Skip to contentSkip to conversation

kernel/fork: beware of __put_task_struct calling context

PATCH Superseded Wander Lairson Costa 4 participants 16 messages May 16, 2023 – Jun 9, 2023

Subsystems (4)

v9 of 10 Superseded Latest is v10 →

lore
Share & tools
lore
#0 lore raw

PATCH v9 1/1 kernel/fork: beware of __put_task_struct calling context

Under PREEMPT_RT, __put_task_struct() indirectly acquires sleeping locks. Therefore, it can't be called from an non-preemptible context.

One practical example is splat inside inactive_task_timer(), which is called in a interrupt context:

CPU: 1 PID: 2848 Comm: life Kdump: loaded Tainted: G W ---------
 Hardware name: HP ProLiant DL388p Gen8, BIOS P70 07/15/2012
 Call Trace:
 dump_stack_lvl+0x57/0x7d
 mark_lock_irq.cold+0x33/0xba
 ? stack_trace_save+0x4b/0x70
 ? save_trace+0x55/0x150
 mark_lock+0x1e7/0x400
 mark_usage+0x11d/0x140
 __lock_acquire+0x30d/0x930
 lock_acquire.part.0+0x9c/0x210
 ? refill_obj_stock+0x3d/0x3a0
 ? rcu_read_lock_sched_held+0x3f/0x70
 ? trace_lock_acquire+0x38/0x140
 ? lock_acquire+0x30/0x80
 ? refill_obj_stock+0x3d/0x3a0
 rt_spin_lock+0x27/0xe0
 ? refill_obj_stock+0x3d/0x3a0
 refill_obj_stock+0x3d/0x3a0
 ? inactive_task_timer+0x1ad/0x340
 kmem_cache_free+0x357/0x560
 inactive_task_timer+0x1ad/0x340
 ? switched_from_dl+0x2d0/0x2d0
 __run_hrtimer+0x8a/0x1a0
 __hrtimer_run_queues+0x91/0x130
 hrtimer_interrupt+0x10f/0x220
 __sysvec_apic_timer_interrupt+0x7b/0xd0
 sysvec_apic_timer_interrupt+0x4f/0xd0
 ? asm_sysvec_apic_timer_interrupt+0xa/0x20
 asm_sysvec_apic_timer_interrupt+0x12/0x20
 RIP: 0033:0x7fff196bf6f5

Instead of calling __put_task_struct() directly, we defer it using call_rcu(). A more natural approach would use a workqueue, but since in PREEMPT_RT, we can't allocate dynamic memory from atomic context, the code would become more complex because we would need to put the work_struct instance in the task_struct and initialize it when we allocate a new task_struct.

Changelog
=========

v1:
* Initial implementation fixing the splat.

v2:
* Isolate the logic in its own function.
* Fix two more cases caught in review.

v3:
* Change __put_task_struct() to handle the issue internally.

v4:
* Explain why call_rcu() is safe to call from interrupt context.

v5:
* Explain why __put_task_struct() doesn't conflict with put_task_sruct_rcu_user.

v6:
* As per Sebastian's review, revert back the implementation of v2 with a distinct function.
* Add a check in put_task_struct() to warning when called from a non-sleepable context.
* Address more call sites.

v7:
* Fix typos.
* Add an explanation why the new function doesn't conflict with delayed_free_task().

v8:
* Bring back v5.
* Fix coding style.

v9:
* Reorganize to not need ___put_task_struct() by Oleg's suggestion.

---
 include/linux/sched/task.h | 28 +++++++++++++++++++++++++++-
 kernel/fork.c              |  8 ++++++++
 2 files changed, 35 insertions(+), 1 deletion(-)
include/linux/sched/task.h +27 -1
index d6c48163c6de..9bcb9535d4e1 100644
@@ -112,10 +112,36 @@ static inline struct task_struct *get_task_struct(struct task_struct *t)
 }
 
 extern void __put_task_struct(struct task_struct *t);
+extern void __put_task_struct_rcu_cb(struct rcu_head *rhp);
 
 static inline void put_task_struct(struct task_struct *t)
 {
-	if (refcount_dec_and_test(&t->usage))
+	if (!refcount_dec_and_test(&t->usage))
+		return;
+
+	/*
+	 * under PREEMPT_RT, we can't call put_task_struct
+	 * in atomic context because it will indirectly
+	 * acquire sleeping locks.
+	 *
+	 * call_rcu() will schedule delayed_put_task_struct_rcu()
+	 * to be called in process context.
+	 *
+	 * __put_task_struct() is called when
+	 * refcount_dec_and_test(&t->usage) succeeds.
+	 *
+	 * This means that it can't "conflict" with
+	 * put_task_struct_rcu_user() which abuses ->rcu the same
+	 * way; rcu_users has a reference so task->usage can't be
+	 * zero after rcu_users 1 -> 0 transition.
+	 *
+	 * delayed_free_task() also uses ->rcu, but it is only called
+	 * when it fails to fork a process. Therefore, there is no
+	 * way it can conflict with put_task_struct().
+	 */
+	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !preemptible())
+		call_rcu(&t->rcu, __put_task_struct_rcu_cb);
+	else
 		__put_task_struct(t);
 }
 
kernel/fork.c +8 -0
index 08969f5aa38d..fd3bb4a554c4 100644
@@ -846,6 +846,14 @@ void __put_task_struct(struct task_struct *tsk)
 }
 EXPORT_SYMBOL_GPL(__put_task_struct);
 
+void __put_task_struct_rcu_cb(struct rcu_head *rhp)
+{
+	struct task_struct *task = container_of(rhp, struct task_struct, rcu);
+
+	__put_task_struct(task);
+}
+EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
+
 void __init __weak arch_task_cache_init(void) { }
 
 /*
Signature · 1 line
2.40.1
#1 lore raw

kernel/fork.c void __put_task_struct(struct task_struct *tsk) · line 855

Quoted 7 lines · Wander Lairson Costa
kernel/fork.c +7 -0
+void __put_task_struct_rcu_cb(struct rcu_head *rhp)
+{
+	struct task_struct *task = container_of(rhp, struct task_struct, rcu);
+
+	__put_task_struct(task);
+}
+EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);

Why does this need to be exported when its only caller is within the main kernel and cannot possibly be built as a module?

maintainer
#2 lore raw
Quoted 11 lines · On Tue, 16 May 2023 20:24:04 +0100 Matthew Wilcox <[email protected]> wrote:

kernel/fork.c void __put_task_struct(struct task_struct *tsk) · line 855

Quoted 7 lines · Wander Lairson Costa
kernel/fork.c +7 -0
+void __put_task_struct_rcu_cb(struct rcu_head *rhp)
+{
+	struct task_struct *task = container_of(rhp, struct task_struct, rcu);
+
+	__put_task_struct(task);
+}
+EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);

Why does this need to be exported when its only caller is within the main kernel and cannot possibly be built as a module?

It's referenced by inlined put_task_struct(), which is called from all over.

However I believe the above definition could be inside #ifdef CONFIG_PREEMPT_RT, to save a scrap of resources?

#3 lore raw
Quoted 14 lines · Andrew Morton
Quoted 11 lines · On Tue, 16 May 2023 20:24:04 +0100 Matthew Wilcox <[email protected]> wrote:

kernel/fork.c void __put_task_struct(struct task_struct *tsk) · line 855

Quoted 7 lines · Wander Lairson Costa
kernel/fork.c +7 -0
+void __put_task_struct_rcu_cb(struct rcu_head *rhp)
+{
+	struct task_struct *task = container_of(rhp, struct task_struct, rcu);
+
+	__put_task_struct(task);
+}
+EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);

Why does this need to be exported when its only caller is within the main kernel and cannot possibly be built as a module?

Quoted 2 lines

It's referenced by inlined put_task_struct(), which is called from all over.

Oh, I missed that put_task_struct() was still inlined. Should it be? It seems quite large now.

Quoted 2 lines

However I believe the above definition could be inside #ifdef CONFIG_PREEMPT_RT, to save a scrap of resources?

maintainer
#4 lore raw
Quoted 3 lines · On Tue, 16 May 2023 22:41:18 +0100 Matthew Wilcox <[email protected]> wrote:

Oh, I missed that put_task_struct() was still inlined. Should it be? It seems quite large now.

It's not significantly worse because of this patch. In fact, it's unchanged for non-RT kernels.

Possibly put_task_struct() *should* be uninlined, because it made the mistake of using the dang refcount stuff, which never saw a byte which it couldn't consume :(

I mean...

fs/open.c renamed from fs/open.c~a +6 -0
@@ -1572,3 +1572,9 @@ int stream_open(struct inode *inode, str
 }
 
 EXPORT_SYMBOL(stream_open);
+
+#include <linux/refcount.h>
+bool foo(refcount_t *r)
+{
+	return refcount_dec_and_test(r);
+}

_

is worth

339 bytes of text for an allmodconfig build
67 bytes of text for an allnoconfig build
77 bytes of text for a defconfig build

#5 lore raw
In reply to Wander Lairson Costa
Quoted 22 lines · Wander Lairson Costa

static inline void put_task_struct(struct task_struct *t) {

include/linux/sched/task.h +18 -1
-	if (refcount_dec_and_test(&t->usage))
+	if (!refcount_dec_and_test(&t->usage))
+		return;
+
+	/*
+	 * under PREEMPT_RT, we can't call put_task_struct
+	 * in atomic context because it will indirectly
+	 * acquire sleeping locks.
+	 *
+	 * call_rcu() will schedule delayed_put_task_struct_rcu()
+	 * to be called in process context.
+	 *
+	 * __put_task_struct() is called when
+	 * refcount_dec_and_test(&t->usage) succeeds.
+	 *
+	 * This means that it can't "conflict" with
+	 * put_task_struct_rcu_user() which abuses ->rcu the same
+	 * way; rcu_users has a reference so task->usage can't be
+	 * zero after rcu_users 1 -> 0 transition.

include/linux/sched/task.h static inline struct task_struct *get_task_struct(struct task_struct *t) · line 146

Quoted 10 lines
include/linux/sched/task.h +8 -0
+	 *
+	 * delayed_free_task() also uses ->rcu, but it is only called
+	 * when it fails to fork a process. Therefore, there is no
+	 * way it can conflict with put_task_struct().
+	 */
+	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !preemptible())
+		call_rcu(&t->rcu, __put_task_struct_rcu_cb);
+	else
 		__put_task_struct(t);
 }

LGTM but we still need to understand the possible problems with CONFIG_PROVE_RAW_LOCK_NESTING ...

Again, I'll try to investigate when I have time although I am not sure I can really help.

Perhaps you too can try to do this ? ;)

Oleg.

#6 lore raw
Quoted 35 lines · Oleg Nesterov
Quoted 22 lines · Wander Lairson Costa

static inline void put_task_struct(struct task_struct *t) {

include/linux/sched/task.h +18 -1
-     if (refcount_dec_and_test(&t->usage))
+     if (!refcount_dec_and_test(&t->usage))
+             return;
+
+     /*
+      * under PREEMPT_RT, we can't call put_task_struct
+      * in atomic context because it will indirectly
+      * acquire sleeping locks.
+      *
+      * call_rcu() will schedule delayed_put_task_struct_rcu()
+      * to be called in process context.
+      *
+      * __put_task_struct() is called when
+      * refcount_dec_and_test(&t->usage) succeeds.
+      *
+      * This means that it can't "conflict" with
+      * put_task_struct_rcu_user() which abuses ->rcu the same
+      * way; rcu_users has a reference so task->usage can't be
+      * zero after rcu_users 1 -> 0 transition.

include/linux/sched/task.h static inline struct task_struct *get_task_struct(struct task_struct *t) · line 146

Quoted 10 lines
include/linux/sched/task.h +8 -0
+      *
+      * delayed_free_task() also uses ->rcu, but it is only called
+      * when it fails to fork a process. Therefore, there is no
+      * way it can conflict with put_task_struct().
+      */
+     if (IS_ENABLED(CONFIG_PREEMPT_RT) && !preemptible())
+             call_rcu(&t->rcu, __put_task_struct_rcu_cb);
+     else
              __put_task_struct(t);
 }
Quoted 6 lines

LGTM but we still need to understand the possible problems with CONFIG_PROVE_RAW_LOCK_NESTING ...

Again, I'll try to investigate when I have time although I am not sure I can really help.

Perhaps you too can try to do this ? ;)

FWIW, I tested this patch with CONFIG_PROVE_LOCK_NESTING in RT and stock kernels. No splat happened.

Quoted 2 lines

Oleg.

#7 lore raw
Quoted 5 lines · Wander Lairson Costa
Quoted 3 lines · Oleg Nesterov

LGTM but we still need to understand the possible problems with CONFIG_PROVE_RAW_LOCK_NESTING ...

Quoted 7 lines
Quoted 4 lines

Again, I'll try to investigate when I have time although I am not sure I can really help.

Perhaps you too can try to do this ? ;)

FWIW, I tested this patch with CONFIG_PROVE_LOCK_NESTING in RT and stock kernels. No splat happened.

Strange... FYI, I am running the kernel with this patch

	diff --git a/kernel/sys.c b/kernel/sys.c
	index 339fee3eff6a..3169cceddf3b 100644
	--- a/kernel/sys.c
	+++ b/kernel/sys.c
	@@ -2412,6 +2412,17 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
	 
		error = 0;
		switch (option) {
	+	case 666: {
	+		static DEFINE_SPINLOCK(l);
	+		static DEFINE_RAW_SPINLOCK(r);
	+
	+		raw_spin_lock(&r);
	+		spin_lock(&l);
	+		spin_unlock(&l);
	+		raw_spin_unlock(&r);
	+
	+		break;
	+	}
		case PR_SET_PDEATHSIG:
			if (!valid_signal(arg2)) {
				error = -EINVAL;

applied (because I am too lazy to compile a module ;) and

	# perl -e 'syscall 157,666'

triggers the lockdep bug

	=============================
	[ BUG: Invalid wait context ]
	6.4.0-rc2-00018-g4d6d4c7f541d-dirty #1176 Not tainted
	-----------------------------
	perl/35 is trying to lock:
	ffffffff81c4cc18 (l){....}-{3:3}, at: __do_sys_prctl+0x21b/0x87b
	other info that might help us debug this:
	context-{5:5}
	...

as expected.

Looks like your testing was wrong... Or maybe you missed another lockdep problem ? Did you check dmesg? Perhaps lockdep detected another bug,say, even at boot time ? In this case debug_locks_off() sets debug_locks = 0 and this disables lockdep.

Oleg.

#8 lore raw
Quoted 40 lines · Oleg Nesterov
Quoted 5 lines · Wander Lairson Costa
Quoted 3 lines · Oleg Nesterov

LGTM but we still need to understand the possible problems with CONFIG_PROVE_RAW_LOCK_NESTING ...

Quoted 7 lines
Quoted 4 lines

Again, I'll try to investigate when I have time although I am not sure I can really help.

Perhaps you too can try to do this ? ;)

FWIW, I tested this patch with CONFIG_PROVE_LOCK_NESTING in RT and stock kernels. No splat happened.

Strange... FYI, I am running the kernel with this patch

        diff --git a/kernel/sys.c b/kernel/sys.c
        index 339fee3eff6a..3169cceddf3b 100644
        --- a/kernel/sys.c
        +++ b/kernel/sys.c
        @@ -2412,6 +2412,17 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,

                error = 0;
                switch (option) {
        +       case 666: {
        +               static DEFINE_SPINLOCK(l);
        +               static DEFINE_RAW_SPINLOCK(r);
        +
        +               raw_spin_lock(&r);
        +               spin_lock(&l);
        +               spin_unlock(&l);
        +               raw_spin_unlock(&r);
        +
        +               break;
        +       }
                case PR_SET_PDEATHSIG:
                        if (!valid_signal(arg2)) {
                                error = -EINVAL;
Quoted 2 lines

applied (because I am too lazy to compile a module ;) and

FWIW, I converted it to a module [1]

Quoted 14 lines
        # perl -e 'syscall 157,666'

triggers the lockdep bug

        =============================
        [ BUG: Invalid wait context ]
        6.4.0-rc2-00018-g4d6d4c7f541d-dirty #1176 Not tainted
        -----------------------------
        perl/35 is trying to lock:
        ffffffff81c4cc18 (l){....}-{3:3}, at: __do_sys_prctl+0x21b/0x87b
        other info that might help us debug this:
        context-{5:5}
        ...
Quoted 2 lines

as expected.

Yeah, I tried it here and I had the same results, but only in the RT kernel. But running the reproducer for put_task_struct(), works fine.

Quoted 6 lines

Looks like your testing was wrong... Or maybe you missed another lockdep problem ? Did you check dmesg? Perhaps lockdep detected another bug,say, even at boot time ? In this case debug_locks_off() sets debug_locks = 0 and this disables lockdep.

Oleg.

#9 lore raw
Quoted 42 lines · Wander Lairson Costa
Quoted 40 lines · Oleg Nesterov
Quoted 5 lines · Wander Lairson Costa
Quoted 3 lines · Oleg Nesterov

LGTM but we still need to understand the possible problems with CONFIG_PROVE_RAW_LOCK_NESTING ...

Quoted 7 lines
Quoted 4 lines

Again, I'll try to investigate when I have time although I am not sure I can really help.

Perhaps you too can try to do this ? ;)

FWIW, I tested this patch with CONFIG_PROVE_LOCK_NESTING in RT and stock kernels. No splat happened.

Strange... FYI, I am running the kernel with this patch

        diff --git a/kernel/sys.c b/kernel/sys.c
        index 339fee3eff6a..3169cceddf3b 100644
        --- a/kernel/sys.c
        +++ b/kernel/sys.c
        @@ -2412,6 +2412,17 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,

                error = 0;
                switch (option) {
        +       case 666: {
        +               static DEFINE_SPINLOCK(l);
        +               static DEFINE_RAW_SPINLOCK(r);
        +
        +               raw_spin_lock(&r);
        +               spin_lock(&l);
        +               spin_unlock(&l);
        +               raw_spin_unlock(&r);
        +
        +               break;
        +       }
                case PR_SET_PDEATHSIG:
                        if (!valid_signal(arg2)) {
                                error = -EINVAL;
Quoted 4 lines
Quoted 2 lines

applied (because I am too lazy to compile a module ;) and

FWIW, I converted it to a module [1]

where is [1] ? not that I think this matters though...

Quoted 14 lines
Quoted 14 lines
        # perl -e 'syscall 157,666'

triggers the lockdep bug

        =============================
        [ BUG: Invalid wait context ]
        6.4.0-rc2-00018-g4d6d4c7f541d-dirty #1176 Not tainted
        -----------------------------
        perl/35 is trying to lock:
        ffffffff81c4cc18 (l){....}-{3:3}, at: __do_sys_prctl+0x21b/0x87b
        other info that might help us debug this:
        context-{5:5}
        ...
Quoted 4 lines
Quoted 2 lines

as expected.

Yeah, I tried it here and I had the same results,

OK,

Quoted 1 line

but only in the RT kernel

this again suggests that your testing was wrong or I am totally confused (quite possible, I know nothing about RT). I did the testing without CONFIG_PREEMPT_RT.

Quoted 1 line

But running the reproducer for put_task_struct(), works fine.

which reproducer ?

Oleg.

#10 lore raw
Quoted 68 lines · Oleg Nesterov
Quoted 42 lines · Wander Lairson Costa
Quoted 40 lines · Oleg Nesterov
Quoted 5 lines · Wander Lairson Costa
Quoted 3 lines · Oleg Nesterov

LGTM but we still need to understand the possible problems with CONFIG_PROVE_RAW_LOCK_NESTING ...

Quoted 7 lines
Quoted 4 lines

Again, I'll try to investigate when I have time although I am not sure I can really help.

Perhaps you too can try to do this ? ;)

FWIW, I tested this patch with CONFIG_PROVE_LOCK_NESTING in RT and stock kernels. No splat happened.

Strange... FYI, I am running the kernel with this patch

        diff --git a/kernel/sys.c b/kernel/sys.c
        index 339fee3eff6a..3169cceddf3b 100644
        --- a/kernel/sys.c
        +++ b/kernel/sys.c
        @@ -2412,6 +2412,17 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,

                error = 0;
                switch (option) {
        +       case 666: {
        +               static DEFINE_SPINLOCK(l);
        +               static DEFINE_RAW_SPINLOCK(r);
        +
        +               raw_spin_lock(&r);
        +               spin_lock(&l);
        +               spin_unlock(&l);
        +               raw_spin_unlock(&r);
        +
        +               break;
        +       }
                case PR_SET_PDEATHSIG:
                        if (!valid_signal(arg2)) {
                                error = -EINVAL;
Quoted 4 lines
Quoted 2 lines

applied (because I am too lazy to compile a module ;) and

FWIW, I converted it to a module [1]

where is [1] ? not that I think this matters though...

Quoted 17 lines
Quoted 16 lines
        # perl -e 'syscall 157,666'

triggers the lockdep bug

        =============================
        [ BUG: Invalid wait context ]
        6.4.0-rc2-00018-g4d6d4c7f541d-dirty #1176 Not tainted
        -----------------------------
        perl/35 is trying to lock:
        ffffffff81c4cc18 (l){....}-{3:3}, at: __do_sys_prctl+0x21b/0x87b
        other info that might help us debug this:
        context-{5:5}
        ...

as expected.

Quoted 9 lines
Quoted 1 line

Yeah, I tried it here and I had the same results,

OK,

Quoted 1 line

but only in the RT kernel

this again suggests that your testing was wrong or I am totally confused (quite possible, I know nothing about RT). I did the testing without CONFIG_PREEMPT_RT.

Hrm, could you please share your .config?

Quoted 4 lines
Quoted 1 line

But running the reproducer for put_task_struct(), works fine.

which reproducer ?

Only now I noticed I didn't add the reproducer to the commit message:

while true; do

    stress-ng --sched deadline --sched-period 1000000000

--sched-runtime 800000000 --sched-deadline 1000000000 --mmapfork 23 -t 20
done

#11 lore raw
Quoted 10 lines · Wander Lairson Costa
Quoted 6 lines · Oleg Nesterov
Quoted 2 lines

but only in the RT kernel

this again suggests that your testing was wrong or I am totally confused (quite possible, I know nothing about RT). I did the testing without CONFIG_PREEMPT_RT.

Hrm, could you please share your .config?

Sure. I do not want to spam the list, I'll send you a private email.

Can you share your kernel module code?

Did you verify that debug_locks != 0 as I asked in my previous email ?

Quoted 5 lines
Quoted 4 lines
Quoted 1 line

But running the reproducer for put_task_struct(), works fine.

which reproducer ?

Quoted 7 lines

Only now I noticed I didn't add the reproducer to the commit message:

while true; do

    stress-ng --sched deadline --sched-period 1000000000

--sched-runtime 800000000 --sched-deadline 1000000000 --mmapfork 23 -t 20
done

Cough ;) I think we need a more simple one to enssure that refcount_sub_and_test(nr, &t->usage) returns true under raw_spin_lock() and then __put_task_struct() actually takes spin_lock().

Oleg.

#15 lore raw
In reply to Wander Lairson Costa
Quoted 10 lines · Wander Lairson Costa
Quoted 6 lines · Oleg Nesterov
Quoted 2 lines

but only in the RT kernel

this again suggests that your testing was wrong or I am totally confused (quite possible, I know nothing about RT). I did the testing without CONFIG_PREEMPT_RT.

Hrm, could you please share your .config?

Sure. I do not want to spam the list, I'll send you a private email.

Can you share your kernel module code?

Did you verify that debug_locks != 0 as I asked in my previous email ?

Quoted 5 lines
Quoted 4 lines
Quoted 1 line

But running the reproducer for put_task_struct(), works fine.

which reproducer ?

Quoted 7 lines

Only now I noticed I didn't add the reproducer to the commit message:

while true; do

    stress-ng --sched deadline --sched-period 1000000000

--sched-runtime 800000000 --sched-deadline 1000000000 --mmapfork 23 -t 20
done

Cough ;) I think we need something more simple to ensure that refcount_sub_and_test(nr, &t->usage) returns true under raw_spin_lock() and then __put_task_struct() actually takes spin_lock().

Oleg.

#12 lore raw
In reply to Oleg Nesterov
Quoted 7 lines · Oleg Nesterov
Quoted 5 lines · Wander Lairson Costa
Quoted 3 lines · Oleg Nesterov
Quoted 2 lines

but only in the RT kernel

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

this again suggests that your testing was wrong or I am totally confused (quite possible, I know nothing about RT). I did the testing without CONFIG_PREEMPT_RT.

Hrm, could you please share your .config?

Sure. I do not want to spam the list, I'll send you a private email.

Thanks. I found an unrelated earlier splat in the console code. That's why I couldn't reproduce it in the stock kernel.

Quoted 2 lines

Can you share your kernel module code?

*facepalm* I forgot to post the link: https://github.com/walac/test-prove-lock/

Quoted 21 lines

Did you verify that debug_locks != 0 as I asked in my previous email ?

Quoted 5 lines
Quoted 4 lines
Quoted 1 line

But running the reproducer for put_task_struct(), works fine.

which reproducer ?

Quoted 7 lines

Only now I noticed I didn't add the reproducer to the commit message:

while true; do

    stress-ng --sched deadline --sched-period 1000000000

--sched-runtime 800000000 --sched-deadline 1000000000 --mmapfork 23 -t 20
done

Cough ;) I think we need a more simple one to enssure that refcount_sub_and_test(nr, &t->usage) returns true under raw_spin_lock() and then __put_task_struct() actually takes spin_lock().

Oleg.

#13 lore raw
Quoted 3 lines · Wander Lairson Costa

Thanks. I found an unrelated earlier splat in the console code. That's why I couldn't reproduce it in the stock kernel.

As expected...

So... Not sure what can I say ;) can you verify that this patch doesn't solve the issues with CONFIG_PROVE_RAW_LOCK_NESTING pointed out by Sebastian? Using stress-ng or anything else.

This is not that bad, unless I am totally confused the current code (without your patch) has the same problem (otherwise we wouldn't need this fix).

But perhaps you can make 2/2 which adds the DEFINE_WAIT_OVERRIDE_MAP() hack as Peter suggested?

Oleg.

#14 lore raw
Quoted 12 lines · Oleg Nesterov
Quoted 3 lines · Wander Lairson Costa

Thanks. I found an unrelated earlier splat in the console code. That's why I couldn't reproduce it in the stock kernel.

As expected...

So... Not sure what can I say ;) can you verify that this patch doesn't solve the issues with CONFIG_PROVE_RAW_LOCK_NESTING pointed out by Sebastian? Using stress-ng or anything else.

I managed to test it without a console. No issues happened in the stock kernel.

Quoted 3 lines

This is not that bad, unless I am totally confused the current code (without your patch) has the same problem (otherwise we wouldn't need this fix).

That's my understanding as well.

Quoted 3 lines

But perhaps you can make 2/2 which adds the DEFINE_WAIT_OVERRIDE_MAP() hack as Peter suggested?

Yes, sure. I would like to get the issue reproduced in practice to make sure I am really fixing the problem. But I can live with that.

Quoted 2 lines

Oleg.