From: Xiubo Li <[email protected]>
The GFP_NOIO means all further allocations will implicitly drop
both __GFP_IO and __GFP_FS flags and so they are safe for both the
IO critical section and the the critical section from the allocation
recursion point of view. Not only the __GFP_IO, which a bit confusing
when reading the code or using the save/restore pair.
Signed-off-by: Xiubo Li <[email protected]>
---
include/linux/sched/mm.h | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
index 4a7944078cc3..9bdc97e52de1 100644
--- a/include/linux/sched/mm.h
+++ b/include/linux/sched/mm.h
@@ -211,10 +211,11 @@ static inline void fs_reclaim_release(gfp_t gfp_mask) { }
* memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
*
* This functions marks the beginning of the GFP_NOIO allocation scope.
- * All further allocations will implicitly drop __GFP_IO flag and so
- * they are safe for the IO critical section from the allocation recursion
- * point of view. Use memalloc_noio_restore to end the scope with flags
- * returned by this function.
+ * All further allocations will implicitly drop __GFP_IO and __GFP_FS
+ * flags and so they are safe for both the IO critical section and the
+ * the critical section from the allocation recursion point of view. Use
+ * memalloc_noio_restore to end the scope with flags returned by this
+ * function.
*
* This function is safe to be used from any context.
*/
--
2.21.0
On Wed 18-09-19 04:58:20, [email protected] wrote:
> From: Xiubo Li <[email protected]>
>
> The GFP_NOIO means all further allocations will implicitly drop
> both __GFP_IO and __GFP_FS flags and so they are safe for both the
> IO critical section and the the critical section from the allocation
> recursion point of view. Not only the __GFP_IO, which a bit confusing
> when reading the code or using the save/restore pair.
Historically GFP_NOIO has always implied GFP_NOFS as well. I can imagine
that this might come as an surprise for somebody not familiar with the
code though. I am wondering whether your update of the documentation
would be better off at __GFP_FS, __GFP_IO resp. GFP_NOFS, GFP_NOIO level.
This interface is simply a way to set a scoped NO{IO,FS} context.
> Signed-off-by: Xiubo Li <[email protected]>
> ---
> include/linux/sched/mm.h | 9 +++++----
> 1 file changed, 5 insertions(+), 4 deletions(-)
>
> diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
> index 4a7944078cc3..9bdc97e52de1 100644
> --- a/include/linux/sched/mm.h
> +++ b/include/linux/sched/mm.h
> @@ -211,10 +211,11 @@ static inline void fs_reclaim_release(gfp_t gfp_mask) { }
> * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
> *
> * This functions marks the beginning of the GFP_NOIO allocation scope.
> - * All further allocations will implicitly drop __GFP_IO flag and so
> - * they are safe for the IO critical section from the allocation recursion
> - * point of view. Use memalloc_noio_restore to end the scope with flags
> - * returned by this function.
> + * All further allocations will implicitly drop __GFP_IO and __GFP_FS
> + * flags and so they are safe for both the IO critical section and the
> + * the critical section from the allocation recursion point of view. Use
> + * memalloc_noio_restore to end the scope with flags returned by this
> + * function.
> *
> * This function is safe to be used from any context.
> */
> --
> 2.21.0
--
Michal Hocko
SUSE Labs
On 2019/9/18 15:25, Michal Hocko wrote:
> On Wed 18-09-19 04:58:20, [email protected] wrote:
>> From: Xiubo Li <[email protected]>
>>
>> The GFP_NOIO means all further allocations will implicitly drop
>> both __GFP_IO and __GFP_FS flags and so they are safe for both the
>> IO critical section and the the critical section from the allocation
>> recursion point of view. Not only the __GFP_IO, which a bit confusing
>> when reading the code or using the save/restore pair.
> Historically GFP_NOIO has always implied GFP_NOFS as well. I can imagine
> that this might come as an surprise for somebody not familiar with the
> code though.
Yeah, it true.
> I am wondering whether your update of the documentation
> would be better off at __GFP_FS, __GFP_IO resp. GFP_NOFS, GFP_NOIO level.
> This interface is simply a way to set a scoped NO{IO,FS} context.
The "Documentation/core-api/gfp_mask-from-fs-io.rst" is already very
detail about them all.
This fixing just means to make sure that it won't surprise someone who
is having a quickly through some code and not familiar much about the
detail. It may make not much sense ?
Thanks,
BRs
Xiubo
>> Signed-off-by: Xiubo Li <[email protected]>
>> ---
>> include/linux/sched/mm.h | 9 +++++----
>> 1 file changed, 5 insertions(+), 4 deletions(-)
>>
>> diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
>> index 4a7944078cc3..9bdc97e52de1 100644
>> --- a/include/linux/sched/mm.h
>> +++ b/include/linux/sched/mm.h
>> @@ -211,10 +211,11 @@ static inline void fs_reclaim_release(gfp_t gfp_mask) { }
>> * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
>> *
>> * This functions marks the beginning of the GFP_NOIO allocation scope.
>> - * All further allocations will implicitly drop __GFP_IO flag and so
>> - * they are safe for the IO critical section from the allocation recursion
>> - * point of view. Use memalloc_noio_restore to end the scope with flags
>> - * returned by this function.
>> + * All further allocations will implicitly drop __GFP_IO and __GFP_FS
>> + * flags and so they are safe for both the IO critical section and the
>> + * the critical section from the allocation recursion point of view. Use
>> + * memalloc_noio_restore to end the scope with flags returned by this
>> + * function.
>> *
>> * This function is safe to be used from any context.
>> */
>> --
>> 2.21.0
On Wed 18-09-19 16:02:52, Xiubo Li wrote:
> On 2019/9/18 15:25, Michal Hocko wrote:
> > On Wed 18-09-19 04:58:20, [email protected] wrote:
> > > From: Xiubo Li <[email protected]>
> > >
> > > The GFP_NOIO means all further allocations will implicitly drop
> > > both __GFP_IO and __GFP_FS flags and so they are safe for both the
> > > IO critical section and the the critical section from the allocation
> > > recursion point of view. Not only the __GFP_IO, which a bit confusing
> > > when reading the code or using the save/restore pair.
> > Historically GFP_NOIO has always implied GFP_NOFS as well. I can imagine
> > that this might come as an surprise for somebody not familiar with the
> > code though.
>
> Yeah, it true.
>
> > I am wondering whether your update of the documentation
> > would be better off at __GFP_FS, __GFP_IO resp. GFP_NOFS, GFP_NOIO level.
> > This interface is simply a way to set a scoped NO{IO,FS} context.
>
> The "Documentation/core-api/gfp_mask-from-fs-io.rst" is already very detail
> about them all.
>
> This fixing just means to make sure that it won't surprise someone who is
> having a quickly through some code and not familiar much about the detail.
> It may make not much sense ?
Ohh, I do not think this would be senseless. I just think that the NOIO
implying NOFS as well should be described at the level where they are
documented rather than the api you have chosen.
--
Michal Hocko
SUSE Labs
On 2019/9/18 16:14, Michal Hocko wrote:
> On Wed 18-09-19 16:02:52, Xiubo Li wrote:
>> On 2019/9/18 15:25, Michal Hocko wrote:
>>> On Wed 18-09-19 04:58:20, [email protected] wrote:
>>>> From: Xiubo Li <[email protected]>
>>>>
>>>> The GFP_NOIO means all further allocations will implicitly drop
>>>> both __GFP_IO and __GFP_FS flags and so they are safe for both the
>>>> IO critical section and the the critical section from the allocation
>>>> recursion point of view. Not only the __GFP_IO, which a bit confusing
>>>> when reading the code or using the save/restore pair.
>>> Historically GFP_NOIO has always implied GFP_NOFS as well. I can imagine
>>> that this might come as an surprise for somebody not familiar with the
>>> code though.
>> Yeah, it true.
>>
>>> I am wondering whether your update of the documentation
>>> would be better off at __GFP_FS, __GFP_IO resp. GFP_NOFS, GFP_NOIO level.
>>> This interface is simply a way to set a scoped NO{IO,FS} context.
>> The "Documentation/core-api/gfp_mask-from-fs-io.rst" is already very detail
>> about them all.
>>
>> This fixing just means to make sure that it won't surprise someone who is
>> having a quickly through some code and not familiar much about the detail.
>> It may make not much sense ?
> Ohh, I do not think this would be senseless. I just think that the NOIO
> implying NOFS as well should be described at the level where they are
> documented rather than the api you have chosen.
Hmm, yeah totally agree :-)
Thanks
BRs