linux.git/include/linux/mlx5, branch master

net/mlx5: SD, prefer sd_group_size from vport context

2026-08-14T19:27:12+00:00

Newer FW reports the SD group size directly in the NIC vport context
via the sd_group_size field, gated by the sd_group_size capability.
Switch sd_init() to source the group size from there and fall back to
the MPIR-based host_buses query only when the cap is absent.
sd_group_size might return 1 in some FW configuration. Add explicit
check to disable SD creation in this case.

While here, rename host_buses to group_size throughout sd.c to follow
the new name on capable FW.

Signed-off-by: Shay Drory 
Reviewed-by: Moshe Shemesh 
Signed-off-by: Tariq Toukan 
Reviewed-by: Simon Horman 
Link: https://patch.msgid.link/20260810093037.3138197-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski

net/mlx5: add debugfs stats for doorbell dma pools

2026-08-06T02:02:15+00:00

Add a debugfs file exposing per-node DMA pool usage for doorbell
allocations.

  # cat /sys/kernel/debug/mlx5//db_dma_pools
  node  block_size  used_blocks  allocated_blocks
     0          64            0                 0
     1          64            0                 0

Signed-off-by: Nimrod Oren 
Signed-off-by: Tariq Toukan 
Link: https://patch.msgid.link/20260803132520.2891860-4-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski

net/mlx5: allocate doorbells from dma pools

2026-08-06T02:02:15+00:00

Allocate doorbells from dma pools instead of the pgdir allocator.
Doorbell records remain cache-line sized coherent DMA allocations, but
their sub-allocation is now handled by the common mlx5 DMA pool
infrastructure.

This also makes doorbell allocation honor the requested NUMA node when
reusing existing backing pages. The old pgdir allocator used the
requested node only when allocating a new pgdir page; later
allocations scanned one global pgdir list and could take any pgdir with
a free entry, even if that page had been allocated for a different
NUMA node. Selecting the per-node DMA pool before sub-allocation keeps
reused doorbell records on pages allocated for the requested node.

Signed-off-by: Nimrod Oren 
Reviewed-by: Dragos Tatulea 
Signed-off-by: Tariq Toukan 
Link: https://patch.msgid.link/20260803132520.2891860-3-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski

net/mlx5: initialize doorbell dma pools

2026-08-06T02:02:15+00:00

Add per-node doorbell dma pool creation and cleanup to mdev lifecycle.

Signed-off-by: Nimrod Oren 
Reviewed-by: Dragos Tatulea 
Signed-off-by: Tariq Toukan 
Link: https://patch.msgid.link/20260803132520.2891860-2-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski

Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net

2026-07-23T21:07:40+00:00

Cross-merge networking fixes after downstream PR (net-7.2-rc5).

Conflicts:

drivers/net/amt.c
  3656a79f94c47 ("amt: re-read skb header pointers after every pull")
  586c4dcf28eb6 ("amt: no longer rely on RTNL in amt_fill_info()")
https://lore.kernel.org/amIaJr3aOQNS_Fvl@sirena.org.uk

Adjacent changes:

drivers/net/geneve.c
  8efb8f8bbb35 ("geneve: require CAP_NET_ADMIN in the device netns for changelink")
  0ba269933f73 ("geneve: convert config to RCU-protected pointer")

Signed-off-by: Jakub Kicinski

net/mlx5: Fix MCIA register buffer overflow on 32 dword reads

2026-07-23T15:19:29+00:00

The MCIA register can return up to 32 dwords (128 bytes) when the device
advertises the mcia_32dwords capability, but struct
mlx5_ifc_mcia_reg_bits only defines dword_0..11, leaving room for just
12 dwords (48 bytes) of data.

mlx5_query_mcia() clamps the read size to mlx5_mcia_max_bytes() and then
memcpy()s that many bytes out of the register, potentially reading past
the end of the 'out' buffer. On kernels built with FORTIFY_SOURCE this
is caught as a buffer overflow while reading the module EEPROM via
ethtool:

  detected buffer overflow in memcpy
  kernel BUG at lib/string_helpers.c:1048!
  RIP: 0010:fortify_panic+0x13/0x20
  Call Trace:
   mlx5_query_mcia.isra.0+0x200/0x210 [mlx5_core]
   mlx5_query_module_eeprom_by_page+0x4a/0xa0 [mlx5_core]
   mlx5e_get_module_eeprom_by_page+0xbb/0x120 [mlx5_core]
   eeprom_prepare_data+0xf3/0x170
   ethnl_default_doit+0xf1/0x3b0

Extend the mcia_reg layout to 32 dwords.

Fixes: 271907ee2f29 ("net/mlx5: Query the maximum MCIA register read size from firmware")
Signed-off-by: Gal Pressman 
Reviewed-by: Alex Lazar 
Signed-off-by: Tariq Toukan 
Link: https://patch.msgid.link/20260717072338.1240582-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski

net/mlx5: Add PSP related fields to the mlx5_ifc

2026-07-16T08:24:02+00:00

This adds:
- misc_parameters_6, containing a few fields for matching PSP headers.
As this is the last misc_parameters field defined, retire the old
optimization added in commit [1] to not touch the reserved part.
- PSP decap action.
- PSP SPI header field pointer.

[1] commit 667cb65ae5ad ("net/mlx5: Don't store reserved part in FTEs
and FGs")

Signed-off-by: Cosmin Ratiu 
Reviewed-by: Dragos Tatulea 
Signed-off-by: Tariq Toukan 
Link: https://patch.msgid.link/20260713084320.1015240-3-tariqt@nvidia.com
Reviewed-by: Jacob Keller 
Signed-off-by: Leon Romanovsky

net/mlx5: Drop redundant esw_cap, reuse e_switch_cap

2026-07-16T08:23:42+00:00

esw_manager_vport_number{,_valid} and merged_eswitch were read through a
separate mlx5_ifc_esw_cap_bits struct, but these bits live in the
e-switch capability that mlx5_ifc_e_switch_cap_bits already describes
(both overlay the same QUERY_HCA_CAP op_mod 0x9 output).

Add esw_manager_vport_number{,_valid} to mlx5_ifc_e_switch_cap_bits at
the same offsets, drop the redundant mlx5_ifc_esw_cap_bits and its
hca_cap_union member, and switch the only user (hws/cmd.c) to
capability.e_switch_cap.

Signed-off-by: Shay Drory 
Reviewed-by: Yevgeny Kliteynik 
Signed-off-by: Tariq Toukan 
Link: https://patch.msgid.link/20260713084320.1015240-2-tariqt@nvidia.com
Reviewed-by: Jacob Keller 
Signed-off-by: Leon Romanovsky

Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

2026-06-18T15:16:21+00:00

Pull rdma updates from Jason Gunthorpe:
 "Many AI driven bug fixes, and several big driver API cleanups

   - Driver bug fixes and minor cleanups in mlx5, hns, rxe, efa, siw,
     rtrs, mana, irdma, mlx4. Commonly error path flows, integer
     arithmetic overflows on unsafe data, out of bounds access, and use
     after free issues under races.

   - Second half of the new udata API for drivers focusing on uAPI
     response

   - bnxt_re supports more options for QP creation that will allow a dv
     path in rdma-core

   - Untangle the module dependencies so drivers don't link to
     ib_uverbs.ko as was originall intended

   - Provide a new way to handle umems with a consistent simplified uAPI
     and update several drivers to use it. This brings dmabuf support to
     more places and more drivers

   - Support for mlx5 rate limit and packet pacing for UD and UC

   - A batch of fixes for the new shared FRMR pools infrastructure"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (148 commits)
  RDMA/irdma: Replace waitqueue and flag with completion
  RDMA/hns: Fix memory leak of bonding resources
  RDMA/rtrs-srv: Bound RDMA-Write length to chunk size in rdma_write_sg
  docs: infiniband: correct name of option to enable the ib_uverbs module
  RDMA/bnxt_re: Reject GET_TOGGLE_MEM when toggle page was not allocated
  RDMA/bnxt_re: Fail DBR related page allocation UAPIs if the feature is disabled
  RDMA/bnxt_re: Avoid repeated requests to allocate WC pages
  RDMA/bnxt_re: Proper rollback if the ioremap fails
  RDMA/bnxt_re: Add a max slot check for SQ
  RDMA/bnxt_re: Avoid displaying the kernel pointer
  RDMA/bnxt_re: Free CQ toggle page after firmware teardown
  RDMA/bnxt_re: Free SRQ toggle page after firmware teardown
  RDMA/bnxt_re: Initialize dpi variable to zero
  ABI: sysfs-class-infiniband: minor cleanup
  RDMA/mlx5: Release the HW‑provided UAR index rather than the SW one
  RDMA/mlx5: Fix undefined shift of user RQ WQE size
  RDMA/mlx5: Remove raw RSS QP restrack tracking
  RDMA/mlx5: Remove DCT restrack tracking
  RDMA/mlx5: Drop FRMR pool handle on UMR revoke failure
  RDMA/core: Add ib_frmr_pool_drop for unrecoverable handles
  ...

Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net

2026-06-11T21:33:35+00:00

Cross-merge networking fixes after downstream PR (net-7.1-rc8).

Conflicts:

drivers/net/ethernet/wangxun/txgbe/txgbe_aml.c
  f67aead16e85 ("net: txgbe: rework service event handling")
  57d39faed4c9 ("net: txgbe: improve functions of AML 40G devices")

net/rds/info.c
  512db8267b73 ("rds: mark snapshot pages dirty in rds_info_getsockopt()")
  6e94eeb2a2a6 ("rds: convert to getsockopt_iter")

Adjacent changes:

include/net/sock.h
  1ee90b77b727 ("net: guard timestamp cmsgs to real error queue skbs")
  f0de88303d5e ("net: make is_skb_wmem() available to modules")

Signed-off-by: Jakub Kicinski