Re: [RFC v2 1/2] mm: Reorganize SLAB freelist randomization

[Thomas Garnier]

On Wed, May 25, 2016 at 6:09 PM, Joonsoo Kim <iamjoonsoo.kim@xxxxxxx> wrote:
> On Tue, May 24, 2016 at 02:15:22PM -0700, Thomas Garnier wrote:
>> This commit reorganizes the previous SLAB freelist randomization to
>> prepare for the SLUB implementation. It moves functions that will be
>> shared to slab_common. It also move the definition of freelist_idx_t in
>> the slab_def header so a similar type can be used for all common
>> functions. The entropy functions are changed to align with the SLUB
>> implementation, now using get_random_* functions.
>
> Could you explain more what's the difference between get_random_*
> and get_random_bytes_arch() and why this change is needed?
>
> And, I think that it should be another patch.
>

Sure. From my test (limited scenario), get_random_bytes_arch was much
slower than get_random_int when the random instructions were not
available. It also seems from looking at both implementation that
get_random_int may provide a bit more entropy at this early stage.

>>
>> Signed-off-by: Thomas Garnier <thgarnie@xxxxxxxxxx>
>> ---
>> Based on 0e01df100b6bf22a1de61b66657502a6454153c5
>> ---
>>  include/linux/slab_def.h | 11 +++++++-
>>  mm/slab.c                | 68 ++----------------------------------------------
>>  mm/slab.h                | 16 ++++++++++++
>>  mm/slab_common.c         | 48 ++++++++++++++++++++++++++++++++++
>>  4 files changed, 76 insertions(+), 67 deletions(-)
>>
>> diff --git a/include/linux/slab_def.h b/include/linux/slab_def.h
>> index 8694f7a..e05a871 100644
>> --- a/include/linux/slab_def.h
>> +++ b/include/linux/slab_def.h
>> @@ -3,6 +3,15 @@
>>
>>  #include <linux/reciprocal_div.h>
>>
>> +#define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \
>> +                             <= SLAB_OBJ_MIN_SIZE) ? 1 : 0)
>> +
>> +#if FREELIST_BYTE_INDEX
>> +typedef unsigned char freelist_idx_t;
>> +#else
>> +typedef unsigned short freelist_idx_t;
>> +#endif
>> +
>
> This is a SLAB specific index size definition and I don't want to export
> it to SLUB. Please use 'void *random_seq' and allocate sizeof(void *)
> memory for each entry. And, then do type casting when suffling in
> SLAB. There is some memory waste but not that much so we can tolerate
> it.
>
> Others look fine to me.
>

Ok, will do. Thanks.

> Thanks.
>