[PATCH v2 0/2] io_uring/io-wq: respect cgroup cpusets

[PATCH v2 0/2] io_uring/io-wq: respect cgroup cpusets

[Felix Moessbauer]

Hi,

this series continues the affinity cleanup work started in
io_uring/sqpoll. It has been successfully tested against the liburing
testsuite (make runtests), liburing @ caae94903d2e201.

The test wq-aff.t succeeds if at least cpu 0,1 are in
the set and fails otherwise. This is expected, as the test wants
to pin on these cpus. I'll send a patch for liburing to skip that test
in case this pre-condition is not met.

Changes since v1:

- rework commit messages (don't use ambient cpus, wq threads are no
  pollers)
- no functional changes

Best regards,
Felix Moessbauer
Siemens AG

Felix Moessbauer (2):
  io_uring/io-wq: do not allow pinning outside of cpuset
  io_uring/io-wq: limit io poller cpuset to ambient one

 io_uring/io-wq.c | 25 +++++++++++++++++++------
 1 file changed, 19 insertions(+), 6 deletions(-)

-- 
2.39.2

[PATCH v2 1/2] io_uring/io-wq: do not allow pinning outside of cpuset

[Felix Moessbauer]

The io worker threads are userland threads that just never exit to the
userland. By that, they are also assigned to a cgroup (the group of the
creating task).

When changing the affinity of the io_wq thread via syscall, we must only
allow cpumasks within the limits defined by the cpuset controller of the
cgroup (if enabled).

Fixes: da64d6db3bd3 ("io_uring: One wqe per wq")
Signed-off-by: Felix Moessbauer <felix.moessbauer@siemens.com>
---
 io_uring/io-wq.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c
index f1e7c670add8..c7055a8895d7 100644
--- a/io_uring/io-wq.c
+++ b/io_uring/io-wq.c
@@ -13,6 +13,7 @@
 #include <linux/slab.h>
 #include <linux/rculist_nulls.h>
 #include <linux/cpu.h>
+#include <linux/cpuset.h>
 #include <linux/task_work.h>
 #include <linux/audit.h>
 #include <linux/mmu_context.h>
@@ -1322,17 +1323,29 @@ static int io_wq_cpu_offline(unsigned int cpu, struct hlist_node *node)
 
 int io_wq_cpu_affinity(struct io_uring_task *tctx, cpumask_var_t mask)
 {
+	cpumask_var_t allowed_mask;
+	int ret = 0;
+
 	if (!tctx || !tctx->io_wq)
 		return -EINVAL;
 
+	if (!alloc_cpumask_var(&allowed_mask, GFP_KERNEL))
+		return -ENOMEM;
+
 	rcu_read_lock();
-	if (mask)
-		cpumask_copy(tctx->io_wq->cpu_mask, mask);
-	else
-		cpumask_copy(tctx->io_wq->cpu_mask, cpu_possible_mask);
+	cpuset_cpus_allowed(tctx->io_wq->task, allowed_mask);
+	if (mask) {
+		if (cpumask_subset(mask, allowed_mask))
+			cpumask_copy(tctx->io_wq->cpu_mask, mask);
+		else
+			ret = -EINVAL;
+	} else {
+		cpumask_copy(tctx->io_wq->cpu_mask, allowed_mask);
+	}
 	rcu_read_unlock();
 
-	return 0;
+	free_cpumask_var(allowed_mask);
+	return ret;
 }
 
 /*
-- 
2.39.2

[PATCH v2 2/2] io_uring/io-wq: limit io poller cpuset to ambient one

[Felix Moessbauer]

The io worker threads are userland threads that just never exit to the
userland. By that, they are also assigned to a cgroup (the group of the
creating task).

When creating a new io worker, this worker should inherit the cpuset
of the cgroup.

Fixes: da64d6db3bd3 ("io_uring: One wqe per wq")
Signed-off-by: Felix Moessbauer <felix.moessbauer@siemens.com>
---
 io_uring/io-wq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c
index c7055a8895d7..a38f36b68060 100644
--- a/io_uring/io-wq.c
+++ b/io_uring/io-wq.c
@@ -1168,7 +1168,7 @@ struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data)
 
 	if (!alloc_cpumask_var(&wq->cpu_mask, GFP_KERNEL))
 		goto err;
-	cpumask_copy(wq->cpu_mask, cpu_possible_mask);
+	cpuset_cpus_allowed(data->task, wq->cpu_mask);
 	wq->acct[IO_WQ_ACCT_BOUND].max_workers = bounded;
 	wq->acct[IO_WQ_ACCT_UNBOUND].max_workers =
 				task_rlimit(current, RLIMIT_NPROC);
-- 
2.39.2

Re: [PATCH v2 2/2] io_uring/io-wq: limit io poller cpuset to ambient one

[Jens Axboe]

On 9/10/24 9:45 AM, Felix Moessbauer wrote:
> The io worker threads are userland threads that just never exit to the
> userland. By that, they are also assigned to a cgroup (the group of the
> creating task).
> 
> When creating a new io worker, this worker should inherit the cpuset
> of the cgroup.

This still has that same ambient usage in the title which I just
cannot make sense of?

-- 
Jens Axboe