There are 3 places where the cpu and node masks of the top cpuset can be initialized in the order they are executed: 1) start_kernel -> cpuset_init() 2) start_kernel -> cgroup_init() -> cpuset_bind() 3) kernel_init_freeable() -> do_basic_setup() -> cpuset_init_smp()
The first cpuset_init() function just sets all the bits in the masks. The last one executed is cpuset_init_smp() which sets up cpu and node masks suitable for v1, but not v2. cpuset_bind() does the right setup for both v1 and v2.
For systems with cgroup v2 setup, cpuset_bind() is called once. For systems with cgroup v1 setup, cpuset_bind() is called twice. It is first called before cpuset_init_smp() in cgroup v2 mode. Then it is called again when cgroup v1 filesystem is mounted in v1 mode after cpuset_init_smp().
[ 2.609781] cpuset_bind() called - v2 = 1 [ 3.079473] cpuset_init_smp() called [ 7.103710] cpuset_bind() called - v2 = 0
As a result, cpu and memory node hot add may fail to update the cpu and node masks of the top cpuset to include the newly added cpu or node in a cgroup v2 environment.
smp_init() is called after the first two init functions. So we don't have a complete list of active cpus and memory nodes until later in cpuset_init_smp() which is the right time to set up effective_cpus and effective_mems.
To fix this problem, the potentially incorrect cpus_allowed & mems_allowed setup in cpuset_init_smp() are removed. For cgroup v2 systems, the initial cpuset_bind() call will set them up correctly. For cgroup v1 systems, the second call to cpuset_bind() will do the right setup.
cc: stable@vger.kernel.org Signed-off-by: Waiman Long longman@redhat.com --- kernel/cgroup/cpuset.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index 9390bfd9f1cd..6bd8f5ef40fe 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -3390,8 +3390,9 @@ static struct notifier_block cpuset_track_online_nodes_nb = { */ void __init cpuset_init_smp(void) { - cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); - top_cpuset.mems_allowed = node_states[N_MEMORY]; + /* + * cpus_allowd/mems_allowed will be properly set up in cpuset_bind(). + */ top_cpuset.old_mems_allowed = top_cpuset.mems_allowed;
cpumask_copy(top_cpuset.effective_cpus, cpu_active_mask);
Hi Waiman,
On Mon, Apr 25, 2022 at 11:55:05AM -0400, Waiman Long wrote:
There are 3 places where the cpu and node masks of the top cpuset can be initialized in the order they are executed:
- start_kernel -> cpuset_init()
- start_kernel -> cgroup_init() -> cpuset_bind()
- kernel_init_freeable() -> do_basic_setup() -> cpuset_init_smp()
The first cpuset_init() function just sets all the bits in the masks. The last one executed is cpuset_init_smp() which sets up cpu and node masks suitable for v1, but not v2. cpuset_bind() does the right setup for both v1 and v2.
For systems with cgroup v2 setup, cpuset_bind() is called once. For systems with cgroup v1 setup, cpuset_bind() is called twice. It is first called before cpuset_init_smp() in cgroup v2 mode. Then it is called again when cgroup v1 filesystem is mounted in v1 mode after cpuset_init_smp().
[ 2.609781] cpuset_bind() called - v2 = 1 [ 3.079473] cpuset_init_smp() called [ 7.103710] cpuset_bind() called - v2 = 0
I run some test, on a server with centOS, this did happen that cpuset_bind() is called twice, first as v2 during kernel boot, and then as v1 post-boot.
However on a QEMU running with a basic debian rootfs image, the second call of cpuset_bind() didn't happen.
As a result, cpu and memory node hot add may fail to update the cpu and node masks of the top cpuset to include the newly added cpu or node in a cgroup v2 environment.
smp_init() is called after the first two init functions. So we don't have a complete list of active cpus and memory nodes until later in cpuset_init_smp() which is the right time to set up effective_cpus and effective_mems.
To fix this problem, the potentially incorrect cpus_allowed & mems_allowed setup in cpuset_init_smp() are removed. For cgroup v2 systems, the initial cpuset_bind() call will set them up correctly. For cgroup v1 systems, the second call to cpuset_bind() will do the right setup.
cc: stable@vger.kernel.org Signed-off-by: Waiman Long longman@redhat.com
kernel/cgroup/cpuset.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index 9390bfd9f1cd..6bd8f5ef40fe 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -3390,8 +3390,9 @@ static struct notifier_block cpuset_track_online_nodes_nb = { */ void __init cpuset_init_smp(void) {
- cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
- top_cpuset.mems_allowed = node_states[N_MEMORY];
So can we keep line cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
and only remove line top_cpuset.mems_allowed = node_states[N_MEMORY]; ?
Thanks, Feng
On 4/25/22 23:23, Feng Tang wrote:
Hi Waiman,
On Mon, Apr 25, 2022 at 11:55:05AM -0400, Waiman Long wrote:
There are 3 places where the cpu and node masks of the top cpuset can be initialized in the order they are executed:
- start_kernel -> cpuset_init()
- start_kernel -> cgroup_init() -> cpuset_bind()
- kernel_init_freeable() -> do_basic_setup() -> cpuset_init_smp()
The first cpuset_init() function just sets all the bits in the masks. The last one executed is cpuset_init_smp() which sets up cpu and node masks suitable for v1, but not v2. cpuset_bind() does the right setup for both v1 and v2.
For systems with cgroup v2 setup, cpuset_bind() is called once. For systems with cgroup v1 setup, cpuset_bind() is called twice. It is first called before cpuset_init_smp() in cgroup v2 mode. Then it is called again when cgroup v1 filesystem is mounted in v1 mode after cpuset_init_smp().
[ 2.609781] cpuset_bind() called - v2 = 1 [ 3.079473] cpuset_init_smp() called [ 7.103710] cpuset_bind() called - v2 = 0
I run some test, on a server with centOS, this did happen that cpuset_bind() is called twice, first as v2 during kernel boot, and then as v1 post-boot.
However on a QEMU running with a basic debian rootfs image, the second call of cpuset_bind() didn't happen.
The first time cpuset_bind() is called in cgroup_init(), the kernel doesn't know if userspace is going to mount v1 or v2 cgroup. By default, it is assumed to be v2. However, if userspace mounts the cgroup v1 filesystem for cpuset, cpuset_bind() will be run at this point by rebind_subsystem() to set up cgroup v1 environment and cpus_allowed/mems_allowed will be correctly set at this point. Mounting the cgroup v2 filesystem, however, does not cause rebind_subsystem() to run and hence cpuset_bind() is not called again.
Is the QEMU setup not mounting any cgroup filesystem at all? If so, does it matter whether v1 or v2 setup is used?
As a result, cpu and memory node hot add may fail to update the cpu and node masks of the top cpuset to include the newly added cpu or node in a cgroup v2 environment.
smp_init() is called after the first two init functions. So we don't have a complete list of active cpus and memory nodes until later in cpuset_init_smp() which is the right time to set up effective_cpus and effective_mems.
To fix this problem, the potentially incorrect cpus_allowed & mems_allowed setup in cpuset_init_smp() are removed. For cgroup v2 systems, the initial cpuset_bind() call will set them up correctly. For cgroup v1 systems, the second call to cpuset_bind() will do the right setup.
cc: stable@vger.kernel.org Signed-off-by: Waiman Long longman@redhat.com
kernel/cgroup/cpuset.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index 9390bfd9f1cd..6bd8f5ef40fe 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -3390,8 +3390,9 @@ static struct notifier_block cpuset_track_online_nodes_nb = { */ void __init cpuset_init_smp(void) {
- cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
- top_cpuset.mems_allowed = node_states[N_MEMORY];
So can we keep line cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
and only remove line top_cpuset.mems_allowed = node_states[N_MEMORY]; ?
That may cause cpusets.cpu to be set incorrectly for systems using cgroup v2. What is really important is that effective_cpus and effective_mems are set correctly.
Cheers, Longman
On Tue, Apr 26, 2022 at 10:58:21PM +0800, Waiman Long wrote:
On 4/25/22 23:23, Feng Tang wrote:
Hi Waiman,
On Mon, Apr 25, 2022 at 11:55:05AM -0400, Waiman Long wrote:
There are 3 places where the cpu and node masks of the top cpuset can be initialized in the order they are executed:
- start_kernel -> cpuset_init()
- start_kernel -> cgroup_init() -> cpuset_bind()
- kernel_init_freeable() -> do_basic_setup() -> cpuset_init_smp()
The first cpuset_init() function just sets all the bits in the masks. The last one executed is cpuset_init_smp() which sets up cpu and node masks suitable for v1, but not v2. cpuset_bind() does the right setup for both v1 and v2.
For systems with cgroup v2 setup, cpuset_bind() is called once. For systems with cgroup v1 setup, cpuset_bind() is called twice. It is first called before cpuset_init_smp() in cgroup v2 mode. Then it is called again when cgroup v1 filesystem is mounted in v1 mode after cpuset_init_smp().
[ 2.609781] cpuset_bind() called - v2 = 1 [ 3.079473] cpuset_init_smp() called [ 7.103710] cpuset_bind() called - v2 = 0
I run some test, on a server with centOS, this did happen that cpuset_bind() is called twice, first as v2 during kernel boot, and then as v1 post-boot.
However on a QEMU running with a basic debian rootfs image, the second call of cpuset_bind() didn't happen.
The first time cpuset_bind() is called in cgroup_init(), the kernel doesn't know if userspace is going to mount v1 or v2 cgroup. By default, it is assumed to be v2. However, if userspace mounts the cgroup v1 filesystem for cpuset, cpuset_bind() will be run at this point by rebind_subsystem() to set up cgroup v1 environment and cpus_allowed/mems_allowed will be correctly set at this point. Mounting the cgroup v2 filesystem, however, does not cause rebind_subsystem() to run and hence cpuset_bind() is not called again.
Is the QEMU setup not mounting any cgroup filesystem at all? If so, does it matter whether v1 or v2 setup is used?
When I got the cpuset binding error report, I tried first on qemu to reproduce and failed (due to there was no memory hotplug), then I reproduced it on a real server. For both system, I used "cgroup_no_v1=all" cmdline parameter to test cgroup-v2, could this be the reason? (TBH, this is the first time I use cgroup-v2).
Here is the info dump:
# mount | grep cgroup tmpfs on /sys/fs/cgroup type tmpfs (ro,nosuid,nodev,noexec,mode=755) cgroup on /sys/fs/cgroup/systemd type cgroup (rw,nosuid,nodev,noexec,relatime,xattr,release_agent=/lib/systemd/systemd-cgroups-agent,name=systemd)
#cat /proc/filesystems | grep cgroup nodev cgroup nodev cgroup2
Thanks, Feng
As a result, cpu and memory node hot add may fail to update the cpu and node masks of the top cpuset to include the newly added cpu or node in a cgroup v2 environment.
smp_init() is called after the first two init functions. So we don't have a complete list of active cpus and memory nodes until later in cpuset_init_smp() which is the right time to set up effective_cpus and effective_mems.
To fix this problem, the potentially incorrect cpus_allowed & mems_allowed setup in cpuset_init_smp() are removed. For cgroup v2 systems, the initial cpuset_bind() call will set them up correctly. For cgroup v1 systems, the second call to cpuset_bind() will do the right setup.
cc: stable@vger.kernel.org Signed-off-by: Waiman Long longman@redhat.com
kernel/cgroup/cpuset.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index 9390bfd9f1cd..6bd8f5ef40fe 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -3390,8 +3390,9 @@ static struct notifier_block cpuset_track_online_nodes_nb = { */ void __init cpuset_init_smp(void) {
- cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
- top_cpuset.mems_allowed = node_states[N_MEMORY];
So can we keep line cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
and only remove line top_cpuset.mems_allowed = node_states[N_MEMORY]; ?
That may cause cpusets.cpu to be set incorrectly for systems using cgroup v2. What is really important is that effective_cpus and effective_mems are set correctly.
Cheers, Longman
On 4/26/22 21:06, Feng Tang wrote:
On Tue, Apr 26, 2022 at 10:58:21PM +0800, Waiman Long wrote:
On 4/25/22 23:23, Feng Tang wrote:
Hi Waiman,
On Mon, Apr 25, 2022 at 11:55:05AM -0400, Waiman Long wrote:
There are 3 places where the cpu and node masks of the top cpuset can be initialized in the order they are executed:
- start_kernel -> cpuset_init()
- start_kernel -> cgroup_init() -> cpuset_bind()
- kernel_init_freeable() -> do_basic_setup() -> cpuset_init_smp()
The first cpuset_init() function just sets all the bits in the masks. The last one executed is cpuset_init_smp() which sets up cpu and node masks suitable for v1, but not v2. cpuset_bind() does the right setup for both v1 and v2.
For systems with cgroup v2 setup, cpuset_bind() is called once. For systems with cgroup v1 setup, cpuset_bind() is called twice. It is first called before cpuset_init_smp() in cgroup v2 mode. Then it is called again when cgroup v1 filesystem is mounted in v1 mode after cpuset_init_smp().
[ 2.609781] cpuset_bind() called - v2 = 1 [ 3.079473] cpuset_init_smp() called [ 7.103710] cpuset_bind() called - v2 = 0I run some test, on a server with centOS, this did happen that cpuset_bind() is called twice, first as v2 during kernel boot, and then as v1 post-boot.
However on a QEMU running with a basic debian rootfs image, the second call of cpuset_bind() didn't happen.
The first time cpuset_bind() is called in cgroup_init(), the kernel doesn't know if userspace is going to mount v1 or v2 cgroup. By default, it is assumed to be v2. However, if userspace mounts the cgroup v1 filesystem for cpuset, cpuset_bind() will be run at this point by rebind_subsystem() to set up cgroup v1 environment and cpus_allowed/mems_allowed will be correctly set at this point. Mounting the cgroup v2 filesystem, however, does not cause rebind_subsystem() to run and hence cpuset_bind() is not called again.
Is the QEMU setup not mounting any cgroup filesystem at all? If so, does it matter whether v1 or v2 setup is used?
When I got the cpuset binding error report, I tried first on qemu to reproduce and failed (due to there was no memory hotplug), then I reproduced it on a real server. For both system, I used "cgroup_no_v1=all" cmdline parameter to test cgroup-v2, could this be the reason? (TBH, this is the first time I use cgroup-v2).
Here is the info dump:
# mount | grep cgroup tmpfs on /sys/fs/cgroup type tmpfs (ro,nosuid,nodev,noexec,mode=755) cgroup on /sys/fs/cgroup/systemd type cgroup (rw,nosuid,nodev,noexec,relatime,xattr,release_agent=/lib/systemd/systemd-cgroups-agent,name=systemd)
#cat /proc/filesystems | grep cgroup nodev cgroup nodev cgroup2
Thanks, Feng
For cgroup v2, cpus_allowed should be set to cpu_possible_mask and mems_allowed to node_possible_map as is done in the first invocation of cpuset_bind(). That is the correct behavior.
Cheers, Longman
On Tue, Apr 26, 2022 at 10:34:21PM -0400, Waiman Long wrote:
On 4/26/22 21:06, Feng Tang wrote:
On Tue, Apr 26, 2022 at 10:58:21PM +0800, Waiman Long wrote:
On 4/25/22 23:23, Feng Tang wrote:
Hi Waiman,
On Mon, Apr 25, 2022 at 11:55:05AM -0400, Waiman Long wrote:
There are 3 places where the cpu and node masks of the top cpuset can be initialized in the order they are executed:
- start_kernel -> cpuset_init()
- start_kernel -> cgroup_init() -> cpuset_bind()
- kernel_init_freeable() -> do_basic_setup() -> cpuset_init_smp()
The first cpuset_init() function just sets all the bits in the masks. The last one executed is cpuset_init_smp() which sets up cpu and node masks suitable for v1, but not v2. cpuset_bind() does the right setup for both v1 and v2.
For systems with cgroup v2 setup, cpuset_bind() is called once. For systems with cgroup v1 setup, cpuset_bind() is called twice. It is first called before cpuset_init_smp() in cgroup v2 mode. Then it is called again when cgroup v1 filesystem is mounted in v1 mode after cpuset_init_smp().
[ 2.609781] cpuset_bind() called - v2 = 1 [ 3.079473] cpuset_init_smp() called [ 7.103710] cpuset_bind() called - v2 = 0I run some test, on a server with centOS, this did happen that cpuset_bind() is called twice, first as v2 during kernel boot, and then as v1 post-boot.
However on a QEMU running with a basic debian rootfs image, the second call of cpuset_bind() didn't happen.
The first time cpuset_bind() is called in cgroup_init(), the kernel doesn't know if userspace is going to mount v1 or v2 cgroup. By default, it is assumed to be v2. However, if userspace mounts the cgroup v1 filesystem for cpuset, cpuset_bind() will be run at this point by rebind_subsystem() to set up cgroup v1 environment and cpus_allowed/mems_allowed will be correctly set at this point. Mounting the cgroup v2 filesystem, however, does not cause rebind_subsystem() to run and hence cpuset_bind() is not called again.
Is the QEMU setup not mounting any cgroup filesystem at all? If so, does it matter whether v1 or v2 setup is used?
When I got the cpuset binding error report, I tried first on qemu to reproduce and failed (due to there was no memory hotplug), then I reproduced it on a real server. For both system, I used "cgroup_no_v1=all" cmdline parameter to test cgroup-v2, could this be the reason? (TBH, this is the first time I use cgroup-v2).
Here is the info dump:
# mount | grep cgroup tmpfs on /sys/fs/cgroup type tmpfs (ro,nosuid,nodev,noexec,mode=755) cgroup on /sys/fs/cgroup/systemd type cgroup (rw,nosuid,nodev,noexec,relatime,xattr,release_agent=/lib/systemd/systemd-cgroups-agent,name=systemd)
#cat /proc/filesystems | grep cgroup nodev cgroup nodev cgroup2
Thanks, Feng
For cgroup v2, cpus_allowed should be set to cpu_possible_mask and mems_allowed to node_possible_map as is done in the first invocation of cpuset_bind(). That is the correct behavior.
OK. For the cgroup v2 mem binding problem with hot-added nodes, I retested today, and it can't be reproduced with this patch. So feel free to add:
Tested-by: Feng Tang feng.tang@intel.com
Thanks, Feng
Cheers, Longman
Hello.
On Mon, Apr 25, 2022 at 11:55:05AM -0400, Waiman Long longman@redhat.com wrote:
smp_init() is called after the first two init functions. So we don't have a complete list of active cpus and memory nodes until later in cpuset_init_smp() which is the right time to set up effective_cpus and effective_mems.
Yes.
setup_arch prefill_possible_map cpuset_init (1) cgroup_init cpuset_bind (2a) ... kernel_init kernel_init_freeable ... cpuset_init_smp (3) ... ... cpuset_bind (2b)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index 9390bfd9f1cd..6bd8f5ef40fe 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -3390,8 +3390,9 @@ static struct notifier_block cpuset_track_online_nodes_nb = { */ void __init cpuset_init_smp(void) {
- cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
- top_cpuset.mems_allowed = node_states[N_MEMORY];
- /*
* cpus_allowd/mems_allowed will be properly set up in cpuset_bind().*/
IIUC, the comment should say
* cpus_allowed/mems_allowed were (v2) or will be (v1) properly set up in cpuset_bind().
(nit)
Reviewed-by: Michal Koutný mkoutny@suse.com
On 4/27/22 09:53, Michal Koutný wrote:
Hello.
On Mon, Apr 25, 2022 at 11:55:05AM -0400, Waiman Long longman@redhat.com wrote:
smp_init() is called after the first two init functions. So we don't have a complete list of active cpus and memory nodes until later in cpuset_init_smp() which is the right time to set up effective_cpus and effective_mems.
Yes.
setup_arch prefill_possible_map cpuset_init (1) cgroup_init cpuset_bind (2a) ... kernel_init kernel_init_freeable ... cpuset_init_smp (3) ... ... cpuset_bind (2b)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index 9390bfd9f1cd..6bd8f5ef40fe 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -3390,8 +3390,9 @@ static struct notifier_block cpuset_track_online_nodes_nb = { */ void __init cpuset_init_smp(void) {
- cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
- top_cpuset.mems_allowed = node_states[N_MEMORY];
- /*
* cpus_allowd/mems_allowed will be properly set up in cpuset_bind().*/IIUC, the comment should say
* cpus_allowed/mems_allowed were (v2) or will be (v1) properly set up in cpuset_bind().(nit)
Reviewed-by: Michal Koutný mkoutny@suse.com
Thanks for the review. I plan to post v3 with updated commit log and comment soon.
Cheers, Longman
linux-stable-mirror@lists.linaro.org
-
Feng Tang -
Michal Koutný -
Waiman Long