When a process exits with numerous large, pinned memory regions consisting
of 4KB pages, the cleanup of the memory region through __ib_umem_release()
may cause soft lockups. This is because unpin_user_page_range_dirty_lock()
is called in a tight loop for unpin and releasing page without yielding the
CPU.
watchdog: BUG: soft lockup - CPU#44 stuck for 26s! [python3:73464]
Kernel panic - not syncing: softlockup: hung tasks
CPU: 44 PID: 73464 Comm: python3 Tainted: G OEL
asm_sysvec_apic_timer_interrupt+0x1b/0x20
RIP: 0010:free_unref_page+0xff/0x190
? free_unref_page+0xe3/0x190
__put_page+0x77/0xe0
put_compound_head+0xed/0x100
unpin_user_page_range_dirty_lock+0xb2/0x180
__ib_umem_release+0x57/0xb0 [ib_core]
ib_umem_release+0x3f/0xd0 [ib_core]
mlx5_ib_dereg_mr+0x2e9/0x440 [mlx5_ib]
ib_dereg_mr_user+0x43/0xb0 [ib_core]
uverbs_free_mr+0x15/0x20 [ib_uverbs]
destroy_hw_idr_uobject+0x21/0x60 [ib_uverbs]
uverbs_destroy_uobject+0x38/0x1b0 [ib_uverbs]
__uverbs_cleanup_ufile+0xd1/0x150 [ib_uverbs]
uverbs_destroy_ufile_hw+0x3f/0x100 [ib_uverbs]
ib_uverbs_close+0x1f/0xb0 [ib_uverbs]
__fput+0x9c/0x280
____fput+0xe/0x20
task_work_run+0x6a/0xb0
do_exit+0x217/0x3c0
do_group_exit+0x3b/0xb0
get_signal+0x150/0x900
arch_do_signal_or_restart+0xde/0x100
exit_to_user_mode_loop+0xc4/0x160
exit_to_user_mode_prepare+0xa0/0xb0
syscall_exit_to_user_mode+0x27/0x50
do_syscall_64+0x63/0xb0
Fix soft lockup issues by incorporating cond_resched() calls within
__ib_umem_release(), and this SG entries are typically grouped in 2MB
chunks on x86_64, adding cond_resched() should has minimal performance
impact.
Signed-off-by: Li RongQing <lirongqing@baidu.com>
Link: https://patch.msgid.link/20251113095317.2628-1-lirongqing@baidu.com
Acked-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
In include/rdma/ib_cm.h:
Correct a typedef's kernel-doc notation by adding the 'typedef' keyword
to it to avoid a warning.
Add a leading " *" to a kernel-doc line to avoid a warning.
Warning: ib_cm.h:289 function parameter 'ib_cm_handler' not described
in 'int'
Warning: ib_cm.h:289 expecting prototype for ib_cm_handler(). Prototype
was for int() instead
Warning: ib_cm.h:484 bad line: connection message in case duplicates
are received.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Link: https://patch.msgid.link/20251112062908.2711007-1-rdunlap@infradead.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Single patch to expose new link mode for 1600Gbps, utilizing 8 lanes at
200Gbps per lane.
Signed-off-by: Leon Romanovsky <leon@kernel.org>
* mlx5-next:
net/mlx5: Expose definition for 1600Gbps link mode
Currently if a user enqueues a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistency cannot be addressed without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6cc ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566 ("workqueue: Add new WQ_PERCPU flag")
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107133626.190952-1-marco.crivellari@suse.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Currently if a user enqueues a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistency cannot be addressed without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6cc ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566 ("workqueue: Add new WQ_PERCPU flag")
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107133306.187939-1-marco.crivellari@suse.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Refactor the _get_prio() function to remove redundant arguments by
reusing the existing flow table attributes struct instead of passing
attributes separately. This improves code clarity and maintainability.
In addition allows downstream patch to add new parameter without
needing to change __get_prio() arguments.
Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Signed-off-by: Edward Srouji <edwards@nvidia.com>
Link: https://patch.msgid.link/20251029-support-other-eswitch-v1-6-98bb707b5d57@nvidia.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
In case of a LAG configuration change the root namespace core device for
all of the LAG slaves to be the core device of the master device for
RDMA_TRANSPORT namespaces, in order to ensure all tables are created
through the master device.
Once the LAG is disabled revert back to the native core device.
Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Signed-off-by: Edward Srouji <edwards@nvidia.com>
Link: https://patch.msgid.link/20251029-support-other-eswitch-v1-4-98bb707b5d57@nvidia.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
When the device in switchdev mode, the RDMA device manages all the
vports which belong to its representors, which can lead to a situation
where the PF that is used to manage the RDMA device isn't the native PF
of some of the vports it manages.
Add infrastructure to allow the master PF to manage all the hardware
resources for the vports under its management.
Whereas currently the only such resource is RDMA TRANSPORT steering
domains.
That is done by adding new FW argument other_eswitch which is passed by
the driver to the FW to allow the master PF to properly manage vports
belonging to other native PF.
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Add mlx5_fs_set_root_dev() function which swaps the root namespace
core device with another one for a given table_type.
It is intended for usage only by RDMA_TRANSPORT tables in case of LAG
configuration, to allow the creation of tables during LAG always
through the LAG master device, which is valid since during LAG the
master is allowed to manage the RDMA_TRANSPORT tables of its slaves.
In addition move the table_type enum to global include to allow its use
in a downstream patch in the RDMA driver.
Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Signed-off-by: Edward Srouji <edwards@nvidia.com>
Link: https://patch.msgid.link/20251029-support-other-eswitch-v1-3-98bb707b5d57@nvidia.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Add other_eswitch support which allows flow tables creation above vports
that reside on different esw managers.
The new flag MLX5_FLOW_TABLE_OTHER_ESWITCH indicates if the
esw_owner_vhca_id attribute is supported.
Note that this is only supported if the Advanced-RDMA cap-
rdma_transport_manager_other_eswitch is set.
And it is the caller responsibility to check that.
Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Signed-off-by: Edward Srouji <edwards@nvidia.com>
Link: https://patch.msgid.link/20251029-support-other-eswitch-v1-2-98bb707b5d57@nvidia.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Correct the kernel-doc comments format to avoid around 35 kernel-doc
warnings:
- use struct keyword to introduce struct kernel-doc comments
- use correct variable name for some struct members
- use correct function name in comments for some functions
- fix spelling in a few comments
- use a ':' instead of '-' to separate struct members from their
descriptions
- add a function name heading for rvt_div_mtu()
This leaves one struct member that is not described:
rdmavt_qp.h:206: warning: Function parameter or struct member 'wq'
not described in 'rvt_krwq'
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Link: https://patch.msgid.link/20251105045127.106822-1-rdunlap@infradead.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
CC: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251101163121.78400-6-marco.crivellari@suse.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
CC: Yishai Hadas <yishaih@nvidia.com>
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251101163121.78400-5-marco.crivellari@suse.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
CC: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251101163121.78400-4-marco.crivellari@suse.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251101163121.78400-3-marco.crivellari@suse.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistency cannot be addressed without refactoring the API.
system_unbound_wq should be the default workqueue so as not to enforce
locality constraints for random work whenever it's not required.
Adding system_dfl_wq to encourage its use when unbound work should be used.
The old system_unbound_wq will be kept for a few release cycles.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251101163121.78400-2-marco.crivellari@suse.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
When a GSI MAD packet is sent on the QP, it will potentially be
retried CMA_MAX_CM_RETRIES times with a timeout value of:
4.096usec * 2 ^ CMA_CM_RESPONSE_TIMEOUT
The above equates to ~64 seconds using the default CMA values.
The cm_id_priv's refcount will be incremented for this period.
Therefore, the timeout value waiting for a cm_id destruction must be
based on the effective timeout of MAD packets. To provide additional
leeway, we add 25% to this timeout and use that instead of the
constant 10 seconds timeout, which may result in false negatives.
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>
Link: https://patch.msgid.link/20251021132738.4179604-1-haakon.bugge@oracle.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Before this patch during either switchdev or legacy mode enablement we
cleared the mac address of vports between changes. This change allows us
to preserve the vports mac address between eswitch mode changes.
Vports hold information for VFs/SFs such as the permanent mac address.
VF/SF mac can be set either by iproute vf interface or devlink function
interface. For no obvious reason we reset it to 0 on switchdev/legacy
mode changes, this patch is fixing that, to align with other vport
information that are never reset, e.g GUID,mtu,promisc mode, etc ..
Signed-off-by: Adithya Jayachandran <ajayachandra@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Acked-by: Leon Romanovsky <leon@kernel.org> # RDMA
Fix 49 kernel-doc warnings in ib_verbs.h:
- Add struct short description for rdma_stat_desc, rdma_hw_stats.
- Fix kernel-doc format for struct members (use ':' instead of '-') for
several structs.
- Don't use "/**" kernel-doc notation for struct members in ib_device_ops
(most members are not documented and most of the kernel-doc was
not formatted correctly).
- Spell function parameters correctly in ib_dma_map_sgtable_attrs(),
ib_device_try_get(), rdma_roce_rescan_device().
- Add kernel-doc for the function parameter in
rdma_flow_label_to_udp_sport().
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Link: https://patch.msgid.link/20251020034320.3011094-1-rdunlap@infradead.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Pull i2c fix from Wolfram Sang:
"One revert because of a regression in the I2C core which has sadly not
showed up during its time in -next"
* tag 'i2c-for-6.18-rc1-hotfix' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
Revert "i2c: boardinfo: Annotate code used in init phase only"
Pull irq fixes from Borislav Petkov:
- Skip interrupt ID 0 in sifive-plic during suspend/resume because
ID 0 is reserved and accessing reserved register space could result
in undefined behavior
- Fix a function's retval check in aspeed-scu-ic
* tag 'irq_urgent_for_v6.18_rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
irqchip/sifive-plic: Avoid interrupt ID 0 handling during suspend/resume
irqchip/aspeed-scu-ic: Fix an IS_ERR() vs NULL check
Pull tracing fixes from Steven Rostedt:
"The previous fix to trace_marker required updating trace_marker_raw as
well. The difference between trace_marker_raw from trace_marker is
that the raw version is for applications to write binary structures
directly into the ring buffer instead of writing ASCII strings. This
is for applications that will read the raw data from the ring buffer
and get the data structures directly. It's a bit quicker than using
the ASCII version.
Unfortunately, it appears that our test suite has several tests that
test writes to the trace_marker file, but lacks any tests to the
trace_marker_raw file (this needs to be remedied). Two issues came
about the update to the trace_marker_raw file that syzbot found:
- Fix tracing_mark_raw_write() to use per CPU buffer
The fix to use the per CPU buffer to copy from user space was
needed for both the trace_maker and trace_maker_raw file.
The fix for reading from user space into per CPU buffers properly
fixed the trace_marker write function, but the trace_marker_raw
file wasn't fixed properly. The user space data was correctly
written into the per CPU buffer, but the code that wrote into the
ring buffer still used the user space pointer and not the per CPU
buffer that had the user space data already written.
- Stop the fortify string warning from writing into trace_marker_raw
After converting the copy_from_user_nofault() into a memcpy(),
another issue appeared. As writes to the trace_marker_raw expects
binary data, the first entry is a 4 byte identifier. The entry
structure is defined as:
struct {
struct trace_entry ent;
int id;
char buf[];
};
The size of this structure is reserved on the ring buffer with:
size = sizeof(*entry) + cnt;
Then it is copied from the buffer into the ring buffer with:
memcpy(&entry->id, buf, cnt);
This use to be a copy_from_user_nofault(), but now converting it to
a memcpy() triggers the fortify-string code, and causes a warning.
The allocated space is actually more than what is copied, as the
cnt used also includes the entry->id portion. Allocating
sizeof(*entry) plus cnt is actually allocating 4 bytes more than
what is needed.
Change the size function to:
size = struct_size(entry, buf, cnt - sizeof(entry->id));
And update the memcpy() to unsafe_memcpy()"
* tag 'trace-v6.18-3' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
tracing: Stop fortify-string from warning in tracing_mark_raw_write()
tracing: Fix tracing_mark_raw_write() to use buf and not ubuf
Pull Kbuild fixes from Nathan Chancellor:
- Fix UAPI types check in headers_check.pl
- Only enable -Werror for hostprogs with CONFIG_WERROR / W=e
- Ignore fsync() error when output of gen_init_cpio is a pipe
- Several little build fixes for recent modules.builtin.modinfo series
* tag 'kbuild-fixes-6.18-1' of git://git.kernel.org/pub/scm/linux/kernel/git/kbuild/linux:
kbuild: Use '--strip-unneeded-symbol' for removing module device table symbols
s390/vmlinux.lds.S: Move .vmlinux.info to end of allocatable sections
kbuild: Add '.rel.*' strip pattern for vmlinux
kbuild: Restore pattern to avoid stripping .rela.dyn from vmlinux
gen_init_cpio: Ignore fsync() returning EINVAL on pipes
scripts/Makefile.extrawarn: Respect CONFIG_WERROR / W=e for hostprogs
kbuild: uapi: Strip comments before size type check
