| Message ID | 20240815155402.3630804-1-shakeel.butt@linux.dev (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [v2] memcg: replace memcg ID idr with xarray | expand |
On Thu, Aug 15, 2024 at 08:54:02AM -0700, Shakeel Butt wrote: > At the moment memcg IDs are managed through IDR which requires external > synchronization mechanisms and makes the allocation code a bit awkward. > Let's switch to xarray and make the code simpler. > > Signed-off-by: Shakeel Butt <shakeel.butt@linux.dev> > Suggested-by: Matthew Wilcox <willy@infradead.org> > Reviewed-by: Roman Gushchin <roman.gushchin@linux.dev> > Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org> > Acked-by: Johannes Weiner <hannes@cmpxchg.org> > Reviewed-by: Muchun Song <muchun.song@linux.dev> > Acked-by: Michal Hocko <mhocko@suse.com> > Signed-off-by: Andrew Morton <akpm@linux-foundation.org> > --- > > Changes since v1: > - Fix error path in mem_cgroup_alloc (Dan Carpenter) > > mm/memcontrol.c | 39 ++++++++++----------------------------- > 1 file changed, 10 insertions(+), 29 deletions(-) > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c > index df84683a0e1c..e8e03a5e1e5e 100644 > --- a/mm/memcontrol.c > +++ b/mm/memcontrol.c > @@ -3408,29 +3408,12 @@ static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) > */ > > #define MEM_CGROUP_ID_MAX ((1UL << MEM_CGROUP_ID_SHIFT) - 1) > -static DEFINE_IDR(mem_cgroup_idr); > -static DEFINE_SPINLOCK(memcg_idr_lock); > - > -static int mem_cgroup_alloc_id(void) > -{ > - int ret; > - > - idr_preload(GFP_KERNEL); > - spin_lock(&memcg_idr_lock); > - ret = idr_alloc(&mem_cgroup_idr, NULL, 1, MEM_CGROUP_ID_MAX + 1, > - GFP_NOWAIT); > - spin_unlock(&memcg_idr_lock); > - idr_preload_end(); > - return ret; > -} > +static DEFINE_XARRAY_ALLOC1(mem_cgroup_ids); > > static void mem_cgroup_id_remove(struct mem_cgroup *memcg) > { > if (memcg->id.id > 0) { > - spin_lock(&memcg_idr_lock); > - idr_remove(&mem_cgroup_idr, memcg->id.id); > - spin_unlock(&memcg_idr_lock); > - > + xa_erase(&mem_cgroup_ids, memcg->id.id); > memcg->id.id = 0; > } > } > @@ -3465,7 +3448,7 @@ static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) > struct mem_cgroup *mem_cgroup_from_id(unsigned short id) > { > WARN_ON_ONCE(!rcu_read_lock_held()); > - return idr_find(&mem_cgroup_idr, id); > + return xa_load(&mem_cgroup_ids, id); > } > > #ifdef CONFIG_SHRINKER_DEBUG > @@ -3558,17 +3541,17 @@ static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent) > struct mem_cgroup *memcg; > int node, cpu; > int __maybe_unused i; > - long error = -ENOMEM; > + long error; > > memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL); > if (!memcg) > - return ERR_PTR(error); > + return ERR_PTR(-ENOMEM); > > - memcg->id.id = mem_cgroup_alloc_id(); > - if (memcg->id.id < 0) { > - error = memcg->id.id; > + error = xa_alloc(&mem_cgroup_ids, &memcg->id.id, NULL, > + XA_LIMIT(1, MEM_CGROUP_ID_MAX), GFP_KERNEL); > + if (error) > goto fail; > - } > + error = -ENOMEM; There is another subtle change here: xa_alloc() returns -EBUSY in the case of the address space exhaustion, while the old code returned -ENOSPC. It's unlikely a big practical problem.
On Thu, Aug 15, 2024 at 07:31:43PM +0000, Roman Gushchin wrote: > There is another subtle change here: xa_alloc() returns -EBUSY in the case > of the address space exhaustion, while the old code returned -ENOSPC. > It's unlikely a big practical problem. I decided that EBUSY was the right errno for this situation; #define EBUSY 16 /* Device or resource busy */ #define ENOSPC 28 /* No space left on device */ ENOSPC seemed wrong; the device isn't out of space.
On Thu 15-08-24 22:16:27, Matthew Wilcox wrote: > On Thu, Aug 15, 2024 at 07:31:43PM +0000, Roman Gushchin wrote: > > There is another subtle change here: xa_alloc() returns -EBUSY in the case > > of the address space exhaustion, while the old code returned -ENOSPC. > > It's unlikely a big practical problem. > > I decided that EBUSY was the right errno for this situation; > > #define EBUSY 16 /* Device or resource busy */ > #define ENOSPC 28 /* No space left on device */ > > ENOSPC seemed wrong; the device isn't out of space. The thing is that this is observable by userspace - mkdir would return a different and potentially unexpected errno. We can try and see whether anybody complains or just translate the error.
On Fri, 16 Aug 2024 09:19:58 +0200 Michal Hocko <mhocko@suse.com> wrote: > On Thu 15-08-24 22:16:27, Matthew Wilcox wrote: > > On Thu, Aug 15, 2024 at 07:31:43PM +0000, Roman Gushchin wrote: > > > There is another subtle change here: xa_alloc() returns -EBUSY in the case > > > of the address space exhaustion, while the old code returned -ENOSPC. > > > It's unlikely a big practical problem. > > > > I decided that EBUSY was the right errno for this situation; > > > > #define EBUSY 16 /* Device or resource busy */ > > #define ENOSPC 28 /* No space left on device */ > > > > ENOSPC seemed wrong; the device isn't out of space. > > The thing is that this is observable by userspace - mkdir would return a > different and potentially unexpected errno. We can try and see whether > anybody complains or just translate the error. The mkdir(2) manpage doesn't list EBUSY. Maybe ENOMEM is close enough.
On Fri 16-08-24 00:43:34, Andrew Morton wrote: > On Fri, 16 Aug 2024 09:19:58 +0200 Michal Hocko <mhocko@suse.com> wrote: > > > On Thu 15-08-24 22:16:27, Matthew Wilcox wrote: > > > On Thu, Aug 15, 2024 at 07:31:43PM +0000, Roman Gushchin wrote: > > > > There is another subtle change here: xa_alloc() returns -EBUSY in the case > > > > of the address space exhaustion, while the old code returned -ENOSPC. > > > > It's unlikely a big practical problem. > > > > > > I decided that EBUSY was the right errno for this situation; > > > > > > #define EBUSY 16 /* Device or resource busy */ > > > #define ENOSPC 28 /* No space left on device */ > > > > > > ENOSPC seemed wrong; the device isn't out of space. > > > > The thing is that this is observable by userspace - mkdir would return a > > different and potentially unexpected errno. We can try and see whether > > anybody complains or just translate the error. > > The mkdir(2) manpage doesn't list EBUSY. Maybe ENOMEM is close enough. it used to report ENOSPC with xarray.
diff --git a/mm/memcontrol.c b/mm/memcontrol.c index df84683a0e1c..e8e03a5e1e5e 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -3408,29 +3408,12 @@ static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) */ #define MEM_CGROUP_ID_MAX ((1UL << MEM_CGROUP_ID_SHIFT) - 1) -static DEFINE_IDR(mem_cgroup_idr); -static DEFINE_SPINLOCK(memcg_idr_lock); - -static int mem_cgroup_alloc_id(void) -{ - int ret; - - idr_preload(GFP_KERNEL); - spin_lock(&memcg_idr_lock); - ret = idr_alloc(&mem_cgroup_idr, NULL, 1, MEM_CGROUP_ID_MAX + 1, - GFP_NOWAIT); - spin_unlock(&memcg_idr_lock); - idr_preload_end(); - return ret; -} +static DEFINE_XARRAY_ALLOC1(mem_cgroup_ids); static void mem_cgroup_id_remove(struct mem_cgroup *memcg) { if (memcg->id.id > 0) { - spin_lock(&memcg_idr_lock); - idr_remove(&mem_cgroup_idr, memcg->id.id); - spin_unlock(&memcg_idr_lock); - + xa_erase(&mem_cgroup_ids, memcg->id.id); memcg->id.id = 0; } } @@ -3465,7 +3448,7 @@ static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) struct mem_cgroup *mem_cgroup_from_id(unsigned short id) { WARN_ON_ONCE(!rcu_read_lock_held()); - return idr_find(&mem_cgroup_idr, id); + return xa_load(&mem_cgroup_ids, id); } #ifdef CONFIG_SHRINKER_DEBUG @@ -3558,17 +3541,17 @@ static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent) struct mem_cgroup *memcg; int node, cpu; int __maybe_unused i; - long error = -ENOMEM; + long error; memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL); if (!memcg) - return ERR_PTR(error); + return ERR_PTR(-ENOMEM); - memcg->id.id = mem_cgroup_alloc_id(); - if (memcg->id.id < 0) { - error = memcg->id.id; + error = xa_alloc(&mem_cgroup_ids, &memcg->id.id, NULL, + XA_LIMIT(1, MEM_CGROUP_ID_MAX), GFP_KERNEL); + if (error) goto fail; - } + error = -ENOMEM; memcg->vmstats = kzalloc(sizeof(struct memcg_vmstats), GFP_KERNEL_ACCOUNT); @@ -3709,9 +3692,7 @@ static int mem_cgroup_css_online(struct cgroup_subsys_state *css) * publish it here at the end of onlining. This matches the * regular ID destruction during offlining. */ - spin_lock(&memcg_idr_lock); - idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); - spin_unlock(&memcg_idr_lock); + xa_store(&mem_cgroup_ids, memcg->id.id, memcg, GFP_KERNEL); return 0; offline_kmem: