Message ID | 20210409231842.8840-6-longman@redhat.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | mm/memcg: Reduce kmemcache memory accounting overhead | expand |
Hi Waiman, I love your patch! Perhaps something to improve: [auto build test WARNING on dennis-percpu/for-next] [also build test WARNING on linus/master v5.12-rc6 next-20210409] [cannot apply to hnaz-linux-mm/master] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in https://git-scm.com/docs/git-format-patch] url: https://github.com/0day-ci/linux/commits/Waiman-Long/mm-memcg-Reduce-kmemcache-memory-accounting-overhead/20210410-071958 base: https://git.kernel.org/pub/scm/linux/kernel/git/dennis/percpu.git for-next config: arm64-randconfig-r031-20210409 (attached as .config) compiler: clang version 13.0.0 (https://github.com/llvm/llvm-project dd453a1389b6a7e6d9214b449d3c54981b1a89b6) reproduce (this is a W=1 build): wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross # install arm64 cross compiling tool for clang build # apt-get install binutils-aarch64-linux-gnu # https://github.com/0day-ci/linux/commit/1f4e22fce44599095a55535301ca83adc5d3a4fe git remote add linux-review https://github.com/0day-ci/linux git fetch --no-tags linux-review Waiman-Long/mm-memcg-Reduce-kmemcache-memory-accounting-overhead/20210410-071958 git checkout 1f4e22fce44599095a55535301ca83adc5d3a4fe # save the attached .config to linux build tree COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross ARCH=arm64 If you fix the issue, kindly add following tag as appropriate Reported-by: kernel test robot <lkp@intel.com> All warnings (new ones prefixed by >>): >> mm/memcontrol.c:3234:10: warning: variable 'stock' is uninitialized when used here [-Wuninitialized] stock = get_obj_stock(flags); ^~~~~~~~~~~~~~~~~~~~ mm/memcontrol.c:2284:16: note: expanded from macro 'get_obj_stock' obj_stock = &stock->task_obj; \ ^~~~~ mm/memcontrol.c:3234:10: note: variable 'stock' is declared here mm/memcontrol.c:2278:2: note: expanded from macro 'get_obj_stock' struct memcg_stock_pcp *stock; \ ^ mm/memcontrol.c:3329:2: warning: variable 'stock' is uninitialized when used here [-Wuninitialized] get_obj_stock(flags); ^~~~~~~~~~~~~~~~~~~~ mm/memcontrol.c:2284:16: note: expanded from macro 'get_obj_stock' obj_stock = &stock->task_obj; \ ^~~~~ mm/memcontrol.c:3329:2: note: variable 'stock' is declared here mm/memcontrol.c:2278:2: note: expanded from macro 'get_obj_stock' struct memcg_stock_pcp *stock; \ ^ mm/memcontrol.c:3371:2: warning: variable 'stock' is uninitialized when used here [-Wuninitialized] get_obj_stock(flags); ^~~~~~~~~~~~~~~~~~~~ mm/memcontrol.c:2284:16: note: expanded from macro 'get_obj_stock' obj_stock = &stock->task_obj; \ ^~~~~ mm/memcontrol.c:3371:2: note: variable 'stock' is declared here mm/memcontrol.c:2278:2: note: expanded from macro 'get_obj_stock' struct memcg_stock_pcp *stock; \ ^ mm/memcontrol.c:3426:2: warning: variable 'stock' is uninitialized when used here [-Wuninitialized] get_obj_stock(flags); ^~~~~~~~~~~~~~~~~~~~ mm/memcontrol.c:2284:16: note: expanded from macro 'get_obj_stock' obj_stock = &stock->task_obj; \ ^~~~~ mm/memcontrol.c:3426:2: note: variable 'stock' is declared here mm/memcontrol.c:2278:2: note: expanded from macro 'get_obj_stock' struct memcg_stock_pcp *stock; \ ^ 4 warnings generated. vim +/stock +3234 mm/memcontrol.c 3227 3228 static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) 3229 { 3230 struct obj_stock *stock; 3231 unsigned long flags; 3232 bool ret = false; 3233 > 3234 stock = get_obj_stock(flags); 3235 3236 stock = current_obj_stock(); 3237 if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { 3238 stock->nr_bytes -= nr_bytes; 3239 ret = true; 3240 } 3241 3242 put_obj_stock(flags); 3243 3244 return ret; 3245 } 3246 --- 0-DAY CI Kernel Test Service, Intel Corporation https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org
On 4/10/21 2:07 AM, kernel test robot wrote: > Hi Waiman, > > I love your patch! Perhaps something to improve: > > [auto build test WARNING on dennis-percpu/for-next] > [also build test WARNING on linus/master v5.12-rc6 next-20210409] > [cannot apply to hnaz-linux-mm/master] > [If your patch is applied to the wrong git tree, kindly drop us a note. > And when submitting patch, we suggest to use '--base' as documented in > https://git-scm.com/docs/git-format-patch] > > url: https://github.com/0day-ci/linux/commits/Waiman-Long/mm-memcg-Reduce-kmemcache-memory-accounting-overhead/20210410-071958 > base: https://git.kernel.org/pub/scm/linux/kernel/git/dennis/percpu.git for-next > config: arm64-randconfig-r031-20210409 (attached as .config) > compiler: clang version 13.0.0 (https://github.com/llvm/llvm-project dd453a1389b6a7e6d9214b449d3c54981b1a89b6) > reproduce (this is a W=1 build): > wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross > chmod +x ~/bin/make.cross > # install arm64 cross compiling tool for clang build > # apt-get install binutils-aarch64-linux-gnu > # https://github.com/0day-ci/linux/commit/1f4e22fce44599095a55535301ca83adc5d3a4fe > git remote add linux-review https://github.com/0day-ci/linux > git fetch --no-tags linux-review Waiman-Long/mm-memcg-Reduce-kmemcache-memory-accounting-overhead/20210410-071958 > git checkout 1f4e22fce44599095a55535301ca83adc5d3a4fe > # save the attached .config to linux build tree > COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross ARCH=arm64 > > If you fix the issue, kindly add following tag as appropriate > Reported-by: kernel test robot <lkp@intel.com> > > All warnings (new ones prefixed by >>): > >>> mm/memcontrol.c:3234:10: warning: variable 'stock' is uninitialized when used here [-Wuninitialized] > stock = get_obj_stock(flags); > ^~~~~~~~~~~~~~~~~~~~ > mm/memcontrol.c:2284:16: note: expanded from macro 'get_obj_stock' > obj_stock = &stock->task_obj; \ > ^~~~~ > mm/memcontrol.c:3234:10: note: variable 'stock' is declared here > mm/memcontrol.c:2278:2: note: expanded from macro 'get_obj_stock' > struct memcg_stock_pcp *stock; \ > ^ > mm/memcontrol.c:3329:2: warning: variable 'stock' is uninitialized when used here [-Wuninitialized] > get_obj_stock(flags); > ^~~~~~~~~~~~~~~~~~~~ > mm/memcontrol.c:2284:16: note: expanded from macro 'get_obj_stock' > obj_stock = &stock->task_obj; \ > ^~~~~ > mm/memcontrol.c:3329:2: note: variable 'stock' is declared here > mm/memcontrol.c:2278:2: note: expanded from macro 'get_obj_stock' > struct memcg_stock_pcp *stock; \ > ^ > mm/memcontrol.c:3371:2: warning: variable 'stock' is uninitialized when used here [-Wuninitialized] > get_obj_stock(flags); > ^~~~~~~~~~~~~~~~~~~~ > mm/memcontrol.c:2284:16: note: expanded from macro 'get_obj_stock' > obj_stock = &stock->task_obj; \ > ^~~~~ > mm/memcontrol.c:3371:2: note: variable 'stock' is declared here > mm/memcontrol.c:2278:2: note: expanded from macro 'get_obj_stock' > struct memcg_stock_pcp *stock; \ > ^ > mm/memcontrol.c:3426:2: warning: variable 'stock' is uninitialized when used here [-Wuninitialized] > get_obj_stock(flags); > ^~~~~~~~~~~~~~~~~~~~ > mm/memcontrol.c:2284:16: note: expanded from macro 'get_obj_stock' > obj_stock = &stock->task_obj; \ > ^~~~~ > mm/memcontrol.c:3426:2: note: variable 'stock' is declared here > mm/memcontrol.c:2278:2: note: expanded from macro 'get_obj_stock' > struct memcg_stock_pcp *stock; \ > ^ > 4 warnings generated. > > > vim +/stock +3234 mm/memcontrol.c > > 3227 > 3228 static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) > 3229 { > 3230 struct obj_stock *stock; > 3231 unsigned long flags; > 3232 bool ret = false; > 3233 >> 3234 stock = get_obj_stock(flags); > 3235 > 3236 stock = current_obj_stock(); > 3237 if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { > 3238 stock->nr_bytes -= nr_bytes; > 3239 ret = true; > 3240 } > 3241 > 3242 put_obj_stock(flags); > 3243 > 3244 return ret; > 3245 } > 3246 > > --- > 0-DAY CI Kernel Test Service, Intel Corporation > https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org My bad, I somehow missed it. I will fix that in the version. Thanks, Longman
On Fri, Apr 09, 2021 at 07:18:42PM -0400, Waiman Long wrote: > Most kmem_cache_alloc() calls are from user context. With instrumentation > enabled, the measured amount of kmem_cache_alloc() calls from non-task > context was about 0.01% of the total. > > The irq disable/enable sequence used in this case to access content > from object stock is slow. To optimize for user context access, there > are now two object stocks for task context and interrupt context access > respectively. > > The task context object stock can be accessed after disabling preemption > which is cheap in non-preempt kernel. The interrupt context object stock > can only be accessed after disabling interrupt. User context code can > access interrupt object stock, but not vice versa. > > The mod_objcg_state() function is also modified to make sure that memcg > and lruvec stat updates are done with interrupted disabled. > > The downside of this change is that there are more data stored in local > object stocks and not reflected in the charge counter and the vmstat > arrays. However, this is a small price to pay for better performance. I agree, the extra memory space is not a significant concern. I'd be more worried about the code complexity, but the result looks nice to me! Acked-by: Roman Gushchin <guro@fb.com> Btw, it seems that the mm tree ran a bit off, so I had to apply this series on top of Linus's tree to review. Please, rebase. Thanks!
On 4/12/21 2:55 PM, Roman Gushchin wrote: > On Fri, Apr 09, 2021 at 07:18:42PM -0400, Waiman Long wrote: >> Most kmem_cache_alloc() calls are from user context. With instrumentation >> enabled, the measured amount of kmem_cache_alloc() calls from non-task >> context was about 0.01% of the total. >> >> The irq disable/enable sequence used in this case to access content >> from object stock is slow. To optimize for user context access, there >> are now two object stocks for task context and interrupt context access >> respectively. >> >> The task context object stock can be accessed after disabling preemption >> which is cheap in non-preempt kernel. The interrupt context object stock >> can only be accessed after disabling interrupt. User context code can >> access interrupt object stock, but not vice versa. >> >> The mod_objcg_state() function is also modified to make sure that memcg >> and lruvec stat updates are done with interrupted disabled. >> >> The downside of this change is that there are more data stored in local >> object stocks and not reflected in the charge counter and the vmstat >> arrays. However, this is a small price to pay for better performance. > I agree, the extra memory space is not a significant concern. > I'd be more worried about the code complexity, but the result looks > nice to me! > > Acked-by: Roman Gushchin <guro@fb.com> > > Btw, it seems that the mm tree ran a bit off, so I had to apply this series > on top of Linus's tree to review. Please, rebase. This patchset is based on the code in Linus' tree. I had applied the patchset to linux-next to see if there was any conflicts. Two of the patches had minor fuzzes around the edge but no actual merge conflict for now. Cheers, Longman
diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 69f728383efe..00c9074e42e5 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -2229,7 +2229,8 @@ struct obj_stock { struct memcg_stock_pcp { struct mem_cgroup *cached; /* this never be root cgroup */ unsigned int nr_pages; - struct obj_stock obj; + struct obj_stock task_obj; + struct obj_stock irq_obj; struct work_struct work; unsigned long flags; @@ -2254,11 +2255,46 @@ static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, } #endif +/* + * Most kmem_cache_alloc() calls are from user context. The irq disable/enable + * sequence used in this case to access content from object stock is slow. + * To optimize for user context access, there are now two object stocks for + * task context and interrupt context access respectively. + * + * The task context object stock can be accessed by disabling preemption only + * which is cheap in non-preempt kernel. The interrupt context object stock + * can only be accessed after disabling interrupt. User context code can + * access interrupt object stock, but not vice versa. + */ static inline struct obj_stock *current_obj_stock(void) { struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); - return &stock->obj; + return in_task() ? &stock->task_obj : &stock->irq_obj; +} + +#define get_obj_stock(flags) \ +({ \ + struct memcg_stock_pcp *stock; \ + struct obj_stock *obj_stock; \ + \ + if (in_task()) { \ + preempt_disable(); \ + (flags) = -1L; \ + obj_stock = &stock->task_obj; \ + } else { \ + local_irq_save(flags); \ + obj_stock = &stock->irq_obj; \ + } \ + obj_stock; \ +}) + +static inline void put_obj_stock(unsigned long flags) +{ + if (flags == -1L) + preempt_enable(); + else + local_irq_restore(flags); } /** @@ -2327,7 +2363,9 @@ static void drain_local_stock(struct work_struct *dummy) local_irq_save(flags); stock = this_cpu_ptr(&memcg_stock); - drain_obj_stock(&stock->obj); + drain_obj_stock(&stock->irq_obj); + if (in_task()) + drain_obj_stock(&stock->task_obj); drain_stock(stock); clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); @@ -3183,7 +3221,7 @@ static inline void mod_objcg_state(struct obj_cgroup *objcg, memcg = obj_cgroup_memcg(objcg); if (pgdat) lruvec = mem_cgroup_lruvec(memcg, pgdat); - __mod_memcg_lruvec_state(memcg, lruvec, idx, nr); + mod_memcg_lruvec_state(memcg, lruvec, idx, nr); rcu_read_unlock(); } @@ -3193,7 +3231,7 @@ static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) unsigned long flags; bool ret = false; - local_irq_save(flags); + stock = get_obj_stock(flags); stock = current_obj_stock(); if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { @@ -3201,7 +3239,7 @@ static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) ret = true; } - local_irq_restore(flags); + put_obj_stock(flags); return ret; } @@ -3254,8 +3292,13 @@ static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, { struct mem_cgroup *memcg; - if (stock->obj.cached_objcg) { - memcg = obj_cgroup_memcg(stock->obj.cached_objcg); + if (in_task() && stock->task_obj.cached_objcg) { + memcg = obj_cgroup_memcg(stock->task_obj.cached_objcg); + if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) + return true; + } + if (stock->irq_obj.cached_objcg) { + memcg = obj_cgroup_memcg(stock->irq_obj.cached_objcg); if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) return true; } @@ -3283,9 +3326,9 @@ static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) { unsigned long flags; - local_irq_save(flags); + get_obj_stock(flags); __refill_obj_stock(objcg, nr_bytes); - local_irq_restore(flags); + put_obj_stock(flags); } static void __mod_obj_stock_state(struct obj_cgroup *objcg, @@ -3325,9 +3368,9 @@ void mod_obj_stock_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, { unsigned long flags; - local_irq_save(flags); + get_obj_stock(flags); __mod_obj_stock_state(objcg, pgdat, idx, nr); - local_irq_restore(flags); + put_obj_stock(flags); } int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) @@ -3380,10 +3423,10 @@ void obj_cgroup_uncharge_mod_state(struct obj_cgroup *objcg, size_t size, { unsigned long flags; - local_irq_save(flags); + get_obj_stock(flags); __refill_obj_stock(objcg, size); __mod_obj_stock_state(objcg, pgdat, idx, -(int)size); - local_irq_restore(flags); + put_obj_stock(flags); } #endif /* CONFIG_MEMCG_KMEM */
Most kmem_cache_alloc() calls are from user context. With instrumentation enabled, the measured amount of kmem_cache_alloc() calls from non-task context was about 0.01% of the total. The irq disable/enable sequence used in this case to access content from object stock is slow. To optimize for user context access, there are now two object stocks for task context and interrupt context access respectively. The task context object stock can be accessed after disabling preemption which is cheap in non-preempt kernel. The interrupt context object stock can only be accessed after disabling interrupt. User context code can access interrupt object stock, but not vice versa. The mod_objcg_state() function is also modified to make sure that memcg and lruvec stat updates are done with interrupted disabled. The downside of this change is that there are more data stored in local object stocks and not reflected in the charge counter and the vmstat arrays. However, this is a small price to pay for better performance. Signed-off-by: Waiman Long <longman@redhat.com> --- mm/memcontrol.c | 71 +++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 57 insertions(+), 14 deletions(-)