mm: provide interface for retrieving kmem_cache name
Subsystems (2)
PATCH 1/1 mm: provide interface for retrieving kmem_cache name
With the restructuring done in commit 9adeaa226988 ("mm, slab: move memcg_cache_params structure to mm/slab.h")
it is no longer possible for code external to mm to access the name of a kmem_cache as struct kmem_cache has effectively become opaque. Having access to the cache name is helpful to kernel testing infrastructure.
Expose a new function kmem_cache_name to mitigate that.
--- include/linux/slab.h | 1 + mm/slab_common.c | 9 +++++++++ 2 files changed, 10 insertions(+)
include/linux/slab.h +1 -0
@@ -702,6 +702,7 @@ static inline void *kzalloc_node(size_t size, gfp_t flags, int node) }unsigned int kmem_cache_size (struct kmem_cache *s); +const char *kmem_cache_name (struct kmem_cache *s);void __init kmem_cache_init_late (void );#if defined(CONFIG_SMP ) && defined(CONFIG_SLAB )
mm/slab_common.c +9 -0
@@ -82,6 +82,15 @@ unsigned int kmem_cache_size(struct kmem_cache *s) }EXPORT_SYMBOL (kmem_cache_size); +/* +* Get the name of a slab object +*/ +const char *kmem_cache_name (struct kmem_cache *s) +{ +return s->name; +} +EXPORT_SYMBOL (kmem_cache_name); +#ifdef CONFIG_DEBUG_VM static int kmem_cache_sanity_check (const char *name,unsigned int size) {
Signature · 1 line
2.20.1
Quoted 9 lines · Knut Omang
With the restructuring done in commit
9adeaa226988("mm, slab: movememcg_cache_paramsstructure tomm/slab.h")it is no longer possible for code external to mm to access the name of a
kmem_cacheas structkmem_cachehas effectively become opaque. Having access to the cache name is helpful to kernel testing infrastructure.Expose a new function
kmem_cache_nameto mitigate that.
Who is going to use that symbol? It is preferred that a user is added in the same patch as the newly added symbol.
Quoted 40 lines
Signed-off-by Knut Omang <[email protected]>--- include/linux/slab.h | 1 + mm/slab_common.c | 9 +++++++++ 2 files changed, 10 insertions(+)include/linux/slab.h +1
-0@@ -702,6 +702,7 @@ static inline void *kzalloc_node(size_t size, gfp_t flags, int node) }unsigned int kmem_cache_size (struct kmem_cache *s); +const char *kmem_cache_name (struct kmem_cache *s);void __init kmem_cache_init_late (void );#if defined(CONFIG_SMP ) && defined(CONFIG_SLAB )mm/slab_common.c +9
-0@@ -82,6 +82,15 @@ unsigned int kmem_cache_size(struct kmem_cache *s) }EXPORT_SYMBOL (kmem_cache_size); +/* +* Get the name of a slab object +*/ +const char *kmem_cache_name (struct kmem_cache *s) +{ +return s->name; +} +EXPORT_SYMBOL (kmem_cache_name); +#ifdef CONFIG_DEBUG_VM static int kmem_cache_sanity_check (const char *name,unsigned int size) {Signature · 1 line
2.20.1
Signature · 2 lines
Michal Hocko SUSE Labs
Quoted 9 lines · Michal Hocko
Quoted 8 lines · Knut Omang
With the restructuring done in commit
9adeaa226988("mm, slab: movememcg_cache_paramsstructure tomm/slab.h")it is no longer possible for code external to mm to access the name of a
kmem_cacheas structkmem_cachehas effectively become opaque. Having access to the cache name is helpful to kernel testing infrastructure.
Quoted 4 lines
Quoted 1 line
Expose a new function
kmem_cache_nameto mitigate that.Who is going to use that symbol? It is preferred that a user is added in the same patch as the newly added symbol.
Yes, I am aware that that's the normal practice,
we're currently using cache->name directly in the kernel unit test framework KTF (https://github.com/oracle/ktf/)
which we are working (https://lkml.org/lkml/2019/8/13/111) to get into the kernel in one form or another.
To me this seems like a natural part of an API for the kmem_cache data structure now that it has in effect become opaque, so it seemed appropriate to get it in close in time to the patch that no longer makes this possible, instead of someone else hitting this down the road.
Thanks,
Knut
Quoted 40 lines
Quoted 40 lines
Signed-off-by Knut Omang <[email protected]>--- include/linux/slab.h | 1 + mm/slab_common.c | 9 +++++++++ 2 files changed, 10 insertions(+)include/linux/slab.h +1
-0@@ -702,6 +702,7 @@ static inline void *kzalloc_node(size_t size, gfp_t flags, int node) }unsigned int kmem_cache_size (struct kmem_cache *s); +const char *kmem_cache_name (struct kmem_cache *s);void __init kmem_cache_init_late (void );#if defined(CONFIG_SMP ) && defined(CONFIG_SLAB )mm/slab_common.c +9
-0@@ -82,6 +82,15 @@ unsigned int kmem_cache_size(struct kmem_cache *s) }EXPORT_SYMBOL (kmem_cache_size); +/* +* Get the name of a slab object +*/ +const char *kmem_cache_name (struct kmem_cache *s) +{ +return s->name; +} +EXPORT_SYMBOL (kmem_cache_name); +#ifdef CONFIG_DEBUG_VM static int kmem_cache_sanity_check (const char *name,unsigned int size) {Signature · 1 line
2.20.1
Quoted 12 lines · Knut Omang
Quoted 11 lines · Michal Hocko
Quoted 9 lines · Knut Omang
With the restructuring done in commit
9adeaa226988("mm, slab: movememcg_cache_paramsstructure tomm/slab.h")it is no longer possible for code external to mm to access the name of a
kmem_cacheas structkmem_cachehas effectively become opaque. Having access to the cache name is helpful to kernel testing infrastructure.Expose a new function
kmem_cache_nameto mitigate that.
Quoted 8 lines
Quoted 2 lines
Who is going to use that symbol? It is preferred that a user is added in the same patch as the newly added symbol.
Yes, I am aware that that's the normal practice,
we're currently usingcache->namedirectly in the kernel unit test framework KTF (https://github.com/oracle/ktf/)
which we are working (https://lkml.org/lkml/2019/8/13/111) to get into the kernel in one form or another.
Please add the export with a patch that really needs it.
Quoted 4 lines
To me this seems like a natural part of an API for the
kmem_cachedata structure now that it has in effect become opaque, so it seemed appropriate to get it in close in time to the patch that no longer makes this possible, instead of someone else hitting this down the road.
Well, this is something for SLAB maintainers but I do not really think the name is something the in kernel code should care about. It is solely for presenting reasonable statistics to the userspace and that code workds just fine AFAIK.
Signature · 2 lines
Michal Hocko SUSE Labs
David Rientjes
reviewerQuoted 16 lines · Michal Hocko
Quoted 15 lines · Knut Omang
Quoted 13 lines · Michal Hocko
Quoted 9 lines · Knut Omang
With the restructuring done in commit
9adeaa226988("mm, slab: movememcg_cache_paramsstructure tomm/slab.h")it is no longer possible for code external to mm to access the name of a
kmem_cacheas structkmem_cachehas effectively become opaque. Having access to the cache name is helpful to kernel testing infrastructure.Expose a new function
kmem_cache_nameto mitigate that.Who is going to use that symbol? It is preferred that a user is added in the same patch as the newly added symbol.
Quoted 8 lines
Quoted 5 lines
Yes, I am aware that that's the normal practice,
we're currently usingcache->namedirectly in the kernel unit test framework KTF (https://github.com/oracle/ktf/)
which we are working (https://lkml.org/lkml/2019/8/13/111) to get into the kernel in one form or another.Please add the export with a patch that really needs it.
Agree, this patch could be added as a predecessor to the series at
https://lkml.org/lkml/2019/8/13/111 which would use it. If it's obvious why using the kmem cache name is useful in that series, acking this will be a no brainer.
Quoted 7 lines · Michal Hocko
Quoted 6 lines · Knut Omang
Quoted 5 lines · Michal Hocko
Quoted 4 lines · Knut Omang
With the restructuring done in commit
9adeaa226988("mm, slab: movememcg_cache_paramsstructure tomm/slab.h")it is no longer possible for code external to mm to access
That patch only affected the memcg_cache_params structure and not kmem_cache.
And I do not see any references to the memcg_cache_param?
The fields that all allocators need to expose are listed in the struct kmme_cache definition in linux/mm/slab.h.
Quoted 8 lines · Christopher Lameter
Quoted 6 lines · Michal Hocko
Quoted 5 lines · Knut Omang
Quoted 4 lines · Michal Hocko
Quoted 3 lines · Knut Omang
With the restructuring done in commit
9adeaa226988("mm, slab: movememcg_cache_paramsstructure tomm/slab.h")
Quoted 6 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
it is no longer possible for code external to mm to access
That patch only affected the
memcg_cache_paramsstructure and notkmem_cache.And I do not see any references to the
memcg_cache_param?
Good point, I should have made explicit reference to it.
It gets inlined into kmem_cache with CONFIG_SLUB if CONFIG_MEMCG is set (include/linux/slub_def.h, line 112)
Quoted 2 lines
The fields that all allocators need to expose are listed in the struct
kmme_cachedefinition inlinux/mm/slab.h.
So I take that kmem_cache::name was still intended to be public, just that that broke due to the inlining of struct memcg_cache_param in slub_def.h?
Knut
Quoted 14 lines · Knut Omang
Quoted 13 lines · Christopher Lameter
Quoted 7 lines · Michal Hocko
Quoted 6 lines · Knut Omang
Quoted 5 lines · Michal Hocko
Quoted 4 lines · Knut Omang
With the restructuring done in commit
9adeaa226988("mm, slab: movememcg_cache_paramsstructure tomm/slab.h")it is no longer possible for code external to mm to access
That patch only affected the
memcg_cache_paramsstructure and notkmem_cache.
Quoted 6 lines
Quoted 1 line
And I do not see any references to the
memcg_cache_param?Good point, I should have made explicit reference to it.
It gets inlined into
kmem_cachewithCONFIG_SLUBifCONFIG_MEMCGis set (include/linux/slub_def.h, line 112)
Yes but that does not affect the "name" field on line 105
Quoted 6 lines
Quoted 2 lines
The fields that all allocators need to expose are listed in the struct
kmme_cachedefinition inlinux/mm/slab.h.So I take that
kmem_cache::namewas still intended to be public, just that that broke due to the inlining of structmemcg_cache_paraminslub_def.h?
The patch did not change the name field. I am not sure what is broken?
Maybe you need memcg_params to compose a name of the memcg with the slab name?
Quoted 28 lines · Christopher Lameter
Quoted 14 lines · Knut Omang
Quoted 13 lines · Christopher Lameter
Quoted 7 lines · Michal Hocko
Quoted 6 lines · Knut Omang
Quoted 5 lines · Michal Hocko
Quoted 4 lines · Knut Omang
With the restructuring done in commit
9adeaa226988("mm, slab: movememcg_cache_paramsstructure tomm/slab.h")it is no longer possible for code external to mm to access
That patch only affected the
memcg_cache_paramsstructure and notkmem_cache.Quoted 6 lines
Quoted 1 line
And I do not see any references to the
memcg_cache_param?Good point, I should have made explicit reference to it.
It gets inlined into
kmem_cachewithCONFIG_SLUBifCONFIG_MEMCGis set (include/linux/slub_def.h, line 112)Yes but that does not affect the "name" field on line 105
Quoted 3 lines
Quoted 2 lines
The fields that all allocators need to expose are listed in the struct
kmme_cachedefinition inlinux/mm/slab.h.
Quoted 8 lines
Quoted 3 lines
So I take that
kmem_cache::namewas still intended to be public, just that that broke due to the inlining of structmemcg_cache_paraminslub_def.h?The patch did not change the name field. I am not sure what is broken?
Maybe you need
memcg_paramsto compose a name of the memcg with the slab name?
No, it's just that the slub definition of struct kmem_cache won't compile because the definition of struct memcg_cache_params is no longer available.
Knut