diff options
author | Dean Luick <dean.luick@intel.com> | 2016-02-18 19:12:08 (GMT) |
---|---|---|
committer | Doug Ledford <dledford@redhat.com> | 2016-03-11 01:45:43 (GMT) |
commit | 4c9e7aacb6a6334168a81b83819fb5cb088d2fb3 (patch) | |
tree | c91e46bc9ccc4ddc2db104c16e5bc48f5aeaa217 /drivers/staging | |
parent | 6b5c5213e57453c228f7695d5d889aa4c84272c3 (diff) | |
download | linux-4c9e7aacb6a6334168a81b83819fb5cb088d2fb3.tar.xz |
staging/rdma/hfi1: Fix xmit discard error weight
Count only the errors that apply to xmit discards. Update
the comment to better explain the limitations of the count.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Diffstat (limited to 'drivers/staging')
-rw-r--r-- | drivers/staging/rdma/hfi1/chip.c | 26 |
1 files changed, 21 insertions, 5 deletions
diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c index 93bf465..6e44d52 100644 --- a/drivers/staging/rdma/hfi1/chip.c +++ b/drivers/staging/rdma/hfi1/chip.c @@ -5566,12 +5566,28 @@ static void handle_send_egress_err_info(struct hfi1_devdata *dd, int weight, i; /* - * Count all, in case multiple bits are set. Reminder: - * since there is only one info register for many sources, - * these may be attributed to the wrong VL if they occur - * too close together. + * Count all applicable bits as individual errors and + * attribute them to the packet that triggered this handler. + * This may not be completely accurate due to limitations + * on the available hardware error information. There is + * a single information register and any number of error + * packets may have occurred and contributed to it before + * this routine is called. This means that: + * a) If multiple packets with the same error occur before + * this routine is called, earlier packets are missed. + * There is only a single bit for each error type. + * b) Errors may not be attributed to the correct VL. + * The driver is attributing all bits in the info register + * to the packet that triggered this call, but bits + * could be an accumulation of different packets with + * different VLs. + * c) A single error packet may have multiple counts attached + * to it. There is no way for the driver to know if + * multiple bits set in the info register are due to a + * single packet or multiple packets. The driver assumes + * multiple packets. */ - weight = hweight64(info); + weight = hweight64(info & PORT_DISCARD_EGRESS_ERRS); for (i = 0; i < weight; i++) { __count_port_discards(ppd); if (vl >= 0 && vl < TXE_NUM_DATA_VL) |