Re: [PATCH] workqueue: missing idr_preload_end() in worker_pool_assign_id()

Re: [PATCH] workqueue: missing idr_preload_end() in worker_pool_assign_id()

[Tejun Heo]

(cc'ing Stephen and linux-next)

Hello, Artem.

On Wed, Mar 13, 2013 at 09:51:32PM +0400, Artem Savkov wrote:
> Added missing idr_preload_end() call in worker_pool_assign_id().
> Without it preemption stays disabled resulting in lots of "scheduling while
> atomic" BUGs during boot.
...
> Introduced in "workqueue: convert to idr_alloc()"

That patch doesn't use idr_preload().  It looks like the issue is
introduced during linux-next merge of wq/for-3.10 and idr patches in
-mm.  Stephen, can you please add idr_preload_end() to the merge
patch?

Once the idr patches land in Linus' tree, I'll resolve the conflict
from wq tree side.

Thanks.

-- 
tejun

Re: [PATCH] workqueue: missing idr_preload_end() in worker_pool_assign_id()

[Stephen Rothwell]

[-- Attachment #1: Type: text/plain, Size: 1218 bytes --]

Hi Tejun,

On Wed, 13 Mar 2013 10:57:45 -0700 Tejun Heo <tj@kernel.org> wrote:
>
> (cc'ing Stephen and linux-next)
> 
> Hello, Artem.
> 
> On Wed, Mar 13, 2013 at 09:51:32PM +0400, Artem Savkov wrote:
> > Added missing idr_preload_end() call in worker_pool_assign_id().
> > Without it preemption stays disabled resulting in lots of "scheduling while
> > atomic" BUGs during boot.
> ...
> > Introduced in "workqueue: convert to idr_alloc()"
> 
> That patch doesn't use idr_preload().  It looks like the issue is
> introduced during linux-next merge of wq/for-3.10 and idr patches in
> -mm.  Stephen, can you please add idr_preload_end() to the merge
> patch?

Oops, sorry about that.  I assume it needs to be added just after the
spin_unlock_irq() but still inside the loop?

> Once the idr patches land in Linus' tree, I'll resolve the conflict
> from wq tree side.

Or you could do what Linus prefers and just tell him how to resolve the
conflict and thereby avoid a back merge or rebase (or provide him with a
separate branch that does the back merge with resolution in addition to
the unmerged branch to pull).

-- 
Cheers,
Stephen Rothwell                    sfr@canb.auug.org.au

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH] workqueue: missing idr_preload_end() in worker_pool_assign_id()

[Tejun Heo]

Hello,

On Wed, Mar 13, 2013 at 2:51 PM, Stephen Rothwell <sfr@canb.auug.org.au> wrote:
>> That patch doesn't use idr_preload().  It looks like the issue is
>> introduced during linux-next merge of wq/for-3.10 and idr patches in
>> -mm.  Stephen, can you please add idr_preload_end() to the merge
>> patch?
>
> Oops, sorry about that.  I assume it needs to be added just after the
> spin_unlock_irq() but still inside the loop?

Yeap. Andrew already has the change, I think.

>> Once the idr patches land in Linus' tree, I'll resolve the conflict
>> from wq tree side.
>
> Or you could do what Linus prefers and just tell him how to resolve the
> conflict and thereby avoid a back merge or rebase (or provide him with a
> separate branch that does the back merge with resolution in addition to
> the unmerged branch to pull).

That part of code is gonna see more changes and I don't wanna build on
top of the deprecated interface, so the back-pull is actually
justified - wq tree actually wants the receive the particular change
to build on top.

Thanks.

-- 
tejun