linux-stable.git/drivers/infiniband, branch v6.2.13

RDMA/core: Fix GID entry ref leak when create_ah fails

2023-04-20T10:36:55+00:00

[ Upstream commit aca3b0fa3d04b40c96934d86cc224cccfa7ea8e0 ]

If AH create request fails, release sgid_attr to avoid GID entry
referrence leak reported while releasing GID table

Fixes: 1a1f460ff151 ("RDMA: Hold the sgid_attr inside the struct ib_ah/qp")
Link: https://lore.kernel.org/r/20230401063424.342204-1-saravanan.vajravel@broadcom.com
Reviewed-by: Selvin Xavier 
Signed-off-by: Saravanan Vajravel 
Signed-off-by: Jason Gunthorpe 
Signed-off-by: Sasha Levin

RDMA/cma: Allow UD qp_type to join multicast only

2023-04-20T10:36:53+00:00

[ Upstream commit 58e84f6b3e84e46524b7e5a916b53c1ad798bc8f ]

As for multicast:
- The SIDR is the only mode that makes sense;
- Besides PS_UDP, other port spaces like PS_IB is also allowed, as it is
  UD compatible. In this case qkey also needs to be set [1].

This patch allows only UD qp_type to join multicast, and set qkey to
default if it's not set, to fix an uninit-value error: the ib->rec.qkey
field is accessed without being initialized.

=====================================================
BUG: KMSAN: uninit-value in cma_set_qkey drivers/infiniband/core/cma.c:510 [inline]
BUG: KMSAN: uninit-value in cma_make_mc_event+0xb73/0xe00 drivers/infiniband/core/cma.c:4570
 cma_set_qkey drivers/infiniband/core/cma.c:510 [inline]
 cma_make_mc_event+0xb73/0xe00 drivers/infiniband/core/cma.c:4570
 cma_iboe_join_multicast drivers/infiniband/core/cma.c:4782 [inline]
 rdma_join_multicast+0x2b83/0x30a0 drivers/infiniband/core/cma.c:4814
 ucma_process_join+0xa76/0xf60 drivers/infiniband/core/ucma.c:1479
 ucma_join_multicast+0x1e3/0x250 drivers/infiniband/core/ucma.c:1546
 ucma_write+0x639/0x6d0 drivers/infiniband/core/ucma.c:1732
 vfs_write+0x8ce/0x2030 fs/read_write.c:588
 ksys_write+0x28c/0x520 fs/read_write.c:643
 __do_sys_write fs/read_write.c:655 [inline]
 __se_sys_write fs/read_write.c:652 [inline]
 __ia32_sys_write+0xdb/0x120 fs/read_write.c:652
 do_syscall_32_irqs_on arch/x86/entry/common.c:114 [inline]
 __do_fast_syscall_32+0x96/0xf0 arch/x86/entry/common.c:180
 do_fast_syscall_32+0x34/0x70 arch/x86/entry/common.c:205
 do_SYSENTER_32+0x1b/0x20 arch/x86/entry/common.c:248
 entry_SYSENTER_compat_after_hwframe+0x4d/0x5c

Local variable ib.i created at:
cma_iboe_join_multicast drivers/infiniband/core/cma.c:4737 [inline]
rdma_join_multicast+0x586/0x30a0 drivers/infiniband/core/cma.c:4814
ucma_process_join+0xa76/0xf60 drivers/infiniband/core/ucma.c:1479

CPU: 0 PID: 29874 Comm: syz-executor.3 Not tainted 5.16.0-rc3-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
=====================================================

[1] https://lore.kernel.org/linux-rdma/20220117183832.GD84788@nvidia.com/

Fixes: b5de0c60cc30 ("RDMA/cma: Fix use after free race in roce multicast join")
Reported-by: syzbot+8fcbb77276d43cc8b693@syzkaller.appspotmail.com
Signed-off-by: Mark Zhang 
Link: https://lore.kernel.org/r/58a4a98323b5e6b1282e83f6b76960d06e43b9fa.1679309909.git.leon@kernel.org
Signed-off-by: Leon Romanovsky 
Signed-off-by: Sasha Levin

RDMA/erdma: Defer probing if netdevice can not be found

2023-04-20T10:36:53+00:00

[ Upstream commit 6bd1bca858f1734a75572a788213d1e1143f2f0a ]

ERDMA device may be probed before its associated netdevice, returning
-EPROBE_DEFER allows OS try to probe erdma device later.

Fixes: d55e6fb4803c ("RDMA/erdma: Add the erdma module")
Signed-off-by: Cheng Xu 
Link: https://lore.kernel.org/r/20230320084652.16807-5-chengyou@linux.alibaba.com
Signed-off-by: Leon Romanovsky 
Signed-off-by: Sasha Levin

RDMA/erdma: Inline mtt entries into WQE if supported

2023-04-20T10:36:53+00:00

[ Upstream commit 0dd83a4d7756713f81990d6c5547500f212a1190 ]

The max inline mtt count supported is ERDMA_MAX_INLINE_MTT_ENTRIES.
When mr->mem.mtt_nents == ERDMA_MAX_INLINE_MTT_ENTRIES, inline mtt
is also supported, fix it.

Fixes: 155055771704 ("RDMA/erdma: Add verbs implementation")
Signed-off-by: Cheng Xu 
Link: https://lore.kernel.org/r/20230320084652.16807-4-chengyou@linux.alibaba.com
Signed-off-by: Leon Romanovsky 
Signed-off-by: Sasha Levin

RDMA/erdma: Update default EQ depth to 4096 and max_send_wr to 8192

2023-04-20T10:36:53+00:00

[ Upstream commit 6256aa9ae955d10ec73a434533ca62034eff1b76 ]

Max EQ depth of hardware is 32K, the current default EQ depth is too small
for some applications, so change the default depth to 4096.
Max send WRs the hardware can support is 8K, but the driver limits the
value to 4K. Remove this limitation.

Fixes: be3cff0f242d ("RDMA/erdma: Add the hardware related definitions")
Fixes: db23ae64caac ("RDMA/erdma: Add verbs header file")
Signed-off-by: Cheng Xu 
Link: https://lore.kernel.org/r/20230320084652.16807-3-chengyou@linux.alibaba.com
Signed-off-by: Leon Romanovsky 
Signed-off-by: Sasha Levin

RDMA/erdma: Fix some typos

2023-04-20T10:36:53+00:00

[ Upstream commit 3fe26c0493e4c2da4b7d8ba8c975a6f48fb75ec2 ]

FAA is short for atomic fetch and add, not FAD. Fix this.

Fixes: 0ca9c2e2844a ("RDMA/erdma: Implement atomic operations support")
Signed-off-by: Cheng Xu 
Link: https://lore.kernel.org/r/20230320084652.16807-2-chengyou@linux.alibaba.com
Signed-off-by: Leon Romanovsky 
Signed-off-by: Sasha Levin

IB/mlx5: Add support for 400G_8X lane speed

2023-04-20T10:36:53+00:00

[ Upstream commit 88c9483faf15ada14eca82714114656893063458 ]

Currently, when driver queries PTYS to report which link speed is being
used on its RoCE ports, it does not check the case of having 400Gbps
transmitted over 8 lanes. Thus it fails to report the said speed and
instead it defaults to report 10G over 4 lanes.

Add a check for the said speed when querying PTYS and report it back
correctly when needed.

Fixes: 08e8676f1607 ("IB/mlx5: Add support for 50Gbps per lane link modes")
Signed-off-by: Maher Sanalla 
Reviewed-by: Aya Levin 
Reviewed-by: Saeed Mahameed 
Link: https://lore.kernel.org/r/ec9040548d119d22557d6a4b4070d6f421701fd4.1678973994.git.leon@kernel.org
Signed-off-by: Leon Romanovsky 
Signed-off-by: Sasha Levin

RDMA/irdma: Add ipv4 check to irdma_find_listener()

2023-04-20T10:36:52+00:00

[ Upstream commit e4522c097ec10f23ea0933e9e69d4fa9d8ae9441 ]

Add ipv4 check to irdma_find_listener(). Otherwise the function
incorrectly finds and returns a listener with a different addr family for
the zero IP addr, if a listener with a zero IP addr and the same port as
the one searched for has already been created.

Fixes: 146b9756f14c ("RDMA/irdma: Add connection manager")
Signed-off-by: Tatyana Nikolova 
Signed-off-by: Shiraz Saleem 
Link: https://lore.kernel.org/r/20230315145231.931-5-shiraz.saleem@intel.com
Signed-off-by: Leon Romanovsky 
Signed-off-by: Sasha Levin

RDMA/irdma: Increase iWARP CM default rexmit count

2023-04-20T10:36:52+00:00

[ Upstream commit 8385a875c9eecc429b2f72970efcbb0e5cb5b547 ]

When running perftest with large number of connections in iWARP mode, the
passive side could be slow to respond. Increase the rexmit counter default
to allow scaling connections.

Fixes: 146b9756f14c ("RDMA/irdma: Add connection manager")
Signed-off-by: Mustafa Ismail 
Signed-off-by: Shiraz Saleem 
Link: https://lore.kernel.org/r/20230315145231.931-4-shiraz.saleem@intel.com
Signed-off-by: Leon Romanovsky 
Signed-off-by: Sasha Levin

RDMA/irdma: Fix memory leak of PBLE objects

2023-04-20T10:36:52+00:00

[ Upstream commit b69a6979dbaa2453675fe9c71bdc2497fedb11f9 ]

On rmmod of irdma, the PBLE object memory is not being freed. PBLE object
memory are not statically pre-allocated at function initialization time
unlike other HMC objects. PBLEs objects and the Segment Descriptors (SD)
for it can be dynamically allocated during scale up and SD's remain
allocated till function deinitialization.

Fix this leak by adding IRDMA_HMC_IW_PBLE to the iw_hmc_obj_types[] table
and skip pbles in irdma_create_hmc_obj but not in irdma_del_hmc_objects().

Fixes: 44d9e52977a1 ("RDMA/irdma: Implement device initialization definitions")
Signed-off-by: Mustafa Ismail 
Signed-off-by: Shiraz Saleem 
Link: https://lore.kernel.org/r/20230315145231.931-3-shiraz.saleem@intel.com
Signed-off-by: Leon Romanovsky 
Signed-off-by: Sasha Levin