diff options
author | Mike Marciniszyn <mike.marciniszyn@intel.com> | 2013-05-30 22:25:25 (GMT) |
---|---|---|
committer | Roland Dreier <roland@purestorage.com> | 2013-06-22 00:19:38 (GMT) |
commit | 8469ba39a6b77917e8879680aed17229bf72f263 (patch) | |
tree | f064dfaf3387f3ed5d59d69941b90825fe628729 /drivers/infiniband/hw/qib/qib_iba7322.c | |
parent | fedaf4ffc224a194e2d13a3ec2abe5df0bc94258 (diff) | |
download | linux-8469ba39a6b77917e8879680aed17229bf72f263.tar.xz |
IB/qib: Add DCA support
This patch adds DCA cache warming for systems that support DCA.
The code uses cpu affinity notification to react to an affinity change
from a user mode program like irqbalance and (re-)program the chip
accordingly. This notification avoids reading the current cpu on every
interrupt.
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
[ Add Kconfig dependency on SMP && GENERIC_HARDIRQS to avoid failure to
build due to undefined struct irq_affinity_notify. - Roland ]
Signed-off-by: Roland Dreier <roland@purestorage.com>
Diffstat (limited to 'drivers/infiniband/hw/qib/qib_iba7322.c')
-rw-r--r-- | drivers/infiniband/hw/qib/qib_iba7322.c | 334 |
1 files changed, 322 insertions, 12 deletions
diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c index 3f6b21e..46ffea0 100644 --- a/drivers/infiniband/hw/qib/qib_iba7322.c +++ b/drivers/infiniband/hw/qib/qib_iba7322.c @@ -44,6 +44,9 @@ #include <linux/module.h> #include <rdma/ib_verbs.h> #include <rdma/ib_smi.h> +#ifdef CONFIG_INFINIBAND_QIB_DCA +#include <linux/dca.h> +#endif #include "qib.h" #include "qib_7322_regs.h" @@ -519,6 +522,14 @@ static const u8 qib_7322_physportstate[0x20] = { [0x17] = IB_PHYSPORTSTATE_CFG_TRAIN }; +#ifdef CONFIG_INFINIBAND_QIB_DCA +struct qib_irq_notify { + int rcv; + void *arg; + struct irq_affinity_notify notify; +}; +#endif + struct qib_chip_specific { u64 __iomem *cregbase; u64 *cntrs; @@ -546,6 +557,12 @@ struct qib_chip_specific { u32 lastbuf_for_pio; u32 stay_in_freeze; u32 recovery_ports_initted; +#ifdef CONFIG_INFINIBAND_QIB_DCA + u32 dca_ctrl; + int rhdr_cpu[18]; + int sdma_cpu[2]; + u64 dca_rcvhdr_ctrl[5]; /* B, C, D, E, F */ +#endif struct qib_msix_entry *msix_entries; unsigned long *sendchkenable; unsigned long *sendgrhchk; @@ -642,28 +659,76 @@ static struct { irq_handler_t handler; int lsb; int port; /* 0 if not port-specific, else port # */ + int dca; } irq_table[] = { - { "", qib_7322intr, -1, 0 }, + { "", qib_7322intr, -1, 0, 0 }, { " (buf avail)", qib_7322bufavail, - SYM_LSB(IntStatus, SendBufAvail), 0 }, + SYM_LSB(IntStatus, SendBufAvail), 0, 0}, { " (sdma 0)", sdma_intr, - SYM_LSB(IntStatus, SDmaInt_0), 1 }, + SYM_LSB(IntStatus, SDmaInt_0), 1, 1 }, { " (sdma 1)", sdma_intr, - SYM_LSB(IntStatus, SDmaInt_1), 2 }, + SYM_LSB(IntStatus, SDmaInt_1), 2, 1 }, { " (sdmaI 0)", sdma_idle_intr, - SYM_LSB(IntStatus, SDmaIdleInt_0), 1 }, + SYM_LSB(IntStatus, SDmaIdleInt_0), 1, 1}, { " (sdmaI 1)", sdma_idle_intr, - SYM_LSB(IntStatus, SDmaIdleInt_1), 2 }, + SYM_LSB(IntStatus, SDmaIdleInt_1), 2, 1}, { " (sdmaP 0)", sdma_progress_intr, - SYM_LSB(IntStatus, SDmaProgressInt_0), 1 }, + SYM_LSB(IntStatus, SDmaProgressInt_0), 1, 1 }, { " (sdmaP 1)", sdma_progress_intr, - SYM_LSB(IntStatus, SDmaProgressInt_1), 2 }, + SYM_LSB(IntStatus, SDmaProgressInt_1), 2, 1 }, { " (sdmaC 0)", sdma_cleanup_intr, - SYM_LSB(IntStatus, SDmaCleanupDone_0), 1 }, + SYM_LSB(IntStatus, SDmaCleanupDone_0), 1, 0 }, { " (sdmaC 1)", sdma_cleanup_intr, - SYM_LSB(IntStatus, SDmaCleanupDone_1), 2 }, + SYM_LSB(IntStatus, SDmaCleanupDone_1), 2 , 0}, }; +#ifdef CONFIG_INFINIBAND_QIB_DCA + +static const struct dca_reg_map { + int shadow_inx; + int lsb; + u64 mask; + u16 regno; +} dca_rcvhdr_reg_map[] = { + { 0, SYM_LSB(DCACtrlB, RcvHdrq0DCAOPH), + ~SYM_MASK(DCACtrlB, RcvHdrq0DCAOPH) , KREG_IDX(DCACtrlB) }, + { 0, SYM_LSB(DCACtrlB, RcvHdrq1DCAOPH), + ~SYM_MASK(DCACtrlB, RcvHdrq1DCAOPH) , KREG_IDX(DCACtrlB) }, + { 0, SYM_LSB(DCACtrlB, RcvHdrq2DCAOPH), + ~SYM_MASK(DCACtrlB, RcvHdrq2DCAOPH) , KREG_IDX(DCACtrlB) }, + { 0, SYM_LSB(DCACtrlB, RcvHdrq3DCAOPH), + ~SYM_MASK(DCACtrlB, RcvHdrq3DCAOPH) , KREG_IDX(DCACtrlB) }, + { 1, SYM_LSB(DCACtrlC, RcvHdrq4DCAOPH), + ~SYM_MASK(DCACtrlC, RcvHdrq4DCAOPH) , KREG_IDX(DCACtrlC) }, + { 1, SYM_LSB(DCACtrlC, RcvHdrq5DCAOPH), + ~SYM_MASK(DCACtrlC, RcvHdrq5DCAOPH) , KREG_IDX(DCACtrlC) }, + { 1, SYM_LSB(DCACtrlC, RcvHdrq6DCAOPH), + ~SYM_MASK(DCACtrlC, RcvHdrq6DCAOPH) , KREG_IDX(DCACtrlC) }, + { 1, SYM_LSB(DCACtrlC, RcvHdrq7DCAOPH), + ~SYM_MASK(DCACtrlC, RcvHdrq7DCAOPH) , KREG_IDX(DCACtrlC) }, + { 2, SYM_LSB(DCACtrlD, RcvHdrq8DCAOPH), + ~SYM_MASK(DCACtrlD, RcvHdrq8DCAOPH) , KREG_IDX(DCACtrlD) }, + { 2, SYM_LSB(DCACtrlD, RcvHdrq9DCAOPH), + ~SYM_MASK(DCACtrlD, RcvHdrq9DCAOPH) , KREG_IDX(DCACtrlD) }, + { 2, SYM_LSB(DCACtrlD, RcvHdrq10DCAOPH), + ~SYM_MASK(DCACtrlD, RcvHdrq10DCAOPH) , KREG_IDX(DCACtrlD) }, + { 2, SYM_LSB(DCACtrlD, RcvHdrq11DCAOPH), + ~SYM_MASK(DCACtrlD, RcvHdrq11DCAOPH) , KREG_IDX(DCACtrlD) }, + { 3, SYM_LSB(DCACtrlE, RcvHdrq12DCAOPH), + ~SYM_MASK(DCACtrlE, RcvHdrq12DCAOPH) , KREG_IDX(DCACtrlE) }, + { 3, SYM_LSB(DCACtrlE, RcvHdrq13DCAOPH), + ~SYM_MASK(DCACtrlE, RcvHdrq13DCAOPH) , KREG_IDX(DCACtrlE) }, + { 3, SYM_LSB(DCACtrlE, RcvHdrq14DCAOPH), + ~SYM_MASK(DCACtrlE, RcvHdrq14DCAOPH) , KREG_IDX(DCACtrlE) }, + { 3, SYM_LSB(DCACtrlE, RcvHdrq15DCAOPH), + ~SYM_MASK(DCACtrlE, RcvHdrq15DCAOPH) , KREG_IDX(DCACtrlE) }, + { 4, SYM_LSB(DCACtrlF, RcvHdrq16DCAOPH), + ~SYM_MASK(DCACtrlF, RcvHdrq16DCAOPH) , KREG_IDX(DCACtrlF) }, + { 4, SYM_LSB(DCACtrlF, RcvHdrq17DCAOPH), + ~SYM_MASK(DCACtrlF, RcvHdrq17DCAOPH) , KREG_IDX(DCACtrlF) }, +}; +#endif + /* ibcctrl bits */ #define QLOGIC_IB_IBCC_LINKINITCMD_DISABLE 1 /* cycle through TS1/TS2 till OK */ @@ -686,6 +751,13 @@ static void write_7322_init_portregs(struct qib_pportdata *); static void setup_7322_link_recovery(struct qib_pportdata *, u32); static void check_7322_rxe_status(struct qib_pportdata *); static u32 __iomem *qib_7322_getsendbuf(struct qib_pportdata *, u64, u32 *); +#ifdef CONFIG_INFINIBAND_QIB_DCA +static void qib_setup_dca(struct qib_devdata *dd); +static void setup_dca_notifier(struct qib_devdata *dd, + struct qib_msix_entry *m); +static void reset_dca_notifier(struct qib_devdata *dd, + struct qib_msix_entry *m); +#endif /** * qib_read_ureg32 - read 32-bit virtualized per-context register @@ -2558,6 +2630,162 @@ static void qib_setup_7322_setextled(struct qib_pportdata *ppd, u32 on) qib_write_kreg_port(ppd, krp_rcvpktledcnt, ledblink); } +#ifdef CONFIG_INFINIBAND_QIB_DCA + +static int qib_7322_notify_dca(struct qib_devdata *dd, unsigned long event) +{ + switch (event) { + case DCA_PROVIDER_ADD: + if (dd->flags & QIB_DCA_ENABLED) + break; + if (!dca_add_requester(&dd->pcidev->dev)) { + qib_devinfo(dd->pcidev, "DCA enabled\n"); + dd->flags |= QIB_DCA_ENABLED; + qib_setup_dca(dd); + } + break; + case DCA_PROVIDER_REMOVE: + if (dd->flags & QIB_DCA_ENABLED) { + dca_remove_requester(&dd->pcidev->dev); + dd->flags &= ~QIB_DCA_ENABLED; + dd->cspec->dca_ctrl = 0; + qib_write_kreg(dd, KREG_IDX(DCACtrlA), + dd->cspec->dca_ctrl); + } + break; + } + return 0; +} + +static void qib_update_rhdrq_dca(struct qib_ctxtdata *rcd, int cpu) +{ + struct qib_devdata *dd = rcd->dd; + struct qib_chip_specific *cspec = dd->cspec; + + if (!(dd->flags & QIB_DCA_ENABLED)) + return; + if (cspec->rhdr_cpu[rcd->ctxt] != cpu) { + const struct dca_reg_map *rmp; + + cspec->rhdr_cpu[rcd->ctxt] = cpu; + rmp = &dca_rcvhdr_reg_map[rcd->ctxt]; + cspec->dca_rcvhdr_ctrl[rmp->shadow_inx] &= rmp->mask; + cspec->dca_rcvhdr_ctrl[rmp->shadow_inx] |= + (u64) dca3_get_tag(&dd->pcidev->dev, cpu) << rmp->lsb; + qib_devinfo(dd->pcidev, + "Ctxt %d cpu %d dca %llx\n", rcd->ctxt, cpu, + (long long) cspec->dca_rcvhdr_ctrl[rmp->shadow_inx]); + qib_write_kreg(dd, rmp->regno, + cspec->dca_rcvhdr_ctrl[rmp->shadow_inx]); + cspec->dca_ctrl |= SYM_MASK(DCACtrlA, RcvHdrqDCAEnable); + qib_write_kreg(dd, KREG_IDX(DCACtrlA), cspec->dca_ctrl); + } +} + +static void qib_update_sdma_dca(struct qib_pportdata *ppd, int cpu) +{ + struct qib_devdata *dd = ppd->dd; + struct qib_chip_specific *cspec = dd->cspec; + unsigned pidx = ppd->port - 1; + + if (!(dd->flags & QIB_DCA_ENABLED)) + return; + if (cspec->sdma_cpu[pidx] != cpu) { + cspec->sdma_cpu[pidx] = cpu; + cspec->dca_rcvhdr_ctrl[4] &= ~(ppd->hw_pidx ? + SYM_MASK(DCACtrlF, SendDma1DCAOPH) : + SYM_MASK(DCACtrlF, SendDma0DCAOPH)); + cspec->dca_rcvhdr_ctrl[4] |= + (u64) dca3_get_tag(&dd->pcidev->dev, cpu) << + (ppd->hw_pidx ? + SYM_LSB(DCACtrlF, SendDma1DCAOPH) : + SYM_LSB(DCACtrlF, SendDma0DCAOPH)); + qib_devinfo(dd->pcidev, + "sdma %d cpu %d dca %llx\n", ppd->hw_pidx, cpu, + (long long) cspec->dca_rcvhdr_ctrl[4]); + qib_write_kreg(dd, KREG_IDX(DCACtrlF), + cspec->dca_rcvhdr_ctrl[4]); + cspec->dca_ctrl |= ppd->hw_pidx ? + SYM_MASK(DCACtrlA, SendDMAHead1DCAEnable) : + SYM_MASK(DCACtrlA, SendDMAHead0DCAEnable); + qib_write_kreg(dd, KREG_IDX(DCACtrlA), cspec->dca_ctrl); + } +} + +static void qib_setup_dca(struct qib_devdata *dd) +{ + struct qib_chip_specific *cspec = dd->cspec; + int i; + + for (i = 0; i < ARRAY_SIZE(cspec->rhdr_cpu); i++) + cspec->rhdr_cpu[i] = -1; + for (i = 0; i < ARRAY_SIZE(cspec->sdma_cpu); i++) + cspec->sdma_cpu[i] = -1; + cspec->dca_rcvhdr_ctrl[0] = + (1ULL << SYM_LSB(DCACtrlB, RcvHdrq0DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlB, RcvHdrq1DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlB, RcvHdrq2DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlB, RcvHdrq3DCAXfrCnt)); + cspec->dca_rcvhdr_ctrl[1] = + (1ULL << SYM_LSB(DCACtrlC, RcvHdrq4DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlC, RcvHdrq5DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlC, RcvHdrq6DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlC, RcvHdrq7DCAXfrCnt)); + cspec->dca_rcvhdr_ctrl[2] = + (1ULL << SYM_LSB(DCACtrlD, RcvHdrq8DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlD, RcvHdrq9DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlD, RcvHdrq10DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlD, RcvHdrq11DCAXfrCnt)); + cspec->dca_rcvhdr_ctrl[3] = + (1ULL << SYM_LSB(DCACtrlE, RcvHdrq12DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlE, RcvHdrq13DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlE, RcvHdrq14DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlE, RcvHdrq15DCAXfrCnt)); + cspec->dca_rcvhdr_ctrl[4] = + (1ULL << SYM_LSB(DCACtrlF, RcvHdrq16DCAXfrCnt)) | + (1ULL << SYM_LSB(DCACtrlF, RcvHdrq17DCAXfrCnt)); + for (i = 0; i < ARRAY_SIZE(cspec->sdma_cpu); i++) + qib_write_kreg(dd, KREG_IDX(DCACtrlB) + i, + cspec->dca_rcvhdr_ctrl[i]); + for (i = 0; i < cspec->num_msix_entries; i++) + setup_dca_notifier(dd, &cspec->msix_entries[i]); +} + +static void qib_irq_notifier_notify(struct irq_affinity_notify *notify, + const cpumask_t *mask) +{ + struct qib_irq_notify *n = + container_of(notify, struct qib_irq_notify, notify); + int cpu = cpumask_first(mask); + + if (n->rcv) { + struct qib_ctxtdata *rcd = (struct qib_ctxtdata *)n->arg; + qib_update_rhdrq_dca(rcd, cpu); + } else { + struct qib_pportdata *ppd = (struct qib_pportdata *)n->arg; + qib_update_sdma_dca(ppd, cpu); + } +} + +static void qib_irq_notifier_release(struct kref *ref) +{ + struct qib_irq_notify *n = + container_of(ref, struct qib_irq_notify, notify.kref); + struct qib_devdata *dd; + + if (n->rcv) { + struct qib_ctxtdata *rcd = (struct qib_ctxtdata *)n->arg; + dd = rcd->dd; + } else { + struct qib_pportdata *ppd = (struct qib_pportdata *)n->arg; + dd = ppd->dd; + } + qib_devinfo(dd->pcidev, + "release on HCA notify 0x%p n 0x%p\n", ref, n); + kfree(n); +} +#endif + /* * Disable MSIx interrupt if enabled, call generic MSIx code * to cleanup, and clear pending MSIx interrupts. @@ -2575,6 +2803,9 @@ static void qib_7322_nomsix(struct qib_devdata *dd) dd->cspec->num_msix_entries = 0; for (i = 0; i < n; i++) { +#ifdef CONFIG_INFINIBAND_QIB_DCA + reset_dca_notifier(dd, &dd->cspec->msix_entries[i]); +#endif irq_set_affinity_hint( dd->cspec->msix_entries[i].msix.vector, NULL); free_cpumask_var(dd->cspec->msix_entries[i].mask); @@ -2602,6 +2833,15 @@ static void qib_setup_7322_cleanup(struct qib_devdata *dd) { int i; +#ifdef CONFIG_INFINIBAND_QIB_DCA + if (dd->flags & QIB_DCA_ENABLED) { + dca_remove_requester(&dd->pcidev->dev); + dd->flags &= ~QIB_DCA_ENABLED; + dd->cspec->dca_ctrl = 0; + qib_write_kreg(dd, KREG_IDX(DCACtrlA), dd->cspec->dca_ctrl); + } +#endif + qib_7322_free_irq(dd); kfree(dd->cspec->cntrs); kfree(dd->cspec->sendchkenable); @@ -3068,6 +3308,53 @@ static irqreturn_t sdma_cleanup_intr(int irq, void *data) return IRQ_HANDLED; } +#ifdef CONFIG_INFINIBAND_QIB_DCA + +static void reset_dca_notifier(struct qib_devdata *dd, struct qib_msix_entry *m) +{ + if (!m->dca) + return; + qib_devinfo(dd->pcidev, + "Disabling notifier on HCA %d irq %d\n", + dd->unit, + m->msix.vector); + irq_set_affinity_notifier( + m->msix.vector, + NULL); + m->notifier = NULL; +} + +static void setup_dca_notifier(struct qib_devdata *dd, struct qib_msix_entry *m) +{ + struct qib_irq_notify *n; + + if (!m->dca) + return; + n = kzalloc(sizeof(*n), GFP_KERNEL); + if (n) { + int ret; + + m->notifier = n; + n->notify.irq = m->msix.vector; + n->notify.notify = qib_irq_notifier_notify; + n->notify.release = qib_irq_notifier_release; + n->arg = m->arg; + n->rcv = m->rcv; + qib_devinfo(dd->pcidev, + "set notifier irq %d rcv %d notify %p\n", + n->notify.irq, n->rcv, &n->notify); + ret = irq_set_affinity_notifier( + n->notify.irq, + &n->notify); + if (ret) { + m->notifier = NULL; + kfree(n); + } + } +} + +#endif + /* * Set up our chip-specific interrupt handler. * The interrupt type has already been setup, so @@ -3149,6 +3436,9 @@ try_intx: void *arg; u64 val; int lsb, reg, sh; +#ifdef CONFIG_INFINIBAND_QIB_DCA + int dca = 0; +#endif dd->cspec->msix_entries[msixnum]. name[sizeof(dd->cspec->msix_entries[msixnum].name) - 1] @@ -3161,6 +3451,9 @@ try_intx: arg = dd->pport + irq_table[i].port - 1; } else arg = dd; +#ifdef CONFIG_INFINIBAND_QIB_DCA + dca = irq_table[i].dca; +#endif lsb = irq_table[i].lsb; handler = irq_table[i].handler; snprintf(dd->cspec->msix_entries[msixnum].name, @@ -3178,6 +3471,9 @@ try_intx: continue; if (qib_krcvq01_no_msi && ctxt < 2) continue; +#ifdef CONFIG_INFINIBAND_QIB_DCA + dca = 1; +#endif lsb = QIB_I_RCVAVAIL_LSB + ctxt; handler = qib_7322pintr; snprintf(dd->cspec->msix_entries[msixnum].name, @@ -3203,6 +3499,11 @@ try_intx: goto try_intx; } dd->cspec->msix_entries[msixnum].arg = arg; +#ifdef CONFIG_INFINIBAND_QIB_DCA + dd->cspec->msix_entries[msixnum].dca = dca; + dd->cspec->msix_entries[msixnum].rcv = + handler == qib_7322pintr; +#endif if (lsb >= 0) { reg = lsb / IBA7322_REDIRECT_VEC_PER_REG; sh = (lsb % IBA7322_REDIRECT_VEC_PER_REG) * @@ -6885,6 +7186,9 @@ struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *pdev, dd->f_sdma_init_early = qib_7322_sdma_init_early; dd->f_writescratch = writescratch; dd->f_tempsense_rd = qib_7322_tempsense_rd; +#ifdef CONFIG_INFINIBAND_QIB_DCA + dd->f_notify_dca = qib_7322_notify_dca; +#endif /* * Do remaining PCIe setup and save PCIe values in dd. * Any error printing is already done by the init code. @@ -6921,7 +7225,7 @@ struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *pdev, actual_cnt -= dd->num_pports; tabsize = actual_cnt; - dd->cspec->msix_entries = kmalloc(tabsize * + dd->cspec->msix_entries = kzalloc(tabsize * sizeof(struct qib_msix_entry), GFP_KERNEL); if (!dd->cspec->msix_entries) { qib_dev_err(dd, "No memory for MSIx table\n"); @@ -6941,7 +7245,13 @@ struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *pdev, /* clear diagctrl register, in case diags were running and crashed */ qib_write_kreg(dd, kr_hwdiagctrl, 0); - +#ifdef CONFIG_INFINIBAND_QIB_DCA + if (!dca_add_requester(&pdev->dev)) { + qib_devinfo(dd->pcidev, "DCA enabled\n"); + dd->flags |= QIB_DCA_ENABLED; + qib_setup_dca(dd); + } +#endif goto bail; bail_cleanup: |