[PATCH] workqueue: Fix alignment calculation in alloc_cwqs()

[PATCH] workqueue: Fix alignment calculation in alloc_cwqs()

[David Howells]

In the MN10300 arch, we occasionally see an assertion being tripped in
alloc_cwqs() at the following line:

	/* just in case, make sure it's actually aligned */
  --->	BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align));
	return wq->cpu_wq.v ? 0 : -ENOMEM;

The values are:

	wa->cpu_wq.v => 0x902776e0
	align => 0x100

and align is calculated by the following:

	const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS,
				   __alignof__(unsigned long long));

which is wrong.  __alignof__() returns its value in bytes, but:

	1 << WORK_STRUCT_FLAG_BITS

returns the value in bits.  It needs dividing by the number of bits in a byte.

Reported-by: Akira Takeuchi <takeuchi.akr@jp.panasonic.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Mark Salter <msalter@redhat.com>
cc: Tejun Heo <tj@kernel.org>
---

 kernel/workqueue.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)


diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 30acdb7..e29ebd4 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2766,8 +2766,9 @@ static int alloc_cwqs(struct workqueue_struct *wq)
 	 * unsigned long long.
 	 */
 	const size_t size = sizeof(struct cpu_workqueue_struct);
-	const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS,
-				   __alignof__(unsigned long long));
+	const size_t align =
+		max_t(size_t, 1 << (WORK_STRUCT_FLAG_BITS - BITS_PER_BYTE),
+		      __alignof__(unsigned long long));
 #ifdef CONFIG_SMP
 	bool percpu = !(wq->flags & WQ_UNBOUND);
 #else

Re: [Linux-am33-list] [PATCH] workqueue: Fix alignment calculation in alloc_cwqs()

[David Howells]

David Howells <dhowells@redhat.com> wrote:

> +	const size_t align =
> +		max_t(size_t, 1 << (WORK_STRUCT_FLAG_BITS - BITS_PER_BYTE),
> +		      __alignof__(unsigned long long));

Actually, no, that's not right either.  Need to subtract log2(BITS_PER_BYTE).

David

Re: [PATCH] workqueue: Fix alignment calculation in alloc_cwqs()

[Linus Torvalds]

On Mon, Oct 25, 2010 at 2:27 PM, David Howells <dhowells@redhat.com> wrote:
> In the MN10300 arch, we occasionally see an assertion being tripped in
> alloc_cwqs() at the following line:
>
>        /* just in case, make sure it's actually aligned */
>  --->  BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align));
>        return wq->cpu_wq.v ? 0 : -ENOMEM;
>
> The values are:
>
>        wa->cpu_wq.v => 0x902776e0
>        align => 0x100
>
> and align is calculated by the following:
>
>        const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS,
>                                   __alignof__(unsigned long long));
>
> which is wrong.  __alignof__() returns its value in bytes, but:
>
>        1 << WORK_STRUCT_FLAG_BITS
>
> returns the value in bits.  It needs dividing by the number of bits in a byte.

No it doesn't. Those bits really require that many bytes of alignment.

Think about it: if the low 8 bits of the pointer are used for flags,
then the actual pointer itself needs to be aligned to a 256-byte
boundary.

So the code is right. If needs to ask for "1 << WORK_STRUCT_FLAG_BITS"
alignment, and if it doesn't get it (because the allocator is somehow
broken - percpu allocator issues?), things will break.

                     Linus

Re: [PATCH] workqueue: Fix alignment calculation in alloc_cwqs()

[David Howells]

Linus Torvalds <torvalds@linux-foundation.org> wrote:

> No it doesn't. Those bits really require that many bytes of alignment.
> 
> Think about it: if the low 8 bits of the pointer are used for flags,
> then the actual pointer itself needs to be aligned to a 256-byte
> boundary.
> 
> So the code is right. If needs to ask for "1 << WORK_STRUCT_FLAG_BITS"
> alignment, and if it doesn't get it (because the allocator is somehow
> broken - percpu allocator issues?), things will break.

That's a good point.

David