From: Yishai Hadas <yishaih@nvidia.com>
To: Leon Romanovsky <leon@kernel.org>
Cc: <jgg@ziepe.ca>, <linux-rdma@vger.kernel.org>,
<netdev@vger.kernel.org>, <saeedm@nvidia.com>,
<tariqt@nvidia.com>, <mbloch@nvidia.com>, <kuba@kernel.org>,
<pabeni@redhat.com>, <edumazet@kernel.org>, <davem@davemloft.net>,
<andrew+netdev@lunn.ch>, <dtatulea@nvidia.com>,
<moshe@nvidia.com>, <maorg@nvidia.com>
Subject: Re: [PATCH V1 rdma-next 22/24] RDMA/mlx5: Clamp log_max_* shifts in query_device and ucontext paths
Date: Mon, 16 Nov 2026 19:26:01 +0200 [thread overview]
Message-ID: <a94c46ea-42f0-4b92-a880-daff9fb918fb@nvidia.com> (raw)
In-Reply-To: <20261116090653.GD11319@unreal>
On 16/11/2026 11:06, Leon Romanovsky wrote:
> On Tue, Nov 10, 2026 at 03:43:18PM +0200, Yishai Hadas wrote:
>> The device, port and ucontext attributes and the SRQ size limit are
>> computed as 1 << cap from log_max_* FW capabilities. An untrusted NIC
>> can report 31 or more, which makes the shift undefined. The same
>> applies to 1ull << log_pg_sz with a value of 64 or more, and to the
>> max_tso computed as 1 << max_lso_cap, which can be 31.
>>
>> Use MLX5_CAP_GEN_LOGSZ_U32() and MLX5_CAP_GEN_LOGSZ_U64() for these
>> values, and BIT() for max_tso. As the limits can now reach 2^31,
>> compute max_qp_wr in 64 bits and saturate the max_res_rd_atom and
>> max_total_mcast_qp_attach products at U32_MAX. Cap max_cqe at 2^30 - 1,
>> the largest CQ size the create path accepts.
>>
>> The SRQ size limit then lets the rounded up SRQ size overflow the int
>> srq->msrq.max, and the SRQ buffer size exceed the int passed to
>> create_srq_user() and create_srq_kernel(). Reject both above INT_MAX,
>> computing the buffer size with check_mul_overflow().
>>
>> Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
>> ---
>> drivers/infiniband/hw/mlx5/mad.c | 2 +-
>> drivers/infiniband/hw/mlx5/main.c | 65 +++++++++++++++++--------------
>> drivers/infiniband/hw/mlx5/srq.c | 14 +++++--
>> 3 files changed, 46 insertions(+), 35 deletions(-)
>
> <...>
>
>> + /* The CQ size is a power of two that fits in an int */
>> + props->max_cqe = min_t(u32, MLX5_CAP_GEN_LOGSZ_U32(mdev, log_max_cq_sz),
>> + rounddown_pow_of_two(INT_MAX)) - 1;
>
> <...>
>
>> + props->max_res_rd_atom = min_t(u64, (u64)props->max_qp_rd_atom * props->max_qp,
>> + U32_MAX);
>
> Avoid min_t() where possible: it silently casts operands to the specified
> type instead of checking that their types match.
OK, in V2 there will be no min_t(). The products will use
check_mul_overflow() on the u32 fields, saturating at U32_MAX, and
max_cqe will use min().
>
> <...>
>
>> + props->max_total_mcast_qp_attach =
>> + min_t(u64, (u64)props->max_mcast_qp_attach * props->max_mcast_grp,
>> + U32_MAX);
>
> <...>
>
>> + /* srq->msrq.max is an int */
>> + max = roundup_pow_of_two(init_attr->attr.max_wr + 1);
>> + if (max > INT_MAX)
>
> Please fix types instead of doing this check.
OK, in V2 srq->msrq.max will be a u32, so the rounded up size always
fits and this check goes away.
>
>> + return -EINVAL;
>> + srq->msrq.max = max;
>> srq->msrq.max_gs = init_attr->attr.max_sge;
>>
>> desc_size = sizeof(struct mlx5_wqe_srq_next_seg) +
>> @@ -238,8 +243,9 @@ int mlx5_ib_create_srq(struct ib_srq *ib_srq,
>> srq->msrq.max_avail_gather = (desc_size - sizeof(struct mlx5_wqe_srq_next_seg)) /
>> sizeof(struct mlx5_wqe_data_seg);
>> srq->msrq.wqe_shift = ilog2(desc_size);
>> - buf_size = srq->msrq.max * desc_size;
>> - if (buf_size < desc_size)
>> + /* create_srq_user() and create_srq_kernel() take an int */
>> + if (check_mul_overflow((size_t)srq->msrq.max, desc_size, &buf_size) ||
>> + buf_size > INT_MAX)
>
> Same
The size ends as an int in mlx5_frag_buf_alloc_node(), shared by all
mlx5 users, so fixing the types in mlx5_ib would only move this check
there. So we kept it where the size is computed.
Yishai
next prev parent reply other threads:[~2026-11-16 17:26 UTC|newest]
Thread overview: 37+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-10 13:42 [PATCH V1 rdma-next 00/24] mlx5: Adversarial mode for confidential computing guests Yishai Hadas
2026-11-10 13:42 ` [PATCH V1 rdma-next 01/24] net/mlx5: Add mlx5_trigger_health_work_force() Yishai Hadas
2026-11-10 13:42 ` [PATCH V1 rdma-next 02/24] net/mlx5: Derive adversarial mode from CC guest memory encryption Yishai Hadas
2026-11-16 9:45 ` Leon Romanovsky
2026-11-10 13:42 ` [PATCH V1 rdma-next 03/24] net/mlx5: Add capability filtering in adversarial mode Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 04/24] net/mlx5: Catch too low max number of EQs Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 05/24] net/mlx5: Add macros for reading log-based caps Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 06/24] RDMA/mlx5: Reject QP transport types not enabled by FW capability Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 07/24] RDMA/mlx5: Require IB_QPT_DRIVER for the DC vendor QP types Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 08/24] RDMA/mlx5: Derive mlx5_ib_adversarial_mode global from mlx5_core adversarial API Yishai Hadas
2026-11-16 8:57 ` Leon Romanovsky
2026-11-16 14:12 ` Yishai Hadas
2026-11-16 14:21 ` Leon Romanovsky
2026-11-10 13:43 ` [PATCH V1 rdma-next 09/24] RDMA/core: Suppress kverbs_provider for adversarial_mode devices Yishai Hadas
2026-11-16 8:58 ` Leon Romanovsky
2026-11-16 14:14 ` Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 10/24] RDMA/mlx5: Remove duplicate mlx5_ib_umr_context struct Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 11/24] RDMA/mlx5: Introduce per-CQ poll_one hook in mlx5_ib_cq Yishai Hadas
2026-11-16 9:01 ` Leon Romanovsky
2026-11-16 15:57 ` Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 12/24] RDMA/mlx5: Add shadow FIFO infrastructure for UMR completions Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 13/24] RDMA/mlx5: Verify UMR CQ polling against a hostile NIC Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 14/24] RDMA/mlx5: Keep the UMR QP failed once a forged CQE is detected Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 15/24] RDMA/mlx5: Block generic CQ poll path in adversarial mode Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 16/24] RDMA/core: Set the CQ comp_handler after its poll context is set up Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 17/24] RDMA/mlx5: Ignore completion events for a kernel CQ that is not ready Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 18/24] RDMA/mlx5: Validate EQE port and GSI port before indexing devr.ports[] Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 19/24] RDMA/mlx5: Validate firmware RQ/SQ state before sqrq_trans[] lookup Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 20/24] RDMA/mlx5: Clamp log_max_cq_sz shift in CQ size checks Yishai Hadas
2026-11-16 9:35 ` Leon Romanovsky
2026-11-16 16:28 ` Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 21/24] RDMA/mlx5: Clamp log_max_* shifts in QP and WQ limit checks Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 22/24] RDMA/mlx5: Clamp log_max_* shifts in query_device and ucontext paths Yishai Hadas
2026-11-16 9:06 ` Leon Romanovsky
2026-11-16 17:26 ` Yishai Hadas [this message]
2026-11-10 13:43 ` [PATCH V1 rdma-next 23/24] RDMA/core: Report and enforce adversarial mode via GET_CONTEXT Yishai Hadas
2026-11-10 13:43 ` [PATCH V1 rdma-next 24/24] RDMA/mlx5: Advertise adversarial mode via ib_device::adversarial_mode Yishai Hadas
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=a94c46ea-42f0-4b92-a880-daff9fb918fb@nvidia.com \
--to=yishaih@nvidia.com \
--cc=andrew+netdev@lunn.ch \
--cc=davem@davemloft.net \
--cc=dtatulea@nvidia.com \
--cc=edumazet@kernel.org \
--cc=jgg@ziepe.ca \
--cc=kuba@kernel.org \
--cc=leon@kernel.org \
--cc=linux-rdma@vger.kernel.org \
--cc=maorg@nvidia.com \
--cc=mbloch@nvidia.com \
--cc=moshe@nvidia.com \
--cc=netdev@vger.kernel.org \
--cc=pabeni@redhat.com \
--cc=saeedm@nvidia.com \
--cc=tariqt@nvidia.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).