RDMA/bnxt_re: Fix input validation and
Subsystems (2)
Selvin Xavier
maintainerv3 cover RDMA/bnxt_re: Fix input validation and
This series fixes some of the bugs found while auditing bnxt_re's handling of user-supplied values and some generic bug fixes
Please review and apply the series.
Thanks,
Selvin Xavier
v2 -> v3:
- Drop the VM_MAYEXEC clearing patch from this series. Will work on a core level series which implements this in other drivers also
- Dropped rdev lifetime race patch from v2 series as it is reporting few more review comments to be handled
- New fix to serialize dcb_wq against async notifier
- Removed is_in_used flags related patch to be posted as a separate rcfw series
v1 -> v2:
- Handle review comment about clearing VM_MAYEXEC vma flag
Selvin Xavier (8):
RDMA/bnxt_re: Fix integer overflow in send payload size computation RDMA/bnxt_re: Detect wrong sge_len passed for inline RDMA/bnxt_re: Validate SRQ max_sge at create time RDMA/bnxt_re: Initialize wqe.flags in bnxt_re_post_srq_recv() RDMA/bnxt_re: Validate num_sge in bnxt_re_post_srq_recv() RDMA/bnxt_re: Serialize dcb_wq access against async notifier RDMA/bnxt_re: Fix the PD and DPI table size
RDMA/bnxt_re: Use bnxt_ext_stats_supported for counter count selection
drivers/infiniband/hw/bnxt_re/bnxt_re.h | 4 ++ drivers/infiniband/hw/bnxt_re/hw_counters.c | 6 +- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 65 ++++++++++++++------- drivers/infiniband/hw/bnxt_re/main.c | 24 ++++++-- drivers/infiniband/hw/bnxt_re/qplib_res.c | 9 +-- drivers/infiniband/hw/bnxt_re/qplib_sp.c | 2 +- 6 files changed, 77 insertions(+), 33 deletions(-)
Signature · 1 line
2.39.3
Selvin Xavier
maintainerPATCH v3 1/8 RDMA/bnxt_re: Fix integer overflow in send payload size computation
bnxt_re_build_sgl() summed SGE lengths into a signed int, which could overflow. Make it return u32, and have bnxt_re_copy_wr_payload() report the size via a u32 out-parameter with the return value carrying only the error status, instead of overloading a signed int with both.
--- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 44 +++++++++++++----------- 1 file changed, 24 insertions(+), 20 deletions(-)
drivers/infiniband/hw/bnxt_re/ib_verbs.c +24 -20
@@ -165,10 +165,11 @@ static void bnxt_re_check_and_set_relaxed_ordering(struct bnxt_re_dev *rdev, qplib_mr->flags |=CMDQ_REGISTER_MR_FLAGS_ENABLE_RO ; } -static int bnxt_re_build_sgl (struct ib_sge *ib_sg_list, +static u32 bnxt_re_build_sgl (struct ib_sge *ib_sg_list,struct bnxt_qplib_sge *sg_list,int num) { -int i, total =0 ; +u32 total =0 ; +int i;for (i =0 ; i < num; i++) { sg_list[i].addr = ib_sg_list[i].addr;
@@ -3185,17 +3186,22 @@ static int bnxt_re_copy_inline_data(struct bnxt_re_dev *rdev,static int bnxt_re_copy_wr_payload (struct bnxt_re_dev *rdev,const struct ib_send_wr *wr, -struct bnxt_qplib_swqe *wqe) +struct bnxt_qplib_swqe *wqe, +u32 *payload_sz) { -int payload_sz =0 ; +int rc; -if (wr->send_flags &IB_SEND_INLINE ) - payload_sz =bnxt_re_copy_inline_data (rdev, wr, wqe); -else - payload_sz =bnxt_re_build_sgl (wr->sg_list, wqe->sg_list, - wqe->num_sge); +if (wr->send_flags &IB_SEND_INLINE ) { + rc =bnxt_re_copy_inline_data (rdev, wr, wqe); +if (rc <0 ) +return rc; + *payload_sz = rc; + }else { + *payload_sz =bnxt_re_build_sgl (wr->sg_list, wqe->sg_list, + wqe->num_sge); + } -return payload_sz; +return 0 ; }static void bnxt_ud_qp_hw_stall_workaround (struct bnxt_re_qp *qp)
@@ -3218,7 +3224,8 @@ static int bnxt_re_post_send_shadow_qp(struct bnxt_re_dev *rdev,struct bnxt_re_qp *qp, const struct ib_send_wr *wr) { - int rc = 0, payload_sz = 0; + int rc = 0; +u32 payload_sz =0 ;unsigned long flags;spin_lock_irqsave (&qp->sq_lock, flags);
@@ -3234,11 +3241,9 @@ static int bnxt_re_post_send_shadow_qp(struct bnxt_re_dev *rdev, goto bad; } - payload_sz =bnxt_re_copy_wr_payload (qp->rdev, wr, &wqe); -if (payload_sz <0 ) { - rc = -EINVAL ; + rc =bnxt_re_copy_wr_payload (qp->rdev, wr, &wqe, &payload_sz); +if (rc) goto bad; - } wqe.wr_id = wr->wr_id; wqe.type =BNXT_QPLIB_SWQE_TYPE_SEND ;
@@ -3279,7 +3284,8 @@ int bnxt_re_post_send(struct ib_qp *ib_qp, const struct ib_send_wr *wr, { struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);struct bnxt_qplib_swqe wqe; -int rc =0 , payload_sz =0 ; +int rc =0 ; +u32 payload_sz =0 ;unsigned long flags;spin_lock_irqsave (&qp->sq_lock, flags);
@@ -3296,11 +3302,9 @@ int bnxt_re_post_send(struct ib_qp *ib_qp, const struct ib_send_wr *wr, goto bad; } - payload_sz =bnxt_re_copy_wr_payload (qp->rdev, wr, &wqe); -if (payload_sz <0 ) { - rc = -EINVAL ; + rc =bnxt_re_copy_wr_payload (qp->rdev, wr, &wqe, &payload_sz); +if (rc) goto bad; - } wqe.wr_id = wr->wr_id;switch (wr->opcode) {
Signature · 1 line
2.39.3
[email protected]
AI reviewRe: [PATCH for-rc v3 1/8] RDMA/bnxt_re: Fix integer overflow in send payload size computation
Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 7 lines
bnxt_re_build_sgl()summed SGE lengths into a signed int, which could overflow. Make it return u32, and havebnxt_re_copy_wr_payload()report the size via a u32 out-parameter with the return value carrying only the error status, instead of overloading a signed int with both.Fixes 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")Signed-off-by Selvin Xavier <[email protected]>
Sashiko has reviewed this patch and found no issues. It looks great!
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Selvin Xavier
maintainerPATCH v3 2/8 RDMA/bnxt_re: Detect wrong sge_len passed for inline
In reply to Selvin XavierAvoid handling wrong sge_len by adding extra check to see if the passed length is more than the inline size supported.
--- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/infiniband/hw/bnxt_re/ib_verbs.c +3 -2
@@ -3169,8 +3169,9 @@ static int bnxt_re_copy_inline_data(struct bnxt_re_dev *rdev, wr->sg_list[i].addr; sge_len = wr->sg_list[i].length; -if ((sge_len + wqe->inline_len) > -BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH ) { +if (sge_len >BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH || + ((sge_len + wqe->inline_len) > +BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH )) {ibdev_err (&rdev->ibdev,"Inline data size requested > supported value" );return -EINVAL ;
Signature · 1 line
2.39.3
[email protected]
AI reviewRe: [PATCH for-rc v3 2/8] RDMA/bnxt_re: Detect wrong sge_len passed for inline
Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 7 lines
Avoid handling wrong
sge_lenby adding extra check to see if the passed length is more than the inline size supported.Fixes 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")Reviewed-by Kalesh AP <[email protected]>Signed-off-by Selvin Xavier <[email protected]>
Sashiko has reviewed this patch and found no issues. It looks great!
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2
Selvin Xavier
maintainerPATCH v3 3/8 RDMA/bnxt_re: Validate SRQ max_sge at create time
In reply to Selvin Xavierbnxt_re_create_srq() stored attr.max_sge into srq->qplib_srq.max_sge unvalidated, which defeats the num_sge check in bnxt_re_post_srq_recv() since that check compares against this same attacker-chosen value. Reject max_sge > dev_attr->max_srq_sges at create time, and clamp dev_attr->max_srq_sges to BNXT_STATIC_MAX_SGE since SRQ WQEs use the fixed-size SGE array.
--- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 8 ++++++++ drivers/infiniband/hw/bnxt_re/qplib_sp.c | 2 +- 2 files changed, 9 insertions(+), 1 deletion(-)
drivers/infiniband/hw/bnxt_re/ib_verbs.c +8 -0
@@ -2236,6 +2236,14 @@ int bnxt_re_create_srq(struct ib_srq *ib_srq, goto exit; } +if (srq_init_attr->attr.max_sge > dev_attr->max_srq_sges) { +ibdev_err (&rdev->ibdev, +"Create SRQ failed - max_sge %d exceeds supported %d" , + srq_init_attr->attr.max_sge, dev_attr->max_srq_sges); + rc = -EINVAL ; + goto exit; + } +if (srq_init_attr->srq_type !=IB_SRQT_BASIC ) { rc = -EOPNOTSUPP ; goto exit;
drivers/infiniband/hw/bnxt_re/qplib_sp.c +1 -1
@@ -160,7 +160,7 @@ int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw) attr->max_srq =le16_to_cpu (sb->max_srq); attr->max_srq_wqes =le32_to_cpu (sb->max_srq_wr) -1 ; - attr->max_srq_sges = sb->max_srq_sge; + attr->max_srq_sges =min_t (u32, sb->max_srq_sge,BNXT_STATIC_MAX_SGE ); attr->max_pkey =1 ; attr->max_inline_data = attr->max_qp_sges *sizeof (struct sq_sge );if (!bnxt_qplib_is_chip_gen_p7 (rcfw->res->cctx))
Signature · 1 line
2.39.3
[email protected]
AI reviewRe: [PATCH for-rc v3 3/8] RDMA/bnxt_re: Validate SRQ max_sge at create time
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Low] bnxt_re: format string mismatch prints u32 as signed --
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] bnxt_re: format string mismatch prints u32 as signed
--
commit c1caa6b4fbeb062c8f585d11a5ad5310dce0f434
Author: Selvin Xavier <[email protected]>
RDMA/bnxt_re: Validate SRQ max_sge at create time
bnxt_re_create_srq() stored attr.max_sge into srq->qplib_srq.max_sge unvalidated, which defeats the num_sge check in bnxt_re_post_srq_recv() since that check compares against this same attacker-chosen value. Reject max_sge > dev_attr->max_srq_sges at create time, and clamp dev_attr->max_srq_sges to BNXT_STATIC_MAX_SGE since SRQ WQEs use the fixed-size SGE array.
drivers/infiniband/hw/bnxt_re/ib_verbs.c int bnxt_re_create_srq(struct ib_srq *ib_srq, · line 2242
Quoted 12 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c +4
-0@@ -2236,6 +2236,14 @@ int bnxt_re_create_srq(struct ib_srq *ib_srq, goto exit; } +if (srq_init_attr->attr.max_sge > dev_attr->max_srq_sges) { +ibdev_err (&rdev->ibdev, +"Create SRQ failed - max_sge %d exceeds supported %d" , + srq_init_attr->attr.max_sge, dev_attr->max_srq_sges);
[Severity: Low]
Will this print confusing error logs when an invalid size is passed?
Since both srq_init_attr->attr.max_sge and dev_attr->max_srq_sges are u32, the %d format specifier will interpret them as signed integers. This means large invalid values (like those >= 0x80000000) will be logged as negative numbers rather than their actual unsigned magnitude. Should these be printed with %u instead?
Quoted 7 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c +4
-0+ rc = -EINVAL ; + goto exit; + } +if (srq_init_attr->srq_type !=IB_SRQT_BASIC ) { rc = -EOPNOTSUPP ; goto exit;
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3
Selvin Xavier
maintainerPATCH v3 4/8 RDMA/bnxt_re: Initialize wqe.flags in bnxt_re_post_srq_recv()
In reply to Selvin Xavierwqe is declared once above the WR loop and never zeroed, so wqe.flags holds indeterminate stack contents on the first WR and the previous WR's stale value thereafter. This propagates straight into the hardware-visible srqe->flags in bnxt_qplib_post_srq_recv().
Most values of the wqe struct is set except flags. So initialize flags also.
--- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 1 + 1 file changed, 1 insertion(+)
drivers/infiniband/hw/bnxt_re/ib_verbs.c +1 -0
@@ -2397,6 +2397,7 @@ int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr,spin_lock_irqsave (&srq->lock, flags);while (wr) {/* Transcribe each ib_recv_wr to qplib_swqe */ + wqe.flags =0 ; wqe.num_sge = wr->num_sge;bnxt_re_build_sgl (wr->sg_list, wqe.sg_list, wr->num_sge); wqe.wr_id = wr->wr_id;
Signature · 1 line
2.39.3
[email protected]
AI reviewRe: [PATCH for-rc v3 4/8] RDMA/bnxt_re: Initialize wqe.flags in bnxt_re_post_srq_recv()
Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 10 lines
wqe is declared once above the WR loop and never zeroed, so wqe.flags holds indeterminate stack contents on the first WR and the previous WR's stale value thereafter. This propagates straight into the hardware-visible
srqe->flagsinbnxt_qplib_post_srq_recv().Most values of the wqe struct is set except flags. So initialize flags also.
Fixes 37cb11acf1f7 ("RDMA/bnxt_re: Add SRQ support for Broadcom adapters")Signed-off-by Selvin Xavier <[email protected]>
Sashiko has reviewed this patch and found no issues. It looks great!
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4
Selvin Xavier
maintainerPATCH v3 5/8 RDMA/bnxt_re: Validate num_sge in bnxt_re_post_srq_recv()
In reply to Selvin XavierAdd a check if max sge for SRQ is exceeding the device maximum.
--- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 7 +++++++ 1 file changed, 7 insertions(+)
drivers/infiniband/hw/bnxt_re/ib_verbs.c +7 -0
@@ -2399,6 +2399,13 @@ int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr,/* Transcribe each ib_recv_wr to qplib_swqe */ wqe.flags =0 ; wqe.num_sge = wr->num_sge; +if (wr->num_sge > srq->qplib_srq.max_sge) { +ibdev_err (&srq->rdev->ibdev, +"Limit exceeded for SRQ Receive SGEs" ); + rc = -EINVAL ; + *bad_wr = wr; +break ; + }bnxt_re_build_sgl (wr->sg_list, wqe.sg_list, wr->num_sge); wqe.wr_id = wr->wr_id; wqe.type =BNXT_QPLIB_SWQE_TYPE_RECV ;
Signature · 1 line
2.39.3
[email protected]
AI reviewRe: [PATCH for-rc v3 5/8] RDMA/bnxt_re: Validate num_sge in bnxt_re_post_srq_recv()
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] rdma/bnxt_re: un-ratelimited error log in spinlock allows DoS --
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] rdma/bnxt_re: un-ratelimited error log in spinlock allows DoS
--
commit bb7479cd7b3a22e9b0e6b2862ab2b9133790528b
Author: Selvin Xavier <[email protected]>
RDMA/bnxt_re: Validate num_sge in bnxt_re_post_srq_recv()
Add a check if max sge for SRQ is exceeding the device maximum.
drivers/infiniband/hw/bnxt_re/ib_verbs.c int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr, · line 2404
Quoted 11 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c +3
-0@@ -2399,6 +2399,13 @@ int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr,/* Transcribe each ib_recv_wr to qplib_swqe */ wqe.flags =0 ; wqe.num_sge = wr->num_sge; +if (wr->num_sge > srq->qplib_srq.max_sge) { +ibdev_err (&srq->rdev->ibdev, +"Limit exceeded for SRQ Receive SGEs" );
[Severity: High]
Is it safe to use an un-ratelimited ibdev_err() here while holding the srq->lock spinlock with interrupts disabled in bnxt_re_post_srq_recv()?
Since an unprivileged user can repeatedly invoke the RDMA uverbs API with a wr->num_sge larger than srq->qplib_srq.max_sge, could this regression allow a local user to flood the kernel log buffer and potentially cause soft lockups?
Quoted 7 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c +4
-0+ rc = -EINVAL ; + *bad_wr = wr; +break ; + }bnxt_re_build_sgl (wr->sg_list, wqe.sg_list, wr->num_sge); wqe.wr_id = wr->wr_id; wqe.type =BNXT_QPLIB_SWQE_TYPE_RECV ;
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=5
Selvin Xavier
maintainerRe: [PATCH for-rc v3 5/8] RDMA/bnxt_re: Validate num_sge in bnxt_re_post_srq_recv()
Quoted 28 lines · On Tue, Sep 22, 2026 at 6:37 AM <[email protected]> wrote:
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] rdma/bnxt_re: un-ratelimited error log in spinlock allows DoS
--commit
bb7479cd7b3a22e9b0e6b2862ab2b9133790528b
Author: Selvin Xavier <[email protected]>RDMA/bnxt_re: Validate
num_sgeinbnxt_re_post_srq_recv()Add a check if max sge for SRQ is exceeding the device maximum.
drivers/infiniband/hw/bnxt_re/ib_verbs.c int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr, · line 2404
Quoted 11 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c +3
-0@@ -2399,6 +2399,13 @@ int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr,/* Transcribe each ib_recv_wr to qplib_swqe */ wqe.flags =0 ; wqe.num_sge = wr->num_sge; +if (wr->num_sge > srq->qplib_srq.max_sge) { +ibdev_err (&srq->rdev->ibdev, +"Limit exceeded for SRQ Receive SGEs" );[Severity: High]
Is it safe to use an un-ratelimitedibdev_err()here while holding thesrq->lockspinlock with interrupts disabled inbnxt_re_post_srq_recv()?
Quoted 3 lines
Since an unprivileged user can repeatedly invoke the RDMA uverbs API with a
wr->num_sgelarger thansrq->qplib_srq.max_sge, could this regression allow a local user to flood the kernel log buffer and potentially cause soft lockups?
It is possible to reach the kernel driver using an under privilaged
user, but there is no support in rdma-core/libbnxt_re to call ibv_cmd_post_srq_recv and this code path will not be reached as of now. So I think this can be ignored.
Quoted 11 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr, · line 2411
Quoted 8 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c +4
-0+ rc = -EINVAL ; + *bad_wr = wr; +break ; + }bnxt_re_build_sgl (wr->sg_list, wqe.sg_list, wr->num_sge); wqe.wr_id = wr->wr_id; wqe.type =BNXT_QPLIB_SWQE_TYPE_RECV ;--
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=5
Re: [PATCH for-rc v3 5/8] RDMA/bnxt_re: Validate num_sge in bnxt_re_post_srq_recv()
Quoted 29 lines · Selvin Xavier
Quoted 28 lines · On Tue, Sep 22, 2026 at 6:37 AM <[email protected]> wrote:
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] rdma/bnxt_re: un-ratelimited error log in spinlock allows DoS
--commit
bb7479cd7b3a22e9b0e6b2862ab2b9133790528b
Author: Selvin Xavier <[email protected]>RDMA/bnxt_re: Validate
num_sgeinbnxt_re_post_srq_recv()Add a check if max sge for SRQ is exceeding the device maximum.
drivers/infiniband/hw/bnxt_re/ib_verbs.c int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr, · line 2404
Quoted 11 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c +3
-0@@ -2399,6 +2399,13 @@ int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr,/* Transcribe each ib_recv_wr to qplib_swqe */ wqe.flags =0 ; wqe.num_sge = wr->num_sge; +if (wr->num_sge > srq->qplib_srq.max_sge) { +ibdev_err (&srq->rdev->ibdev, +"Limit exceeded for SRQ Receive SGEs" );[Severity: High]
Is it safe to use an un-ratelimitedibdev_err()here while holding thesrq->lockspinlock with interrupts disabled inbnxt_re_post_srq_recv()?
Quoted 7 lines
Quoted 3 lines
Since an unprivileged user can repeatedly invoke the RDMA uverbs API with a
wr->num_sgelarger thansrq->qplib_srq.max_sge, could this regression allow a local user to flood the kernel log buffer and potentially cause soft lockups?It is possible to reach the kernel driver using an under privilaged
user, but there is no support in rdma-core/libbnxt_re to callibv_cmd_post_srq_recvand this code path will not be reached as of now. So I think this can be ignored.
We cannot assume that rdma-core is trusted. Anyone can build it locally and use it to interact with the kernel. In addition, syzkaller uses our UAPI structures and commands directly, without involving rdma-core.
Thanks
Quoted 11 lines
Quoted 11 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr, · line 2411
Quoted 8 lines
drivers/infiniband/hw/bnxt_re/ib_verbs.c +4
-0+ rc = -EINVAL ; + *bad_wr = wr; +break ; + }bnxt_re_build_sgl (wr->sg_list, wqe.sg_list, wr->num_sge); wqe.wr_id = wr->wr_id; wqe.type =BNXT_QPLIB_SWQE_TYPE_RECV ;--
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=5
Selvin Xavier
maintainerPATCH v3 6/8 RDMA/bnxt_re: Serialize dcb_wq access against async notifier
In reply to Selvin Xavierbnxt_re_async_notifier() runs in NAPI/softirq context via bnxt_en's RCU-protected ULP ops dispatch, never under rtnl_lock, so it isn't serialized against bnxt_re_uninit_dcb_wq() destroying rdev->dcb_wq. A concurrent notifier call can queue_work() on a workqueue that is being, or has just been, destroyed.
Add a spinlock scoped to dcb_wq: the notifier takes it before checking dcb_wq and queuing work, and bnxt_re_uninit_dcb_wq() takes it to atomically clear dcb_wq before destroying it. Initialize the lock at rdev allocation so it is valid on every teardown path, including bnxt_re_dev_init()'s early failure labels.
--- drivers/infiniband/hw/bnxt_re/bnxt_re.h | 4 ++++ drivers/infiniband/hw/bnxt_re/main.c | 24 ++++++++++++++++++++---- 2 files changed, 24 insertions(+), 4 deletions(-)
drivers/infiniband/hw/bnxt_re/bnxt_re.h +4 -0
@@ -216,6 +216,10 @@ struct bnxt_re_dev {unsigned long event_bitmap;struct bnxt_qplib_cc_param cc_param;struct workqueue_struct *dcb_wq; +/* Protects dcb_wq against bnxt_re_uninit_dcb_wq() destroying it +* concurrently with bnxt_re_async_notifier() queuing work on it. +*/ +spinlock_t dcb_lock;struct dentry *cc_config;struct bnxt_re_dbg_cc_config_params *cc_config_params;struct dentry *cq_coal_cfg;
drivers/infiniband/hw/bnxt_re/main.c +20 -4
@@ -367,9 +367,15 @@ static int bnxt_re_init_dcb_wq(struct bnxt_re_dev *rdev)static void bnxt_re_uninit_dcb_wq (struct bnxt_re_dev *rdev) { -if (!rdev->dcb_wq) -return ; -destroy_workqueue (rdev->dcb_wq); +struct workqueue_struct *dcb_wq; + +spin_lock_bh (&rdev->dcb_lock); + dcb_wq = rdev->dcb_wq; + rdev->dcb_wq =NULL ; +spin_unlock_bh (&rdev->dcb_lock); + +if (dcb_wq) +destroy_workqueue (dcb_wq); }static void bnxt_re_dcb_wq_task (struct work_struct *work)
@@ -424,14 +430,23 @@ static void bnxt_re_async_notifier(void *handle, struct hwrm_async_event_cmpl *cswitch (event_id) {case ASYNC_EVENT_CMPL_EVENT_ID_DCB_CONFIG_CHANGE : +spin_lock (&rdev->dcb_lock); +if (!rdev->dcb_wq) { +spin_unlock (&rdev->dcb_lock); +break ; + } + dcb_work =kzalloc_obj (*dcb_work,GFP_ATOMIC ); -if (!dcb_work) +if (!dcb_work) { +spin_unlock (&rdev->dcb_lock);break ; + } dcb_work->rdev = rdev;memcpy (&dcb_work->cmpl, cmpl,sizeof (*cmpl));INIT_WORK (&dcb_work->work, bnxt_re_dcb_wq_task);queue_work (rdev->dcb_wq, &dcb_work->work); +spin_unlock (&rdev->dcb_lock);break ;default :break ;
@@ -1448,6 +1463,7 @@ static struct bnxt_re_dev *bnxt_re_dev_add(struct auxiliary_device *adev,INIT_LIST_HEAD (&rdev->qp_list);mutex_init (&rdev->qp_lock);mutex_init (&rdev->pacing.dbq_lock); +spin_lock_init (&rdev->dcb_lock);atomic_set (&rdev->stats.res.qp_count,0 );atomic_set (&rdev->stats.res.cq_count,0 );atomic_set (&rdev->stats.res.srq_count,0 );
Signature · 1 line
2.39.3
[email protected]
AI reviewRe: [PATCH for-rc v3 6/8] RDMA/bnxt_re: Serialize dcb_wq access against async notifier
Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 6 lines
bnxt_re_async_notifier()runs in NAPI/softirq context via bnxt_en's RCU-protected ULP ops dispatch, never underrtnl_lock, so it isn't serialized againstbnxt_re_uninit_dcb_wq()destroyingrdev->dcb_wq. A concurrent notifier call canqueue_work()on a workqueue that is being, or has just been, destroyed.
Quoted 8 lines
Add a spinlock scoped to
dcb_wq: the notifier takes it before checkingdcb_wqand queuing work, andbnxt_re_uninit_dcb_wq()takes it to atomically cleardcb_wqbefore destroying it. Initialize the lock at rdev allocation so it is valid on every teardown path, including bnxt_re_dev_init()'s early failure labels.Fixes 51dc5312dcd9 ("RDMA/bnxt_re: Add support to handle DCB_CONFIG_CHANGE event")Signed-off-by Selvin Xavier <[email protected]>
Sashiko has reviewed this patch and found no issues. It looks great!
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=6
Selvin Xavier
maintainerPATCH v3 7/8 RDMA/bnxt_re: Fix the PD and DPI table size
In reply to Selvin XavierThe PD and DPI bitmaps were sized as max >> 3 (bytes), but bitmap ops (set_bit(), clear_bit(), find_first_bit(), test_and_set_bit()) operate on whole unsigned long words, so whenever max isn't a multiple of BITS_PER_LONG,
the buffer under-allocates and the top word's bitops read/write past the end of the kmalloc()'d buffer. Most exposed on the DPI table, since dpit->max comes from the firmware-reported dev_attr->max_dpi with no alignment guarantee.
Fix the size of both allocations withBITS_TO_LONGS(max) * sizeof(unsigned long).
--- drivers/infiniband/hw/bnxt_re/qplib_res.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-)
drivers/infiniband/hw/bnxt_re/qplib_res.c +5 -4
@@ -45,6 +45,7 @@#include <linux/dma-mapping.h> #include <linux/if_vlan.h> #include <linux/vmalloc.h> +#include <linux/bitops.h> #include <rdma/ib_verbs.h> #include <rdma/iter.h>
@@ -668,9 +669,9 @@ static int bnxt_qplib_alloc_pd_tbl(struct bnxt_qplib_res *res, {u32 bytes; - bytes = max >>3 ; + bytes =BITS_TO_LONGS (max) *sizeof (unsigned long );if (!bytes) - bytes =1 ; + bytes =sizeof (unsigned long ); pdt->tbl =kmalloc (bytes,GFP_KERNEL );if (!pdt->tbl)return -ENOMEM ;
@@ -848,9 +849,9 @@ static int bnxt_qplib_alloc_dpi_tbl(struct bnxt_qplib_res *res,if (!dpit->app_tbl)return -ENOMEM ; - bytes = dpit->max >>3 ; + bytes =BITS_TO_LONGS (dpit->max) *sizeof (unsigned long );if (!bytes) - bytes =1 ; + bytes =sizeof (unsigned long ); dpit->tbl =kmalloc (bytes,GFP_KERNEL );if (!dpit->tbl) {
Signature · 1 line
2.39.3
[email protected]
AI reviewRe: [PATCH for-rc v3 7/8] RDMA/bnxt_re: Fix the PD and DPI table size
Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 9 lines
The PD and DPI bitmaps were sized as max >> 3 (bytes), but bitmap ops (
set_bit(),clear_bit(),find_first_bit(),test_and_set_bit()) operate on whole unsigned long words, so whenever max isn't a multiple ofBITS_PER_LONG,
the buffer under-allocates and the top word's bitops read/write past the end of the kmalloc()'d buffer. Most exposed on the DPI table, sincedpit->maxcomes from the firmware-reporteddev_attr->max_dpiwith no alignment guarantee.
Quoted 5 lines
Fix the size of both allocations with
BITS_TO_LONGS(max)* sizeof(unsigned long).Fixes 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")Signed-off-by Selvin Xavier <[email protected]>
Sashiko has reviewed this patch and found no issues. It looks great!
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=7
Selvin Xavier
maintainerPATCH v3 8/8 RDMA/bnxt_re: Use bnxt_ext_stats_supported for counter count selection
In reply to Selvin XavierSelecting between BNXT_RE_NUM_EXT_COUNTERS and BNXT_RE_NUM_STD_COUNTERS relied only on bnxt_qplib_is_chip_gen_p5_p7(), which doesn't account for the extended stats capability flag or the PF/VF restriction. Use bnxt_ext_stats_supported() instead, matching the check already used to populate the extended stats, so the counter count stays consistent with what gets filled in.
--- drivers/infiniband/hw/bnxt_re/hw_counters.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/infiniband/hw/bnxt_re/hw_counters.c +4 -2
@@ -413,7 +413,8 @@ int bnxt_re_ib_get_hw_stats(struct ib_device *ibdev, } done: -return bnxt_qplib_is_chip_gen_p5_p7 (rdev->chip_ctx) ? +return bnxt_ext_stats_supported (rdev->chip_ctx, rdev->dev_attr->dev_cap_flags, + rdev->is_virtfn) ?BNXT_RE_NUM_EXT_COUNTERS :BNXT_RE_NUM_STD_COUNTERS ; }
@@ -423,7 +424,8 @@ struct rdma_hw_stats *bnxt_re_ib_alloc_hw_port_stats(struct ib_device *ibdev,struct bnxt_re_dev *rdev =to_bnxt_re_dev (ibdev, ibdev);int num_counters =0 ; -if (bnxt_qplib_is_chip_gen_p5_p7 (rdev->chip_ctx)) +if (bnxt_ext_stats_supported (rdev->chip_ctx, rdev->dev_attr->dev_cap_flags, + rdev->is_virtfn)) num_counters =BNXT_RE_NUM_EXT_COUNTERS ;else num_counters =BNXT_RE_NUM_STD_COUNTERS ;
Signature · 1 line
2.39.3
[email protected]
AI reviewRe: [PATCH for-rc v3 8/8] RDMA/bnxt_re: Use bnxt_ext_stats_supported for counter count selection
Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 11 lines
Selecting between
BNXT_RE_NUM_EXT_COUNTERSandBNXT_RE_NUM_STD_COUNTERSrelied only onbnxt_qplib_is_chip_gen_p5_p7(), which doesn't account for the extended stats capability flag or the PF/VF restriction. Usebnxt_ext_stats_supported()instead, matching the check already used to populate the extended stats, so the counter count stays consistent with what gets filled in.Fixes 8238c7bd8420 ("RDMA/bnxt_re: Fix the statistics for Gen P7 VF")Reviewed-by Kalesh AP <[email protected]>Signed-off-by Selvin Xavier <[email protected]>
Sashiko has reviewed this patch and found no issues. It looks great!
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=8
Re: (subset) [PATCH for-rc v3 0/8] RDMA/bnxt_re: Fix input validation and
In reply to Selvin XavierQuoted 9 lines · Selvin Xavier
This series fixes some of the bugs found while auditing bnxt_re's handling of user-supplied values and some generic bug fixes
Please review and apply the series.
Thanks,
Selvin Xavier[...]
Applied, thanks!
[1/8] RDMA/bnxt_re: Fix integer overflow in send payload size computation
https://git.kernel.org/rdma/rdma/c/2cb09585a33dd5 [2/8] RDMA/bnxt_re: Detect wrong sge_len passed for inline https://git.kernel.org/rdma/rdma/c/30fb34c69bd3b6 [3/8] RDMA/bnxt_re: Validate SRQ max_sge at create time https://git.kernel.org/rdma/rdma/c/ebd9388a1180e3 [4/8] RDMA/bnxt_re: Initialize wqe.flags in bnxt_re_post_srq_recv() https://git.kernel.org/rdma/rdma/c/41c2d60d7fee40 [6/8] RDMA/bnxt_re: Serialize dcb_wq access against async notifier https://git.kernel.org/rdma/rdma/c/9254983623f7ff [7/8] RDMA/bnxt_re: Fix the PD and DPI table size https://git.kernel.org/rdma/rdma/c/a99861b65221f9 [8/8] RDMA/bnxt_re: Use bnxt_ext_stats_supported for counter count selection https://git.kernel.org/rdma/rdma/c/856bfaa78cce2c
Best regards,
Signature · 1 line
Leon Romanovsky <[email protected]>
Quoted 19 lines · Selvin Xavier
This series fixes some of the bugs found while auditing bnxt_re's handling of user-supplied values and some generic bug fixes
Please review and apply the series.
Thanks,
Selvin Xavierv2 -> v3:
- Drop theVM_MAYEXECclearing patch from this series. Will work on a core level series which implements this in other drivers also
- Dropped rdev lifetime race patch from v2 series as it is reporting few more review comments to be handled
- New fix to serializedcb_wqagainst async notifier
- Removedis_in_usedflags related patch to be posted as a separate rcfw series
v1 -> v2:
- Handle review comment about clearingVM_MAYEXECvma flag
Quoted 8 lines
Selvin Xavier (8):
RDMA/bnxt_re: Fix integer overflow in send payload size computation RDMA/bnxt_re: Detect wrongsge_lenpassed for inline RDMA/bnxt_re: Validate SRQmax_sgeat create time RDMA/bnxt_re: Initialize wqe.flags inbnxt_re_post_srq_recv()RDMA/bnxt_re: Serializedcb_wqaccess against async notifier RDMA/bnxt_re: Fix the PD and DPI table size
RDMA/bnxt_re: Usebnxt_ext_stats_supportedfor counter count selection
I have applied all the patches above.
Quoted 1 line
RDMA/bnxt_re: Validate
num_sgeinbnxt_re_post_srq_recv()
This needs some more work. It can be improved commit message, or code to make sashiko happy.
Thanks
Selvin Xavier
maintainerQuoted 30 lines · Leon Romanovsky
Quoted 19 lines · Selvin Xavier
This series fixes some of the bugs found while auditing bnxt_re's handling of user-supplied values and some generic bug fixes
Please review and apply the series.
Thanks,
Selvin Xavierv2 -> v3:
- Drop theVM_MAYEXECclearing patch from this series. Will work on a core level series which implements this in other drivers also
- Dropped rdev lifetime race patch from v2 series as it is reporting few more review comments to be handled
- New fix to serializedcb_wqagainst async notifier
- Removedis_in_usedflags related patch to be posted as a separate rcfw series
v1 -> v2:
- Handle review comment about clearingVM_MAYEXECvma flagQuoted 8 lines
Selvin Xavier (8):
RDMA/bnxt_re: Fix integer overflow in send payload size computation RDMA/bnxt_re: Detect wrongsge_lenpassed for inline RDMA/bnxt_re: Validate SRQmax_sgeat create time RDMA/bnxt_re: Initialize wqe.flags inbnxt_re_post_srq_recv()RDMA/bnxt_re: Serializedcb_wqaccess against async notifier RDMA/bnxt_re: Fix the PD and DPI table size
RDMA/bnxt_re: Usebnxt_ext_stats_supportedfor counter count selection
Quoted 8 lines
I have applied all the patches above.
Quoted 1 line
RDMA/bnxt_re: Validate
num_sgeinbnxt_re_post_srq_recv()This needs some more work. It can be improved commit message, or code to make sashiko happy.
Thanks
Sure. will repost it.
Leon Romanovsky
AppliedRe: (subset) [PATCH for-rc v3 0/8] RDMA/bnxt_re: Fix input validation and
In reply to Selvin XavierQuoted 9 lines · Selvin Xavier
This series fixes some of the bugs found while auditing bnxt_re's handling of user-supplied values and some generic bug fixes
Please review and apply the series.
Thanks,
Selvin Xavier[...]
Applied, thanks!
[1/8] RDMA/bnxt_re: Fix integer overflow in send payload size computation
https://git.kernel.org/rdma/rdma/c/3483cbbd748716 [2/8] RDMA/bnxt_re: Detect wrong sge_len passed for inline https://git.kernel.org/rdma/rdma/c/81551ece0864c3 [3/8] RDMA/bnxt_re: Validate SRQ max_sge at create time https://git.kernel.org/rdma/rdma/c/1894ed9a664d19 [4/8] RDMA/bnxt_re: Initialize wqe.flags in bnxt_re_post_srq_recv() https://git.kernel.org/rdma/rdma/c/4d8168a3b7c933 [6/8] RDMA/bnxt_re: Serialize dcb_wq access against async notifier https://git.kernel.org/rdma/rdma/c/81caf9240dfb08 [7/8] RDMA/bnxt_re: Fix the PD and DPI table size https://git.kernel.org/rdma/rdma/c/aa6c2ef821a6ba [8/8] RDMA/bnxt_re: Use bnxt_ext_stats_supported for counter count selection https://git.kernel.org/rdma/rdma/c/7cc8ccb157a476
Best regards,
Signature · 1 line
Leon Romanovsky <[email protected]>