Re: [PATCH v2] mm/slab_common: move dma-kmalloc caches creation into new_kmalloc_cache()

Re: [PATCH v2] mm/slab_common: move dma-kmalloc caches creation into new_kmalloc_cache()

[Hyeonggon Yoo]

On Mon, Apr 11, 2022 at 01:25:11AM +0900, Ohhoon Kwon wrote:
> There are four types of kmalloc_caches: KMALLOC_NORMAL, KMALLOC_CGROUP,
> KMALLOC_RECLAIM, and KMALLOC_DMA. While the first three types are
> created using new_kmalloc_cache(), KMALLOC_DMA caches are created in a
> separate logic. Let KMALLOC_DMA caches be also created using
> new_kmalloc_cache(), to enhance readability.
> 
> Historically, there were only KMALLOC_NORMAL caches and KMALLOC_DMA
> caches in the first place, and they were initialized in two separate
> logics. However, when KMALLOC_RECLAIM was introduced in v4.20 via
> commit 1291523f2c1d ("mm, slab/slub: introduce kmalloc-reclaimable
> caches") and KMALLOC_CGROUP was introduced in v5.14 via
> commit 494c1dfe855e ("mm: memcg/slab: create a new set of kmalloc-cg-<n>
> caches"), their creations were merged with KMALLOC_NORMAL's only.
> KMALLOC_DMA creation logic should be merged with them, too.
> 
> By merging KMALLOC_DMA initialization with other types, the following
> two changes might occur:
> 1. The order dma-kmalloc-<n> caches added in slab_cache list may be
> sorted by size. i.e. the order they appear in /proc/slabinfo may change
> as well.
> 2. slab_state will be set to UP after KMALLOC_DMA is created.
> In case of slub, freelist randomization is dependent on slab_state>=UP,
> and therefore KMALLOC_DMA cache's freelist will not be randomized in
> creation, but will be deferred to init_freelist_randomization().
> 
> Co-developed-by: JaeSang Yoo <jsyoo5b@gmail.com>
> Signed-off-by: JaeSang Yoo <jsyoo5b@gmail.com>
> Signed-off-by: Ohhoon Kwon <ohkwon1043@gmail.com>
> ---
>  mm/slab_common.c | 18 +++---------------
>  1 file changed, 3 insertions(+), 15 deletions(-)
> 
> diff --git a/mm/slab_common.c b/mm/slab_common.c
> index 6ee64d6208b3..a959d247c27b 100644
> --- a/mm/slab_common.c
> +++ b/mm/slab_common.c
> @@ -849,6 +849,8 @@ new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
>  			return;
>  		}
>  		flags |= SLAB_ACCOUNT;
> +	} else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
> +		flags |= SLAB_CACHE_DMA;
>  	}
>  
>  	kmalloc_caches[type][idx] = create_kmalloc_cache(
> @@ -877,7 +879,7 @@ void __init create_kmalloc_caches(slab_flags_t flags)
>  	/*
>  	 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
>  	 */
> -	for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
> +	for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
>  		for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
>  			if (!kmalloc_caches[type][i])
>  				new_kmalloc_cache(i, type, flags);
> @@ -898,20 +900,6 @@ void __init create_kmalloc_caches(slab_flags_t flags)
>  
>  	/* Kmalloc array is now usable */
>  	slab_state = UP;
> -
> -#ifdef CONFIG_ZONE_DMA
> -	for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
> -		struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
> -
> -		if (s) {
> -			kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
> -				kmalloc_info[i].name[KMALLOC_DMA],
> -				kmalloc_info[i].size,
> -				SLAB_CACHE_DMA | flags, 0,
> -				kmalloc_info[i].size);
> -		}
> -	}
> -#endif
>  }
>  #endif /* !CONFIG_SLOB */
>

Reviewed-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>

BTW this patch may conflict with [1] (not merged yet)

[1] https://lore.kernel.org/linux-mm/20220405135758.774016-9-catalin.marinas@arm.com/

Thanks!

> -- 
> 2.25.1
> 

-- 
Thanks,
Hyeonggon

Re: [PATCH v2] mm/slab_common: move dma-kmalloc caches creation into new_kmalloc_cache()

[Vlastimil Babka]

On 4/11/22 13:07, Hyeonggon Yoo wrote:
> On Mon, Apr 11, 2022 at 01:25:11AM +0900, Ohhoon Kwon wrote:
>> There are four types of kmalloc_caches: KMALLOC_NORMAL, KMALLOC_CGROUP,
>> KMALLOC_RECLAIM, and KMALLOC_DMA. While the first three types are
>> created using new_kmalloc_cache(), KMALLOC_DMA caches are created in a
>> separate logic. Let KMALLOC_DMA caches be also created using
>> new_kmalloc_cache(), to enhance readability.
>> 
>> Historically, there were only KMALLOC_NORMAL caches and KMALLOC_DMA
>> caches in the first place, and they were initialized in two separate
>> logics. However, when KMALLOC_RECLAIM was introduced in v4.20 via
>> commit 1291523f2c1d ("mm, slab/slub: introduce kmalloc-reclaimable
>> caches") and KMALLOC_CGROUP was introduced in v5.14 via
>> commit 494c1dfe855e ("mm: memcg/slab: create a new set of kmalloc-cg-<n>
>> caches"), their creations were merged with KMALLOC_NORMAL's only.
>> KMALLOC_DMA creation logic should be merged with them, too.
>> 
>> By merging KMALLOC_DMA initialization with other types, the following
>> two changes might occur:
>> 1. The order dma-kmalloc-<n> caches added in slab_cache list may be
>> sorted by size. i.e. the order they appear in /proc/slabinfo may change
>> as well.
>> 2. slab_state will be set to UP after KMALLOC_DMA is created.
>> In case of slub, freelist randomization is dependent on slab_state>=UP,
>> and therefore KMALLOC_DMA cache's freelist will not be randomized in
>> creation, but will be deferred to init_freelist_randomization().
>> 
>> Co-developed-by: JaeSang Yoo <jsyoo5b@gmail.com>
>> Signed-off-by: JaeSang Yoo <jsyoo5b@gmail.com>
>> Signed-off-by: Ohhoon Kwon <ohkwon1043@gmail.com>
>> ---
>>  mm/slab_common.c | 18 +++---------------
>>  1 file changed, 3 insertions(+), 15 deletions(-)
>> 
>> diff --git a/mm/slab_common.c b/mm/slab_common.c
>> index 6ee64d6208b3..a959d247c27b 100644
>> --- a/mm/slab_common.c
>> +++ b/mm/slab_common.c
>> @@ -849,6 +849,8 @@ new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
>>  			return;
>>  		}
>>  		flags |= SLAB_ACCOUNT;
>> +	} else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
>> +		flags |= SLAB_CACHE_DMA;
>>  	}
>>  
>>  	kmalloc_caches[type][idx] = create_kmalloc_cache(
>> @@ -877,7 +879,7 @@ void __init create_kmalloc_caches(slab_flags_t flags)
>>  	/*
>>  	 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
>>  	 */
>> -	for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
>> +	for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
>>  		for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
>>  			if (!kmalloc_caches[type][i])
>>  				new_kmalloc_cache(i, type, flags);
>> @@ -898,20 +900,6 @@ void __init create_kmalloc_caches(slab_flags_t flags)
>>  
>>  	/* Kmalloc array is now usable */
>>  	slab_state = UP;
>> -
>> -#ifdef CONFIG_ZONE_DMA
>> -	for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
>> -		struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
>> -
>> -		if (s) {
>> -			kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
>> -				kmalloc_info[i].name[KMALLOC_DMA],
>> -				kmalloc_info[i].size,
>> -				SLAB_CACHE_DMA | flags, 0,
>> -				kmalloc_info[i].size);
>> -		}
>> -	}
>> -#endif
>>  }
>>  #endif /* !CONFIG_SLOB */
>>
> 
> Reviewed-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>

Thanks.
Added to:
https://git.kernel.org/pub/scm/linux/kernel/git/vbabka/slab.git/log/?h=for-5.19/refactor

> BTW this patch may conflict with [1] (not merged yet)
> 
> [1] https://lore.kernel.org/linux-mm/20220405135758.774016-9-catalin.marinas@arm.com/

FYI Catalin, might want to base v2 on the above once the crypto side is
solved. At cursory look it shouldn't be a significant conflict.

> Thanks!
> 
>> -- 
>> 2.25.1
>> 
> 

Re: [PATCH v2] mm/slab_common: move dma-kmalloc caches creation into new_kmalloc_cache()

[Catalin Marinas]

On Mon, Apr 11, 2022 at 05:31:51PM +0200, Vlastimil Babka wrote:
> On 4/11/22 13:07, Hyeonggon Yoo wrote:
> > On Mon, Apr 11, 2022 at 01:25:11AM +0900, Ohhoon Kwon wrote:
> >> @@ -898,20 +900,6 @@ void __init create_kmalloc_caches(slab_flags_t flags)
> >>  
> >>  	/* Kmalloc array is now usable */
> >>  	slab_state = UP;
> >> -
> >> -#ifdef CONFIG_ZONE_DMA
> >> -	for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
> >> -		struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
> >> -
> >> -		if (s) {
> >> -			kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
> >> -				kmalloc_info[i].name[KMALLOC_DMA],
> >> -				kmalloc_info[i].size,
> >> -				SLAB_CACHE_DMA | flags, 0,
> >> -				kmalloc_info[i].size);
> >> -		}
> >> -	}
> >> -#endif
> >>  }
> >>  #endif /* !CONFIG_SLOB */
> > 
> > Reviewed-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>
> 
> Thanks.
> Added to:
> https://git.kernel.org/pub/scm/linux/kernel/git/vbabka/slab.git/log/?h=for-5.19/refactor
> 
> > BTW this patch may conflict with [1] (not merged yet)
> > 
> > [1] https://lore.kernel.org/linux-mm/20220405135758.774016-9-catalin.marinas@arm.com/
> 
> FYI Catalin, might want to base v2 on the above once the crypto side is
> solved. At cursory look it shouldn't be a significant conflict.

Thanks for the heads-up. I did wonder why this was a separate loop.

-- 
Catalin

Re: [PATCH v2] mm/slab_common: move dma-kmalloc caches creation into new_kmalloc_cache()

[David Rientjes]

On Mon, 11 Apr 2022, Ohhoon Kwon wrote:

> There are four types of kmalloc_caches: KMALLOC_NORMAL, KMALLOC_CGROUP,
> KMALLOC_RECLAIM, and KMALLOC_DMA. While the first three types are
> created using new_kmalloc_cache(), KMALLOC_DMA caches are created in a
> separate logic. Let KMALLOC_DMA caches be also created using
> new_kmalloc_cache(), to enhance readability.
> 
> Historically, there were only KMALLOC_NORMAL caches and KMALLOC_DMA
> caches in the first place, and they were initialized in two separate
> logics. However, when KMALLOC_RECLAIM was introduced in v4.20 via
> commit 1291523f2c1d ("mm, slab/slub: introduce kmalloc-reclaimable
> caches") and KMALLOC_CGROUP was introduced in v5.14 via
> commit 494c1dfe855e ("mm: memcg/slab: create a new set of kmalloc-cg-<n>
> caches"), their creations were merged with KMALLOC_NORMAL's only.
> KMALLOC_DMA creation logic should be merged with them, too.
> 
> By merging KMALLOC_DMA initialization with other types, the following
> two changes might occur:
> 1. The order dma-kmalloc-<n> caches added in slab_cache list may be
> sorted by size. i.e. the order they appear in /proc/slabinfo may change
> as well.
> 2. slab_state will be set to UP after KMALLOC_DMA is created.
> In case of slub, freelist randomization is dependent on slab_state>=UP,
> and therefore KMALLOC_DMA cache's freelist will not be randomized in
> creation, but will be deferred to init_freelist_randomization().
> 
> Co-developed-by: JaeSang Yoo <jsyoo5b@gmail.com>
> Signed-off-by: JaeSang Yoo <jsyoo5b@gmail.com>
> Signed-off-by: Ohhoon Kwon <ohkwon1043@gmail.com>

Acked-by: David Rientjes <rientjes@google.com>