sched: Use kfree_rcu() in do_set_cpus_allowed()
PATCH v3 1/1 sched: Use kfree_rcu() in do_set_cpus_allowed()
Commit 851a723e45d1 ("sched: Always clear user_cpus_ptr in do_set_cpus_allowed()") may call kfree() if user_cpus_ptr was previously set. Unfortunately, some of the callers of do_set_cpus_allowed() may have pi_lock held when calling it. So the following splats may be printed especially when running with a PREEMPT_RT kernel:
WARNING: possible circular locking dependency detected
BUG: sleeping function called from invalid context
To avoid these problems, kfree_rcu() is used instead. An internal cpumask_rcuhead union is created for the sole purpose of facilitating the use of kfree_rcu() to free the cpumask.
--- kernel/sched/core.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-)
[v3: Fix build problem reported by kernel test robot]
kernel/sched/core.c +17 -3
@@ -2604,9 +2604,19 @@ void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) .user_mask = NULL, .flags = SCA_USER, /* clear the user requested mask */ }; +union cpumask_rcuhead { +cpumask_t cpumask; +struct rcu_head rcu; + };__do_set_cpus_allowed (p, &ac); -kfree (ac.user_mask); + +/* +* Because this is called with p->pi_lock held, it is not possible +* to use kfree() here (when PREEMPT_RT=y), therefore punt to using +* kfree_rcu(). +*/ +kfree_rcu ((union cpumask_rcuhead *)ac.user_mask, rcu); } int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src,
@@ -8220,7 +8230,7 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)struct affinity_context ac;struct cpumask *user_mask;struct task_struct *p; -int retval; +int retval, size;rcu_read_lock ();
@@ -8253,7 +8263,11 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)if (retval) goto out_put_task; - user_mask =kmalloc (cpumask_size (),GFP_KERNEL ); +/* +* See do_set_cpus_allowed() for the rcu_head usage. +*/ + size =max_t (int,cpumask_size (),sizeof (struct rcu_head )); + user_mask =kmalloc (size,GFP_KERNEL );if (!user_mask) { retval = -ENOMEM ; goto out_put_task;
Signature · 1 line
2.31.1
Quoted 59 lines · Waiman Long
Commit
851a723e45d1("sched: Always clearuser_cpus_ptrindo_set_cpus_allowed()") may callkfree()ifuser_cpus_ptrwas previously set. Unfortunately, some of the callers ofdo_set_cpus_allowed()may havepi_lockheld when calling it. So the following splats may be printed especially when running with aPREEMPT_RTkernel:WARNING: possible circular locking dependency detected
BUG: sleeping function called from invalid contextTo avoid these problems,
kfree_rcu()is used instead. An internalcpumask_rcuheadunion is created for the sole purpose of facilitating the use ofkfree_rcu()to free the cpumask.Fixes 851a723e45d1 ("sched: Always clear user_cpus_ptr in do_set_cpus_allowed()")Suggested-by Peter Zijlstra <[email protected]>Signed-off-by Waiman Long <[email protected]>--- kernel/sched/core.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-)[v3: Fix build problem reported by kernel test robot]
kernel/sched/core.c +12
-2@@ -2604,9 +2604,19 @@ void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) .user_mask = NULL, .flags = SCA_USER, /* clear the user requested mask */ }; +union cpumask_rcuhead { +cpumask_t cpumask; +struct rcu_head rcu; + };__do_set_cpus_allowed (p, &ac); -kfree (ac.user_mask); + +/* +* Because this is called with p->pi_lock held, it is not possible +* to use kfree() here (when PREEMPT_RT=y), therefore punt to using +* kfree_rcu(). +*/ +kfree_rcu ((union cpumask_rcuhead *)ac.user_mask, rcu); } int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src,@@ -8220,7 +8230,7 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)struct affinity_context ac;struct cpumask *user_mask;struct task_struct *p; -int retval; +int retval, size;rcu_read_lock ();@@ -8253,7 +8263,11 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)if (retval) goto out_put_task;
kernel/sched/core.c long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) · line 8273
Quoted 10 lines
kernel/sched/core.c +5
-1- user_mask =kmalloc (cpumask_size (),GFP_KERNEL ); +/* +* See do_set_cpus_allowed() for the rcu_head usage. +*/ + size =max_t (int,cpumask_size (),sizeof (struct rcu_head )); + user_mask =kmalloc (size,GFP_KERNEL );if (!user_mask) { retval = -ENOMEM ; goto out_put_task;
Hi Peter,
Is this patch good enough to be merged as commit 851a723e45d1 ("sched: Always clear user_cpus_ptr in do_set_cpus_allowed()") is now in the Linus' tree?
Thanks,
Longman
Peter Zijlstra
maintainerQuoted 59 lines · Waiman Long
Commit
851a723e45d1("sched: Always clearuser_cpus_ptrindo_set_cpus_allowed()") may callkfree()ifuser_cpus_ptrwas previously set. Unfortunately, some of the callers ofdo_set_cpus_allowed()may havepi_lockheld when calling it. So the following splats may be printed especially when running with aPREEMPT_RTkernel:WARNING: possible circular locking dependency detected
BUG: sleeping function called from invalid contextTo avoid these problems,
kfree_rcu()is used instead. An internalcpumask_rcuheadunion is created for the sole purpose of facilitating the use ofkfree_rcu()to free the cpumask.Fixes 851a723e45d1 ("sched: Always clear user_cpus_ptr in do_set_cpus_allowed()")Suggested-by Peter Zijlstra <[email protected]>Signed-off-by Waiman Long <[email protected]>--- kernel/sched/core.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-)[v3: Fix build problem reported by kernel test robot]
kernel/sched/core.c +12
-2@@ -2604,9 +2604,19 @@ void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) .user_mask = NULL, .flags = SCA_USER, /* clear the user requested mask */ }; +union cpumask_rcuhead { +cpumask_t cpumask; +struct rcu_head rcu; + };__do_set_cpus_allowed (p, &ac); -kfree (ac.user_mask); + +/* +* Because this is called with p->pi_lock held, it is not possible +* to use kfree() here (when PREEMPT_RT=y), therefore punt to using +* kfree_rcu(). +*/ +kfree_rcu ((union cpumask_rcuhead *)ac.user_mask, rcu); } int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src,@@ -8220,7 +8230,7 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)struct affinity_context ac;struct cpumask *user_mask;struct task_struct *p; -int retval; +int retval, size;rcu_read_lock ();@@ -8253,7 +8263,11 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)if (retval) goto out_put_task;
kernel/sched/core.c long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) · line 8273
Quoted 10 lines
kernel/sched/core.c +5
-1- user_mask =kmalloc (cpumask_size (),GFP_KERNEL ); +/* +* See do_set_cpus_allowed() for the rcu_head usage. +*/ + size =max_t (int,cpumask_size (),sizeof (struct rcu_head )); + user_mask =kmalloc (size,GFP_KERNEL );if (!user_mask) { retval = -ENOMEM ; goto out_put_task;
AFAICT you forgot dup_user_cpus_ptr().
Quoted 61 lines · Peter Zijlstra
Quoted 60 lines · Waiman Long
Commit
851a723e45d1("sched: Always clearuser_cpus_ptrindo_set_cpus_allowed()") may callkfree()ifuser_cpus_ptrwas previously set. Unfortunately, some of the callers ofdo_set_cpus_allowed()may havepi_lockheld when calling it. So the following splats may be printed especially when running with aPREEMPT_RTkernel:WARNING: possible circular locking dependency detected
BUG: sleeping function called from invalid contextTo avoid these problems,
kfree_rcu()is used instead. An internalcpumask_rcuheadunion is created for the sole purpose of facilitating the use ofkfree_rcu()to free the cpumask.Fixes 851a723e45d1 ("sched: Always clear user_cpus_ptr in do_set_cpus_allowed()")Suggested-by Peter Zijlstra <[email protected]>Signed-off-by Waiman Long <[email protected]>--- kernel/sched/core.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-)[v3: Fix build problem reported by kernel test robot]
kernel/sched/core.c +12
-2@@ -2604,9 +2604,19 @@ void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) .user_mask = NULL, .flags = SCA_USER, /* clear the user requested mask */ }; +union cpumask_rcuhead { +cpumask_t cpumask; +struct rcu_head rcu; + };__do_set_cpus_allowed (p, &ac); -kfree (ac.user_mask); + +/* +* Because this is called with p->pi_lock held, it is not possible +* to use kfree() here (when PREEMPT_RT=y), therefore punt to using +* kfree_rcu(). +*/ +kfree_rcu ((union cpumask_rcuhead *)ac.user_mask, rcu); } int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src,@@ -8220,7 +8230,7 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)struct affinity_context ac;struct cpumask *user_mask;struct task_struct *p; -int retval; +int retval, size;rcu_read_lock ();@@ -8253,7 +8263,11 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)if (retval) goto out_put_task;
Quoted 10 lines
kernel/sched/core.c long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) · line 8273
Quoted 9 lines
kernel/sched/core.c +5
-1- user_mask =kmalloc (cpumask_size (),GFP_KERNEL ); +/* +* See do_set_cpus_allowed() for the rcu_head usage. +*/ + size =max_t (int,cpumask_size (),sizeof (struct rcu_head )); + user_mask =kmalloc (size,GFP_KERNEL );if (!user_mask) { retval = -ENOMEM ; goto out_put_task;AFAICT you forgot
dup_user_cpus_ptr().
I haven't received any response from you for a while. So it is just a ping. Of course, I am aware that there is another dup_user_cpus_ptr() patch ouststanding. I will of course talk about that when you respond. I also have a pending rwsem patch series waiting for your review:-)
Cheers,
Longman
Peter Zijlstra
maintainerQuoted 21 lines · Waiman Long
Quoted 12 lines
Quoted 12 lines
kernel/sched/core.c +2
-2@@ -8220,7 +8230,7 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)struct affinity_context ac;struct cpumask *user_mask;struct task_struct *p; -int retval; +int retval, size;rcu_read_lock ();@@ -8253,7 +8263,11 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)if (retval) goto out_put_task; - user_mask =kmalloc (cpumask_size (),GFP_KERNEL ); +/* Quoted 8 lines
kernel/sched/core.c long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) · line 8273
Quoted 7 lines
kernel/sched/core.c +4
-0+* See do_set_cpus_allowed() for the rcu_head usage. +*/ + size =max_t (int,cpumask_size (),sizeof (struct rcu_head )); + user_mask =kmalloc (size,GFP_KERNEL );if (!user_mask) { retval = -ENOMEM ; goto out_put_task;AFAICT you forgot
dup_user_cpus_ptr().
Quoted 1 line
I haven't received any response from you for a while. So it is just a ping.
I was out sick :/
Quoted 3 lines
Of course, I am aware that there is another
dup_user_cpus_ptr()patch ouststanding. I will of course talk about that when you respond. I also have a pending rwsem patch series waiting for your review:-)
The point was that dup_user_cpus_ptr() also does an allocation and needs to allocate the possibly bigger size too, no?
Quoted 22 lines · Peter Zijlstra
Quoted 20 lines · Waiman Long
Quoted 20 lines
Quoted 9 lines
kernel/sched/core.c +1
-1@@ -8220,7 +8230,7 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)struct affinity_context ac;struct cpumask *user_mask;struct task_struct *p; -int retval; +int retval, size;rcu_read_lock ();@@ -8253,7 +8263,11 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)if (retval)kernel/sched/core.c long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) · line 8272
Quoted 10 lines
kernel/sched/core.c +5
-1goto out_put_task; - user_mask =kmalloc (cpumask_size (),GFP_KERNEL ); +/* +* See do_set_cpus_allowed() for the rcu_head usage. +*/ + size =max_t (int,cpumask_size (),sizeof (struct rcu_head )); + user_mask =kmalloc (size,GFP_KERNEL );if (!user_mask) { retval = -ENOMEM ; goto out_put_task;AFAICT you forgot
dup_user_cpus_ptr().
Quoted 2 lines
Quoted 1 line
I haven't received any response from you for a while. So it is just a ping.
I was out sick :/
I am sorry to hear that. Hope you are all right now.
Quoted 6 lines
Quoted 4 lines
Of course, I am aware that there is another
dup_user_cpus_ptr()patch ouststanding. I will of course talk about that when you respond. I also have a pending rwsem patch series waiting for your review:-)The point was that
dup_user_cpus_ptr()also does an allocation and needs to allocate the possibly bigger size too, no?
Oh, yes. I missed that. I will combine the two patches into a patch series.
Thanks,
Longman