_addr[2]; + guid[3] = 0xff; + guid[4] = 0xfe; + guid[5] = mac_addr[3]; + guid[6] = mac_addr[4]; + guid[7] = mac_addr[5]; + sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL); + memcpy(&sgid->raw[8], guid, sizeof(guid)); + + /* Update LL2 */ + rc = dev->ops->roce_ll2_set_mac_filter(dev->cdev, + dev->gsi_ll2_mac_address, + dev->ndev->dev_addr); + + ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr); + + qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE); + + if (rc) + DP_ERR(dev, "Error updating mac filter\n"); +} + +/* event handling via NIC driver ensures that all the NIC specific + * initialization done before RoCE driver notifies + * event to stack. + */ +static void qedr_notify(struct qedr_dev *dev, enum qede_roce_event event) +{ + switch (event) { + case QEDE_UP: + qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE); + break; + case QEDE_DOWN: + qedr_close(dev); + break; + case QEDE_CLOSE: + qedr_shutdown(dev); + break; + case QEDE_CHANGE_ADDR: + qedr_mac_address_change(dev); + break; + default: + pr_err("Event not supported\n"); + } +} + +static struct qedr_driver qedr_drv = { + .name = "qedr_driver", + .add = qedr_add, + .remove = qedr_remove, + .notify = qedr_notify, +}; + +static int __init qedr_init_module(void) +{ + return qede_roce_register_driver(&qedr_drv); +} + +static void __exit qedr_exit_module(void) +{ + qede_roce_unregister_driver(&qedr_drv); +} + +module_init(qedr_init_module); +module_exit(qedr_exit_module); diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h new file mode 100644 index 0000000..620badd --- /dev/null +++ b/drivers/infiniband/hw/qedr/qedr.h @@ -0,0 +1,495 @@ +/* QLogic qedr NIC Driver + * Copyright (c) 2015-2016 QLogic Corporation + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and /or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef __QEDR_H__ +#define __QEDR_H__ + +#include <linux/pci.h> +#include <rdma/ib_addr.h> +#include <linux/qed/qed_if.h> +#include <linux/qed/qed_chain.h> +#include <linux/qed/qed_roce_if.h> +#include <linux/qed/qede_roce.h> +#include "qedr_hsi.h" + +#define QEDR_MODULE_VERSION "8.10.10.0" +#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA" +#define DP_NAME(dev) ((dev)->ibdev.name) + +#define DP_DEBUG(dev, module, fmt, ...) \ + pr_debug("(%s) " module ": " fmt, \ + DP_NAME(dev) ? DP_NAME(dev) : "", ## __VA_ARGS__) + +#define QEDR_MSG_INIT "INIT" +#define QEDR_MSG_MISC "MISC" +#define QEDR_MSG_CQ " CQ" +#define QEDR_MSG_MR " MR" +#define QEDR_MSG_RQ " RQ" +#define QEDR_MSG_SQ " SQ" +#define QEDR_MSG_QP " QP" +#define QEDR_MSG_GSI " GSI" + +#define QEDR_CQ_MAGIC_NUMBER (0x11223344) + +struct qedr_dev; + +struct qedr_cnq { + struct qedr_dev *dev; + struct qed_chain pbl; + struct qed_sb_info *sb; + char name[32]; + u64 n_comp; + __le16 *hw_cons_ptr; + u8 index; +}; + +#define QEDR_MAX_SGID 128 + +struct qedr_device_attr { + u32 vendor_id; + u32 vendor_part_id; + u32 hw_ver; + u64 fw_ver; + u64 node_guid; + u64 sys_image_guid; + u8 max_cnq; + u8 max_sge; + u16 max_inline; + u32 max_sqe; + u32 max_rqe; + u8 max_qp_resp_rd_atomic_resc; + u8 max_qp_req_rd_atomic_resc; + u64 max_dev_resp_rd_atomic_resc; + u32 max_cq; + u32 max_qp; + u32 max_mr; + u64 max_mr_size; + u32 max_cqe; + u32 max_mw; + u32 max_fmr; + u32 max_mr_mw_fmr_pbl; + u64 max_mr_mw_fmr_size; + u32 max_pd; + u32 max_ah; + u8 max_pkey; + u32 max_srq; + u32 max_srq_wr; + u8 max_srq_sge; + u8 max_stats_queues; + u32 dev_caps; + + u64 page_size_caps; + u8 dev_ack_delay; + u32 reserved_lkey; + u32 bad_pkey_counter; + struct qed_rdma_events events; +}; + +struct qedr_dev { + struct ib_device ibdev; + struct qed_dev *cdev; + struct pci_dev *pdev; + struct net_device *ndev; + + enum ib_atomic_cap atomic_cap; + + void *rdma_ctx; + struct qedr_device_attr attr; + + const struct qed_rdma_ops *ops; + struct qed_int_info int_info; + + struct qed_sb_info *sb_array; + struct qedr_cnq *cnq_array; + int num_cnq; + int sb_start; + + void __iomem *db_addr; + u64 db_phys_addr; + u32 db_size; + u16 dpi; + + union ib_gid *sgid_tbl; + + /* Lock for sgid table */ + spinlock_t sgid_lock; + + u64 guid; + + u32 dp_module; + u8 dp_level; + u8 num_hwfns; + uint wq_multiplier; + u8 gsi_ll2_mac_address[ETH_ALEN]; + int gsi_qp_created; + struct qedr_cq *gsi_sqcq; + struct qedr_cq *gsi_rqcq; + struct qedr_qp *gsi_qp; +}; + +#define QEDR_MAX_SQ_PBL (0x8000) +#define QEDR_MAX_SQ_PBL_ENTRIES (0x10000 / sizeof(void *)) +#define QEDR_SQE_ELEMENT_SIZE (sizeof(struct rdma_sq_sge)) +#define QEDR_MAX_SQE_ELEMENTS_PER_SQE (ROCE_REQ_MAX_SINGLE_SQ_WQE_SIZE / \ + QEDR_SQE_ELEMENT_SIZE) +#define QEDR_MAX_SQE_ELEMENTS_PER_PAGE ((RDMA_RING_PAGE_SIZE) / \ + QEDR_SQE_ELEMENT_SIZE) +#define QEDR_MAX_SQE ((QEDR_MAX_SQ_PBL_ENTRIES) *\ + (RDMA_RING_PAGE_SIZE) / \ + (QEDR_SQE_ELEMENT_SIZE) /\ + (QEDR_MAX_SQE_ELEMENTS_PER_SQE)) +/* RQ */ +#define QEDR_MAX_RQ_PBL (0x2000) +#define QEDR_MAX_RQ_PBL_ENTRIES (0x10000 / sizeof(void *)) +#define QEDR_RQE_ELEMENT_SIZE (sizeof(struct rdma_rq_sge)) +#define QEDR_MAX_RQE_ELEMENTS_PER_RQE (RDMA_MAX_SGE_PER_RQ_WQE) +#define QEDR_MAX_RQE_ELEMENTS_PER_PAGE ((RDMA_RING_PAGE_SIZE) / \ + QEDR_RQE_ELEMENT_SIZE) +#define QEDR_MAX_RQE ((QEDR_MAX_RQ_PBL_ENTRIES) *\ + (RDMA_RING_PAGE_SIZE) / \ + (QEDR_RQE_ELEMENT_SIZE) /\ + (QEDR_MAX_RQE_ELEMENTS_PER_RQE)) + +#define QEDR_CQE_SIZE (sizeof(union rdma_cqe)) +#define QEDR_MAX_CQE_PBL_SIZE (512 * 1024) +#define QEDR_MAX_CQE_PBL_ENTRIES (((QEDR_MAX_CQE_PBL_SIZE) / \ + sizeof(u64)) - 1) +#define QEDR_MAX_CQES ((u32)((QEDR_MAX_CQE_PBL_ENTRIES) * \ + (QED_CHAIN_PAGE_SIZE) / QEDR_CQE_SIZE)) + +#define QEDR_ROCE_MAX_CNQ_SIZE (0x4000) + +#define QEDR_MAX_PORT (1) + +#define QEDR_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME) + +#define QEDR_ROCE_PKEY_MAX 1 +#define QEDR_ROCE_PKEY_TABLE_LEN 1 +#define QEDR_ROCE_PKEY_DEFAULT 0xffff + +struct qedr_pbl { + struct list_head list_entry; + void *va; + dma_addr_t pa; +}; + +struct qedr_ucontext { + struct ib_ucontext ibucontext; + struct qedr_dev *dev; + struct qedr_pd *pd; + u64 dpi_addr; + u64 dpi_phys_addr; + u32 dpi_size; + u16 dpi; + + struct list_head mm_head; + + /* Lock to protect mm list */ + struct mutex mm_list_lock; +}; + +union db_prod64 { + struct rdma_pwm_val32_data data; + u64 raw; +}; + +enum qedr_cq_type { + QEDR_CQ_TYPE_GSI, + QEDR_CQ_TYPE_KERNEL, + QEDR_CQ_TYPE_USER, +}; + +struct qedr_pbl_info { + u32 num_pbls; + u32 num_pbes; + u32 pbl_size; + u32 pbe_size; + bool two_layered; +}; + +struct qedr_userq { + struct ib_umem *umem; + struct qedr_pbl_info pbl_info; + struct qedr_pbl *pbl_tbl; + u64 buf_addr; + size_t buf_len; +}; + +struct qedr_cq { + struct ib_cq ibcq; + + enum qedr_cq_type cq_type; + u32 sig; + + u16 icid; + + /* Lock to protect completion handler */ + spinlock_t comp_handler_lock; + + /* Lock to protect multiplem CQ's */ + spinlock_t cq_lock; + u8 arm_flags; + struct qed_chain pbl; + + void __iomem *db_addr; + union db_prod64 db; + + u8 pbl_toggle; + union rdma_cqe *latest_cqe; + union rdma_cqe *toggle_cqe; + + u32 cq_cons; + + struct qedr_userq q; +}; + +struct qedr_pd { + struct ib_pd ibpd; + u32 pd_id; + struct qedr_ucontext *uctx; +}; + +struct qedr_mm { + struct { + u64 phy_addr; + unsigned long len; + } key; + struct list_head entry; +}; + +union db_prod32 { + struct rdma_pwm_val16_data data; + u32 raw; +}; + +struct qedr_qp_hwq_info { + /* WQE Elements */ + struct qed_chain pbl; + u64 p_phys_addr_tbl; + u32 max_sges; + + /* WQE */ + u16 prod; + u16 cons; + u16 wqe_cons; + u16 gsi_cons; + u16 max_wr; + + /* DB */ + void __iomem *db; + union db_prod32 db_data; +}; + +#define QEDR_INC_SW_IDX(p_info, index) \ + do { \ + p_info->index = (p_info->index + 1) & \ + qed_chain_get_capacity(p_info->pbl) \ + } while (0) + +enum qedr_qp_err_bitmap { + QEDR_QP_ERR_SQ_FULL = 1, + QEDR_QP_ERR_RQ_FULL = 2, + QEDR_QP_ERR_BAD_SR = 4, + QEDR_QP_ERR_BAD_RR = 8, + QEDR_QP_ERR_SQ_PBL_FULL = 16, + QEDR_QP_ERR_RQ_PBL_FULL = 32, +}; + +struct qedr_qp { + struct ib_qp ibqp; /* must be first */ + struct qedr_dev *dev; + + struct qedr_qp_hwq_info sq; + struct qedr_qp_hwq_info rq; + + u32 max_inline_data; + + /* Lock for QP's */ + spinlock_t q_lock; + struct qedr_cq *sq_cq; + struct qedr_cq *rq_cq; + struct qedr_srq *srq; + enum qed_roce_qp_state state; + u32 id; + struct qedr_pd *pd; + enum ib_qp_type qp_type; + struct qed_rdma_qp *qed_qp; + u32 qp_id; + u16 icid; + u16 mtu; + int sgid_idx; + u32 rq_psn; + u32 sq_psn; + u32 qkey; + u32 dest_qp_num; + + /* Relevant to qps created from kernel space only (ULPs) */ + u8 prev_wqe_size; + u16 wqe_cons; + u32 err_bitmap; + bool signaled; + + /* SQ shadow */ + struct { + u64 wr_id; + enum ib_wc_opcode opcode; + u32 bytes_len; + u8 wqe_size; + bool signaled; + dma_addr_t icrc_mapping; + u32 *icrc; + struct qedr_mr *mr; + } *wqe_wr_id; + + /* RQ shadow */ + struct { + u64 wr_id; + struct ib_sge sg_list[RDMA_MAX_SGE_PER_RQ_WQE]; + u8 wqe_size; + + u8 smac[ETH_ALEN]; + u16 vlan_id; + int rc; + } *rqe_wr_id; + + /* Relevant to qps created from user space only (applications) */ + struct qedr_userq usq; + struct qedr_userq urq; +}; + +struct qedr_ah { + struct ib_ah ibah; + struct ib_ah_attr attr; +}; + +enum qedr_mr_type { + QEDR_MR_USER, + QEDR_MR_KERNEL, + QEDR_MR_DMA, + QEDR_MR_FRMR, +}; + +struct mr_info { + struct qedr_pbl *pbl_table; + struct qedr_pbl_info pbl_info; + struct list_head free_pbl_list; + struct list_head inuse_pbl_list; + u32 completed; + u32 completed_handled; +}; + +struct qedr_mr { + struct ib_mr ibmr; + struct ib_umem *umem; + + struct qed_rdma_register_tid_in_params hw_mr; + enum qedr_mr_type type; + + struct qedr_dev *dev; + struct mr_info info; + + u64 *pages; + u32 npages; +}; + +#define SET_FIELD2(value, name, flag) ((value) |= ((flag) << (name ## _SHIFT))) + +#define QEDR_RESP_IMM (RDMA_CQE_RESPONDER_IMM_FLG_MASK << \ + RDMA_CQE_RESPONDER_IMM_FLG_SHIFT) +#define QEDR_RESP_RDMA (RDMA_CQE_RESPONDER_RDMA_FLG_MASK << \ + RDMA_CQE_RESPONDER_RDMA_FLG_SHIFT) +#define QEDR_RESP_RDMA_IMM (QEDR_RESP_IMM | QEDR_RESP_RDMA) + +static inline void qedr_inc_sw_cons(struct qedr_qp_hwq_info *info) +{ + info->cons = (info->cons + 1) % info->max_wr; + info->wqe_cons++; +} + +static inline void qedr_inc_sw_prod(struct qedr_qp_hwq_info *info) +{ + info->prod = (info->prod + 1) % info->max_wr; +} + +static inline int qedr_get_dmac(struct qedr_dev *dev, + struct ib_ah_attr *ah_attr, u8 *mac_addr) +{ + union ib_gid zero_sgid = { { 0 } }; + struct in6_addr in6; + + if (!memcmp(&ah_attr->grh.dgid, &zero_sgid, sizeof(union ib_gid))) { + DP_ERR(dev, "Local port GID not supported\n"); + eth_zero_addr(mac_addr); + return -EINVAL; + } + + memcpy(&in6, ah_attr->grh.dgid.raw, sizeof(in6)); + ether_addr_copy(mac_addr, ah_attr->dmac); + + return 0; +} + +static inline +struct qedr_ucontext *get_qedr_ucontext(struct ib_ucontext *ibucontext) +{ + return container_of(ibucontext, struct qedr_ucontext, ibucontext); +} + +static inline struct qedr_dev *get_qedr_dev(struct ib_device *ibdev) +{ + return container_of(ibdev, struct qedr_dev, ibdev); +} + +static inline struct qedr_pd *get_qedr_pd(struct ib_pd *ibpd) +{ + return container_of(ibpd, struct qedr_pd, ibpd); +} + +static inline struct qedr_cq *get_qedr_cq(struct ib_cq *ibcq) +{ + return container_of(ibcq, struct qedr_cq, ibcq); +} + +static inline struct qedr_qp *get_qedr_qp(struct ib_qp *ibqp) +{ + return container_of(ibqp, struct qedr_qp, ibqp); +} + +static inline struct qedr_ah *get_qedr_ah(struct ib_ah *ibah) +{ + return container_of(ibah, struct qedr_ah, ibah); +} + +static inline struct qedr_mr *get_qedr_mr(struct ib_mr *ibmr) +{ + return container_of(ibmr, struct qedr_mr, ibmr); +} +#endif diff --git a/drivers/infiniband/hw/qedr/qedr_cm.c b/drivers/infiniband/hw/qedr/qedr_cm.c new file mode 100644 index 0000000..63890eb --- /dev/null +++ b/drivers/infiniband/hw/qedr/qedr_cm.c @@ -0,0 +1,622 @@ +/* QLogic qedr NIC Driver + * Copyright (c) 2015-2016 QLogic Corporation + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and /or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include <linux/dma-mapping.h> +#include <linux/crc32.h> +#include <linux/iommu.h> +#include <net/ip.h> +#include <net/ipv6.h> +#include <net/udp.h> + +#include <rdma/ib_verbs.h> +#include <rdma/ib_user_verbs.h> +#include <rdma/iw_cm.h> +#include <rdma/ib_umem.h> +#include <rdma/ib_addr.h> +#include <rdma/ib_cache.h> + +#include "qedr_hsi.h" +#include <linux/qed/qed_if.h> +#include <linux/qed/qed_roce_if.h> +#include "qedr.h" +#include "qedr_hsi.h" +#include "verbs.h" +#include <rdma/qedr-abi.h> +#include "qedr_hsi.h" +#include "qedr_cm.h" + +void qedr_inc_sw_gsi_cons(struct qedr_qp_hwq_info *info) +{ + info->gsi_cons = (info->gsi_cons + 1) % info->max_wr; +} + +void qedr_store_gsi_qp_cq(struct qedr_dev *dev, struct qedr_qp *qp, + struct ib_qp_init_attr *attrs) +{ + dev->gsi_qp_created = 1; + dev->gsi_sqcq = get_qedr_cq(attrs->send_cq); + dev->gsi_rqcq = get_qedr_cq(attrs->recv_cq); + dev->gsi_qp = qp; +} + +void qedr_ll2_tx_cb(void *_qdev, struct qed_roce_ll2_packet *pkt) +{ + struct qedr_dev *dev = (struct qedr_dev *)_qdev; + struct qedr_cq *cq = dev->gsi_sqcq; + struct qedr_qp *qp = dev->gsi_qp; + unsigned long flags; + + DP_DEBUG(dev, QEDR_MSG_GSI, + "LL2 TX CB: gsi_sqcq=%p, gsi_rqcq=%p, gsi_cons=%d, ibcq_comp=%s\n", + dev->gsi_sqcq, dev->gsi_rqcq, qp->sq.gsi_cons, + cq->ibcq.comp_handler ? "Yes" : "No"); + + dma_free_coherent(&dev->pdev->dev, pkt->header.len, pkt->header.vaddr, + pkt->header.baddr); + kfree(pkt); + + spin_lock_irqsave(&qp->q_lock, flags); + qedr_inc_sw_gsi_cons(&qp->sq); + spin_unlock_irqrestore(&qp->q_lock, flags); + + if (cq->ibcq.comp_handler) { + spin_lock_irqsave(&cq->comp_handler_lock, flags); + (*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context); + spin_unlock_irqrestore(&cq->comp_handler_lock, flags); + } +} + +void qedr_ll2_rx_cb(void *_dev, struct qed_roce_ll2_packet *pkt, + struct qed_roce_ll2_rx_params *params) +{ + struct qedr_dev *dev = (struct qedr_dev *)_dev; + struct qedr_cq *cq = dev->gsi_rqcq; + struct qedr_qp *qp = dev->gsi_qp; + unsigned long flags; + + spin_lock_irqsave(&qp->q_lock, flags); + + qp->rqe_wr_id[qp->rq.gsi_cons].rc = params->rc; + qp->rqe_wr_id[qp->rq.gsi_cons].vlan_id = params->vlan_id; + qp->rqe_wr_id[qp->rq.gsi_cons].sg_list[0].length = pkt->payload[0].len; + ether_addr_copy(qp->rqe_wr_id[qp->rq.gsi_cons].smac, params->smac); + + qedr_inc_sw_gsi_cons(&qp->rq); + + spin_unlock_irqrestore(&qp->q_lock, flags); + + if (cq->ibcq.comp_handler) { + spin_lock_irqsave(&cq->comp_handler_lock, flags); + (*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context); + spin_unlock_irqrestore(&cq->comp_handler_lock, flags); + } +} + +static void qedr_destroy_gsi_cq(struct qedr_dev *dev, + struct ib_qp_init_attr *attrs) +{ + struct qed_rdma_destroy_cq_in_params iparams; + struct qed_rdma_destroy_cq_out_params oparams; + struct qedr_cq *cq; + + cq = get_qedr_cq(attrs->send_cq); + iparams.icid = cq->icid; + dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams); + dev->ops->common->chain_free(dev->cdev, &cq->pbl); + + cq = get_qedr_cq(attrs->recv_cq); + /* if a dedicated recv_cq was used, delete it too */ + if (iparams.icid != cq->icid) { + iparams.icid = cq->icid; + dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams); + dev->ops->common->chain_free(dev->cdev, &cq->pbl); + } +} + +static inline int qedr_check_gsi_qp_attrs(struct qedr_dev *dev, + struct ib_qp_init_attr *attrs) +{ + if (attrs->cap.max_recv_sge > QEDR_GSI_MAX_RECV_SGE) { + DP_ERR(dev, + " create gsi qp: failed. max_recv_sge is larger the max %d>%d\n", + attrs->cap.max_recv_sge, QEDR_GSI_MAX_RECV_SGE); + return -EINVAL; + } + + if (attrs->cap.max_recv_wr > QEDR_GSI_MAX_RECV_WR) { + DP_ERR(dev, + " create gsi qp: failed. max_recv_wr is too large %d>%d\n", + attrs->cap.max_recv_wr, QEDR_GSI_MAX_RECV_WR); + return -EINVAL; + } + + if (attrs->cap.max_send_wr > QEDR_GSI_MAX_SEND_WR) { + DP_ERR(dev, + " create gsi qp: failed. max_send_wr is too large %d>%d\n", + attrs->cap.max_send_wr, QEDR_GSI_MAX_SEND_WR); + return -EINVAL; + } + + return 0; +} + +struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev, + struct ib_qp_init_attr *attrs, + struct qedr_qp *qp) +{ + struct qed_roce_ll2_params ll2_params; + int rc; + + rc = qedr_check_gsi_qp_attrs(dev, attrs); + if (rc) + return ERR_PTR(rc); + + /* configure and start LL2 */ + memset(&ll2_params, 0, sizeof(ll2_params)); + ll2_params.max_tx_buffers = attrs->cap.max_send_wr; + ll2_params.max_rx_buffers = attrs->cap.max_recv_wr; + ll2_params.cbs.tx_cb = qedr_ll2_tx_cb; + ll2_params.cbs.rx_cb = qedr_ll2_rx_cb; + ll2_params.cb_cookie = (void *)dev; + ll2_params.mtu = dev->ndev->mtu; + ether_addr_copy(ll2_params.mac_address, dev->ndev->dev_addr); + rc = dev->ops->roce_ll2_start(dev->cdev, &ll2_params); + if (rc) { + DP_ERR(dev, "create gsi qp: failed on ll2 start. rc=%d\n", rc); + return ERR_PTR(rc); + } + + /* create QP */ + qp->ibqp.qp_num = 1; + qp->rq.max_wr = attrs->cap.max_recv_wr; + qp->sq.max_wr = attrs->cap.max_send_wr; + + qp->rqe_wr_id = kcalloc(qp->rq.max_wr, sizeof(*qp->rqe_wr_id), + GFP_KERNEL); + if (!qp->rqe_wr_id) + goto err; + qp->wqe_wr_id = kcalloc(qp->sq.max_wr, sizeof(*qp->wqe_wr_id), + GFP_KERNEL); + if (!qp->wqe_wr_id) + goto err; + + qedr_store_gsi_qp_cq(dev, qp, attrs); + ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr); + + /* the GSI CQ is handled by the driver so remove it from the FW */ + qedr_destroy_gsi_cq(dev, attrs); + dev->gsi_rqcq->cq_type = QEDR_CQ_TYPE_GSI; + dev->gsi_rqcq->cq_type = QEDR_CQ_TYPE_GSI; + + DP_DEBUG(dev, QEDR_MSG_GSI, "created GSI QP %p\n", qp); + + return &qp->ibqp; + +err: + kfree(qp->rqe_wr_id); + + rc = dev->ops->roce_ll2_stop(dev->cdev); + if (rc) + DP_ERR(dev, "create gsi qp: failed destroy on create\n"); + + return ERR_PTR(-ENOMEM); +} + +int qedr_destroy_gsi_qp(struct qedr_dev *dev) +{ + int rc; + + rc = dev->ops->roce_ll2_stop(dev->cdev); + if (rc) + DP_ERR(dev, "destroy gsi qp: failed (rc=%d)\n", rc); + else + DP_DEBUG(dev, QEDR_MSG_GSI, "destroy gsi qp: success\n"); + + return rc; +} + +#define QEDR_MAX_UD_HEADER_SIZE (100) +#define QEDR_GSI_QPN (1) +static inline int qedr_gsi_build_header(struct qedr_dev *dev, + struct qedr_qp *qp, + struct ib_send_wr *swr, + struct ib_ud_header *udh, + int *roce_mode) +{ + bool has_vlan = false, has_grh_ipv6 = true; + struct ib_ah_attr *ah_attr = &get_qedr_ah(ud_wr(swr)->ah)->attr; + struct ib_global_route *grh = &ah_attr->grh; + union ib_gid sgid; + int send_size = 0; + u16 vlan_id = 0; + u16 ether_type; + struct ib_gid_attr sgid_attr; + int rc; + int ip_ver = 0; + + bool has_udp = false; + int i; + + send_size = 0; + for (i = 0; i < swr->num_sge; ++i) + send_size += swr->sg_list[i].length; + + rc = ib_get_cached_gid(qp->ibqp.device, ah_attr->port_num, + grh->sgid_index, &sgid, &sgid_attr); + if (rc) { + DP_ERR(dev, + "gsi post send: failed to get cached GID (port=%d, ix=%d)\n", + ah_attr->port_num, grh->sgid_index); + return rc; + } + + vlan_id = rdma_vlan_dev_vlan_id(sgid_attr.ndev); + if (vlan_id < VLAN_CFI_MASK) + has_vlan = true; + if (sgid_attr.ndev) + dev_put(sgid_attr.ndev); + + if (!memcmp(&sgid, &zgid, sizeof(sgid))) { + DP_ERR(dev, "gsi post send: GID not found GID index %d\n", + ah_attr->grh.sgid_index); + return -ENOENT; + } + + has_udp = (sgid_attr.gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP); + if (!has_udp) { + /* RoCE v1 */ + ether_type = ETH_P_ROCE; + *roce_mode = ROCE_V1; + } else if (ipv6_addr_v4mapped((struct in6_addr *)&sgid)) { + /* RoCE v2 IPv4 */ + ip_ver = 4; + ether_type = ETH_P_IP; + has_grh_ipv6 = false; + *roce_mode = ROCE_V2_IPV4; + } else { + /* RoCE v2 IPv6 */ + ip_ver = 6; + ether_type = ETH_P_IPV6; + *roce_mode = ROCE_V2_IPV6; + } + + rc = ib_ud_header_init(send_size, false, true, has_vlan, + has_grh_ipv6, ip_ver, has_udp, 0, udh); + if (rc) { + DP_ERR(dev, "gsi post send: failed to init header\n"); + return rc; + } + + /* ENET + VLAN headers */ + ether_addr_copy(udh->eth.dmac_h, ah_attr->dmac); + ether_addr_copy(udh->eth.smac_h, dev->ndev->dev_addr); + if (has_vlan) { + udh->eth.type = htons(ETH_P_8021Q); + udh->vlan.tag = htons(vlan_id); + udh->vlan.type = htons(ether_type); + } else { + udh->eth.type = htons(ether_type); + } + + /* BTH */ + udh->bth.solicited_event = !!(swr->send_flags & IB_SEND_SOLICITED); + udh->bth.pkey = QEDR_ROCE_PKEY_DEFAULT; + udh->bth.destination_qpn = htonl(ud_wr(swr)->remote_qpn); + udh->bth.psn = htonl((qp->sq_psn++) & ((1 << 24) - 1)); + udh->bth.opcode = IB_OPCODE_UD_SEND_ONLY; + + /* DETH */ + udh->deth.qkey = htonl(0x80010000); + udh->deth.source_qpn = htonl(QEDR_GSI_QPN); + + if (has_grh_ipv6) { + /* GRH / IPv6 header */ + udh->grh.traffic_class = grh->traffic_class; + udh->grh.flow_label = grh->flow_label; + udh->grh.hop_limit = grh->hop_limit; + udh->grh.destination_gid = grh->dgid; + memcpy(&udh->grh.source_gid.raw, &sgid.raw, + sizeof(udh->grh.source_gid.raw)); + } else { + /* IPv4 header */ + u32 ipv4_addr; + + udh->ip4.protocol = IPPROTO_UDP; + udh->ip4.tos = htonl(ah_attr->grh.flow_label); + udh->ip4.frag_off = htons(IP_DF); + udh->ip4.ttl = ah_attr->grh.hop_limit; + + ipv4_addr = qedr_get_ipv4_from_gid(sgid.raw); + udh->ip4.saddr = ipv4_addr; + ipv4_addr = qedr_get_ipv4_from_gid(ah_attr->grh.dgid.raw); + udh->ip4.daddr = ipv4_addr; + /* note: checksum is calculated by the device */ + } + + /* UDP */ + if (has_udp) { + udh->udp.sport = htons(QEDR_ROCE_V2_UDP_SPORT); + udh->udp.dport = htons(ROCE_V2_UDP_DPORT); + udh->udp.csum = 0; + /* UDP length is untouched hence is zero */ + } + return 0; +} + +static inline int qedr_gsi_build_packet(struct qedr_dev *dev, + struct qedr_qp *qp, + struct ib_send_wr *swr, + struct qed_roce_ll2_packet **p_packet) +{ + u8 ud_header_buffer[QEDR_MAX_UD_HEADER_SIZE]; + struct qed_roce_ll2_packet *packet; + struct pci_dev *pdev = dev->pdev; + int roce_mode, header_size; + struct ib_ud_header udh; + int i, rc; + + *p_packet = NULL; + + rc = qedr_gsi_build_header(dev, qp, swr, &udh, &roce_mode); + if (rc) + return rc; + + header_size = ib_ud_header_pack(&udh, &ud_header_buffer); + + packet = kzalloc(sizeof(*packet), GFP_ATOMIC); + if (!packet) + return -ENOMEM; + + packet->header.vaddr = dma_alloc_coherent(&pdev->dev, header_size, + &packet->header.baddr, + GFP_ATOMIC); + if (!packet->header.vaddr) { + kfree(packet); + return -ENOMEM; + } + + if (ether_addr_equal(udh.eth.smac_h, udh.eth.dmac_h)) + packet->tx_dest = QED_ROCE_LL2_TX_DEST_NW; + else + packet->tx_dest = QED_ROCE_LL2_TX_DEST_LB; + + packet->roce_mode = roce_mode; + memcpy(packet->header.vaddr, ud_header_buffer, header_size); + packet->header.len = header_size; + packet->n_seg = swr->num_sge; + for (i = 0; i < packet->n_seg; i++) { + packet->payload[i].baddr = swr->sg_list[i].addr; + packet->payload[i].len = swr->sg_list[i].length; + } + + *p_packet = packet; + + return 0; +} + +int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + struct qed_roce_ll2_packet *pkt = NULL; + struct qedr_qp *qp = get_qedr_qp(ibqp); + struct qed_roce_ll2_tx_params params; + struct qedr_dev *dev = qp->dev; + unsigned long flags; + int rc; + + if (qp->state != QED_ROCE_QP_STATE_RTS) { + *bad_wr = wr; + DP_ERR(dev, + "gsi post recv: failed to post rx buffer. state is %d and not QED_ROCE_QP_STATE_RTS\n", + qp->state); + return -EINVAL; + } + + if (wr->num_sge > RDMA_MAX_SGE_PER_SQ_WQE) { + DP_ERR(dev, "gsi post send: num_sge is too large (%d>%d)\n", + wr->num_sge, RDMA_MAX_SGE_PER_SQ_WQE); + rc = -EINVAL; + goto err; + } + + if (wr->opcode != IB_WR_SEND) { + DP_ERR(dev, + "gsi post send: failed due to unsupported opcode %d\n", + wr->opcode); + rc = -EINVAL; + goto err; + } + + memset(¶ms, 0, sizeof(params)); + + spin_lock_irqsave(&qp->q_lock, flags); + + rc = qedr_gsi_build_packet(dev, qp, wr, &pkt); + if (rc) { + spin_unlock_irqrestore(&qp->q_lock, flags); + goto err; + } + + rc = dev->ops->roce_ll2_tx(dev->cdev, pkt, ¶ms); + if (!rc) { + qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id; + qedr_inc_sw_prod(&qp->sq); + DP_DEBUG(qp->dev, QEDR_MSG_GSI, + "gsi post send: opcode=%d, in_irq=%ld, irqs_disabled=%d, wr_id=%llx\n", + wr->opcode, in_irq(), irqs_disabled(), wr->wr_id); + } else { + if (rc == QED_ROCE_TX_HEAD_FAILURE) { + /* TX failed while posting header - release resources */ + dma_free_coherent(&dev->pdev->dev, pkt->header.len, + pkt->header.vaddr, pkt->header.baddr); + kfree(pkt); + } else if (rc == QED_ROCE_TX_FRAG_FAILURE) { + /* NTD since TX failed while posting a fragment. We will + * release the resources on TX callback + */ + } + + DP_ERR(dev, "gsi post send: failed to transmit (rc=%d)\n", rc); + rc = -EAGAIN; + *bad_wr = wr; + } + + spin_unlock_irqrestore(&qp->q_lock, flags); + + if (wr->next) { + DP_ERR(dev, + "gsi post send: failed second WR. Only one WR may be passed at a time\n"); + *bad_wr = wr->next; + rc = -EINVAL; + } + + return rc; + +err: + *bad_wr = wr; + return rc; +} + +int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, + struct ib_recv_wr **bad_wr) +{ + struct qedr_dev *dev = get_qedr_dev(ibqp->device); + struct qedr_qp *qp = get_qedr_qp(ibqp); + struct qed_roce_ll2_buffer buf; + unsigned long flags; + int status = 0; + int rc; + + if ((qp->state != QED_ROCE_QP_STATE_RTR) && + (qp->state != QED_ROCE_QP_STATE_RTS)) { + *bad_wr = wr; + DP_ERR(dev, + "gsi post recv: failed to post rx buffer. state is %d and not QED_ROCE_QP_STATE_RTR/S\n", + qp->state); + return -EINVAL; + } + + memset(&buf, 0, sizeof(buf)); + + spin_lock_irqsave(&qp->q_lock, flags); + + while (wr) { + if (wr->num_sge > QEDR_GSI_MAX_RECV_SGE) { + DP_ERR(dev, + "gsi post recv: failed to post rx buffer. too many sges %d>%d\n", + wr->num_sge, QEDR_GSI_MAX_RECV_SGE); + goto err; + } + + buf.baddr = wr->sg_list[0].addr; + buf.len = wr->sg_list[0].length; + + rc = dev->ops->roce_ll2_post_rx_buffer(dev->cdev, &buf, 0, 1); + if (rc) { + DP_ERR(dev, + "gsi post recv: failed to post rx buffer (rc=%d)\n", + rc); + goto err; + } + + memset(&qp->rqe_wr_id[qp->rq.prod], 0, + sizeof(qp->rqe_wr_id[qp->rq.prod])); + qp->rqe_wr_id[qp->rq.prod].sg_list[0] = wr->sg_list[0]; + qp->rqe_wr_id[qp->rq.prod].wr_id = wr->wr_id; + + qedr_inc_sw_prod(&qp->rq); + + wr = wr->next; + } + + spin_unlock_irqrestore(&qp->q_lock, flags); + + return status; +err: + spin_unlock_irqrestore(&qp->q_lock, flags); + *bad_wr = wr; + return -ENOMEM; +} + +int qedr_gsi_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc) +{ + struct qedr_dev *dev = get_qedr_dev(ibcq->device); + struct qedr_cq *cq = get_qedr_cq(ibcq); + struct qedr_qp *qp = dev->gsi_qp; + unsigned long flags; + int i = 0; + + spin_lock_irqsave(&cq->cq_lock, flags); + + while (i < num_entries && qp->rq.cons != qp->rq.gsi_cons) { + memset(&wc[i], 0, sizeof(*wc)); + + wc[i].qp = &qp->ibqp; + wc[i].wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id; + wc[i].opcode = IB_WC_RECV; + wc[i].pkey_index = 0; + wc[i].status = (qp->rqe_wr_id[qp->rq.cons].rc) ? + IB_WC_GENERAL_ERR : IB_WC_SUCCESS; + /* 0 - currently only one recv sg is supported */ + wc[i].byte_len = qp->rqe_wr_id[qp->rq.cons].sg_list[0].length; + wc[i].wc_flags |= IB_WC_GRH | IB_WC_IP_CSUM_OK; + ether_addr_copy(wc[i].smac, qp->rqe_wr_id[qp->rq.cons].smac); + wc[i].wc_flags |= IB_WC_WITH_SMAC; + if (qp->rqe_wr_id[qp->rq.cons].vlan_id) { + wc[i].wc_flags |= IB_WC_WITH_VLAN; + wc[i].vlan_id = qp->rqe_wr_id[qp->rq.cons].vlan_id; + } + + qedr_inc_sw_cons(&qp->rq); + i++; + } + + while (i < num_entries && qp->sq.cons != qp->sq.gsi_cons) { + memset(&wc[i], 0, sizeof(*wc)); + + wc[i].qp = &qp->ibqp; + wc[i].wr_id = qp->wqe_wr_id[qp->sq.cons].wr_id; + wc[i].opcode = IB_WC_SEND; + wc[i].status = IB_WC_SUCCESS; + + qedr_inc_sw_cons(&qp->sq); + i++; + } + + spin_unlock_irqrestore(&cq->cq_lock, flags); + + DP_DEBUG(dev, QEDR_MSG_GSI, + "gsi poll_cq: requested entries=%d, actual=%d, qp->rq.cons=%d, qp->rq.gsi_cons=%x, qp->sq.cons=%d, qp->sq.gsi_cons=%d, qp_num=%d\n", + num_entries, i, qp->rq.cons, qp->rq.gsi_cons, qp->sq.cons, + qp->sq.gsi_cons, qp->ibqp.qp_num); + + return i; +} diff --git a/drivers/infiniband/hw/qedr/qedr_cm.h b/drivers/infiniband/hw/qedr/qedr_cm.h new file mode 100644 index 0000000..9ba6e15 --- /dev/null +++ b/drivers/infiniband/hw/qedr/qedr_cm.h @@ -0,0 +1,61 @@ +/* QLogic qedr NIC Driver + * Copyright (c) 2015-2016 QLogic Corporation + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and /or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef LINUX_QEDR_CM_H_ +#define LINUX_QEDR_CM_H_ + +#define QEDR_GSI_MAX_RECV_WR (4096) +#define QEDR_GSI_MAX_SEND_WR (4096) + +#define QEDR_GSI_MAX_RECV_SGE (1) /* LL2 FW limitation */ + +#define ETH_P_ROCE (0x8915) +#define QEDR_ROCE_V2_UDP_SPORT (0000) + +static inline u32 qedr_get_ipv4_from_gid(u8 *gid) +{ + return *(u32 *)(void *)&gid[12]; +} + +/* RDMA CM */ +int qedr_gsi_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc); +int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, + struct ib_recv_wr **bad_wr); +int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, + struct ib_send_wr **bad_wr); +struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev, + struct ib_qp_init_attr *attrs, + struct qedr_qp *qp); +void qedr_store_gsi_qp_cq(struct qedr_dev *dev, + struct qedr_qp *qp, struct ib_qp_init_attr *attrs); +int qedr_destroy_gsi_qp(struct qedr_dev *dev); +void qedr_inc_sw_gsi_cons(struct qedr_qp_hwq_info *info); +#endif diff --git a/drivers/infiniband/hw/qedr/qedr_hsi.h b/drivers/infiniband/hw/qedr/qedr_hsi.h new file mode 100644 index 0000000..66d2752 --- /dev/null +++ b/drivers/infiniband/hw/qedr/qedr_hsi.h @@ -0,0 +1,56 @@ +/* QLogic qedr NIC Driver + * Copyright (c) 2015-2016 QLogic Corporation + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and /or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef __QED_HSI_ROCE__ +#define __QED_HSI_ROCE__ + +#include <linux/qed/common_hsi.h> +#include <linux/qed/roce_common.h> +#include "qedr_hsi_rdma.h" + +/* Affiliated asynchronous events / errors enumeration */ +enum roce_async_events_type { + ROCE_ASYNC_EVENT_NONE = 0, + ROCE_ASYNC_EVENT_COMM_EST = 1, + ROCE_ASYNC_EVENT_SQ_DRAINED, + ROCE_ASYNC_EVENT_SRQ_LIMIT, + ROCE_ASYNC_EVENT_LAST_WQE_REACHED, + ROCE_ASYNC_EVENT_CQ_ERR, + ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR, + ROCE_ASYNC_EVENT_LOCAL_CATASTROPHIC_ERR, + ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR, + ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR, + ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR, + ROCE_ASYNC_EVENT_SRQ_EMPTY, + MAX_ROCE_ASYNC_EVENTS_TYPE +}; + +#endif /* __QED_HSI_ROCE__ */ diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h new file mode 100644 index 0000000..5c98d20 --- /dev/null +++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h @@ -0,0 +1,748 @@ +/* QLogic qedr NIC Driver + * Copyright (c) 2015-2016 QLogic Corporation + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and /or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef __QED_HSI_RDMA__ +#define __QED_HSI_RDMA__ + +#include <linux/qed/rdma_common.h> + +/* rdma completion notification queue element */ +struct rdma_cnqe { + struct regpair cq_handle; +}; + +struct rdma_cqe_responder { + struct regpair srq_wr_id; + struct regpair qp_handle; + __le32 imm_data_or_inv_r_Key; + __le32 length; + __le32 imm_data_hi; + __le16 rq_cons; + u8 flags; +#define RDMA_CQE_RESPONDER_TOGGLE_BIT_MASK 0x1 +#define RDMA_CQE_RESPONDER_TOGGLE_BIT_SHIFT 0 +#define RDMA_CQE_RESPONDER_TYPE_MASK 0x3 +#define RDMA_CQE_RESPONDER_TYPE_SHIFT 1 +#define RDMA_CQE_RESPONDER_INV_FLG_MASK 0x1 +#define RDMA_CQE_RESPONDER_INV_FLG_SHIFT 3 +#define RDMA_CQE_RESPONDER_IMM_FLG_MASK 0x1 +#define RDMA_CQE_RESPONDER_IMM_FLG_SHIFT 4 +#define RDMA_CQE_RESPONDER_RDMA_FLG_MASK 0x1 +#define RDMA_CQE_RESPONDER_RDMA_FLG_SHIFT 5 +#define RDMA_CQE_RESPONDER_RESERVED2_MASK 0x3 +#define RDMA_CQE_RESPONDER_RESERVED2_SHIFT 6 + u8 status; +}; + +struct rdma_cqe_requester { + __le16 sq_cons; + __le16 reserved0; + __le32 reserved1; + struct regpair qp_handle; + struct regpair reserved2; + __le32 reserved3; + __le16 reserved4; + u8 flags; +#define RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK 0x1 +#define RDMA_CQE_REQUESTER_TOGGLE_BIT_SHIFT 0 +#define RDMA_CQE_REQUESTER_TYPE_MASK 0x3 +#define RDMA_CQE_REQUESTER_TYPE_SHIFT 1 +#define RDMA_CQE_REQUESTER_RESERVED5_MASK 0x1F +#define RDMA_CQE_REQUESTER_RESERVED5_SHIFT 3 + u8 status; +}; + +struct rdma_cqe_common { + struct regpair reserved0; + struct regpair qp_handle; + __le16 reserved1[7]; + u8 flags; +#define RDMA_CQE_COMMON_TOGGLE_BIT_MASK 0x1 +#define RDMA_CQE_COMMON_TOGGLE_BIT_SHIFT 0 +#define RDMA_CQE_COMMON_TYPE_MASK 0x3 +#define RDMA_CQE_COMMON_TYPE_SHIFT 1 +#define RDMA_CQE_COMMON_RESERVED2_MASK 0x1F +#define RDMA_CQE_COMMON_RESERVED2_SHIFT 3 + u8 status; +}; + +/* rdma completion queue element */ +union rdma_cqe { + struct rdma_cqe_responder resp; + struct rdma_cqe_requester req; + struct rdma_cqe_common cmn; +}; + +/* * CQE requester status enumeration */ +enum rdma_cqe_requester_status_enum { + RDMA_CQE_REQ_STS_OK, + RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR, + RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR, + RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR, + RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR, + RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR, + RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR, + RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR, + RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR, + RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR, + RDMA_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR, + RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR, + MAX_RDMA_CQE_REQUESTER_STATUS_ENUM +}; + +/* CQE responder status enumeration */ +enum rdma_cqe_responder_status_enum { + RDMA_CQE_RESP_STS_OK, + RDMA_CQE_RESP_STS_LOCAL_ACCESS_ERR, + RDMA_CQE_RESP_STS_LOCAL_LENGTH_ERR, + RDMA_CQE_RESP_STS_LOCAL_QP_OPERATION_ERR, + RDMA_CQE_RESP_STS_LOCAL_PROTECTION_ERR, + RDMA_CQE_RESP_STS_MEMORY_MGT_OPERATION_ERR, + RDMA_CQE_RESP_STS_REMOTE_INVALID_REQUEST_ERR, + RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR, + MAX_RDMA_CQE_RESPONDER_STATUS_ENUM +}; + +/* CQE type enumeration */ +enum rdma_cqe_type { + RDMA_CQE_TYPE_REQUESTER, + RDMA_CQE_TYPE_RESPONDER_RQ, + RDMA_CQE_TYPE_RESPONDER_SRQ, + RDMA_CQE_TYPE_INVALID, + MAX_RDMA_CQE_TYPE +}; + +struct rdma_sq_sge { + __le32 length; + struct regpair addr; + __le32 l_key; +}; + +struct rdma_rq_sge { + struct regpair addr; + __le32 length; + __le32 flags; +#define RDMA_RQ_SGE_L_KEY_MASK 0x3FFFFFF +#define RDMA_RQ_SGE_L_KEY_SHIFT 0 +#define RDMA_RQ_SGE_NUM_SGES_MASK 0x7 +#define RDMA_RQ_SGE_NUM_SGES_SHIFT 26 +#define RDMA_RQ_SGE_RESERVED0_MASK 0x7 +#define RDMA_RQ_SGE_RESERVED0_SHIFT 29 +}; + +struct rdma_srq_sge { + struct regpair addr; + __le32 length; + __le32 l_key; +}; + +/* Rdma doorbell data for SQ and RQ */ +struct rdma_pwm_val16_data { + __le16 icid; + __le16 value; +}; + +union rdma_pwm_val16_data_union { + struct rdma_pwm_val16_data as_struct; + __le32 as_dword; +}; + +/* Rdma doorbell data for CQ */ +struct rdma_pwm_val32_data { + __le16 icid; + u8 agg_flags; + u8 params; +#define RDMA_PWM_VAL32_DATA_AGG_CMD_MASK 0x3 +#define RDMA_PWM_VAL32_DATA_AGG_CMD_SHIFT 0 +#define RDMA_PWM_VAL32_DATA_BYPASS_EN_MASK 0x1 +#define RDMA_PWM_VAL32_DATA_BYPASS_EN_SHIFT 2 +#define RDMA_PWM_VAL32_DATA_RESERVED_MASK 0x1F +#define RDMA_PWM_VAL32_DATA_RESERVED_SHIFT 3 + __le32 value; +}; + +/* DIF Block size options */ +enum rdma_dif_block_size { + RDMA_DIF_BLOCK_512 = 0, + RDMA_DIF_BLOCK_4096 = 1, + MAX_RDMA_DIF_BLOCK_SIZE +}; + +/* DIF CRC initial value */ +enum rdma_dif_crc_seed { + RDMA_DIF_CRC_SEED_0000 = 0, + RDMA_DIF_CRC_SEED_FFFF = 1, + MAX_RDMA_DIF_CRC_SEED +}; + +/* RDMA DIF Error Result Structure */ +struct rdma_dif_error_result { + __le32 error_intervals; + __le32 dif_error_1st_interval; + u8 flags; +#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_CRC_MASK 0x1 +#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_CRC_SHIFT 0 +#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_APP_TAG_MASK 0x1 +#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_APP_TAG_SHIFT 1 +#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_REF_TAG_MASK 0x1 +#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_REF_TAG_SHIFT 2 +#define RDMA_DIF_ERROR_RESULT_RESERVED0_MASK 0xF +#define RDMA_DIF_ERROR_RESULT_RESERVED0_SHIFT 3 +#define RDMA_DIF_ERROR_RESULT_TOGGLE_BIT_MASK 0x1 +#define RDMA_DIF_ERROR_RESULT_TOGGLE_BIT_SHIFT 7 + u8 reserved1[55]; +}; + +/* DIF IO direction */ +enum rdma_dif_io_direction_flg { + RDMA_DIF_DIR_RX = 0, + RDMA_DIF_DIR_TX = 1, + MAX_RDMA_DIF_IO_DIRECTION_FLG +}; + +/* RDMA DIF Runt Result Structure */ +struct rdma_dif_runt_result { + __le16 guard_tag; + __le16 reserved[3]; +}; + +/* Memory window type enumeration */ +enum rdma_mw_type { + RDMA_MW_TYPE_1, + RDMA_MW_TYPE_2A, + MAX_RDMA_MW_TYPE +}; + +struct rdma_sq_atomic_wqe { + __le32 reserved1; + __le32 length; + __le32 xrc_srq; + u8 req_type; + u8 flags; +#define RDMA_SQ_ATOMIC_WQE_COMP_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_COMP_FLG_SHIFT 0 +#define RDMA_SQ_ATOMIC_WQE_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_ATOMIC_WQE_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_ATOMIC_WQE_SE_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_SE_FLG_SHIFT 3 +#define RDMA_SQ_ATOMIC_WQE_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_ATOMIC_WQE_DIF_ON_HOST_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_DIF_ON_HOST_FLG_SHIFT 5 +#define RDMA_SQ_ATOMIC_WQE_RESERVED0_MASK 0x3 +#define RDMA_SQ_ATOMIC_WQE_RESERVED0_SHIFT 6 + u8 wqe_size; + u8 prev_wqe_size; + struct regpair remote_va; + __le32 r_key; + __le32 reserved2; + struct regpair cmp_data; + struct regpair swap_data; +}; + +/* First element (16 bytes) of atomic wqe */ +struct rdma_sq_atomic_wqe_1st { + __le32 reserved1; + __le32 length; + __le32 xrc_srq; + u8 req_type; + u8 flags; +#define RDMA_SQ_ATOMIC_WQE_1ST_COMP_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_1ST_COMP_FLG_SHIFT 0 +#define RDMA_SQ_ATOMIC_WQE_1ST_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_1ST_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_ATOMIC_WQE_1ST_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_1ST_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_ATOMIC_WQE_1ST_SE_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_1ST_SE_FLG_SHIFT 3 +#define RDMA_SQ_ATOMIC_WQE_1ST_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_ATOMIC_WQE_1ST_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_ATOMIC_WQE_1ST_RESERVED0_MASK 0x7 +#define RDMA_SQ_ATOMIC_WQE_1ST_RESERVED0_SHIFT 5 + u8 wqe_size; + u8 prev_wqe_size; +}; + +/* Second element (16 bytes) of atomic wqe */ +struct rdma_sq_atomic_wqe_2nd { + struct regpair remote_va; + __le32 r_key; + __le32 reserved2; +}; + +/* Third element (16 bytes) of atomic wqe */ +struct rdma_sq_atomic_wqe_3rd { + struct regpair cmp_data; + struct regpair swap_data; +}; + +struct rdma_sq_bind_wqe { + struct regpair addr; + __le32 l_key; + u8 req_type; + u8 flags; +#define RDMA_SQ_BIND_WQE_COMP_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_COMP_FLG_SHIFT 0 +#define RDMA_SQ_BIND_WQE_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_BIND_WQE_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_BIND_WQE_SE_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_SE_FLG_SHIFT 3 +#define RDMA_SQ_BIND_WQE_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_BIND_WQE_RESERVED0_MASK 0x7 +#define RDMA_SQ_BIND_WQE_RESERVED0_SHIFT 5 + u8 wqe_size; + u8 prev_wqe_size; + u8 bind_ctrl; +#define RDMA_SQ_BIND_WQE_ZERO_BASED_MASK 0x1 +#define RDMA_SQ_BIND_WQE_ZERO_BASED_SHIFT 0 +#define RDMA_SQ_BIND_WQE_MW_TYPE_MASK 0x1 +#define RDMA_SQ_BIND_WQE_MW_TYPE_SHIFT 1 +#define RDMA_SQ_BIND_WQE_RESERVED1_MASK 0x3F +#define RDMA_SQ_BIND_WQE_RESERVED1_SHIFT 2 + u8 access_ctrl; +#define RDMA_SQ_BIND_WQE_REMOTE_READ_MASK 0x1 +#define RDMA_SQ_BIND_WQE_REMOTE_READ_SHIFT 0 +#define RDMA_SQ_BIND_WQE_REMOTE_WRITE_MASK 0x1 +#define RDMA_SQ_BIND_WQE_REMOTE_WRITE_SHIFT 1 +#define RDMA_SQ_BIND_WQE_ENABLE_ATOMIC_MASK 0x1 +#define RDMA_SQ_BIND_WQE_ENABLE_ATOMIC_SHIFT 2 +#define RDMA_SQ_BIND_WQE_LOCAL_READ_MASK 0x1 +#define RDMA_SQ_BIND_WQE_LOCAL_READ_SHIFT 3 +#define RDMA_SQ_BIND_WQE_LOCAL_WRITE_MASK 0x1 +#define RDMA_SQ_BIND_WQE_LOCAL_WRITE_SHIFT 4 +#define RDMA_SQ_BIND_WQE_RESERVED2_MASK 0x7 +#define RDMA_SQ_BIND_WQE_RESERVED2_SHIFT 5 + u8 reserved3; + u8 length_hi; + __le32 length_lo; + __le32 parent_l_key; + __le32 reserved4; +}; + +/* First element (16 bytes) of bind wqe */ +struct rdma_sq_bind_wqe_1st { + struct regpair addr; + __le32 l_key; + u8 req_type; + u8 flags; +#define RDMA_SQ_BIND_WQE_1ST_COMP_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_1ST_COMP_FLG_SHIFT 0 +#define RDMA_SQ_BIND_WQE_1ST_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_1ST_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_BIND_WQE_1ST_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_1ST_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_BIND_WQE_1ST_SE_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_1ST_SE_FLG_SHIFT 3 +#define RDMA_SQ_BIND_WQE_1ST_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_BIND_WQE_1ST_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_BIND_WQE_1ST_RESERVED0_MASK 0x7 +#define RDMA_SQ_BIND_WQE_1ST_RESERVED0_SHIFT 5 + u8 wqe_size; + u8 prev_wqe_size; +}; + +/* Second element (16 bytes) of bind wqe */ +struct rdma_sq_bind_wqe_2nd { + u8 bind_ctrl; +#define RDMA_SQ_BIND_WQE_2ND_ZERO_BASED_MASK 0x1 +#define RDMA_SQ_BIND_WQE_2ND_ZERO_BASED_SHIFT 0 +#define RDMA_SQ_BIND_WQE_2ND_MW_TYPE_MASK 0x1 +#define RDMA_SQ_BIND_WQE_2ND_MW_TYPE_SHIFT 1 +#define RDMA_SQ_BIND_WQE_2ND_RESERVED1_MASK 0x3F +#define RDMA_SQ_BIND_WQE_2ND_RESERVED1_SHIFT 2 + u8 access_ctrl; +#define RDMA_SQ_BIND_WQE_2ND_REMOTE_READ_MASK 0x1 +#define RDMA_SQ_BIND_WQE_2ND_REMOTE_READ_SHIFT 0 +#define RDMA_SQ_BIND_WQE_2ND_REMOTE_WRITE_MASK 0x1 +#define RDMA_SQ_BIND_WQE_2ND_REMOTE_WRITE_SHIFT 1 +#define RDMA_SQ_BIND_WQE_2ND_ENABLE_ATOMIC_MASK 0x1 +#define RDMA_SQ_BIND_WQE_2ND_ENABLE_ATOMIC_SHIFT 2 +#define RDMA_SQ_BIND_WQE_2ND_LOCAL_READ_MASK 0x1 +#define RDMA_SQ_BIND_WQE_2ND_LOCAL_READ_SHIFT 3 +#define RDMA_SQ_BIND_WQE_2ND_LOCAL_WRITE_MASK 0x1 +#define RDMA_SQ_BIND_WQE_2ND_LOCAL_WRITE_SHIFT 4 +#define RDMA_SQ_BIND_WQE_2ND_RESERVED2_MASK 0x7 +#define RDMA_SQ_BIND_WQE_2ND_RESERVED2_SHIFT 5 + u8 reserved3; + u8 length_hi; + __le32 length_lo; + __le32 parent_l_key; + __le32 reserved4; +}; + +/* Structure with only the SQ WQE common + * fields. Size is of one SQ element (16B) + */ +struct rdma_sq_common_wqe { + __le32 reserved1[3]; + u8 req_type; + u8 flags; +#define RDMA_SQ_COMMON_WQE_COMP_FLG_MASK 0x1 +#define RDMA_SQ_COMMON_WQE_COMP_FLG_SHIFT 0 +#define RDMA_SQ_COMMON_WQE_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_COMMON_WQE_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_COMMON_WQE_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_COMMON_WQE_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_COMMON_WQE_SE_FLG_MASK 0x1 +#define RDMA_SQ_COMMON_WQE_SE_FLG_SHIFT 3 +#define RDMA_SQ_COMMON_WQE_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_COMMON_WQE_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_COMMON_WQE_RESERVED0_MASK 0x7 +#define RDMA_SQ_COMMON_WQE_RESERVED0_SHIFT 5 + u8 wqe_size; + u8 prev_wqe_size; +}; + +struct rdma_sq_fmr_wqe { + struct regpair addr; + __le32 l_key; + u8 req_type; + u8 flags; +#define RDMA_SQ_FMR_WQE_COMP_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_COMP_FLG_SHIFT 0 +#define RDMA_SQ_FMR_WQE_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_FMR_WQE_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_FMR_WQE_SE_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_SE_FLG_SHIFT 3 +#define RDMA_SQ_FMR_WQE_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_FMR_WQE_DIF_ON_HOST_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_DIF_ON_HOST_FLG_SHIFT 5 +#define RDMA_SQ_FMR_WQE_RESERVED0_MASK 0x3 +#define RDMA_SQ_FMR_WQE_RESERVED0_SHIFT 6 + u8 wqe_size; + u8 prev_wqe_size; + u8 fmr_ctrl; +#define RDMA_SQ_FMR_WQE_PAGE_SIZE_LOG_MASK 0x1F +#define RDMA_SQ_FMR_WQE_PAGE_SIZE_LOG_SHIFT 0 +#define RDMA_SQ_FMR_WQE_ZERO_BASED_MASK 0x1 +#define RDMA_SQ_FMR_WQE_ZERO_BASED_SHIFT 5 +#define RDMA_SQ_FMR_WQE_BIND_EN_MASK 0x1 +#define RDMA_SQ_FMR_WQE_BIND_EN_SHIFT 6 +#define RDMA_SQ_FMR_WQE_RESERVED1_MASK 0x1 +#define RDMA_SQ_FMR_WQE_RESERVED1_SHIFT 7 + u8 access_ctrl; +#define RDMA_SQ_FMR_WQE_REMOTE_READ_MASK 0x1 +#define RDMA_SQ_FMR_WQE_REMOTE_READ_SHIFT 0 +#define RDMA_SQ_FMR_WQE_REMOTE_WRITE_MASK 0x1 +#define RDMA_SQ_FMR_WQE_REMOTE_WRITE_SHIFT 1 +#define RDMA_SQ_FMR_WQE_ENABLE_ATOMIC_MASK 0x1 +#define RDMA_SQ_FMR_WQE_ENABLE_ATOMIC_SHIFT 2 +#define RDMA_SQ_FMR_WQE_LOCAL_READ_MASK 0x1 +#define RDMA_SQ_FMR_WQE_LOCAL_READ_SHIFT 3 +#define RDMA_SQ_FMR_WQE_LOCAL_WRITE_MASK 0x1 +#define RDMA_SQ_FMR_WQE_LOCAL_WRITE_SHIFT 4 +#define RDMA_SQ_FMR_WQE_RESERVED2_MASK 0x7 +#define RDMA_SQ_FMR_WQE_RESERVED2_SHIFT 5 + u8 reserved3; + u8 length_hi; + __le32 length_lo; + struct regpair pbl_addr; + __le32 dif_base_ref_tag; + __le16 dif_app_tag; + __le16 dif_app_tag_mask; + __le16 dif_runt_crc_value; + __le16 dif_flags; +#define RDMA_SQ_FMR_WQE_DIF_IO_DIRECTION_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_DIF_IO_DIRECTION_FLG_SHIFT 0 +#define RDMA_SQ_FMR_WQE_DIF_BLOCK_SIZE_MASK 0x1 +#define RDMA_SQ_FMR_WQE_DIF_BLOCK_SIZE_SHIFT 1 +#define RDMA_SQ_FMR_WQE_DIF_RUNT_VALID_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_DIF_RUNT_VALID_FLG_SHIFT 2 +#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_CRC_GUARD_MASK 0x1 +#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_CRC_GUARD_SHIFT 3 +#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_REF_TAG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_REF_TAG_SHIFT 4 +#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_APP_TAG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_APP_TAG_SHIFT 5 +#define RDMA_SQ_FMR_WQE_DIF_CRC_SEED_MASK 0x1 +#define RDMA_SQ_FMR_WQE_DIF_CRC_SEED_SHIFT 6 +#define RDMA_SQ_FMR_WQE_RESERVED4_MASK 0x1FF +#define RDMA_SQ_FMR_WQE_RESERVED4_SHIFT 7 + __le32 Reserved5; +}; + +/* First element (16 bytes) of fmr wqe */ +struct rdma_sq_fmr_wqe_1st { + struct regpair addr; + __le32 l_key; + u8 req_type; + u8 flags; +#define RDMA_SQ_FMR_WQE_1ST_COMP_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_1ST_COMP_FLG_SHIFT 0 +#define RDMA_SQ_FMR_WQE_1ST_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_1ST_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_FMR_WQE_1ST_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_1ST_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_FMR_WQE_1ST_SE_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_1ST_SE_FLG_SHIFT 3 +#define RDMA_SQ_FMR_WQE_1ST_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_1ST_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_FMR_WQE_1ST_DIF_ON_HOST_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_1ST_DIF_ON_HOST_FLG_SHIFT 5 +#define RDMA_SQ_FMR_WQE_1ST_RESERVED0_MASK 0x3 +#define RDMA_SQ_FMR_WQE_1ST_RESERVED0_SHIFT 6 + u8 wqe_size; + u8 prev_wqe_size; +}; + +/* Second element (16 bytes) of fmr wqe */ +struct rdma_sq_fmr_wqe_2nd { + u8 fmr_ctrl; +#define RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG_MASK 0x1F +#define RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG_SHIFT 0 +#define RDMA_SQ_FMR_WQE_2ND_ZERO_BASED_MASK 0x1 +#define RDMA_SQ_FMR_WQE_2ND_ZERO_BASED_SHIFT 5 +#define RDMA_SQ_FMR_WQE_2ND_BIND_EN_MASK 0x1 +#define RDMA_SQ_FMR_WQE_2ND_BIND_EN_SHIFT 6 +#define RDMA_SQ_FMR_WQE_2ND_RESERVED1_MASK 0x1 +#define RDMA_SQ_FMR_WQE_2ND_RESERVED1_SHIFT 7 + u8 access_ctrl; +#define RDMA_SQ_FMR_WQE_2ND_REMOTE_READ_MASK 0x1 +#define RDMA_SQ_FMR_WQE_2ND_REMOTE_READ_SHIFT 0 +#define RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE_MASK 0x1 +#define RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE_SHIFT 1 +#define RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC_MASK 0x1 +#define RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC_SHIFT 2 +#define RDMA_SQ_FMR_WQE_2ND_LOCAL_READ_MASK 0x1 +#define RDMA_SQ_FMR_WQE_2ND_LOCAL_READ_SHIFT 3 +#define RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE_MASK 0x1 +#define RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE_SHIFT 4 +#define RDMA_SQ_FMR_WQE_2ND_RESERVED2_MASK 0x7 +#define RDMA_SQ_FMR_WQE_2ND_RESERVED2_SHIFT 5 + u8 reserved3; + u8 length_hi; + __le32 length_lo; + struct regpair pbl_addr; +}; + +/* Third element (16 bytes) of fmr wqe */ +struct rdma_sq_fmr_wqe_3rd { + __le32 dif_base_ref_tag; + __le16 dif_app_tag; + __le16 dif_app_tag_mask; + __le16 dif_runt_crc_value; + __le16 dif_flags; +#define RDMA_SQ_FMR_WQE_3RD_DIF_IO_DIRECTION_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_3RD_DIF_IO_DIRECTION_FLG_SHIFT 0 +#define RDMA_SQ_FMR_WQE_3RD_DIF_BLOCK_SIZE_MASK 0x1 +#define RDMA_SQ_FMR_WQE_3RD_DIF_BLOCK_SIZE_SHIFT 1 +#define RDMA_SQ_FMR_WQE_3RD_DIF_RUNT_VALID_FLG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_3RD_DIF_RUNT_VALID_FLG_SHIFT 2 +#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_CRC_GUARD_MASK 0x1 +#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_CRC_GUARD_SHIFT 3 +#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_REF_TAG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_REF_TAG_SHIFT 4 +#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_APP_TAG_MASK 0x1 +#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_APP_TAG_SHIFT 5 +#define RDMA_SQ_FMR_WQE_3RD_DIF_CRC_SEED_MASK 0x1 +#define RDMA_SQ_FMR_WQE_3RD_DIF_CRC_SEED_SHIFT 6 +#define RDMA_SQ_FMR_WQE_3RD_RESERVED4_MASK 0x1FF +#define RDMA_SQ_FMR_WQE_3RD_RESERVED4_SHIFT 7 + __le32 Reserved5; +}; + +struct rdma_sq_local_inv_wqe { + struct regpair reserved; + __le32 inv_l_key; + u8 req_type; + u8 flags; +#define RDMA_SQ_LOCAL_INV_WQE_COMP_FLG_MASK 0x1 +#define RDMA_SQ_LOCAL_INV_WQE_COMP_FLG_SHIFT 0 +#define RDMA_SQ_LOCAL_INV_WQE_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_LOCAL_INV_WQE_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_LOCAL_INV_WQE_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_LOCAL_INV_WQE_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_LOCAL_INV_WQE_SE_FLG_MASK 0x1 +#define RDMA_SQ_LOCAL_INV_WQE_SE_FLG_SHIFT 3 +#define RDMA_SQ_LOCAL_INV_WQE_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_LOCAL_INV_WQE_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_LOCAL_INV_WQE_DIF_ON_HOST_FLG_MASK 0x1 +#define RDMA_SQ_LOCAL_INV_WQE_DIF_ON_HOST_FLG_SHIFT 5 +#define RDMA_SQ_LOCAL_INV_WQE_RESERVED0_MASK 0x3 +#define RDMA_SQ_LOCAL_INV_WQE_RESERVED0_SHIFT 6 + u8 wqe_size; + u8 prev_wqe_size; +}; + +struct rdma_sq_rdma_wqe { + __le32 imm_data; + __le32 length; + __le32 xrc_srq; + u8 req_type; + u8 flags; +#define RDMA_SQ_RDMA_WQE_COMP_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_COMP_FLG_SHIFT 0 +#define RDMA_SQ_RDMA_WQE_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_RDMA_WQE_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_RDMA_WQE_SE_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_SE_FLG_SHIFT 3 +#define RDMA_SQ_RDMA_WQE_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_RDMA_WQE_DIF_ON_HOST_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_DIF_ON_HOST_FLG_SHIFT 5 +#define RDMA_SQ_RDMA_WQE_RESERVED0_MASK 0x3 +#define RDMA_SQ_RDMA_WQE_RESERVED0_SHIFT 6 + u8 wqe_size; + u8 prev_wqe_size; + struct regpair remote_va; + __le32 r_key; + u8 dif_flags; +#define RDMA_SQ_RDMA_WQE_DIF_BLOCK_SIZE_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_DIF_BLOCK_SIZE_SHIFT 0 +#define RDMA_SQ_RDMA_WQE_DIF_FIRST_RDMA_IN_IO_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_DIF_FIRST_RDMA_IN_IO_FLG_SHIFT 1 +#define RDMA_SQ_RDMA_WQE_DIF_LAST_RDMA_IN_IO_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_DIF_LAST_RDMA_IN_IO_FLG_SHIFT 2 +#define RDMA_SQ_RDMA_WQE_RESERVED1_MASK 0x1F +#define RDMA_SQ_RDMA_WQE_RESERVED1_SHIFT 3 + u8 reserved2[3]; +}; + +/* First element (16 bytes) of rdma wqe */ +struct rdma_sq_rdma_wqe_1st { + __le32 imm_data; + __le32 length; + __le32 xrc_srq; + u8 req_type; + u8 flags; +#define RDMA_SQ_RDMA_WQE_1ST_COMP_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_1ST_COMP_FLG_SHIFT 0 +#define RDMA_SQ_RDMA_WQE_1ST_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_1ST_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_RDMA_WQE_1ST_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_1ST_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_RDMA_WQE_1ST_SE_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_1ST_SE_FLG_SHIFT 3 +#define RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_RDMA_WQE_1ST_DIF_ON_HOST_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_1ST_DIF_ON_HOST_FLG_SHIFT 5 +#define RDMA_SQ_RDMA_WQE_1ST_RESERVED0_MASK 0x3 +#define RDMA_SQ_RDMA_WQE_1ST_RESERVED0_SHIFT 6 + u8 wqe_size; + u8 prev_wqe_size; +}; + +/* Second element (16 bytes) of rdma wqe */ +struct rdma_sq_rdma_wqe_2nd { + struct regpair remote_va; + __le32 r_key; + u8 dif_flags; +#define RDMA_SQ_RDMA_WQE_2ND_DIF_BLOCK_SIZE_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_2ND_DIF_BLOCK_SIZE_SHIFT 0 +#define RDMA_SQ_RDMA_WQE_2ND_DIF_FIRST_SEGMENT_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_2ND_DIF_FIRST_SEGMENT_FLG_SHIFT 1 +#define RDMA_SQ_RDMA_WQE_2ND_DIF_LAST_SEGMENT_FLG_MASK 0x1 +#define RDMA_SQ_RDMA_WQE_2ND_DIF_LAST_SEGMENT_FLG_SHIFT 2 +#define RDMA_SQ_RDMA_WQE_2ND_RESERVED1_MASK 0x1F +#define RDMA_SQ_RDMA_WQE_2ND_RESERVED1_SHIFT 3 + u8 reserved2[3]; +}; + +/* SQ WQE req type enumeration */ +enum rdma_sq_req_type { + RDMA_SQ_REQ_TYPE_SEND, + RDMA_SQ_REQ_TYPE_SEND_WITH_IMM, + RDMA_SQ_REQ_TYPE_SEND_WITH_INVALIDATE, + RDMA_SQ_REQ_TYPE_RDMA_WR, + RDMA_SQ_REQ_TYPE_RDMA_WR_WITH_IMM, + RDMA_SQ_REQ_TYPE_RDMA_RD, + RDMA_SQ_REQ_TYPE_ATOMIC_CMP_AND_SWAP, + RDMA_SQ_REQ_TYPE_ATOMIC_ADD, + RDMA_SQ_REQ_TYPE_LOCAL_INVALIDATE, + RDMA_SQ_REQ_TYPE_FAST_MR, + RDMA_SQ_REQ_TYPE_BIND, + RDMA_SQ_REQ_TYPE_INVALID, + MAX_RDMA_SQ_REQ_TYPE +}; + +struct rdma_sq_send_wqe { + __le32 inv_key_or_imm_data; + __le32 length; + __le32 xrc_srq; + u8 req_type; + u8 flags; +#define RDMA_SQ_SEND_WQE_COMP_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_COMP_FLG_SHIFT 0 +#define RDMA_SQ_SEND_WQE_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_SEND_WQE_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_SEND_WQE_SE_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_SE_FLG_SHIFT 3 +#define RDMA_SQ_SEND_WQE_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_SEND_WQE_DIF_ON_HOST_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_DIF_ON_HOST_FLG_SHIFT 5 +#define RDMA_SQ_SEND_WQE_RESERVED0_MASK 0x3 +#define RDMA_SQ_SEND_WQE_RESERVED0_SHIFT 6 + u8 wqe_size; + u8 prev_wqe_size; + __le32 reserved1[4]; +}; + +struct rdma_sq_send_wqe_1st { + __le32 inv_key_or_imm_data; + __le32 length; + __le32 xrc_srq; + u8 req_type; + u8 flags; +#define RDMA_SQ_SEND_WQE_1ST_COMP_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_1ST_COMP_FLG_SHIFT 0 +#define RDMA_SQ_SEND_WQE_1ST_RD_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_1ST_RD_FENCE_FLG_SHIFT 1 +#define RDMA_SQ_SEND_WQE_1ST_INV_FENCE_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_1ST_INV_FENCE_FLG_SHIFT 2 +#define RDMA_SQ_SEND_WQE_1ST_SE_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_1ST_SE_FLG_SHIFT 3 +#define RDMA_SQ_SEND_WQE_1ST_INLINE_FLG_MASK 0x1 +#define RDMA_SQ_SEND_WQE_1ST_INLINE_FLG_SHIFT 4 +#define RDMA_SQ_SEND_WQE_1ST_RESERVED0_MASK 0x7 +#define RDMA_SQ_SEND_WQE_1ST_RESERVED0_SHIFT 5 + u8 wqe_size; + u8 prev_wqe_size; +}; + +struct rdma_sq_send_wqe_2st { + __le32 reserved1[4]; +}; + +#endif /* __QED_HSI_RDMA__ */ diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c new file mode 100644 index 0000000..a615142 --- /dev/null +++ b/drivers/infiniband/hw/qedr/verbs.c @@ -0,0 +1,3547 @@ +/* QLogic qedr NIC Driver + * Copyright (c) 2015-2016 QLogic Corporation + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and /or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include <linux/dma-mapping.h> +#include <linux/crc32.h> +#include <net/ip.h> +#include <net/ipv6.h> +#include <net/udp.h> +#include <linux/iommu.h> + +#include <rdma/ib_verbs.h> +#include <rdma/ib_user_verbs.h> +#include <rdma/iw_cm.h> +#include <rdma/ib_umem.h> +#include <rdma/ib_addr.h> +#include <rdma/ib_cache.h> + +#include "qedr_hsi.h" +#include <linux/qed/qed_if.h> +#include "qedr.h" +#include "verbs.h" +#include <rdma/qedr-abi.h> +#include "qedr_cm.h" + +#define DB_ADDR_SHIFT(addr) ((addr) << DB_PWM_ADDR_OFFSET_SHIFT) + +int qedr_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey) +{ + if (index > QEDR_ROCE_PKEY_TABLE_LEN) + return -EINVAL; + + *pkey = QEDR_ROCE_PKEY_DEFAULT; + return 0; +} + +int qedr_query_gid(struct ib_device *ibdev, u8 port, int index, + union ib_gid *sgid) +{ + struct qedr_dev *dev = get_qedr_dev(ibdev); + int rc = 0; + + if (!rdma_cap_roce_gid_table(ibdev, port)) + return -ENODEV; + + rc = ib_get_cached_gid(ibdev, port, index, sgid, NULL); + if (rc == -EAGAIN) { + memcpy(sgid, &zgid, sizeof(*sgid)); + return 0; + } + + DP_DEBUG(dev, QEDR_MSG_INIT, "query gid: index=%d %llx:%llx\n", index, + sgid->global.interface_id, sgid->global.subnet_prefix); + + return rc; +} + +int qedr_add_gid(struct ib_device *device, u8 port_num, + unsigned int index, const union ib_gid *gid, + const struct ib_gid_attr *attr, void **context) +{ + if (!rdma_cap_roce_gid_table(device, port_num)) + return -EINVAL; + + if (port_num > QEDR_MAX_PORT) + return -EINVAL; + + if (!context) + return -EINVAL; + + return 0; +} + +int qedr_del_gid(struct ib_device *device, u8 port_num, + unsigned int index, void **context) +{ + if (!rdma_cap_roce_gid_table(device, port_num)) + return -EINVAL; + + if (port_num > QEDR_MAX_PORT) + return -EINVAL; + + if (!context) + return -EINVAL; + + return 0; +} + +int qedr_query_device(struct ib_device *ibdev, + struct ib_device_attr *attr, struct ib_udata *udata) +{ + struct qedr_dev *dev = get_qedr_dev(ibdev); + struct qedr_device_attr *qattr = &dev->attr; + + if (!dev->rdma_ctx) { + DP_ERR(dev, + "qedr_query_device called with invalid params rdma_ctx=%p\n", + dev->rdma_ctx); + return -EINVAL; + } + + memset(attr, 0, sizeof(*attr)); + + attr->fw_ver = qattr->fw_ver; + attr->sys_image_guid = qattr->sys_image_guid; + attr->max_mr_size = qattr->max_mr_size; + attr->page_size_cap = qattr->page_size_caps; + attr->vendor_id = qattr->vendor_id; + attr->vendor_part_id = qattr->vendor_part_id; + attr->hw_ver = qattr->hw_ver; + attr->max_qp = qattr->max_qp; + attr->max_qp_wr = max_t(u32, qattr->max_sqe, qattr->max_rqe); + attr->device_cap_flags = IB_DEVICE_CURR_QP_STATE_MOD | + IB_DEVICE_RC_RNR_NAK_GEN | + IB_DEVICE_LOCAL_DMA_LKEY | IB_DEVICE_MEM_MGT_EXTENSIONS; + + attr->max_sge = qattr->max_sge; + attr->max_sge_rd = qattr->max_sge; + attr->max_cq = qattr->max_cq; + attr->max_cqe = qattr->max_cqe; + attr->max_mr = qattr->max_mr; + attr->max_mw = qattr->max_mw; + attr->max_pd = qattr->max_pd; + attr->atomic_cap = dev->atomic_cap; + attr->max_fmr = qattr->max_fmr; + attr->max_map_per_fmr = 16; + attr->max_qp_init_rd_atom = + 1 << (fls(qattr->max_qp_req_rd_atomic_resc) - 1); + attr->max_qp_rd_atom = + min(1 << (fls(qattr->max_qp_resp_rd_atomic_resc) - 1), + attr->max_qp_init_rd_atom); + + attr->max_srq = qattr->max_srq; + attr->max_srq_sge = qattr->max_srq_sge; + attr->max_srq_wr = qattr->max_srq_wr; + + attr->local_ca_ack_delay = qattr->dev_ack_delay; + attr->max_fast_reg_page_list_len = qattr->max_mr / 8; + attr->max_pkeys = QEDR_ROCE_PKEY_MAX; + attr->max_ah = qattr->max_ah; + + return 0; +} + +#define QEDR_SPEED_SDR (1) +#define QEDR_SPEED_DDR (2) +#define QEDR_SPEED_QDR (4) +#define QEDR_SPEED_FDR10 (8) +#define QEDR_SPEED_FDR (16) +#define QEDR_SPEED_EDR (32) + +static inline void get_link_speed_and_width(int speed, u8 *ib_speed, + u8 *ib_width) +{ + switch (speed) { + case 1000: + *ib_speed = QEDR_SPEED_SDR; + *ib_width = IB_WIDTH_1X; + break; + case 10000: + *ib_speed = QEDR_SPEED_QDR; + *ib_width = IB_WIDTH_1X; + break; + + case 20000: + *ib_speed = QEDR_SPEED_DDR; + *ib_width = IB_WIDTH_4X; + break; + + case 25000: + *ib_speed = QEDR_SPEED_EDR; + *ib_width = IB_WIDTH_1X; + break; + + case 40000: + *ib_speed = QEDR_SPEED_QDR; + *ib_width = IB_WIDTH_4X; + break; + + case 50000: + *ib_speed = QEDR_SPEED_QDR; + *ib_width = IB_WIDTH_4X; + break; + + case 100000: + *ib_speed = QEDR_SPEED_EDR; + *ib_width = IB_WIDTH_4X; + break; + + default: + /* Unsupported */ + *ib_speed = QEDR_SPEED_SDR; + *ib_width = IB_WIDTH_1X; + } +} + +int qedr_query_port(struct ib_device *ibdev, u8 port, struct ib_port_attr *attr) +{ + struct qedr_dev *dev; + struct qed_rdma_port *rdma_port; + + dev = get_qedr_dev(ibdev); + if (port > 1) { + DP_ERR(dev, "invalid_port=0x%x\n", port); + return -EINVAL; + } + + if (!dev->rdma_ctx) { + DP_ERR(dev, "rdma_ctx is NULL\n"); + return -EINVAL; + } + + rdma_port = dev->ops->rdma_query_port(dev->rdma_ctx); + memset(attr, 0, sizeof(*attr)); + + if (rdma_port->port_state == QED_RDMA_PORT_UP) { + attr->state = IB_PORT_ACTIVE; + attr->phys_state = 5; + } else { + attr->state = IB_PORT_DOWN; + attr->phys_state = 3; + } + attr->max_mtu = IB_MTU_4096; + attr->active_mtu = iboe_get_mtu(dev->ndev->mtu); + attr->lid = 0; + attr->lmc = 0; + attr->sm_lid = 0; + attr->sm_sl = 0; + attr->port_cap_flags = IB_PORT_IP_BASED_GIDS; + attr->gid_tbl_len = QEDR_MAX_SGID; + attr->pkey_tbl_len = QEDR_ROCE_PKEY_TABLE_LEN; + attr->bad_pkey_cntr = rdma_port->pkey_bad_counter; + attr->qkey_viol_cntr = 0; + get_link_speed_and_width(rdma_port->link_speed, + &attr->active_speed, &attr->active_width); + attr->max_msg_sz = rdma_port->max_msg_size; + attr->max_vl_num = 4; + + return 0; +} + +int qedr_modify_port(struct ib_device *ibdev, u8 port, int mask, + struct ib_port_modify *props) +{ + struct qedr_dev *dev; + + dev = get_qedr_dev(ibdev); + if (port > 1) { + DP_ERR(dev, "invalid_port=0x%x\n", port); + return -EINVAL; + } + + return 0; +} + +static int qedr_add_mmap(struct qedr_ucontext *uctx, u64 phy_addr, + unsigned long len) +{ + struct qedr_mm *mm; + + mm = kzalloc(sizeof(*mm), GFP_KERNEL); + if (!mm) + return -ENOMEM; + + mm->key.phy_addr = phy_addr; + /* This function might be called with a length which is not a multiple + * of PAGE_SIZE, while the mapping is PAGE_SIZE grained and the kernel + * forces this granularity by increasing the requested size if needed. + * When qedr_mmap is called, it will search the list with the updated + * length as a key. To prevent search failures, the length is rounded up + * in advance to PAGE_SIZE. + */ + mm->key.len = roundup(len, PAGE_SIZE); + INIT_LIST_HEAD(&mm->entry); + + mutex_lock(&uctx->mm_list_lock); + list_add(&mm->entry, &uctx->mm_head); + mutex_unlock(&uctx->mm_list_lock); + + DP_DEBUG(uctx->dev, QEDR_MSG_MISC, + "added (addr=0x%llx,len=0x%lx) for ctx=%p\n", + (unsigned long long)mm->key.phy_addr, + (unsigned long)mm->key.len, uctx); + + return 0; +} + +static bool qedr_search_mmap(struct qedr_ucontext *uctx, u64 phy_addr, + unsigned long len) +{ + bool found = false; + struct qedr_mm *mm; + + mutex_lock(&uctx->mm_list_lock); + list_for_each_entry(mm, &uctx->mm_head, entry) { + if (len != mm->key.len || phy_addr != mm->key.phy_addr) + continue; + + found = true; + break; + } + mutex_unlock(&uctx->mm_list_lock); + DP_DEBUG(uctx->dev, QEDR_MSG_MISC, + "searched for (addr=0x%llx,len=0x%lx) for ctx=%p, result=%d\n", + mm->key.phy_addr, mm->key.len, uctx, found); + + return found; +} + +struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *ibdev, + struct ib_udata *udata) +{ + int rc; + struct qedr_ucontext *ctx; + struct qedr_alloc_ucontext_resp uresp; + struct qedr_dev *dev = get_qedr_dev(ibdev); + struct qed_rdma_add_user_out_params oparams; + + if (!udata) + return ERR_PTR(-EFAULT); + + ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); + if (!ctx) + return ERR_PTR(-ENOMEM); + + rc = dev->ops->rdma_add_user(dev->rdma_ctx, &oparams); + if (rc) { + DP_ERR(dev, + "failed to allocate a DPI for a new RoCE application, rc=%d. To overcome this consider to increase the number of DPIs, increase the doorbell BAR size or just close unnecessary RoCE applications. In order to increase the number of DPIs consult the qedr readme\n", + rc); + goto err; + } + + ctx->dpi = oparams.dpi; + ctx->dpi_addr = oparams.dpi_addr; + ctx->dpi_phys_addr = oparams.dpi_phys_addr; + ctx->dpi_size = oparams.dpi_size; + INIT_LIST_HEAD(&ctx->mm_head); + mutex_init(&ctx->mm_list_lock); + + memset(&uresp, 0, sizeof(uresp)); + + uresp.db_pa = ctx->dpi_phys_addr; + uresp.db_size = ctx->dpi_size; + uresp.max_send_wr = dev->attr.max_sqe; + uresp.max_recv_wr = dev->attr.max_rqe; + uresp.max_srq_wr = dev->attr.max_srq_wr; + uresp.sges_per_send_wr = QEDR_MAX_SQE_ELEMENTS_PER_SQE; + uresp.sges_per_recv_wr = QEDR_MAX_RQE_ELEMENTS_PER_RQE; + uresp.sges_per_srq_wr = dev->attr.max_srq_sge; + uresp.max_cqes = QEDR_MAX_CQES; + + rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp)); + if (rc) + goto err; + + ctx->dev = dev; + + rc = qedr_add_mmap(ctx, ctx->dpi_phys_addr, ctx->dpi_size); + if (rc) + goto err; + + DP_DEBUG(dev, QEDR_MSG_INIT, "Allocating user context %p\n", + &ctx->ibucontext); + return &ctx->ibucontext; + +err: + kfree(ctx); + return ERR_PTR(rc); +} + +int qedr_dealloc_ucontext(struct ib_ucontext *ibctx) +{ + struct qedr_ucontext *uctx = get_qedr_ucontext(ibctx); + struct qedr_mm *mm, *tmp; + int status = 0; + + DP_DEBUG(uctx->dev, QEDR_MSG_INIT, "Deallocating user context %p\n", + uctx); + uctx->dev->ops->rdma_remove_user(uctx->dev->rdma_ctx, uctx->dpi); + + list_for_each_entry_safe(mm, tmp, &uctx->mm_head, entry) { + DP_DEBUG(uctx->dev, QEDR_MSG_MISC, + "deleted (addr=0x%llx,len=0x%lx) for ctx=%p\n", + mm->key.phy_addr, mm->key.len, uctx); + list_del(&mm->entry); + kfree(mm); + } + + kfree(uctx); + return status; +} + +int qedr_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) +{ + struct qedr_ucontext *ucontext = get_qedr_ucontext(context); + struct qedr_dev *dev = get_qedr_dev(context->device); + unsigned long vm_page = vma->vm_pgoff << PAGE_SHIFT; + u64 unmapped_db = dev->db_phys_addr; + unsigned long len = (vma->vm_end - vma->vm_start); + int rc = 0; + bool found; + + DP_DEBUG(dev, QEDR_MSG_INIT, + "qedr_mmap called vm_page=0x%lx vm_pgoff=0x%lx unmapped_db=0x%llx db_size=%x, len=%lx\n", + vm_page, vma->vm_pgoff, unmapped_db, dev->db_size, len); + if (vma->vm_start & (PAGE_SIZE - 1)) { + DP_ERR(dev, "Vma_start not page aligned = %ld\n", + vma->vm_start); + return -EINVAL; + } + + found = qedr_search_mmap(ucontext, vm_page, len); + if (!found) { + DP_ERR(dev, "Vma_pgoff not found in mapped array = %ld\n", + vma->vm_pgoff); + return -EINVAL; + } + + DP_DEBUG(dev, QEDR_MSG_INIT, "Mapping doorbell bar\n"); + + if ((vm_page >= unmapped_db) && (vm_page <= (unmapped_db + + dev->db_size))) { + DP_DEBUG(dev, QEDR_MSG_INIT, "Mapping doorbell bar\n"); + if (vma->vm_flags & VM_READ) { + DP_ERR(dev, "Trying to map doorbell bar for read\n"); + return -EPERM; + } + + vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot); + + rc = io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff, + PAGE_SIZE, vma->vm_page_prot); + } else { + DP_DEBUG(dev, QEDR_MSG_INIT, "Mapping chains\n"); + rc = remap_pfn_range(vma, vma->vm_start, + vma->vm_pgoff, len, vma->vm_page_prot); + } + DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_mmap return code: %d\n", rc); + return rc; +} + +struct ib_pd *qedr_alloc_pd(struct ib_device *ibdev, + struct ib_ucontext *context, struct ib_udata *udata) +{ + struct qedr_dev *dev = get_qedr_dev(ibdev); + struct qedr_ucontext *uctx = NULL; + struct qedr_alloc_pd_uresp uresp; + struct qedr_pd *pd; + u16 pd_id; + int rc; + + DP_DEBUG(dev, QEDR_MSG_INIT, "Function called from: %s\n", + (udata && context) ? "User Lib" : "Kernel"); + + if (!dev->rdma_ctx) { + DP_ERR(dev, "invlaid RDMA context\n"); + return ERR_PTR(-EINVAL); + } + + pd = kzalloc(sizeof(*pd), GFP_KERNEL); + if (!pd) + return ERR_PTR(-ENOMEM); + + dev->ops->rdma_alloc_pd(dev->rdma_ctx, &pd_id); + + uresp.pd_id = pd_id; + pd->pd_id = pd_id; + + if (udata && context) { + rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp)); + if (rc) + DP_ERR(dev, "copy error pd_id=0x%x.\n", pd_id); + uctx = get_qedr_ucontext(context); + uctx->pd = pd; + pd->uctx = uctx; + } + + return &pd->ibpd; +} + +int qedr_dealloc_pd(struct ib_pd *ibpd) +{ + struct qedr_dev *dev = get_qedr_dev(ibpd->device); + struct qedr_pd *pd = get_qedr_pd(ibpd); + + if (!pd) + pr_err("Invalid PD received in dealloc_pd\n"); + + DP_DEBUG(dev, QEDR_MSG_INIT, "Deallocating PD %d\n", pd->pd_id); + dev->ops->rdma_dealloc_pd(dev->rdma_ctx, pd->pd_id); + + kfree(pd); + + return 0; +} + +static void qedr_free_pbl(struct qedr_dev *dev, + struct qedr_pbl_info *pbl_info, struct qedr_pbl *pbl) +{ + struct pci_dev *pdev = dev->pdev; + int i; + + for (i = 0; i < pbl_info->num_pbls; i++) { + if (!pbl[i].va) + continue; + dma_free_coherent(&pdev->dev, pbl_info->pbl_size, + pbl[i].va, pbl[i].pa); + } + + kfree(pbl); +} + +#define MIN_FW_PBL_PAGE_SIZE (4 * 1024) +#define MAX_FW_PBL_PAGE_SIZE (64 * 1024) + +#define NUM_PBES_ON_PAGE(_page_size) (_page_size / sizeof(u64)) +#define MAX_PBES_ON_PAGE NUM_PBES_ON_PAGE(MAX_FW_PBL_PAGE_SIZE) +#define MAX_PBES_TWO_LAYER (MAX_PBES_ON_PAGE * MAX_PBES_ON_PAGE) + +static struct qedr_pbl *qedr_alloc_pbl_tbl(struct qedr_dev *dev, + struct qedr_pbl_info *pbl_info, + gfp_t flags) +{ + struct pci_dev *pdev = dev->pdev; + struct qedr_pbl *pbl_table; + dma_addr_t *pbl_main_tbl; + dma_addr_t pa; + void *va; + int i; + + pbl_table = kcalloc(pbl_info->num_pbls, sizeof(*pbl_table), flags); + if (!pbl_table) + return ERR_PTR(-ENOMEM); + + for (i = 0; i < pbl_info->num_pbls; i++) { + va = dma_alloc_coherent(&pdev->dev, pbl_info->pbl_size, + &pa, flags); + if (!va) + goto err; + + memset(va, 0, pbl_info->pbl_size); + pbl_table[i].va = va; + pbl_table[i].pa = pa; + } + + /* Two-Layer PBLs, if we have more than one pbl we need to initialize + * the first one with physical pointers to all of the rest + */ + pbl_main_tbl = (dma_addr_t *)pbl_table[0].va; + for (i = 0; i < pbl_info->num_pbls - 1; i++) + pbl_main_tbl[i] = pbl_table[i + 1].pa; + + return pbl_table; + +err: + for (i--; i >= 0; i--) + dma_free_coherent(&pdev->dev, pbl_info->pbl_size, + pbl_table[i].va, pbl_table[i].pa); + + qedr_free_pbl(dev, pbl_info, pbl_table); + + return ERR_PTR(-ENOMEM); +} + +static int qedr_prepare_pbl_tbl(struct qedr_dev *dev, + struct qedr_pbl_info *pbl_info, + u32 num_pbes, int two_layer_capable) +{ + u32 pbl_capacity; + u32 pbl_size; + u32 num_pbls; + + if ((num_pbes > MAX_PBES_ON_PAGE) && two_layer_capable) { + if (num_pbes > MAX_PBES_TWO_LAYER) { + DP_ERR(dev, "prepare pbl table: too many pages %d\n", + num_pbes); + return -EINVAL; + } + + /* calculate required pbl page size */ + pbl_size = MIN_FW_PBL_PAGE_SIZE; + pbl_capacity = NUM_PBES_ON_PAGE(pbl_size) * + NUM_PBES_ON_PAGE(pbl_size); + + while (pbl_capacity < num_pbes) { + pbl_size *= 2; + pbl_capacity = pbl_size / sizeof(u64); + pbl_capacity = pbl_capacity * pbl_capacity; + } + + num_pbls = DIV_ROUND_UP(num_pbes, NUM_PBES_ON_PAGE(pbl_size)); + num_pbls++; /* One for the layer0 ( points to the pbls) */ + pbl_info->two_layered = true; + } else { + /* One layered PBL */ + num_pbls = 1; + pbl_size = max_t(u32, MIN_FW_PBL_PAGE_SIZE, + roundup_pow_of_two((num_pbes * sizeof(u64)))); + pbl_info->two_layered = false; + } + + pbl_info->num_pbls = num_pbls; + pbl_info->pbl_size = pbl_size; + pbl_info->num_pbes = num_pbes; + + DP_DEBUG(dev, QEDR_MSG_MR, + "prepare pbl table: num_pbes=%d, num_pbls=%d, pbl_size=%d\n", + pbl_info->num_pbes, pbl_info->num_pbls, pbl_info->pbl_size); + + return 0; +} + +static void qedr_populate_pbls(struct qedr_dev *dev, struct ib_umem *umem, + struct qedr_pbl *pbl, + struct qedr_pbl_info *pbl_info) +{ + int shift, pg_cnt, pages, pbe_cnt, total_num_pbes = 0; + struct qedr_pbl *pbl_tbl; + struct scatterlist *sg; + struct regpair *pbe; + int entry; + u32 addr; + + if (!pbl_info->num_pbes) + return; + + /* If we have a two layered pbl, the first pbl points to the rest + * of the pbls and the first entry lays on the second pbl in the table + */ + if (pbl_info->two_layered) + pbl_tbl = &pbl[1]; + else + pbl_tbl = pbl; + + pbe = (struct regpair *)pbl_tbl->va; + if (!pbe) { + DP_ERR(dev, "cannot populate PBL due to a NULL PBE\n"); + return; + } + + pbe_cnt = 0; + + shift = ilog2(umem->page_size); + + for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { + pages = sg_dma_len(sg) >> shift; + for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) { + /* store the page address in pbe */ + pbe->lo = cpu_to_le32(sg_dma_address(sg) + + umem->page_size * pg_cnt); + addr = upper_32_bits(sg_dma_address(sg) + + umem->page_size * pg_cnt); + pbe->hi = cpu_to_le32(addr); + pbe_cnt++; + total_num_pbes++; + pbe++; + + if (total_num_pbes == pbl_info->num_pbes) + return; + + /* If the given pbl is full storing the pbes, + * move to next pbl. + */ + if (pbe_cnt == (pbl_info->pbl_size / sizeof(u64))) { + pbl_tbl++; + pbe = (struct regpair *)pbl_tbl->va; + pbe_cnt = 0; + } + } + } +} + +static int qedr_copy_cq_uresp(struct qedr_dev *dev, + struct qedr_cq *cq, struct ib_udata *udata) +{ + struct qedr_create_cq_uresp uresp; + int rc; + + memset(&uresp, 0, sizeof(uresp)); + + uresp.db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_UCM_RDMA_CQ_CONS_32BIT); + uresp.icid = cq->icid; + + rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp)); + if (rc) + DP_ERR(dev, "copy error cqid=0x%x.\n", cq->icid); + + return rc; +} + +static void consume_cqe(struct qedr_cq *cq) +{ + if (cq->latest_cqe == cq->toggle_cqe) + cq->pbl_toggle ^= RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK; + + cq->latest_cqe = qed_chain_consume(&cq->pbl); +} + +static inline int qedr_align_cq_entries(int entries) +{ + u64 size, aligned_size; + + /* We allocate an extra entry that we don't report to the FW. */ + size = (entries + 1) * QEDR_CQE_SIZE; + aligned_size = ALIGN(size, PAGE_SIZE); + + return aligned_size / QEDR_CQE_SIZE; +} + +static inline int qedr_init_user_queue(struct ib_ucontext *ib_ctx, + struct qedr_dev *dev, + struct qedr_userq *q, + u64 buf_addr, size_t buf_len, + int access, int dmasync) +{ + int page_cnt; + int rc; + + q->buf_addr = buf_addr; + q->buf_len = buf_len; + q->umem = ib_umem_get(ib_ctx, q->buf_addr, q->buf_len, access, dmasync); + if (IS_ERR(q->umem)) { + DP_ERR(dev, "create user queue: failed ib_umem_get, got %ld\n", + PTR_ERR(q->umem)); + return PTR_ERR(q->umem); + } + + page_cnt = ib_umem_page_count(q->umem); + rc = qedr_prepare_pbl_tbl(dev, &q->pbl_info, page_cnt, 0); + if (rc) + goto err0; + + q->pbl_tbl = qedr_alloc_pbl_tbl(dev, &q->pbl_info, GFP_KERNEL); + if (IS_ERR_OR_NULL(q->pbl_tbl)) + goto err0; + + qedr_populate_pbls(dev, q->umem, q->pbl_tbl, &q->pbl_info); + + return 0; + +err0: + ib_umem_release(q->umem); + + return rc; +} + +static inline void qedr_init_cq_params(struct qedr_cq *cq, + struct qedr_ucontext *ctx, + struct qedr_dev *dev, int vector, + int chain_entries, int page_cnt, + u64 pbl_ptr, + struct qed_rdma_create_cq_in_params + *params) +{ + memset(params, 0, sizeof(*params)); + params->cq_handle_hi = upper_32_bits((uintptr_t)cq); + params->cq_handle_lo = lower_32_bits((uintptr_t)cq); + params->cnq_id = vector; + params->cq_size = chain_entries - 1; + params->dpi = (ctx) ? ctx->dpi : dev->dpi; + params->pbl_num_pages = page_cnt; + params->pbl_ptr = pbl_ptr; + params->pbl_two_level = 0; +} + +static void doorbell_cq(struct qedr_cq *cq, u32 cons, u8 flags) +{ + /* Flush data before signalling doorbell */ + wmb(); + cq->db.data.agg_flags = flags; + cq->db.data.value = cpu_to_le32(cons); + writeq(cq->db.raw, cq->db_addr); + + /* Make sure write would stick */ + mmiowb(); +} + +int qedr_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags) +{ + struct qedr_cq *cq = get_qedr_cq(ibcq); + unsigned long sflags; + + if (cq->cq_type == QEDR_CQ_TYPE_GSI) + return 0; + + spin_lock_irqsave(&cq->cq_lock, sflags); + + cq->arm_flags = 0; + + if (flags & IB_CQ_SOLICITED) + cq->arm_flags |= DQ_UCM_ROCE_CQ_ARM_SE_CF_CMD; + + if (flags & IB_CQ_NEXT_COMP) + cq->arm_flags |= DQ_UCM_ROCE_CQ_ARM_CF_CMD; + + doorbell_cq(cq, cq->cq_cons - 1, cq->arm_flags); + + spin_unlock_irqrestore(&cq->cq_lock, sflags); + + return 0; +} + +struct ib_cq *qedr_create_cq(struct ib_device *ibdev, + const struct ib_cq_init_attr *attr, + struct ib_ucontext *ib_ctx, struct ib_udata *udata) +{ + struct qedr_ucontext *ctx = get_qedr_ucontext(ib_ctx); + struct qed_rdma_destroy_cq_out_params destroy_oparams; + struct qed_rdma_destroy_cq_in_params destroy_iparams; + struct qedr_dev *dev = get_qedr_dev(ibdev); + struct qed_rdma_create_cq_in_params params; + struct qedr_create_cq_ureq ureq; + int vector = attr->comp_vector; + int entries = attr->cqe; + struct qedr_cq *cq; + int chain_entries; + int page_cnt; + u64 pbl_ptr; + u16 icid; + int rc; + + DP_DEBUG(dev, QEDR_MSG_INIT, + "create_cq: called from %s. entries=%d, vector=%d\n", + udata ? "User Lib" : "Kernel", entries, vector); + + if (entries > QEDR_MAX_CQES) { + DP_ERR(dev, + "create cq: the number of entries %d is too high. Must be equal or below %d.\n", + entries, QEDR_MAX_CQES); + return ERR_PTR(-EINVAL); + } + + chain_entries = qedr_align_cq_entries(entries); + chain_entries = min_t(int, chain_entries, QEDR_MAX_CQES); + + cq = kzalloc(sizeof(*cq), GFP_KERNEL); + if (!cq) + return ERR_PTR(-ENOMEM); + + if (udata) { + memset(&ureq, 0, sizeof(ureq)); + if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) { + DP_ERR(dev, + "create cq: problem copying data from user space\n"); + goto err0; + } + + if (!ureq.len) { + DP_ERR(dev, + "create cq: cannot create a cq with 0 entries\n"); + goto err0; + } + + cq->cq_type = QEDR_CQ_TYPE_USER; + + rc = qedr_init_user_queue(ib_ctx, dev, &cq->q, ureq.addr, + ureq.len, IB_ACCESS_LOCAL_WRITE, 1); + if (rc) + goto err0; + + pbl_ptr = cq->q.pbl_tbl->pa; + page_cnt = cq->q.pbl_info.num_pbes; + } else { + cq->cq_type = QEDR_CQ_TYPE_KERNEL; + + rc = dev->ops->common->chain_alloc(dev->cdev, + QED_CHAIN_USE_TO_CONSUME, + QED_CHAIN_MODE_PBL, + QED_CHAIN_CNT_TYPE_U32, + chain_entries, + sizeof(union rdma_cqe), + &cq->pbl); + if (rc) + goto err1; + + page_cnt = qed_chain_get_page_cnt(&cq->pbl); + pbl_ptr = qed_chain_get_pbl_phys(&cq->pbl); + } + + qedr_init_cq_params(cq, ctx, dev, vector, chain_entries, page_cnt, + pbl_ptr, ¶ms); + + rc = dev->ops->rdma_create_cq(dev->rdma_ctx, ¶ms, &icid); + if (rc) + goto err2; + + cq->icid = icid; + cq->sig = QEDR_CQ_MAGIC_NUMBER; + spin_lock_init(&cq->cq_lock); + + if (ib_ctx) { + rc = qedr_copy_cq_uresp(dev, cq, udata); + if (rc) + goto err3; + } else { + /* Generate doorbell address. */ + cq->db_addr = dev->db_addr + + DB_ADDR_SHIFT(DQ_PWM_OFFSET_UCM_RDMA_CQ_CONS_32BIT); + cq->db.data.icid = cq->icid; + cq->db.data.params = DB_AGG_CMD_SET << + RDMA_PWM_VAL32_DATA_AGG_CMD_SHIFT; + + /* point to the very last element, passing it we will toggle */ + cq->toggle_cqe = qed_chain_get_last_elem(&cq->pbl); + cq->pbl_toggle = RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK; + cq->latest_cqe = NULL; + consume_cqe(cq); + cq->cq_cons = qed_chain_get_cons_idx_u32(&cq->pbl); + } + + DP_DEBUG(dev, QEDR_MSG_CQ, + "create cq: icid=0x%0x, addr=%p, size(entries)=0x%0x\n", + cq->icid, cq, params.cq_size); + + return &cq->ibcq; + +err3: + destroy_iparams.icid = cq->icid; + dev->ops->rdma_destroy_cq(dev->rdma_ctx, &destroy_iparams, + &destroy_oparams); +err2: + if (udata) + qedr_free_pbl(dev, &cq->q.pbl_info, cq->q.pbl_tbl); + else + dev->ops->common->chain_free(dev->cdev, &cq->pbl); +err1: + if (udata) + ib_umem_release(cq->q.umem); +err0: + kfree(cq); + return ERR_PTR(-EINVAL); +} + +int qedr_resize_cq(struct ib_cq *ibcq, int new_cnt, struct ib_udata *udata) +{ + struct qedr_dev *dev = get_qedr_dev(ibcq->device); + struct qedr_cq *cq = get_qedr_cq(ibcq); + + DP_ERR(dev, "cq %p RESIZE NOT SUPPORTED\n", cq); + + return 0; +} + +int qedr_destroy_cq(struct ib_cq *ibcq) +{ + struct qedr_dev *dev = get_qedr_dev(ibcq->device); + struct qed_rdma_destroy_cq_out_params oparams; + struct qed_rdma_destroy_cq_in_params iparams; + struct qedr_cq *cq = get_qedr_cq(ibcq); + + DP_DEBUG(dev, QEDR_MSG_CQ, "destroy cq: cq_id %d", cq->icid); + + /* GSIs CQs are handled by driver, so they don't exist in the FW */ + if (cq->cq_type != QEDR_CQ_TYPE_GSI) { + iparams.icid = cq->icid; + dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams); + dev->ops->common->chain_free(dev->cdev, &cq->pbl); + } + + if (ibcq->uobject && ibcq->uobject->context) { + qedr_free_pbl(dev, &cq->q.pbl_info, cq->q.pbl_tbl); + ib_umem_release(cq->q.umem); + } + + kfree(cq); + + return 0; +} + +static inline int get_gid_info_from_table(struct ib_qp *ibqp, + struct ib_qp_attr *attr, + int attr_mask, + struct qed_rdma_modify_qp_in_params + *qp_params) +{ + enum rdma_network_type nw_type; + struct ib_gid_attr gid_attr; + union ib_gid gid; + u32 ipv4_addr; + int rc = 0; + int i; + + rc = ib_get_cached_gid(ibqp->device, attr->ah_attr.port_num, + attr->ah_attr.grh.sgid_index, &gid, &gid_attr); + if (rc) + return rc; + + if (!memcmp(&gid, &zgid, sizeof(gid))) + return -ENOENT; + + if (gid_attr.ndev) { + qp_params->vlan_id = rdma_vlan_dev_vlan_id(gid_attr.ndev); + + dev_put(gid_attr.ndev); + nw_type = ib_gid_to_network_type(gid_attr.gid_type, &gid); + switch (nw_type) { + case RDMA_NETWORK_IPV6: + memcpy(&qp_params->sgid.bytes[0], &gid.raw[0], + sizeof(qp_params->sgid)); + memcpy(&qp_params->dgid.bytes[0], + &attr->ah_attr.grh.dgid, + sizeof(qp_params->dgid)); + qp_params->roce_mode = ROCE_V2_IPV6; + SET_FIELD(qp_params->modify_flags, + QED_ROCE_MODIFY_QP_VALID_ROCE_MODE, 1); + break; + case RDMA_NETWORK_IB: + memcpy(&qp_params->sgid.bytes[0], &gid.raw[0], + sizeof(qp_params->sgid)); + memcpy(&qp_params->dgid.bytes[0], + &attr->ah_attr.grh.dgid, + sizeof(qp_params->dgid)); + qp_params->roce_mode = ROCE_V1; + break; + case RDMA_NETWORK_IPV4: + memset(&qp_params->sgid, 0, sizeof(qp_params->sgid)); + memset(&qp_params->dgid, 0, sizeof(qp_params->dgid)); + ipv4_addr = qedr_get_ipv4_from_gid(gid.raw); + qp_params->sgid.ipv4_addr = ipv4_addr; + ipv4_addr = + qedr_get_ipv4_from_gid(attr->ah_attr.grh.dgid.raw); + qp_params->dgid.ipv4_addr = ipv4_addr; + SET_FIELD(qp_params->modify_flags, + QED_ROCE_MODIFY_QP_VALID_ROCE_MODE, 1); + qp_params->roce_mode = ROCE_V2_IPV4; + break; + } + } + + for (i = 0; i < 4; i++) { + qp_params->sgid.dwords[i] = ntohl(qp_params->sgid.dwords[i]); + qp_params->dgid.dwords[i] = ntohl(qp_params->dgid.dwords[i]); + } + + if (qp_params->vlan_id >= VLAN_CFI_MASK) + qp_params->vlan_id = 0; + + return 0; +} + +static void qedr_cleanup_user_sq(struct qedr_dev *dev, struct qedr_qp *qp) +{ + qedr_free_pbl(dev, &qp->usq.pbl_info, qp->usq.pbl_tbl); + ib_umem_release(qp->usq.umem); +} + +static void qedr_cleanup_user_rq(struct qedr_dev *dev, struct qedr_qp *qp) +{ + qedr_free_pbl(dev, &qp->urq.pbl_info, qp->urq.pbl_tbl); + ib_umem_release(qp->urq.umem); +} + +static void qedr_cleanup_kernel_sq(struct qedr_dev *dev, struct qedr_qp *qp) +{ + dev->ops->common->chain_free(dev->cdev, &qp->sq.pbl); + kfree(qp->wqe_wr_id); +} + +static void qedr_cleanup_kernel_rq(struct qedr_dev *dev, struct qedr_qp *qp) +{ + dev->ops->common->chain_free(dev->cdev, &qp->rq.pbl); + kfree(qp->rqe_wr_id); +} + +static int qedr_check_qp_attrs(struct ib_pd *ibpd, struct qedr_dev *dev, + struct ib_qp_init_attr *attrs) +{ + struct qedr_device_attr *qattr = &dev->attr; + + /* QP0... attrs->qp_type == IB_QPT_GSI */ + if (attrs->qp_type != IB_QPT_RC && attrs->qp_type != IB_QPT_GSI) { + DP_DEBUG(dev, QEDR_MSG_QP, + "create qp: unsupported qp type=0x%x requested\n", + attrs->qp_type); + return -EINVAL; + } + + if (attrs->cap.max_send_wr > qattr->max_sqe) { + DP_ERR(dev, + "create qp: cannot create a SQ with %d elements (max_send_wr=0x%x)\n", + attrs->cap.max_send_wr, qattr->max_sqe); + return -EINVAL; + } + + if (attrs->cap.max_inline_data > qattr->max_inline) { + DP_ERR(dev, + "create qp: unsupported inline data size=0x%x requested (max_inline=0x%x)\n", + attrs->cap.max_inline_data, qattr->max_inline); + return -EINVAL; + } + + if (attrs->cap.max_send_sge > qattr->max_sge) { + DP_ERR(dev, + "create qp: unsupported send_sge=0x%x requested (max_send_sge=0x%x)\n", + attrs->cap.max_send_sge, qattr->max_sge); + return -EINVAL; + } + + if (attrs->cap.max_recv_sge > qattr->max_sge) { + DP_ERR(dev, + "create qp: unsupported recv_sge=0x%x requested (max_recv_sge=0x%x)\n", + attrs->cap.max_recv_sge, qattr->max_sge); + return -EINVAL; + } + + /* Unprivileged user space cannot create special QP */ + if (ibpd->uobject && attrs->qp_type == IB_QPT_GSI) { + DP_ERR(dev, + "create qp: userspace can't create special QPs of type=0x%x\n", + attrs->qp_type); + return -EINVAL; + } + + return 0; +} + +static void qedr_copy_rq_uresp(struct qedr_create_qp_uresp *uresp, + struct qedr_qp *qp) +{ + uresp->rq_db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD); + uresp->rq_icid = qp->icid; +} + +static void qedr_copy_sq_uresp(struct qedr_create_qp_uresp *uresp, + struct qedr_qp *qp) +{ + uresp->sq_db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD); + uresp->sq_icid = qp->icid + 1; +} + +static int qedr_copy_qp_uresp(struct qedr_dev *dev, + struct qedr_qp *qp, struct ib_udata *udata) +{ + struct qedr_create_qp_uresp uresp; + int rc; + + memset(&uresp, 0, sizeof(uresp)); + qedr_copy_sq_uresp(&uresp, qp); + qedr_copy_rq_uresp(&uresp, qp); + + uresp.atomic_supported = dev->atomic_cap != IB_ATOMIC_NONE; + uresp.qp_id = qp->qp_id; + + rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp)); + if (rc) + DP_ERR(dev, + "create qp: failed a copy to user space with qp icid=0x%x.\n", + qp->icid); + + return rc; +} + +static void qedr_set_qp_init_params(struct qedr_dev *dev, + struct qedr_qp *qp, + struct qedr_pd *pd, + struct ib_qp_init_attr *attrs) +{ + qp->pd = pd; + + spin_lock_init(&qp->q_lock); + + qp->qp_type = attrs->qp_type; + qp->max_inline_data = attrs->cap.max_inline_data; + qp->sq.max_sges = attrs->cap.max_send_sge; + qp->state = QED_ROCE_QP_STATE_RESET; + qp->signaled = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR) ? true : false; + qp->sq_cq = get_qedr_cq(attrs->send_cq); + qp->rq_cq = get_qedr_cq(attrs->recv_cq); + qp->dev = dev; + + DP_DEBUG(dev, QEDR_MSG_QP, + "QP params:\tpd = %d, qp_type = %d, max_inline_data = %d, state = %d, signaled = %d, use_srq=%d\n", + pd->pd_id, qp->qp_type, qp->max_inline_data, + qp->state, qp->signaled, (attrs->srq) ? 1 : 0); + DP_DEBUG(dev, QEDR_MSG_QP, + "SQ params:\tsq_max_sges = %d, sq_cq_id = %d\n", + qp->sq.max_sges, qp->sq_cq->icid); + qp->rq.max_sges = attrs->cap.max_recv_sge; + DP_DEBUG(dev, QEDR_MSG_QP, + "RQ params:\trq_max_sges = %d, rq_cq_id = %d\n", + qp->rq.max_sges, qp->rq_cq->icid); +} + +static inline void +qedr_init_qp_user_params(struct qed_rdma_create_qp_in_params *params, + struct qedr_create_qp_ureq *ureq) +{ + /* QP handle to be written in CQE */ + params->qp_handle_lo = ureq->qp_handle_lo; + params->qp_handle_hi = ureq->qp_handle_hi; +} + +static inline void +qedr_init_qp_kernel_doorbell_sq(struct qedr_dev *dev, struct qedr_qp *qp) +{ + qp->sq.db = dev->db_addr + + DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD); + qp->sq.db_data.data.icid = qp->icid + 1; +} + +static inline void +qedr_init_qp_kernel_doorbell_rq(struct qedr_dev *dev, struct qedr_qp *qp) +{ + qp->rq.db = dev->db_addr + + DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD); + qp->rq.db_data.data.icid = qp->icid; +} + +static inline int +qedr_init_qp_kernel_params_rq(struct qedr_dev *dev, + struct qedr_qp *qp, struct ib_qp_init_attr *attrs) +{ + /* Allocate driver internal RQ array */ + qp->rqe_wr_id = kcalloc(qp->rq.max_wr, sizeof(*qp->rqe_wr_id), + GFP_KERNEL); + if (!qp->rqe_wr_id) + return -ENOMEM; + + DP_DEBUG(dev, QEDR_MSG_QP, "RQ max_wr set to %d.\n", qp->rq.max_wr); + + return 0; +} + +static inline int +qedr_init_qp_kernel_params_sq(struct qedr_dev *dev, + struct qedr_qp *qp, + struct ib_qp_init_attr *attrs, + struct qed_rdma_create_qp_in_params *params) +{ + u32 temp_max_wr; + + /* Allocate driver internal SQ array */ + temp_max_wr = attrs->cap.max_send_wr * dev->wq_multiplier; + temp_max_wr = min_t(u32, temp_max_wr, dev->attr.max_sqe); + + /* temp_max_wr < attr->max_sqe < u16 so the casting is safe */ + qp->sq.max_wr = (u16)temp_max_wr; + qp->wqe_wr_id = kcalloc(qp->sq.max_wr, sizeof(*qp->wqe_wr_id), + GFP_KERNEL); + if (!qp->wqe_wr_id) + return -ENOMEM; + + DP_DEBUG(dev, QEDR_MSG_QP, "SQ max_wr set to %d.\n", qp->sq.max_wr); + + /* QP handle to be written in CQE */ + params->qp_handle_lo = lower_32_bits((uintptr_t)qp); + params->qp_handle_hi = upper_32_bits((uintptr_t)qp); + + return 0; +} + +static inline int qedr_init_qp_kernel_sq(struct qedr_dev *dev, + struct qedr_qp *qp, + struct ib_qp_init_attr *attrs) +{ + u32 n_sq_elems, n_sq_entries; + int rc; + + /* A single work request may take up to QEDR_MAX_SQ_WQE_SIZE elements in + * the ring. The ring should allow at least a single WR, even if the + * user requested none, due to allocation issues. + */ + n_sq_entries = attrs->cap.max_send_wr; + n_sq_entries = min_t(u32, n_sq_entries, dev->attr.max_sqe); + n_sq_entries = max_t(u32, n_sq_entries, 1); + n_sq_elems = n_sq_entries * QEDR_MAX_SQE_ELEMENTS_PER_SQE; + rc = dev->ops->common->chain_alloc(dev->cdev, + QED_CHAIN_USE_TO_PRODUCE, + QED_CHAIN_MODE_PBL, + QED_CHAIN_CNT_TYPE_U32, + n_sq_elems, + QEDR_SQE_ELEMENT_SIZE, + &qp->sq.pbl); + if (rc) { + DP_ERR(dev, "failed to allocate QP %p SQ\n", qp); + return rc; + } + + DP_DEBUG(dev, QEDR_MSG_SQ, + "SQ Pbl base addr = %llx max_send_wr=%d max_wr=%d capacity=%d, rc=%d\n", + qed_chain_get_pbl_phys(&qp->sq.pbl), attrs->cap.max_send_wr, + n_sq_entries, qed_chain_get_capacity(&qp->sq.pbl), rc); + return 0; +} + +static inline int qedr_init_qp_kernel_rq(struct qedr_dev *dev, + struct qedr_qp *qp, + struct ib_qp_init_attr *attrs) +{ + u32 n_rq_elems, n_rq_entries; + int rc; + + /* A single work request may take up to QEDR_MAX_RQ_WQE_SIZE elements in + * the ring. There ring should allow at least a single WR, even if the + * user requested none, due to allocation issues. + */ + n_rq_entries = max_t(u32, attrs->cap.max_recv_wr, 1); + n_rq_elems = n_rq_entries * QEDR_MAX_RQE_ELEMENTS_PER_RQE; + rc = dev->ops->common->chain_alloc(dev->cdev, + QED_CHAIN_USE_TO_CONSUME_PRODUCE, + QED_CHAIN_MODE_PBL, + QED_CHAIN_CNT_TYPE_U32, + n_rq_elems, + QEDR_RQE_ELEMENT_SIZE, + &qp->rq.pbl); + + if (rc) { + DP_ERR(dev, "failed to allocate memory for QP %p RQ\n", qp); + return -ENOMEM; + } + + DP_DEBUG(dev, QEDR_MSG_RQ, + "RQ Pbl base addr = %llx max_recv_wr=%d max_wr=%d capacity=%d, rc=%d\n", + qed_chain_get_pbl_phys(&qp->rq.pbl), attrs->cap.max_recv_wr, + n_rq_entries, qed_chain_get_capacity(&qp->rq.pbl), rc); + + /* n_rq_entries < u16 so the casting is safe */ + qp->rq.max_wr = (u16)n_rq_entries; + + return 0; +} + +static inline void +qedr_init_qp_in_params_sq(struct qedr_dev *dev, + struct qedr_pd *pd, + struct qedr_qp *qp, + struct ib_qp_init_attr *attrs, + struct ib_udata *udata, + struct qed_rdma_create_qp_in_params *params) +{ + /* QP handle to be written in an async event */ + params->qp_handle_async_lo = lower_32_bits((uintptr_t)qp); + params->qp_handle_async_hi = upper_32_bits((uintptr_t)qp); + + params->signal_all = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR); + params->fmr_and_reserved_lkey = !udata; + params->pd = pd->pd_id; + params->dpi = pd->uctx ? pd->uctx->dpi : dev->dpi; + params->sq_cq_id = get_qedr_cq(attrs->send_cq)->icid; + params->max_sq_sges = 0; + params->stats_queue = 0; + + if (udata) { + params->sq_num_pages = qp->usq.pbl_info.num_pbes; + params->sq_pbl_ptr = qp->usq.pbl_tbl->pa; + } else { + params->sq_num_pages = qed_chain_get_page_cnt(&qp->sq.pbl); + params->sq_pbl_ptr = qed_chain_get_pbl_phys(&qp->sq.pbl); + } +} + +static inline void +qedr_init_qp_in_params_rq(struct qedr_qp *qp, + struct ib_qp_init_attr *attrs, + struct ib_udata *udata, + struct qed_rdma_create_qp_in_params *params) +{ + params->rq_cq_id = get_qedr_cq(attrs->recv_cq)->icid; + params->srq_id = 0; + params->use_srq = false; + + if (udata) { + params->rq_num_pages = qp->urq.pbl_info.num_pbes; + params->rq_pbl_ptr = qp->urq.pbl_tbl->pa; + } else { + params->rq_num_pages = qed_chain_get_page_cnt(&qp->rq.pbl); + params->rq_pbl_ptr = qed_chain_get_pbl_phys(&qp->rq.pbl); + } +} + +static inline void qedr_qp_user_print(struct qedr_dev *dev, struct qedr_qp *qp) +{ + DP_DEBUG(dev, QEDR_MSG_QP, + "create qp: successfully created user QP. qp=%p, sq_addr=0x%llx, sq_len=%zd, rq_addr=0x%llx, rq_len=%zd\n", + qp, qp->usq.buf_addr, qp->usq.buf_len, qp->urq.buf_addr, + qp->urq.buf_len); +} + +static inline int qedr_init_user_qp(struct ib_ucontext *ib_ctx, + struct qedr_dev *dev, + struct qedr_qp *qp, + struct qedr_create_qp_ureq *ureq) +{ + int rc; + + /* SQ - read access only (0), dma sync not required (0) */ + rc = qedr_init_user_queue(ib_ctx, dev, &qp->usq, ureq->sq_addr, + ureq->sq_len, 0, 0); + if (rc) + return rc; + + /* RQ - read access only (0), dma sync not required (0) */ + rc = qedr_init_user_queue(ib_ctx, dev, &qp->urq, ureq->rq_addr, + ureq->rq_len, 0, 0); + + if (rc) + qedr_cleanup_user_sq(dev, qp); + return rc; +} + +static inline int +qedr_init_kernel_qp(struct qedr_dev *dev, + struct qedr_qp *qp, + struct ib_qp_init_attr *attrs, + struct qed_rdma_create_qp_in_params *params) +{ + int rc; + + rc = qedr_init_qp_kernel_sq(dev, qp, attrs); + if (rc) { + DP_ERR(dev, "failed to init kernel QP %p SQ\n", qp); + return rc; + } + + rc = qedr_init_qp_kernel_params_sq(dev, qp, attrs, params); + if (rc) { + dev->ops->common->chain_free(dev->cdev, &qp->sq.pbl); + DP_ERR(dev, "failed to init kernel QP %p SQ params\n", qp); + return rc; + } + + rc = qedr_init_qp_kernel_rq(dev, qp, attrs); + if (rc) { + qedr_cleanup_kernel_sq(dev, qp); + DP_ERR(dev, "failed to init kernel QP %p RQ\n", qp); + return rc; + } + + rc = qedr_init_qp_kernel_params_rq(dev, qp, attrs); + if (rc) { + DP_ERR(dev, "failed to init kernel QP %p RQ params\n", qp); + qedr_cleanup_kernel_sq(dev, qp); + dev->ops->common->chain_free(dev->cdev, &qp->rq.pbl); + return rc; + } + + return rc; +} + +struct ib_qp *qedr_create_qp(struct ib_pd *ibpd, + struct ib_qp_init_attr *attrs, + struct ib_udata *udata) +{ + struct qedr_dev *dev = get_qedr_dev(ibpd->device); + struct qed_rdma_create_qp_out_params out_params; + struct qed_rdma_create_qp_in_params in_params; + struct qedr_pd *pd = get_qedr_pd(ibpd); + struct ib_ucontext *ib_ctx = NULL; + struct qedr_ucontext *ctx = NULL; + struct qedr_create_qp_ureq ureq; + struct qedr_qp *qp; + int rc = 0; + + DP_DEBUG(dev, QEDR_MSG_QP, "create qp: called from %s, pd=%p\n", + udata ? "user library" : "kernel", pd); + + rc = qedr_check_qp_attrs(ibpd, dev, attrs); + if (rc) + return ERR_PTR(rc); + + qp = kzalloc(sizeof(*qp), GFP_KERNEL); + if (!qp) + return ERR_PTR(-ENOMEM); + + if (attrs->srq) + return ERR_PTR(-EINVAL); + + DP_DEBUG(dev, QEDR_MSG_QP, + "create qp: sq_cq=%p, sq_icid=%d, rq_cq=%p, rq_icid=%d\n", + get_qedr_cq(attrs->send_cq), + get_qedr_cq(attrs->send_cq)->icid, + get_qedr_cq(attrs->recv_cq), + get_qedr_cq(attrs->recv_cq)->icid); + + qedr_set_qp_init_params(dev, qp, pd, attrs); + + if (attrs->qp_type == IB_QPT_GSI) { + if (udata) { + DP_ERR(dev, + "create qp: unexpected udata when creating GSI QP\n"); + goto err0; + } + return qedr_create_gsi_qp(dev, attrs, qp); + } + + memset(&in_params, 0, sizeof(in_params)); + + if (udata) { + if (!(udata && ibpd->uobject && ibpd->uobject->context)) + goto err0; + + ib_ctx = ibpd->uobject->context; + ctx = get_qedr_ucontext(ib_ctx); + + memset(&ureq, 0, sizeof(ureq)); + if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) { + DP_ERR(dev, + "create qp: problem copying data from user space\n"); + goto err0; + } + + rc = qedr_init_user_qp(ib_ctx, dev, qp, &ureq); + if (rc) + goto err0; + + qedr_init_qp_user_params(&in_params, &ureq); + } else { + rc = qedr_init_kernel_qp(dev, qp, attrs, &in_params); + if (rc) + goto err0; + } + + qedr_init_qp_in_params_sq(dev, pd, qp, attrs, udata, &in_params); + qedr_init_qp_in_params_rq(qp, attrs, udata, &in_params); + + qp->qed_qp = dev->ops->rdma_create_qp(dev->rdma_ctx, + &in_params, &out_params); + + if (!qp->qed_qp) + goto err1; + + qp->qp_id = out_params.qp_id; + qp->icid = out_params.icid; + qp->ibqp.qp_num = qp->qp_id; + + if (udata) { + rc = qedr_copy_qp_uresp(dev, qp, udata); + if (rc) + goto err2; + + qedr_qp_user_print(dev, qp); + } else { + qedr_init_qp_kernel_doorbell_sq(dev, qp); + qedr_init_qp_kernel_doorbell_rq(dev, qp); + } + + DP_DEBUG(dev, QEDR_MSG_QP, "created %s space QP %p\n", + udata ? "user" : "kernel", qp); + + return &qp->ibqp; + +err2: + rc = dev->ops->rdma_destroy_qp(dev->rdma_ctx, qp->qed_qp); + if (rc) + DP_ERR(dev, "create qp: fatal fault. rc=%d", rc); +err1: + if (udata) { + qedr_cleanup_user_sq(dev, qp); + qedr_cleanup_user_rq(dev, qp); + } else { + qedr_cleanup_kernel_sq(dev, qp); + qedr_cleanup_kernel_rq(dev, qp); + } + +err0: + kfree(qp); + + return ERR_PTR(-EFAULT); +} + +enum ib_qp_state qedr_get_ibqp_state(enum qed_roce_qp_state qp_state) +{ + switch (qp_state) { + case QED_ROCE_QP_STATE_RESET: + return IB_QPS_RESET; + case QED_ROCE_QP_STATE_INIT: + return IB_QPS_INIT; + case QED_ROCE_QP_STATE_RTR: + return IB_QPS_RTR; + case QED_ROCE_QP_STATE_RTS: + return IB_QPS_RTS; + case QED_ROCE_QP_STATE_SQD: + return IB_QPS_SQD; + case QED_ROCE_QP_STATE_ERR: + return IB_QPS_ERR; + case QED_ROCE_QP_STATE_SQE: + return IB_QPS_SQE; + } + return IB_QPS_ERR; +} + +enum qed_roce_qp_state qedr_get_state_from_ibqp(enum ib_qp_state qp_state) +{ + switch (qp_state) { + case IB_QPS_RESET: + return QED_ROCE_QP_STATE_RESET; + case IB_QPS_INIT: + return QED_ROCE_QP_STATE_INIT; + case IB_QPS_RTR: + return QED_ROCE_QP_STATE_RTR; + case IB_QPS_RTS: + return QED_ROCE_QP_STATE_RTS; + case IB_QPS_SQD: + return QED_ROCE_QP_STATE_SQD; + case IB_QPS_ERR: + return QED_ROCE_QP_STATE_ERR; + default: + return QED_ROCE_QP_STATE_ERR; + } +} + +static void qedr_reset_qp_hwq_info(struct qedr_qp_hwq_info *qph) +{ + qed_chain_reset(&qph->pbl); + qph->prod = 0; + qph->cons = 0; + qph->wqe_cons = 0; + qph->db_data.data.value = cpu_to_le16(0); +} + +static int qedr_update_qp_state(struct qedr_dev *dev, + struct qedr_qp *qp, + enum qed_roce_qp_state new_state) +{ + int status = 0; + + if (new_state == qp->state) + return 1; + + switch (qp->state) { + case QED_ROCE_QP_STATE_RESET: + switch (new_state) { + case QED_ROCE_QP_STATE_INIT: + qp->prev_wqe_size = 0; + qedr_reset_qp_hwq_info(&qp->sq); + qedr_reset_qp_hwq_info(&qp->rq); + break; + default: + status = -EINVAL; + break; + }; + break; + case QED_ROCE_QP_STATE_INIT: + switch (new_state) { + case QED_ROCE_QP_STATE_RTR: + /* Update doorbell (in case post_recv was + * done before move to RTR) + */ + wmb(); + writel(qp->rq.db_data.raw, qp->rq.db); + /* Make sure write takes effect */ + mmiowb(); + break; + case QED_ROCE_QP_STATE_ERR: + break; + default: + /* Invalid state change. */ + status = -EINVAL; + break; + }; + break; + case QED_ROCE_QP_STATE_RTR: + /* RTR->XXX */ + switch (new_state) { + case QED_ROCE_QP_STATE_RTS: + break; + case QED_ROCE_QP_STATE_ERR: + break; + default: + /* Invalid state change. */ + status = -EINVAL; + break; + }; + break; + case QED_ROCE_QP_STATE_RTS: + /* RTS->XXX */ + switch (new_state) { + case QED_ROCE_QP_STATE_SQD: + break; + case QED_ROCE_QP_STATE_ERR: + break; + default: + /* Invalid state change. */ + status = -EINVAL; + break; + }; + break; + case QED_ROCE_QP_STATE_SQD: + /* SQD->XXX */ + switch (new_state) { + case QED_ROCE_QP_STATE_RTS: + case QED_ROCE_QP_STATE_ERR: + break; + default: + /* Invalid state change. */ + status = -EINVAL; + break; + }; + break; + case QED_ROCE_QP_STATE_ERR: + /* ERR->XXX */ + switch (new_state) { + case QED_ROCE_QP_STATE_RESET: + break; + default: + status = -EINVAL; + break; + }; + break; + default: + status = -EINVAL; + break; + }; + + return status; +} + +int qedr_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, + int attr_mask, struct ib_udata *udata) +{ + struct qedr_qp *qp = get_qedr_qp(ibqp); + struct qed_rdma_modify_qp_in_params qp_params = { 0 }; + struct qedr_dev *dev = get_qedr_dev(&qp->dev->ibdev); + enum ib_qp_state old_qp_state, new_qp_state; + int rc = 0; + + DP_DEBUG(dev, QEDR_MSG_QP, + "modify qp: qp %p attr_mask=0x%x, state=%d", qp, attr_mask, + attr->qp_state); + + old_qp_state = qedr_get_ibqp_state(qp->state); + if (attr_mask & IB_QP_STATE) + new_qp_state = attr->qp_state; + else + new_qp_state = old_qp_state; + + if (!ib_modify_qp_is_ok + (old_qp_state, new_qp_state, ibqp->qp_type, attr_mask, + IB_LINK_LAYER_ETHERNET)) { + DP_ERR(dev, + "modify qp: invalid attribute mask=0x%x specified for\n" + "qpn=0x%x of type=0x%x old_qp_state=0x%x, new_qp_state=0x%x\n", + attr_mask, qp->qp_id, ibqp->qp_type, old_qp_state, + new_qp_state); + rc = -EINVAL; + goto err; + } + + /* Translate the masks... */ + if (attr_mask & IB_QP_STATE) { + SET_FIELD(qp_params.modify_flags, + QED_RDMA_MODIFY_QP_VALID_NEW_STATE, 1); + qp_params.new_state = qedr_get_state_from_ibqp(attr->qp_state); + } + + if (attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY) + qp_params.sqd_async = true; + + if (attr_mask & IB_QP_PKEY_INDEX) { + SET_FIELD(qp_params.modify_flags, + QED_ROCE_MODIFY_QP_VALID_PKEY, 1); + if (attr->pkey_index >= QEDR_ROCE_PKEY_TABLE_LEN) { + rc = -EINVAL; + goto err; + } + + qp_params.pkey = QEDR_ROCE_PKEY_DEFAULT; + } + + if (attr_mask & IB_QP_QKEY) + qp->qkey = attr->qkey; + + if (attr_mask & IB_QP_ACCESS_FLAGS) { + SET_FIELD(qp_params.modify_flags, + QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN, 1); + qp_params.incoming_rdma_read_en = attr->qp_access_flags & + IB_ACCESS_REMOTE_READ; + qp_params.incoming_rdma_write_en = attr->qp_access_flags & + IB_ACCESS_REMOTE_WRITE; + qp_params.incoming_atomic_en = attr->qp_access_flags & + IB_ACCESS_REMOTE_ATOMIC; + } + + if (attr_mask & (IB_QP_AV | IB_QP_PATH_MTU)) { + if (attr_mask & IB_QP_PATH_MTU) { + if (attr->path_mtu < IB_MTU_256 || + attr->path_mtu > IB_MTU_4096) { + pr_err("error: Only MTU sizes of 256, 512, 1024, 2048 and 4096 are supported by RoCE\n"); + rc = -EINVAL; + goto err; + } + qp->mtu = min(ib_mtu_enum_to_int(attr->path_mtu), + ib_mtu_enum_to_int(iboe_get_mtu + (dev->ndev->mtu))); + } + + if (!qp->mtu) { + qp->mtu = + ib_mtu_enum_to_int(iboe_get_mtu(dev->ndev->mtu)); + pr_err("Fixing zeroed MTU to qp->mtu = %d\n", qp->mtu); + } + + SET_FIELD(qp_params.modify_flags, + QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR, 1); + + qp_params.traffic_class_tos = attr->ah_attr.grh.traffic_class; + qp_params.flow_label = attr->ah_attr.grh.flow_label; + qp_params.hop_limit_ttl = attr->ah_attr.grh.hop_limit; + + qp->sgid_idx = attr->ah_attr.grh.sgid_index; + + rc = get_gid_info_from_table(ibqp, attr, attr_mask, &qp_params); + if (rc) { + DP_ERR(dev, + "modify qp: problems with GID index %d (rc=%d)\n", + attr->ah_attr.grh.sgid_index, rc); + return rc; + } + + rc = qedr_get_dmac(dev, &attr->ah_attr, + qp_params.remote_mac_addr); + if (rc) + return rc; + + qp_params.use_local_mac = true; + ether_addr_copy(qp_params.local_mac_addr, dev->ndev->dev_addr); + + DP_DEBUG(dev, QEDR_MSG_QP, "dgid=%x:%x:%x:%x\n", + qp_params.dgid.dwords[0], qp_params.dgid.dwords[1], + qp_params.dgid.dwords[2], qp_params.dgid.dwords[3]); + DP_DEBUG(dev, QEDR_MSG_QP, "sgid=%x:%x:%x:%x\n", + qp_params.sgid.dwords[0], qp_params.sgid.dwords[1], + qp_params.sgid.dwords[2], qp_params.sgid.dwords[3]); + DP_DEBUG(dev, QEDR_MSG_QP, "remote_mac=[%pM]\n", + qp_params.remote_mac_addr); +; + + qp_params.mtu = qp->mtu; + qp_params.lb_indication = false; + } + + if (!qp_params.mtu) { + /* Stay with current MTU */ + if (qp->mtu) + qp_params.mtu = qp->mtu; + else + qp_params.mtu = + ib_mtu_enum_to_int(iboe_get_mtu(dev->ndev->mtu)); + } + + if (attr_mask & IB_QP_TIMEOUT) { + SET_FIELD(qp_params.modify_flags, + QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT, 1); + + qp_params.ack_timeout = attr->timeout; + if (attr->timeout) { + u32 temp; + + temp = 4096 * (1UL << attr->timeout) / 1000 / 1000; + /* FW requires [msec] */ + qp_params.ack_timeout = temp; + } else { + /* Infinite */ + qp_params.ack_timeout = 0; + } + } + if (attr_mask & IB_QP_RETRY_CNT) { + SET_FIELD(qp_params.modify_flags, + QED_ROCE_MODIFY_QP_VALID_RETRY_CNT, 1); + qp_params.retry_cnt = attr->retry_cnt; + } + + if (attr_mask & IB_QP_RNR_RETRY) { + SET_FIELD(qp_params.modify_flags, + QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT, 1); + qp_params.rnr_retry_cnt = attr->rnr_retry; + } + + if (attr_mask & IB_QP_RQ_PSN) { + SET_FIELD(qp_params.modify_flags, + QED_ROCE_MODIFY_QP_VALID_RQ_PSN, 1); + qp_params.rq_psn = attr->rq_psn; + qp->rq_psn = attr->rq_psn; + } + + if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) { + if (attr->max_rd_atomic > dev->attr.max_qp_req_rd_atomic_resc) { + rc = -EINVAL; + DP_ERR(dev, + "unsupported max_rd_atomic=%d, supported=%d\n", + attr->max_rd_atomic, + dev->attr.max_qp_req_rd_atomic_resc); + goto err; + } + + SET_FIELD(qp_params.modify_flags, + QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ, 1); + qp_params.max_rd_atomic_req = attr->max_rd_atomic; + } + + if (attr_mask & IB_QP_MIN_RNR_TIMER) { + SET_FIELD(qp_params.modify_flags, + QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER, 1); + qp_params.min_rnr_nak_timer = attr->min_rnr_timer; + } + + if (attr_mask & IB_QP_SQ_PSN) { + SET_FIELD(qp_params.modify_flags, + QED_ROCE_MODIFY_QP_VALID_SQ_PSN, 1); + qp_params.sq_psn = attr->sq_psn; + qp->sq_psn = attr->sq_psn; + } + + if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) { + if (attr->max_dest_rd_atomic > + dev->attr.max_qp_resp_rd_atomic_resc) { + DP_ERR(dev, + "unsupported max_dest_rd_atomic=%d, supported=%d\n", + attr->max_dest_rd_atomic, + dev->attr.max_qp_resp_rd_atomic_resc); + + rc = -EINVAL; + goto err; + } + + SET_FIELD(qp_params.modify_flags, + QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP, 1); + qp_params.max_rd_atomic_resp = attr->max_dest_rd_atomic; + } + + if (attr_mask & IB_QP_DEST_QPN) { + SET_FIELD(qp_params.modify_flags, + QED_ROCE_MODIFY_QP_VALID_DEST_QP, 1); + + qp_params.dest_qp = attr->dest_qp_num; + qp->dest_qp_num = attr->dest_qp_num; + } + + if (qp->qp_type != IB_QPT_GSI) + rc = dev->ops->rdma_modify_qp(dev->rdma_ctx, + qp->qed_qp, &qp_params); + + if (attr_mask & IB_QP_STATE) { + if ((qp->qp_type != IB_QPT_GSI) && (!udata)) + qedr_update_qp_state(dev, qp, qp_params.new_state); + qp->state = qp_params.new_state; + } + +err: + return rc; +} + +static int qedr_to_ib_qp_acc_flags(struct qed_rdma_query_qp_out_params *params) +{ + int ib_qp_acc_flags = 0; + + if (params->incoming_rdma_write_en) + ib_qp_acc_flags |= IB_ACCESS_REMOTE_WRITE; + if (params->incoming_rdma_read_en) + ib_qp_acc_flags |= IB_ACCESS_REMOTE_READ; + if (params->incoming_atomic_en) + ib_qp_acc_flags |= IB_ACCESS_REMOTE_ATOMIC; + ib_qp_acc_flags |= IB_ACCESS_LOCAL_WRITE; + return ib_qp_acc_flags; +} + +int qedr_query_qp(struct ib_qp *ibqp, + struct ib_qp_attr *qp_attr, + int attr_mask, struct ib_qp_init_attr *qp_init_attr) +{ + struct qed_rdma_query_qp_out_params params; + struct qedr_qp *qp = get_qedr_qp(ibqp); + struct qedr_dev *dev = qp->dev; + int rc = 0; + + memset(¶ms, 0, sizeof(params)); + + rc = dev->ops->rdma_query_qp(dev->rdma_ctx, qp->qed_qp, ¶ms); + if (rc) + goto err; + + memset(qp_attr, 0, sizeof(*qp_attr)); + memset(qp_init_attr, 0, sizeof(*qp_init_attr)); + + qp_attr->qp_state = qedr_get_ibqp_state(params.state); + qp_attr->cur_qp_state = qedr_get_ibqp_state(params.state); + qp_attr->path_mtu = iboe_get_mtu(params.mtu); + qp_attr->path_mig_state = IB_MIG_MIGRATED; + qp_attr->rq_psn = params.rq_psn; + qp_attr->sq_psn = params.sq_psn; + qp_attr->dest_qp_num = params.dest_qp; + + qp_attr->qp_access_flags = qedr_to_ib_qp_acc_flags(¶ms); + + qp_attr->cap.max_send_wr = qp->sq.max_wr; + qp_attr->cap.max_recv_wr = qp->rq.max_wr; + qp_attr->cap.max_send_sge = qp->sq.max_sges; + qp_attr->cap.max_recv_sge = qp->rq.max_sges; + qp_attr->cap.max_inline_data = qp->max_inline_data; + qp_init_attr->cap = qp_attr->cap; + + memcpy(&qp_attr->ah_attr.grh.dgid.raw[0], ¶ms.dgid.bytes[0], + sizeof(qp_attr->ah_attr.grh.dgid.raw)); + + qp_attr->ah_attr.grh.flow_label = params.flow_label; + qp_attr->ah_attr.grh.sgid_index = qp->sgid_idx; + qp_attr->ah_attr.grh.hop_limit = params.hop_limit_ttl; + qp_attr->ah_attr.grh.traffic_class = params.traffic_class_tos; + + qp_attr->ah_attr.ah_flags = IB_AH_GRH; + qp_attr->ah_attr.port_num = 1; + qp_attr->ah_attr.sl = 0; + qp_attr->timeout = params.timeout; + qp_attr->rnr_retry = params.rnr_retry; + qp_attr->retry_cnt = params.retry_cnt; + qp_attr->min_rnr_timer = params.min_rnr_nak_timer; + qp_attr->pkey_index = params.pkey_index; + qp_attr->port_num = 1; + qp_attr->ah_attr.src_path_bits = 0; + qp_attr->ah_attr.static_rate = 0; + qp_attr->alt_pkey_index = 0; + qp_attr->alt_port_num = 0; + qp_attr->alt_timeout = 0; + memset(&qp_attr->alt_ah_attr, 0, sizeof(qp_attr->alt_ah_attr)); + + qp_attr->sq_draining = (params.state == QED_ROCE_QP_STATE_SQD) ? 1 : 0; + qp_attr->max_dest_rd_atomic = params.max_dest_rd_atomic; + qp_attr->max_rd_atomic = params.max_rd_atomic; + qp_attr->en_sqd_async_notify = (params.sqd_async) ? 1 : 0; + + DP_DEBUG(dev, QEDR_MSG_QP, "QEDR_QUERY_QP: max_inline_data=%d\n", + qp_attr->cap.max_inline_data); + +err: + return rc; +} + +int qedr_destroy_qp(struct ib_qp *ibqp) +{ + struct qedr_qp *qp = get_qedr_qp(ibqp); + struct qedr_dev *dev = qp->dev; + struct ib_qp_attr attr; + int attr_mask = 0; + int rc = 0; + + DP_DEBUG(dev, QEDR_MSG_QP, "destroy qp: destroying %p, qp type=%d\n", + qp, qp->qp_type); + + if (qp->state != (QED_ROCE_QP_STATE_RESET | QED_ROCE_QP_STATE_ERR | + QED_ROCE_QP_STATE_INIT)) { + attr.qp_state = IB_QPS_ERR; + attr_mask |= IB_QP_STATE; + + /* Change the QP state to ERROR */ + qedr_modify_qp(ibqp, &attr, attr_mask, NULL); + } + + if (qp->qp_type != IB_QPT_GSI) { + rc = dev->ops->rdma_destroy_qp(dev->rdma_ctx, qp->qed_qp); + if (rc) + return rc; + } else { + qedr_destroy_gsi_qp(dev); + } + + if (ibqp->uobject && ibqp->uobject->context) { + qedr_cleanup_user_sq(dev, qp); + qedr_cleanup_user_rq(dev, qp); + } else { + qedr_cleanup_kernel_sq(dev, qp); + qedr_cleanup_kernel_rq(dev, qp); + } + + kfree(qp); + + return rc; +} + +struct ib_ah *qedr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr) +{ + struct qedr_ah *ah; + + ah = kzalloc(sizeof(*ah), GFP_ATOMIC); + if (!ah) + return ERR_PTR(-ENOMEM); + + ah->attr = *attr; + + return &ah->ibah; +} + +int qedr_destroy_ah(struct ib_ah *ibah) +{ + struct qedr_ah *ah = get_qedr_ah(ibah); + + kfree(ah); + return 0; +} + +static void free_mr_info(struct qedr_dev *dev, struct mr_info *info) +{ + struct qedr_pbl *pbl, *tmp; + + if (info->pbl_table) + list_add_tail(&info->pbl_table->list_entry, + &info->free_pbl_list); + + if (!list_empty(&info->inuse_pbl_list)) + list_splice(&info->inuse_pbl_list, &info->free_pbl_list); + + list_for_each_entry_safe(pbl, tmp, &info->free_pbl_list, list_entry) { + list_del(&pbl->list_entry); + qedr_free_pbl(dev, &info->pbl_info, pbl); + } +} + +static int init_mr_info(struct qedr_dev *dev, struct mr_info *info, + size_t page_list_len, bool two_layered) +{ + struct qedr_pbl *tmp; + int rc; + + INIT_LIST_HEAD(&info->free_pbl_list); + INIT_LIST_HEAD(&info->inuse_pbl_list); + + rc = qedr_prepare_pbl_tbl(dev, &info->pbl_info, + page_list_len, two_layered); + if (rc) + goto done; + + info->pbl_table = qedr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL); + if (!info->pbl_table) { + rc = -ENOMEM; + goto done; + } + + DP_DEBUG(dev, QEDR_MSG_MR, "pbl_table_pa = %pa\n", + &info->pbl_table->pa); + + /* in usual case we use 2 PBLs, so we add one to free + * list and allocating another one + */ + tmp = qedr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL); + if (!tmp) { + DP_DEBUG(dev, QEDR_MSG_MR, "Extra PBL is not allocated\n"); + goto done; + } + + list_add_tail(&tmp->list_entry, &info->free_pbl_list); + + DP_DEBUG(dev, QEDR_MSG_MR, "extra pbl_table_pa = %pa\n", &tmp->pa); + +done: + if (rc) + free_mr_info(dev, info); + + return rc; +} + +struct ib_mr *qedr_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len, + u64 usr_addr, int acc, struct ib_udata *udata) +{ + struct qedr_dev *dev = get_qedr_dev(ibpd->device); + struct qedr_mr *mr; + struct qedr_pd *pd; + int rc = -ENOMEM; + + pd = get_qedr_pd(ibpd); + DP_DEBUG(dev, QEDR_MSG_MR, + "qedr_register user mr pd = %d start = %lld, len = %lld, usr_addr = %lld, acc = %d\n", + pd->pd_id, start, len, usr_addr, acc); + + if (acc & IB_ACCESS_REMOTE_WRITE && !(acc & IB_ACCESS_LOCAL_WRITE)) + return ERR_PTR(-EINVAL); + + mr = kzalloc(sizeof(*mr), GFP_KERNEL); + if (!mr) + return ERR_PTR(rc); + + mr->type = QEDR_MR_USER; + + mr->umem = ib_umem_get(ibpd->uobject->context, start, len, acc, 0); + if (IS_ERR(mr->umem)) { + rc = -EFAULT; + goto err0; + } + + rc = init_mr_info(dev, &mr->info, ib_umem_page_count(mr->umem), 1); + if (rc) + goto err1; + + qedr_populate_pbls(dev, mr->umem, mr->info.pbl_table, + &mr->info.pbl_info); + + rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid); + if (rc) { + DP_ERR(dev, "roce alloc tid returned an error %d\n", rc); + goto err1; + } + + /* Index only, 18 bit long, lkey = itid << 8 | key */ + mr->hw_mr.tid_type = QED_RDMA_TID_REGISTERED_MR; + mr->hw_mr.key = 0; + mr->hw_mr.pd = pd->pd_id; + mr->hw_mr.local_read = 1; + mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0; + mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0; + mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0; + mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0; + mr->hw_mr.mw_bind = false; + mr->hw_mr.pbl_ptr = mr->info.pbl_table[0].pa; + mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered; + mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size); + mr->hw_mr.page_size_log = ilog2(mr->umem->page_size); + mr->hw_mr.fbo = ib_umem_offset(mr->umem); + mr->hw_mr.length = len; + mr->hw_mr.vaddr = usr_addr; + mr->hw_mr.zbva = false; + mr->hw_mr.phy_mr = false; + mr->hw_mr.dma_mr = false; + + rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr); + if (rc) { + DP_ERR(dev, "roce register tid returned an error %d\n", rc); + goto err2; + } + + mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key; + if (mr->hw_mr.remote_write || mr->hw_mr.remote_read || + mr->hw_mr.remote_atomic) + mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key; + + DP_DEBUG(dev, QEDR_MSG_MR, "register user mr lkey: %x\n", + mr->ibmr.lkey); + return &mr->ibmr; + +err2: + dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid); +err1: + qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table); +err0: + kfree(mr); + return ERR_PTR(rc); +} + +int qedr_dereg_mr(struct ib_mr *ib_mr) +{ + struct qedr_mr *mr = get_qedr_mr(ib_mr); + struct qedr_dev *dev = get_qedr_dev(ib_mr->device); + int rc = 0; + + rc = dev->ops->rdma_deregister_tid(dev->rdma_ctx, mr->hw_mr.itid); + if (rc) + return rc; + + dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid); + + if ((mr->type != QEDR_MR_DMA) && (mr->type != QEDR_MR_FRMR)) + qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table); + + /* it could be user registered memory. */ + if (mr->umem) + ib_umem_release(mr->umem); + + kfree(mr); + + return rc; +} + +struct qedr_mr *__qedr_alloc_mr(struct ib_pd *ibpd, int max_page_list_len) +{ + struct qedr_pd *pd = get_qedr_pd(ibpd); + struct qedr_dev *dev = get_qedr_dev(ibpd->device); + struct qedr_mr *mr; + int rc = -ENOMEM; + + DP_DEBUG(dev, QEDR_MSG_MR, + "qedr_alloc_frmr pd = %d max_page_list_len= %d\n", pd->pd_id, + max_page_list_len); + + mr = kzalloc(sizeof(*mr), GFP_KERNEL); + if (!mr) + return ERR_PTR(rc); + + mr->dev = dev; + mr->type = QEDR_MR_FRMR; + + rc = init_mr_info(dev, &mr->info, max_page_list_len, 1); + if (rc) + goto err0; + + rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid); + if (rc) { + DP_ERR(dev, "roce alloc tid returned an error %d\n", rc); + goto err0; + } + + /* Index only, 18 bit long, lkey = itid << 8 | key */ + mr->hw_mr.tid_type = QED_RDMA_TID_FMR; + mr->hw_mr.key = 0; + mr->hw_mr.pd = pd->pd_id; + mr->hw_mr.local_read = 1; + mr->hw_mr.local_write = 0; + mr->hw_mr.remote_read = 0; + mr->hw_mr.remote_write = 0; + mr->hw_mr.remote_atomic = 0; + mr->hw_mr.mw_bind = false; + mr->hw_mr.pbl_ptr = 0; + mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered; + mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size); + mr->hw_mr.fbo = 0; + mr->hw_mr.length = 0; + mr->hw_mr.vaddr = 0; + mr->hw_mr.zbva = false; + mr->hw_mr.phy_mr = true; + mr->hw_mr.dma_mr = false; + + rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr); + if (rc) { + DP_ERR(dev, "roce register tid returned an error %d\n", rc); + goto err1; + } + + mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key; + mr->ibmr.rkey = mr->ibmr.lkey; + + DP_DEBUG(dev, QEDR_MSG_MR, "alloc frmr: %x\n", mr->ibmr.lkey); + return mr; + +err1: + dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid); +err0: + kfree(mr); + return ERR_PTR(rc); +} + +struct ib_mr *qedr_alloc_mr(struct ib_pd *ibpd, + enum ib_mr_type mr_type, u32 max_num_sg) +{ + struct qedr_dev *dev; + struct qedr_mr *mr; + + if (mr_type != IB_MR_TYPE_MEM_REG) + return ERR_PTR(-EINVAL); + + mr = __qedr_alloc_mr(ibpd, max_num_sg); + + if (IS_ERR(mr)) + return ERR_PTR(-EINVAL); + + dev = mr->dev; + + return &mr->ibmr; +} + +static int qedr_set_page(struct ib_mr *ibmr, u64 addr) +{ + struct qedr_mr *mr = get_qedr_mr(ibmr); + struct qedr_pbl *pbl_table; + struct regpair *pbe; + u32 pbes_in_page; + + if (unlikely(mr->npages == mr->info.pbl_info.num_pbes)) { + DP_ERR(mr->dev, "qedr_set_page failes when %d\n", mr->npages); + return -ENOMEM; + } + + DP_DEBUG(mr->dev, QEDR_MSG_MR, "qedr_set_page pages[%d] = 0x%llx\n", + mr->npages, addr); + + pbes_in_page = mr->info.pbl_info.pbl_size / sizeof(u64); + pbl_table = mr->info.pbl_table + (mr->npages / pbes_in_page); + pbe = (struct regpair *)pbl_table->va; + pbe += mr->npages % pbes_in_page; + pbe->lo = cpu_to_le32((u32)addr); + pbe->hi = cpu_to_le32((u32)upper_32_bits(addr)); + + mr->npages++; + + return 0; +} + +static void handle_completed_mrs(struct qedr_dev *dev, struct mr_info *info) +{ + int work = info->completed - info->completed_handled - 1; + + DP_DEBUG(dev, QEDR_MSG_MR, "Special FMR work = %d\n", work); + while (work-- > 0 && !list_empty(&info->inuse_pbl_list)) { + struct qedr_pbl *pbl; + + /* Free all the page list that are possible to be freed + * (all the ones that were invalidated), under the assumption + * that if an FMR was completed successfully that means that + * if there was an invalidate operation before it also ended + */ + pbl = list_first_entry(&info->inuse_pbl_list, + struct qedr_pbl, list_entry); + list_del(&pbl->list_entry); + list_add_tail(&pbl->list_entry, &info->free_pbl_list); + info->completed_handled++; + } +} + +int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, + int sg_nents, unsigned int *sg_offset) +{ + struct qedr_mr *mr = get_qedr_mr(ibmr); + + mr->npages = 0; + + handle_completed_mrs(mr->dev, &mr->info); + return ib_sg_to_pages(ibmr, sg, sg_nents, NULL, qedr_set_page); +} + +struct ib_mr *qedr_get_dma_mr(struct ib_pd *ibpd, int acc) +{ + struct qedr_dev *dev = get_qedr_dev(ibpd->device); + struct qedr_pd *pd = get_qedr_pd(ibpd); + struct qedr_mr *mr; + int rc; + + mr = kzalloc(sizeof(*mr), GFP_KERNEL); + if (!mr) + return ERR_PTR(-ENOMEM); + + mr->type = QEDR_MR_DMA; + + rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid); + if (rc) { + DP_ERR(dev, "roce alloc tid returned an error %d\n", rc); + goto err1; + } + + /* index only, 18 bit long, lkey = itid << 8 | key */ + mr->hw_mr.tid_type = QED_RDMA_TID_REGISTERED_MR; + mr->hw_mr.pd = pd->pd_id; + mr->hw_mr.local_read = 1; + mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0; + mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0; + mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0; + mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0; + mr->hw_mr.dma_mr = true; + + rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr); + if (rc) { + DP_ERR(dev, "roce register tid returned an error %d\n", rc); + goto err2; + } + + mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key; + if (mr->hw_mr.remote_write || mr->hw_mr.remote_read || + mr->hw_mr.remote_atomic) + mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key; + + DP_DEBUG(dev, QEDR_MSG_MR, "get dma mr: lkey = %x\n", mr->ibmr.lkey); + return &mr->ibmr; + +err2: + dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid); +err1: + kfree(mr); + return ERR_PTR(rc); +} + +static inline int qedr_wq_is_full(struct qedr_qp_hwq_info *wq) +{ + return (((wq->prod + 1) % wq->max_wr) == wq->cons); +} + +static int sge_data_len(struct ib_sge *sg_list, int num_sge) +{ + int i, len = 0; + + for (i = 0; i < num_sge; i++) + len += sg_list[i].length; + + return len; +} + +static void swap_wqe_data64(u64 *p) +{ + int i; + + for (i = 0; i < QEDR_SQE_ELEMENT_SIZE / sizeof(u64); i++, p++) + *p = cpu_to_be64(cpu_to_le64(*p)); +} + +static u32 qedr_prepare_sq_inline_data(struct qedr_dev *dev, + struct qedr_qp *qp, u8 *wqe_size, + struct ib_send_wr *wr, + struct ib_send_wr **bad_wr, u8 *bits, + u8 bit) +{ + u32 data_size = sge_data_len(wr->sg_list, wr->num_sge); + char *seg_prt, *wqe; + int i, seg_siz; + + if (data_size > ROCE_REQ_MAX_INLINE_DATA_SIZE) { + DP_ERR(dev, "Too much inline data in WR: %d\n", data_size); + *bad_wr = wr; + return 0; + } + + if (!data_size) + return data_size; + + *bits |= bit; + + seg_prt = NULL; + wqe = NULL; + seg_siz = 0; + + /* Copy data inline */ + for (i = 0; i < wr->num_sge; i++) { + u32 len = wr->sg_list[i].length; + void *src = (void *)(uintptr_t)wr->sg_list[i].addr; + + while (len > 0) { + u32 cur; + + /* New segment required */ + if (!seg_siz) { + wqe = (char *)qed_chain_produce(&qp->sq.pbl); + seg_prt = wqe; + seg_siz = sizeof(struct rdma_sq_common_wqe); + (*wqe_size)++; + } + + /* Calculate currently allowed length */ + cur = min_t(u32, len, seg_siz); + memcpy(seg_prt, src, cur); + + /* Update segment variables */ + seg_prt += cur; + seg_siz -= cur; + + /* Update sge variables */ + src += cur; + len -= cur; + + /* Swap fully-completed segments */ + if (!seg_siz) + swap_wqe_data64((u64 *)wqe); + } + } + + /* swap last not completed segment */ + if (seg_siz) + swap_wqe_data64((u64 *)wqe); + + return data_size; +} + +#define RQ_SGE_SET(sge, vaddr, vlength, vflags) \ + do { \ + DMA_REGPAIR_LE(sge->addr, vaddr); \ + (sge)->length = cpu_to_le32(vlength); \ + (sge)->flags = cpu_to_le32(vflags); \ + } while (0) + +#define SRQ_HDR_SET(hdr, vwr_id, num_sge) \ + do { \ + DMA_REGPAIR_LE(hdr->wr_id, vwr_id); \ + (hdr)->num_sges = num_sge; \ + } while (0) + +#define SRQ_SGE_SET(sge, vaddr, vlength, vlkey) \ + do { \ + DMA_REGPAIR_LE(sge->addr, vaddr); \ + (sge)->length = cpu_to_le32(vlength); \ + (sge)->l_key = cpu_to_le32(vlkey); \ + } while (0) + +static u32 qedr_prepare_sq_sges(struct qedr_qp *qp, u8 *wqe_size, + struct ib_send_wr *wr) +{ + u32 data_size = 0; + int i; + + for (i = 0; i < wr->num_sge; i++) { + struct rdma_sq_sge *sge = qed_chain_produce(&qp->sq.pbl); + + DMA_REGPAIR_LE(sge->addr, wr->sg_list[i].addr); + sge->l_key = cpu_to_le32(wr->sg_list[i].lkey); + sge->length = cpu_to_le32(wr->sg_list[i].length); + data_size += wr->sg_list[i].length; + } + + if (wqe_size) + *wqe_size += wr->num_sge; + + return data_size; +} + +static u32 qedr_prepare_sq_rdma_data(struct qedr_dev *dev, + struct qedr_qp *qp, + struct rdma_sq_rdma_wqe_1st *rwqe, + struct rdma_sq_rdma_wqe_2nd *rwqe2, + struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + rwqe2->r_key = cpu_to_le32(rdma_wr(wr)->rkey); + DMA_REGPAIR_LE(rwqe2->remote_va, rdma_wr(wr)->remote_addr); + + if (wr->send_flags & IB_SEND_INLINE) { + u8 flags = 0; + + SET_FIELD2(flags, RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG, 1); + return qedr_prepare_sq_inline_data(dev, qp, &rwqe->wqe_size, wr, + bad_wr, &rwqe->flags, flags); + } + + return qedr_prepare_sq_sges(qp, &rwqe->wqe_size, wr); +} + +static u32 qedr_prepare_sq_send_data(struct qedr_dev *dev, + struct qedr_qp *qp, + struct rdma_sq_send_wqe_1st *swqe, + struct rdma_sq_send_wqe_2st *swqe2, + struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + memset(swqe2, 0, sizeof(*swqe2)); + if (wr->send_flags & IB_SEND_INLINE) { + u8 flags = 0; + + SET_FIELD2(flags, RDMA_SQ_SEND_WQE_INLINE_FLG, 1); + return qedr_prepare_sq_inline_data(dev, qp, &swqe->wqe_size, wr, + bad_wr, &swqe->flags, flags); + } + + return qedr_prepare_sq_sges(qp, &swqe->wqe_size, wr); +} + +static int qedr_prepare_reg(struct qedr_qp *qp, + struct rdma_sq_fmr_wqe_1st *fwqe1, + struct ib_reg_wr *wr) +{ + struct qedr_mr *mr = get_qedr_mr(wr->mr); + struct rdma_sq_fmr_wqe_2nd *fwqe2; + + fwqe2 = (struct rdma_sq_fmr_wqe_2nd *)qed_chain_produce(&qp->sq.pbl); + fwqe1->addr.hi = upper_32_bits(mr->ibmr.iova); + fwqe1->addr.lo = lower_32_bits(mr->ibmr.iova); + fwqe1->l_key = wr->key; + + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_READ, + !!(wr->access & IB_ACCESS_REMOTE_READ)); + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE, + !!(wr->access & IB_ACCESS_REMOTE_WRITE)); + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC, + !!(wr->access & IB_ACCESS_REMOTE_ATOMIC)); + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_READ, 1); + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE, + !!(wr->access & IB_ACCESS_LOCAL_WRITE)); + fwqe2->fmr_ctrl = 0; + + SET_FIELD2(fwqe2->fmr_ctrl, RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG, + ilog2(mr->ibmr.page_size) - 12); + + fwqe2->length_hi = 0; + fwqe2->length_lo = mr->ibmr.length; + fwqe2->pbl_addr.hi = upper_32_bits(mr->info.pbl_table->pa); + fwqe2->pbl_addr.lo = lower_32_bits(mr->info.pbl_table->pa); + + qp->wqe_wr_id[qp->sq.prod].mr = mr; + + return 0; +} + +enum ib_wc_opcode qedr_ib_to_wc_opcode(enum ib_wr_opcode opcode) +{ + switch (opcode) { + case IB_WR_RDMA_WRITE: + case IB_WR_RDMA_WRITE_WITH_IMM: + return IB_WC_RDMA_WRITE; + case IB_WR_SEND_WITH_IMM: + case IB_WR_SEND: + case IB_WR_SEND_WITH_INV: + return IB_WC_SEND; + case IB_WR_RDMA_READ: + return IB_WC_RDMA_READ; + case IB_WR_ATOMIC_CMP_AND_SWP: + return IB_WC_COMP_SWAP; + case IB_WR_ATOMIC_FETCH_AND_ADD: + return IB_WC_FETCH_ADD; + case IB_WR_REG_MR: + return IB_WC_REG_MR; + case IB_WR_LOCAL_INV: + return IB_WC_LOCAL_INV; + default: + return IB_WC_SEND; + } +} + +inline bool qedr_can_post_send(struct qedr_qp *qp, struct ib_send_wr *wr) +{ + int wq_is_full, err_wr, pbl_is_full; + struct qedr_dev *dev = qp->dev; + + /* prevent SQ overflow and/or processing of a bad WR */ + err_wr = wr->num_sge > qp->sq.max_sges; + wq_is_full = qedr_wq_is_full(&qp->sq); + pbl_is_full = qed_chain_get_elem_left_u32(&qp->sq.pbl) < + QEDR_MAX_SQE_ELEMENTS_PER_SQE; + if (wq_is_full || err_wr || pbl_is_full) { + if (wq_is_full && !(qp->err_bitmap & QEDR_QP_ERR_SQ_FULL)) { + DP_ERR(dev, + "error: WQ is full. Post send on QP %p failed (this error appears only once)\n", + qp); + qp->err_bitmap |= QEDR_QP_ERR_SQ_FULL; + } + + if (err_wr && !(qp->err_bitmap & QEDR_QP_ERR_BAD_SR)) { + DP_ERR(dev, + "error: WR is bad. Post send on QP %p failed (this error appears only once)\n", + qp); + qp->err_bitmap |= QEDR_QP_ERR_BAD_SR; + } + + if (pbl_is_full && + !(qp->err_bitmap & QEDR_QP_ERR_SQ_PBL_FULL)) { + DP_ERR(dev, + "error: WQ PBL is full. Post send on QP %p failed (this error appears only once)\n", + qp); + qp->err_bitmap |= QEDR_QP_ERR_SQ_PBL_FULL; + } + return false; + } + return true; +} + +int __qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + struct qedr_dev *dev = get_qedr_dev(ibqp->device); + struct qedr_qp *qp = get_qedr_qp(ibqp); + struct rdma_sq_atomic_wqe_1st *awqe1; + struct rdma_sq_atomic_wqe_2nd *awqe2; + struct rdma_sq_atomic_wqe_3rd *awqe3; + struct rdma_sq_send_wqe_2st *swqe2; + struct rdma_sq_local_inv_wqe *iwqe; + struct rdma_sq_rdma_wqe_2nd *rwqe2; + struct rdma_sq_send_wqe_1st *swqe; + struct rdma_sq_rdma_wqe_1st *rwqe; + struct rdma_sq_fmr_wqe_1st *fwqe1; + struct rdma_sq_common_wqe *wqe; + u32 length; + int rc = 0; + bool comp; + + if (!qedr_can_post_send(qp, wr)) { + *bad_wr = wr; + return -ENOMEM; + } + + wqe = qed_chain_produce(&qp->sq.pbl); + qp->wqe_wr_id[qp->sq.prod].signaled = + !!(wr->send_flags & IB_SEND_SIGNALED) || qp->signaled; + + wqe->flags = 0; + SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_SE_FLG, + !!(wr->send_flags & IB_SEND_SOLICITED)); + comp = (!!(wr->send_flags & IB_SEND_SIGNALED)) || qp->signaled; + SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_COMP_FLG, comp); + SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_RD_FENCE_FLG, + !!(wr->send_flags & IB_SEND_FENCE)); + wqe->prev_wqe_size = qp->prev_wqe_size; + + qp->wqe_wr_id[qp->sq.prod].opcode = qedr_ib_to_wc_opcode(wr->opcode); + + switch (wr->opcode) { + case IB_WR_SEND_WITH_IMM: + wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_IMM; + swqe = (struct rdma_sq_send_wqe_1st *)wqe; + swqe->wqe_size = 2; + swqe2 = qed_chain_produce(&qp->sq.pbl); + + swqe->inv_key_or_imm_data = cpu_to_le32(wr->ex.imm_data); + length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2, + wr, bad_wr); + swqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size; + qp->prev_wqe_size = swqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length; + break; + case IB_WR_SEND: + wqe->req_type = RDMA_SQ_REQ_TYPE_SEND; + swqe = (struct rdma_sq_send_wqe_1st *)wqe; + + swqe->wqe_size = 2; + swqe2 = qed_chain_produce(&qp->sq.pbl); + length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2, + wr, bad_wr); + swqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size; + qp->prev_wqe_size = swqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length; + break; + case IB_WR_SEND_WITH_INV: + wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_INVALIDATE; + swqe = (struct rdma_sq_send_wqe_1st *)wqe; + swqe2 = qed_chain_produce(&qp->sq.pbl); + swqe->wqe_size = 2; + swqe->inv_key_or_imm_data = cpu_to_le32(wr->ex.invalidate_rkey); + length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2, + wr, bad_wr); + swqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size; + qp->prev_wqe_size = swqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length; + break; + + case IB_WR_RDMA_WRITE_WITH_IMM: + wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR_WITH_IMM; + rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe; + + rwqe->wqe_size = 2; + rwqe->imm_data = htonl(cpu_to_le32(wr->ex.imm_data)); + rwqe2 = qed_chain_produce(&qp->sq.pbl); + length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2, + wr, bad_wr); + rwqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size; + qp->prev_wqe_size = rwqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length; + break; + case IB_WR_RDMA_WRITE: + wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR; + rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe; + + rwqe->wqe_size = 2; + rwqe2 = qed_chain_produce(&qp->sq.pbl); + length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2, + wr, bad_wr); + rwqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size; + qp->prev_wqe_size = rwqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length; + break; + case IB_WR_RDMA_READ_WITH_INV: + DP_ERR(dev, + "RDMA READ WITH INVALIDATE not supported\n"); + *bad_wr = wr; + rc = -EINVAL; + break; + + case IB_WR_RDMA_READ: + wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_RD; + rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe; + + rwqe->wqe_size = 2; + rwqe2 = qed_chain_produce(&qp->sq.pbl); + length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2, + wr, bad_wr); + rwqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size; + qp->prev_wqe_size = rwqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length; + break; + + case IB_WR_ATOMIC_CMP_AND_SWP: + case IB_WR_ATOMIC_FETCH_AND_ADD: + awqe1 = (struct rdma_sq_atomic_wqe_1st *)wqe; + awqe1->wqe_size = 4; + + awqe2 = qed_chain_produce(&qp->sq.pbl); + DMA_REGPAIR_LE(awqe2->remote_va, atomic_wr(wr)->remote_addr); + awqe2->r_key = cpu_to_le32(atomic_wr(wr)->rkey); + + awqe3 = qed_chain_produce(&qp->sq.pbl); + + if (wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) { + wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_ADD; + DMA_REGPAIR_LE(awqe3->swap_data, + atomic_wr(wr)->compare_add); + } else { + wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_CMP_AND_SWAP; + DMA_REGPAIR_LE(awqe3->swap_data, + atomic_wr(wr)->swap); + DMA_REGPAIR_LE(awqe3->cmp_data, + atomic_wr(wr)->compare_add); + } + + qedr_prepare_sq_sges(qp, NULL, wr); + + qp->wqe_wr_id[qp->sq.prod].wqe_size = awqe1->wqe_size; + qp->prev_wqe_size = awqe1->wqe_size; + break; + + case IB_WR_LOCAL_INV: + iwqe = (struct rdma_sq_local_inv_wqe *)wqe; + iwqe->wqe_size = 1; + + iwqe->req_type = RDMA_SQ_REQ_TYPE_LOCAL_INVALIDATE; + iwqe->inv_l_key = wr->ex.invalidate_rkey; + qp->wqe_wr_id[qp->sq.prod].wqe_size = iwqe->wqe_size; + qp->prev_wqe_size = iwqe->wqe_size; + break; + case IB_WR_REG_MR: + DP_DEBUG(dev, QEDR_MSG_CQ, "REG_MR\n"); + wqe->req_type = RDMA_SQ_REQ_TYPE_FAST_MR; + fwqe1 = (struct rdma_sq_fmr_wqe_1st *)wqe; + fwqe1->wqe_size = 2; + + rc = qedr_prepare_reg(qp, fwqe1, reg_wr(wr)); + if (rc) { + DP_ERR(dev, "IB_REG_MR failed rc=%d\n", rc); + *bad_wr = wr; + break; + } + + qp->wqe_wr_id[qp->sq.prod].wqe_size = fwqe1->wqe_size; + qp->prev_wqe_size = fwqe1->wqe_size; + break; + default: + DP_ERR(dev, "invalid opcode 0x%x!\n", wr->opcode); + rc = -EINVAL; + *bad_wr = wr; + break; + } + + if (*bad_wr) { + u16 value; + + /* Restore prod to its position before + * this WR was processed + */ + value = le16_to_cpu(qp->sq.db_data.data.value); + qed_chain_set_prod(&qp->sq.pbl, value, wqe); + + /* Restore prev_wqe_size */ + qp->prev_wqe_size = wqe->prev_wqe_size; + rc = -EINVAL; + DP_ERR(dev, "POST SEND FAILED\n"); + } + + return rc; +} + +int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + struct qedr_dev *dev = get_qedr_dev(ibqp->device); + struct qedr_qp *qp = get_qedr_qp(ibqp); + unsigned long flags; + int rc = 0; + + *bad_wr = NULL; + + if (qp->qp_type == IB_QPT_GSI) + return qedr_gsi_post_send(ibqp, wr, bad_wr); + + spin_lock_irqsave(&qp->q_lock, flags); + + if ((qp->state == QED_ROCE_QP_STATE_RESET) || + (qp->state == QED_ROCE_QP_STATE_ERR)) { + spin_unlock_irqrestore(&qp->q_lock, flags); + *bad_wr = wr; + DP_DEBUG(dev, QEDR_MSG_CQ, + "QP in wrong state! QP icid=0x%x state %d\n", + qp->icid, qp->state); + return -EINVAL; + } + + if (!wr) { + DP_ERR(dev, "Got an empty post send.\n"); + return -EINVAL; + } + + while (wr) { + rc = __qedr_post_send(ibqp, wr, bad_wr); + if (rc) + break; + + qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id; + + qedr_inc_sw_prod(&qp->sq); + + qp->sq.db_data.data.value++; + + wr = wr->next; + } + + /* Trigger doorbell + * If there was a failure in the first WR then it will be triggered in + * vane. However this is not harmful (as long as the producer value is + * unchanged). For performance reasons we avoid checking for this + * redundant doorbell. + */ + wmb(); + writel(qp->sq.db_data.raw, qp->sq.db); + + /* Make sure write sticks */ + mmiowb(); + + spin_unlock_irqrestore(&qp->q_lock, flags); + + return rc; +} + +int qedr_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, + struct ib_recv_wr **bad_wr) +{ + struct qedr_qp *qp = get_qedr_qp(ibqp); + struct qedr_dev *dev = qp->dev; + unsigned long flags; + int status = 0; + + if (qp->qp_type == IB_QPT_GSI) + return qedr_gsi_post_recv(ibqp, wr, bad_wr); + + spin_lock_irqsave(&qp->q_lock, flags); + + if ((qp->state == QED_ROCE_QP_STATE_RESET) || + (qp->state == QED_ROCE_QP_STATE_ERR)) { + spin_unlock_irqrestore(&qp->q_lock, flags); + *bad_wr = wr; + return -EINVAL; + } + + while (wr) { + int i; + + if (qed_chain_get_elem_left_u32(&qp->rq.pbl) < + QEDR_MAX_RQE_ELEMENTS_PER_RQE || + wr->num_sge > qp->rq.max_sges) { + DP_ERR(dev, "Can't post WR (%d < %d) || (%d > %d)\n", + qed_chain_get_elem_left_u32(&qp->rq.pbl), + QEDR_MAX_RQE_ELEMENTS_PER_RQE, wr->num_sge, + qp->rq.max_sges); + status = -ENOMEM; + *bad_wr = wr; + break; + } + for (i = 0; i < wr->num_sge; i++) { + u32 flags = 0; + struct rdma_rq_sge *rqe = + qed_chain_produce(&qp->rq.pbl); + + /* First one must include the number + * of SGE in the list + */ + if (!i) + SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES, + wr->num_sge); + + SET_FIELD(flags, RDMA_RQ_SGE_L_KEY, + wr->sg_list[i].lkey); + + RQ_SGE_SET(rqe, wr->sg_list[i].addr, + wr->sg_list[i].length, flags); + } + + /* Special case of no sges. FW requires between 1-4 sges... + * in this case we need to post 1 sge with length zero. this is + * because rdma write with immediate consumes an RQ. + */ + if (!wr->num_sge) { + u32 flags = 0; + struct rdma_rq_sge *rqe = + qed_chain_produce(&qp->rq.pbl); + + /* First one must include the number + * of SGE in the list + */ + SET_FIELD(flags, RDMA_RQ_SGE_L_KEY, 0); + SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES, 1); + + RQ_SGE_SET(rqe, 0, 0, flags); + i = 1; + } + + qp->rqe_wr_id[qp->rq.prod].wr_id = wr->wr_id; + qp->rqe_wr_id[qp->rq.prod].wqe_size = i; + + qedr_inc_sw_prod(&qp->rq); + + /* Flush all the writes before signalling doorbell */ + wmb(); + + qp->rq.db_data.data.value++; + + writel(qp->rq.db_data.raw, qp->rq.db); + + /* Make sure write sticks */ + mmiowb(); + + wr = wr->next; + } + + spin_unlock_irqrestore(&qp->q_lock, flags); + + return status; +} + +static int is_valid_cqe(struct qedr_cq *cq, union rdma_cqe *cqe) +{ + struct rdma_cqe_requester *resp_cqe = &cqe->req; + + return (resp_cqe->flags & RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK) == + cq->pbl_toggle; +} + +static struct qedr_qp *cqe_get_qp(union rdma_cqe *cqe) +{ + struct rdma_cqe_requester *resp_cqe = &cqe->req; + struct qedr_qp *qp; + + qp = (struct qedr_qp *)(uintptr_t)HILO_GEN(resp_cqe->qp_handle.hi, + resp_cqe->qp_handle.lo, + u64); + return qp; +} + +static enum rdma_cqe_type cqe_get_type(union rdma_cqe *cqe) +{ + struct rdma_cqe_requester *resp_cqe = &cqe->req; + + return GET_FIELD(resp_cqe->flags, RDMA_CQE_REQUESTER_TYPE); +} + +/* Return latest CQE (needs processing) */ +static union rdma_cqe *get_cqe(struct qedr_cq *cq) +{ + return cq->latest_cqe; +} + +/* In fmr we need to increase the number of fmr completed counter for the fmr + * algorithm determining whether we can free a pbl or not. + * we need to perform this whether the work request was signaled or not. for + * this purpose we call this function from the condition that checks if a wr + * should be skipped, to make sure we don't miss it ( possibly this fmr + * operation was not signalted) + */ +static inline void qedr_chk_if_fmr(struct qedr_qp *qp) +{ + if (qp->wqe_wr_id[qp->sq.cons].opcode == IB_WC_REG_MR) + qp->wqe_wr_id[qp->sq.cons].mr->info.completed++; +} + +static int process_req(struct qedr_dev *dev, struct qedr_qp *qp, + struct qedr_cq *cq, int num_entries, + struct ib_wc *wc, u16 hw_cons, enum ib_wc_status status, + int force) +{ + u16 cnt = 0; + + while (num_entries && qp->sq.wqe_cons != hw_cons) { + if (!qp->wqe_wr_id[qp->sq.cons].signaled && !force) { + qedr_chk_if_fmr(qp); + /* skip WC */ + goto next_cqe; + } + + /* fill WC */ + wc->status = status; + wc->wc_flags = 0; + wc->src_qp = qp->id; + wc->qp = &qp->ibqp; + + wc->wr_id = qp->wqe_wr_id[qp->sq.cons].wr_id; + wc->opcode = qp->wqe_wr_id[qp->sq.cons].opcode; + + switch (wc->opcode) { + case IB_WC_RDMA_WRITE: + wc->byte_len = qp->wqe_wr_id[qp->sq.cons].bytes_len; + break; + case IB_WC_COMP_SWAP: + case IB_WC_FETCH_ADD: + wc->byte_len = 8; + break; + case IB_WC_REG_MR: + qp->wqe_wr_id[qp->sq.cons].mr->info.completed++; + break; + default: + break; + } + + num_entries--; + wc++; + cnt++; +next_cqe: + while (qp->wqe_wr_id[qp->sq.cons].wqe_size--) + qed_chain_consume(&qp->sq.pbl); + qedr_inc_sw_cons(&qp->sq); + } + + return cnt; +} + +static int qedr_poll_cq_req(struct qedr_dev *dev, + struct qedr_qp *qp, struct qedr_cq *cq, + int num_entries, struct ib_wc *wc, + struct rdma_cqe_requester *req) +{ + int cnt = 0; + + switch (req->status) { + case RDMA_CQE_REQ_STS_OK: + cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons, + IB_WC_SUCCESS, 0); + break; + case RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons, + IB_WC_WR_FLUSH_ERR, 0); + break; + default: + /* process all WQE before the cosumer */ + qp->state = QED_ROCE_QP_STATE_ERR; + cnt = process_req(dev, qp, cq, num_entries, wc, + req->sq_cons - 1, IB_WC_SUCCESS, 0); + wc += cnt; + /* if we have extra WC fill it with actual error info */ + if (cnt < num_entries) { + enum ib_wc_status wc_status; + + switch (req->status) { + case RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_BAD_RESP_ERR; + break; + case RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_LOC_LEN_ERR; + break; + case RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_LOC_QP_OP_ERR; + break; + case RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_LOC_PROT_ERR; + break; + case RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_MW_BIND_ERR; + break; + case RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_REM_INV_REQ_ERR; + break; + case RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_REM_ACCESS_ERR; + break; + case RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_REM_OP_ERR; + break; + case RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_RNR_RETRY_EXC_ERR; + break; + case RDMA_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR: + DP_ERR(dev, + "Error: POLL CQ with ROCE_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_RETRY_EXC_ERR; + break; + default: + DP_ERR(dev, + "Error: POLL CQ with IB_WC_GENERAL_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_GENERAL_ERR; + } + cnt += process_req(dev, qp, cq, 1, wc, req->sq_cons, + wc_status, 1); + } + } + + return cnt; +} + +static void __process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp, + struct qedr_cq *cq, struct ib_wc *wc, + struct rdma_cqe_responder *resp, u64 wr_id) +{ + enum ib_wc_status wc_status = IB_WC_SUCCESS; + u8 flags; + + wc->opcode = IB_WC_RECV; + wc->wc_flags = 0; + + switch (resp->status) { + case RDMA_CQE_RESP_STS_LOCAL_ACCESS_ERR: + wc_status = IB_WC_LOC_ACCESS_ERR; + break; + case RDMA_CQE_RESP_STS_LOCAL_LENGTH_ERR: + wc_status = IB_WC_LOC_LEN_ERR; + break; + case RDMA_CQE_RESP_STS_LOCAL_QP_OPERATION_ERR: + wc_status = IB_WC_LOC_QP_OP_ERR; + break; + case RDMA_CQE_RESP_STS_LOCAL_PROTECTION_ERR: + wc_status = IB_WC_LOC_PROT_ERR; + break; + case RDMA_CQE_RESP_STS_MEMORY_MGT_OPERATION_ERR: + wc_status = IB_WC_MW_BIND_ERR; + break; + case RDMA_CQE_RESP_STS_REMOTE_INVALID_REQUEST_ERR: + wc_status = IB_WC_REM_INV_RD_REQ_ERR; + break; + case RDMA_CQE_RESP_STS_OK: + wc_status = IB_WC_SUCCESS; + wc->byte_len = le32_to_cpu(resp->length); + + flags = resp->flags & QEDR_RESP_RDMA_IMM; + + if (flags == QEDR_RESP_RDMA_IMM) + wc->opcode = IB_WC_RECV_RDMA_WITH_IMM; + + if (flags == QEDR_RESP_RDMA_IMM || flags == QEDR_RESP_IMM) { + wc->ex.imm_data = + le32_to_cpu(resp->imm_data_or_inv_r_Key); + wc->wc_flags |= IB_WC_WITH_IMM; + } + break; + default: + wc->status = IB_WC_GENERAL_ERR; + DP_ERR(dev, "Invalid CQE status detected\n"); + } + + /* fill WC */ + wc->status = wc_status; + wc->src_qp = qp->id; + wc->qp = &qp->ibqp; + wc->wr_id = wr_id; +} + +static int process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp, + struct qedr_cq *cq, struct ib_wc *wc, + struct rdma_cqe_responder *resp) +{ + u64 wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id; + + __process_resp_one(dev, qp, cq, wc, resp, wr_id); + + while (qp->rqe_wr_id[qp->rq.cons].wqe_size--) + qed_chain_consume(&qp->rq.pbl); + qedr_inc_sw_cons(&qp->rq); + + return 1; +} + +static int process_resp_flush(struct qedr_qp *qp, struct qedr_cq *cq, + int num_entries, struct ib_wc *wc, u16 hw_cons) +{ + u16 cnt = 0; + + while (num_entries && qp->rq.wqe_cons != hw_cons) { + /* fill WC */ + wc->status = IB_WC_WR_FLUSH_ERR; + wc->wc_flags = 0; + wc->src_qp = qp->id; + wc->byte_len = 0; + wc->wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id; + wc->qp = &qp->ibqp; + num_entries--; + wc++; + cnt++; + while (qp->rqe_wr_id[qp->rq.cons].wqe_size--) + qed_chain_consume(&qp->rq.pbl); + qedr_inc_sw_cons(&qp->rq); + } + + return cnt; +} + +static void try_consume_resp_cqe(struct qedr_cq *cq, struct qedr_qp *qp, + struct rdma_cqe_responder *resp, int *update) +{ + if (le16_to_cpu(resp->rq_cons) == qp->rq.wqe_cons) { + consume_cqe(cq); + *update |= 1; + } +} + +static int qedr_poll_cq_resp(struct qedr_dev *dev, struct qedr_qp *qp, + struct qedr_cq *cq, int num_entries, + struct ib_wc *wc, struct rdma_cqe_responder *resp, + int *update) +{ + int cnt; + + if (resp->status == RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR) { + cnt = process_resp_flush(qp, cq, num_entries, wc, + resp->rq_cons); + try_consume_resp_cqe(cq, qp, resp, update); + } else { + cnt = process_resp_one(dev, qp, cq, wc, resp); + consume_cqe(cq); + *update |= 1; + } + + return cnt; +} + +static void try_consume_req_cqe(struct qedr_cq *cq, struct qedr_qp *qp, + struct rdma_cqe_requester *req, int *update) +{ + if (le16_to_cpu(req->sq_cons) == qp->sq.wqe_cons) { + consume_cqe(cq); + *update |= 1; + } +} + +int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc) +{ + struct qedr_dev *dev = get_qedr_dev(ibcq->device); + struct qedr_cq *cq = get_qedr_cq(ibcq); + union rdma_cqe *cqe = cq->latest_cqe; + u32 old_cons, new_cons; + unsigned long flags; + int update = 0; + int done = 0; + + if (cq->cq_type == QEDR_CQ_TYPE_GSI) + return qedr_gsi_poll_cq(ibcq, num_entries, wc); + + spin_lock_irqsave(&cq->cq_lock, flags); + old_cons = qed_chain_get_cons_idx_u32(&cq->pbl); + while (num_entries && is_valid_cqe(cq, cqe)) { + struct qedr_qp *qp; + int cnt = 0; + + /* prevent speculative reads of any field of CQE */ + rmb(); + + qp = cqe_get_qp(cqe); + if (!qp) { + WARN(1, "Error: CQE QP pointer is NULL. CQE=%p\n", cqe); + break; + } + + wc->qp = &qp->ibqp; + + switch (cqe_get_type(cqe)) { + case RDMA_CQE_TYPE_REQUESTER: + cnt = qedr_poll_cq_req(dev, qp, cq, num_entries, wc, + &cqe->req); + try_consume_req_cqe(cq, qp, &cqe->req, &update); + break; + case RDMA_CQE_TYPE_RESPONDER_RQ: + cnt = qedr_poll_cq_resp(dev, qp, cq, num_entries, wc, + &cqe->resp, &update); + break; + case RDMA_CQE_TYPE_INVALID: + default: + DP_ERR(dev, "Error: invalid CQE type = %d\n", + cqe_get_type(cqe)); + } + num_entries -= cnt; + wc += cnt; + done += cnt; + + cqe = get_cqe(cq); + } + new_cons = qed_chain_get_cons_idx_u32(&cq->pbl); + + cq->cq_cons += new_cons - old_cons; + + if (update) + /* doorbell notifies abount latest VALID entry, + * but chain already point to the next INVALID one + */ + doorbell_cq(cq, cq->cq_cons - 1, cq->arm_flags); + + spin_unlock_irqrestore(&cq->cq_lock, flags); + return done; +} + +int qedr_process_mad(struct ib_device *ibdev, int process_mad_flags, + u8 port_num, + const struct ib_wc *in_wc, + const struct ib_grh *in_grh, + const struct ib_mad_hdr *mad_hdr, + size_t in_mad_size, struct ib_mad_hdr *out_mad, + size_t *out_mad_size, u16 *out_mad_pkey_index) +{ + struct qedr_dev *dev = get_qedr_dev(ibdev); + + DP_DEBUG(dev, QEDR_MSG_GSI, + "QEDR_PROCESS_MAD in_mad %x %x %x %x %x %x %x %x\n", + mad_hdr->attr_id, mad_hdr->base_version, mad_hdr->attr_mod, + mad_hdr->class_specific, mad_hdr->class_version, + mad_hdr->method, mad_hdr->mgmt_class, mad_hdr->status); + return IB_MAD_RESULT_SUCCESS; +} + +int qedr_port_immutable(struct ib_device *ibdev, u8 port_num, + struct ib_port_immutable *immutable) +{ + struct ib_port_attr attr; + int err; + + err = qedr_query_port(ibdev, port_num, &attr); + if (err) + return err; + + immutable->pkey_tbl_len = attr.pkey_tbl_len; + immutable->gid_tbl_len = attr.gid_tbl_len; + immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE | + RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP; + immutable->max_mad_size = IB_MGMT_MAD_SIZE; + + return 0; +} diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h new file mode 100644 index 0000000..a9b5e67 --- /dev/null +++ b/drivers/infiniband/hw/qedr/verbs.h @@ -0,0 +1,101 @@ +/* QLogic qedr NIC Driver + * Copyright (c) 2015-2016 QLogic Corporation + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and /or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef __QEDR_VERBS_H__ +#define __QEDR_VERBS_H__ + +int qedr_query_device(struct ib_device *ibdev, + struct ib_device_attr *attr, struct ib_udata *udata); +int qedr_query_port(struct ib_device *, u8 port, struct ib_port_attr *props); +int qedr_modify_port(struct ib_device *, u8 port, int mask, + struct ib_port_modify *props); + +int qedr_query_gid(struct ib_device *, u8 port, int index, union ib_gid *gid); + +int qedr_query_pkey(struct ib_device *, u8 port, u16 index, u16 *pkey); + +struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *, struct ib_udata *); +int qedr_dealloc_ucontext(struct ib_ucontext *); + +int qedr_mmap(struct ib_ucontext *, struct vm_area_struct *vma); +int qedr_del_gid(struct ib_device *device, u8 port_num, + unsigned int index, void **context); +int qedr_add_gid(struct ib_device *device, u8 port_num, + unsigned int index, const union ib_gid *gid, + const struct ib_gid_attr *attr, void **context); +struct ib_pd *qedr_alloc_pd(struct ib_device *, + struct ib_ucontext *, struct ib_udata *); +int qedr_dealloc_pd(struct ib_pd *pd); + +struct ib_cq *qedr_create_cq(struct ib_device *ibdev, + const struct ib_cq_init_attr *attr, + struct ib_ucontext *ib_ctx, + struct ib_udata *udata); +int qedr_resize_cq(struct ib_cq *, int cqe, struct ib_udata *); +int qedr_destroy_cq(struct ib_cq *); +int qedr_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags); +struct ib_qp *qedr_create_qp(struct ib_pd *, struct ib_qp_init_attr *attrs, + struct ib_udata *); +int qedr_modify_qp(struct ib_qp *, struct ib_qp_attr *attr, + int attr_mask, struct ib_udata *udata); +int qedr_query_qp(struct ib_qp *, struct ib_qp_attr *qp_attr, + int qp_attr_mask, struct ib_qp_init_attr *); +int qedr_destroy_qp(struct ib_qp *ibqp); + +struct ib_ah *qedr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr); +int qedr_destroy_ah(struct ib_ah *ibah); + +int qedr_dereg_mr(struct ib_mr *); +struct ib_mr *qedr_get_dma_mr(struct ib_pd *, int acc); + +struct ib_mr *qedr_reg_user_mr(struct ib_pd *, u64 start, u64 length, + u64 virt, int acc, struct ib_udata *); + +int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, + int sg_nents, unsigned int *sg_offset); + +struct ib_mr *qedr_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type, + u32 max_num_sg); +int qedr_poll_cq(struct ib_cq *, int num_entries, struct ib_wc *wc); +int qedr_post_send(struct ib_qp *, struct ib_send_wr *, + struct ib_send_wr **bad_wr); +int qedr_post_recv(struct ib_qp *, struct ib_recv_wr *, + struct ib_recv_wr **bad_wr); +int qedr_process_mad(struct ib_device *ibdev, int process_mad_flags, + u8 port_num, const struct ib_wc *in_wc, + const struct ib_grh *in_grh, + const struct ib_mad_hdr *in_mad, + size_t in_mad_size, struct ib_mad_hdr *out_mad, + size_t *out_mad_size, u16 *out_mad_pkey_index); + +int qedr_port_immutable(struct ib_device *ibdev, u8 port_num, + struct ib_port_immutable *immutable); +#endif |