[PATCH] staging: lustre: o2iblnd: Stop MLX5 triggering a dump_cqe

[PATCH] staging: lustre: o2iblnd: Stop MLX5 triggering a dump_cqe

[Doug Oucharek]

We have found that MLX5 will trigger a dump_cqe if we don't
invalidate the rkey on a newly alloated MR for FastReg usage.

This fix just tags the MR as invalid on its creation if we are
using FastReg and that will force it to do an invalidate of the
rkey on first usage.

Signed-off-by: Doug Oucharek <doug.s.oucharek@intel.com>
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-8752
Reviewed-on: https://review.whamcloud.com/24306
Reviewed-by: Dmitry Eremin <dmitry.eremin@intel.com>
Reviewed-by: Amir Shehata <amir.shehata@intel.com>
Reviewed-by: James Simmons <uja.ornl@yahoo.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: Doug Oucharek <dougso@me.com>
---
drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c
index 7ae2955..00ebc61 100644
--- a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c
+++ b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c
@@ -1384,7 +1384,10 @@ static int kiblnd_alloc_freg_pool(struct kib_fmr_poolset *fps, struct kib_fmr_po
			goto out_middle;
		}

-		frd->frd_valid = true;
+		/* There appears to be a bug in MLX5 code where you must
+		 * invalidate the rkey of a new FastReg pool before first
+		 * using it. Thus, I am marking the FRD invalid here. */
+		frd->frd_valid = false;

		list_add_tail(&frd->frd_list, &fpo->fast_reg.fpo_pool_list);
		fpo->fast_reg.fpo_pool_size++;
-- 
1.8.3.1

_______________________________________________
devel mailing list
devel@linuxdriverproject.org
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel

Re: [PATCH] staging: lustre: o2iblnd: Stop MLX5 triggering a dump_cqe

[Dan Carpenter]

I don't really understand this patch...

On Fri, Mar 16, 2018 at 04:40:21PM -0700, Doug Oucharek wrote:
> We have found that MLX5 will trigger a dump_cqe if we don't
> invalidate the rkey on a newly alloated MR for FastReg usage.
> 
> This fix just tags the MR as invalid on its creation if we are
> using FastReg and that will force it to do an invalidate of the
> rkey on first usage.

This paragraph makes the change seem like a limited workaround for a
bug in the MLX5 code.  Why can't the MLX5 code be fixed instead?

Looking at the patch it doesn't seem like a limitted solution at all.
Now frd->frd_valid is *always* set to false.  Why don't we instead just
delete ->frd_valid along with the newly dead code?

regards,
dan carpenter