[openib-general] [PATCH] RFC: srp filesystem data corruption problem/work-around

[Michael S. Tsirkin]

Hi, Roland!

For some time now we've been trying to debug a data corruption failure that
occurs when working with Mellanox SRP target.  We could not yet pin-point
exactly where the problem is (debugging suggestions are wellcome) but the
corruption seems to occur only when the target gets a request where the virtual
address is not 0.

Until there's a better understanding of the issue, I've come up with the
following simple patch that will use indirect mode for this case.

Could you comment on this please?

---

Add work-around for data corruption observed with Mellanox targets
when VA != 0.

Signed-off-by: Michael S. Tsirkin <mst at mellanox.co.il>

diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index 8f472e7..6a8b286 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -75,6 +75,12 @@ module_param(topspin_workarounds, int, 0
 MODULE_PARM_DESC(topspin_workarounds,
 		 "Enable workarounds for Topspin/Cisco SRP target bugs if != 0");
 
+static int mellanox_workarounds = 1;
+
+module_param(mellanox_workarounds, int, 0444);
+MODULE_PARM_DESC(mellanox_workarounds,
+		 "Enable workarounds for Mellanox SRP target bugs if != 0");
+
 static const u8 topspin_oui[3] = { 0x00, 0x05, 0xad };
 
 static void srp_add_one(struct ib_device *device);
@@ -581,6 +587,10 @@ static int srp_map_fmr(struct srp_device
 	if (!dev->fmr_pool)
 		return -ENODEV;
 
+	if (mellanox_workarounds && 
+		sg_dma_address(&scat[0]) & ~dev->fmr_page_mask)
+		return -EINVAL;
+
 	len = page_cnt = 0;
 	for (i = 0; i < sg_cnt; ++i) {
 		if (sg_dma_address(&scat[i]) & ~dev->fmr_page_mask) {

-- 
The question of whether a computer can think is no more interesting than the
question of whether a submarine can swim. -- Edsger Wybe Dijkstra

----- End forwarded message -----

-- 
MST