On 11/17/24 13:56, Xiongwei Song wrote:
> It seems there is a conflict between fc0eac57d08c ("slab: pull kmem_cache_open()
> into do_kmem_cache_create()") and 543c5bde72e9 ("mm/slab: Allow cache creation
> to proceed even if sysfs registration fails"). The err is initialized
No conflict, commit fc0eac57d08c is from 6.12-rc1, there's no merge here.
I've introduced the bug by trying to improve Hyeonggon's patch and he
quickly corrected me [1] so the buggy version never even made it to -next.
But here I managed to resurrect it before the git pull.
[1] https://lore.kernel.org/all/195242b4-5471-419a-a350-08fd246973f0@xxxxxxx/
> to -EINVAL in
> the entry of do_kmem_cache_create(), sysfs_slab_add() call can assign
> a value to
> err normally, but the assignment to err variable has been removed in
> Hyeonggon's
> patch, so do_kmem_cache_create() returns -EINVAL normally. We probably can do
> the following fix:
>
> diff --git a/mm/slub.c b/mm/slub.c
> index 73eea67a306b..19630a2da8e1 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -6199,9 +6199,11 @@ int do_kmem_cache_create(struct kmem_cache *s,
> const char *name,
> if (!alloc_kmem_cache_cpus(s))
> goto out;
>
> + /* The error code is not needed any more. */
> + err = 0;
> +
> /* Mutex is not taken during early boot */
> if (slab_state <= UP) {
> - err = 0;
That was indeed the fix.
> goto out;
> }
>
> Regards,
> Xiongwei
>
> On Sun, Nov 17, 2024 at 11:42 AM Sasha Levin <sashal@xxxxxxxxxx> wrote:
>>
>> On Sat, Nov 16, 2024 at 09:43:07PM +0100, Vlastimil Babka wrote:
>> >On 11/16/24 16:35, Sasha Levin wrote:
>> >> [ Obviously I fat-fingered it and didn't add Christian or Al ]
>> >
>> >I have found the problem and looks like I managed to force push an older
>> >broken version of a branch, possibly due to switching between two computers.
>> >Serves me well for amending in some last minute R-b tags. Doing git diff
>> >@{u} to check for unexpected suprirses before pushing the result didn't help
>> >this time, either I forgot or was blind.
>> >
>> >I have deleted the slab-for-6.13 signed tag and pushed the fixed branch only
>> >to -next. Thanks a lot Sasha for catching this early and please drop the
>> >merge from the for-linus tree.
>>
>> I've dropped the merge and tests are passing now, thanks!
>>
>> --
>> Thanks,
>> Sasha
>>
