Re: [PATCH 4.9 12/16] net/rds: dma_map_sg is entitled to merge entries
From: Pavel Machek <hidden>
Date: 2021-09-03 06:50:40
Also in:
stable
Hi!
[ Upstream commit fb4b1373dcab086d0619c29310f0466a0b2ceb8a ]
Function "dma_map_sg" is entitled to merge adjacent entries
and return a value smaller than what was passed as "nents".
Subsequently "ib_map_mr_sg" needs to work with this value ("sg_dma_len")
rather than the original "nents" parameter ("sg_len").
This old RDS bug was exposed and reliably causes kernel panics
(using RDMA operations "rds-stress -D") on x86_64 starting with:
commit c588072bba6b ("iommu/vt-d: Convert intel iommu driver to the iommu ops")
Simply put: Linux 5.11 and later.I see this queued for 4.19 and 5.10 where "iommu/vt-d: Convert intel iommu driver to the iommu ops" is not present. It may be okay for older kernels, too, but I wanted to double-check. Best regards, Pavel
quoted hunk ↗ jump to hunk
+++ b/net/rds/ib_frmr.c@@ -112,9 +112,9 @@ static int rds_ib_post_reg_frmr(struct rds_ib_mr *ibmr) cpu_relax(); } - ret = ib_map_mr_sg_zbva(frmr->mr, ibmr->sg, ibmr->sg_len, + ret = ib_map_mr_sg_zbva(frmr->mr, ibmr->sg, ibmr->sg_dma_len, &off, PAGE_SIZE); - if (unlikely(ret != ibmr->sg_len)) + if (unlikely(ret != ibmr->sg_dma_len)) return ret < 0 ? ret : -EINVAL; /* Perform a WR for the fast_reg_mr. Each individual page
-- DENX Software Engineering GmbH, Managing Director: Wolfgang Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Attachments
- signature.asc [application/pgp-signature] 181 bytes