Message ID | 1486770080-25007-1-git-send-email-hoeun.ryu@gmail.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Sat 11-02-17 08:40:38, Hoeun Ryu wrote: > Using virtually mapped stack, kernel stacks are allocated via vmalloc. > In the current implementation, two stacks per cpu can be cached when > tasks are freed and the cached stacks are used again in task duplications. > but the cached stacks may remain unfreed even when cpu are offline. > By adding a cpu hotplug callback to free the cached stacks when a cpu > goes offline, the pages of the cached stacks are not wasted. > > Signed-off-by: Hoeun Ryu <hoeun.ryu@gmail.com> Acked-by: Michal Hocko <mhocko@suse.com> > --- > v4: > use CPUHP_BP_PREPARE_DYN state for cpuhp setup > fix minor coding style > v3: > fix misuse of per-cpu api > fix location of function definition within CONFIG_VMAP_STACK > v2: > remove cpuhp callback for `startup`, only `teardown` callback is installed. > > kernel/fork.c | 23 +++++++++++++++++++++++ > 1 file changed, 23 insertions(+) > > diff --git a/kernel/fork.c b/kernel/fork.c > index 937ba59..61634d7 100644 > --- a/kernel/fork.c > +++ b/kernel/fork.c > @@ -168,6 +168,24 @@ void __weak arch_release_thread_stack(unsigned long *stack) > */ > #define NR_CACHED_STACKS 2 > static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); > + > +static int free_vm_stack_cache(unsigned int cpu) > +{ > + struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); > + int i; > + > + for (i = 0; i < NR_CACHED_STACKS; i++) { > + struct vm_struct *vm_stack = cached_vm_stacks[i]; > + > + if (!vm_stack) > + continue; > + > + vfree(vm_stack->addr); > + cached_vm_stacks[i] = NULL; > + } > + > + return 0; > +} > #endif > > static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node) > @@ -456,6 +474,11 @@ void __init fork_init(void) > for (i = 0; i < UCOUNT_COUNTS; i++) { > init_user_ns.ucount_max[i] = max_threads/2; > } > + > +#ifdef CONFIG_VMAP_STACK > + cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vmstack_cache", > + NULL, free_vm_stack_cache); > +#endif > } > > int __weak arch_dup_task_struct(struct task_struct *dst, > -- > 2.7.4 >
On Sat, 11 Feb 2017, Hoeun Ryu wrote: > #define NR_CACHED_STACKS 2 > static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); > + > +static int free_vm_stack_cache(unsigned int cpu) > +{ > + struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); > + int i; > + > + for (i = 0; i < NR_CACHED_STACKS; i++) { > + struct vm_struct *vm_stack = cached_vm_stacks[i]; > + > + if (!vm_stack) > + continue; > + > + vfree(vm_stack->addr); > + cached_vm_stacks[i] = NULL; > + } > + > + return 0; > +} > #endif > > static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node) > @@ -456,6 +474,11 @@ void __init fork_init(void) > for (i = 0; i < UCOUNT_COUNTS; i++) { > init_user_ns.ucount_max[i] = max_threads/2; > } > + > +#ifdef CONFIG_VMAP_STACK > + cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vmstack_cache", > + NULL, free_vm_stack_cache); > +#endif If you wrap that in vm_stack_cache_init() and have a stub for !VMAP_STACK (add an #else to the #ifdef VMAP_STACK section above) then you spare the ugly ifdef in fork_init() Thanks, tglx
> On Feb 11, 2017, at 5:31 PM, Thomas Gleixner <tglx@linutronix.de> wrote: > >> On Sat, 11 Feb 2017, Hoeun Ryu wrote: >> #define NR_CACHED_STACKS 2 >> static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); >> + >> +static int free_vm_stack_cache(unsigned int cpu) >> +{ >> + struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); >> + int i; >> + >> + for (i = 0; i < NR_CACHED_STACKS; i++) { >> + struct vm_struct *vm_stack = cached_vm_stacks[i]; >> + >> + if (!vm_stack) >> + continue; >> + >> + vfree(vm_stack->addr); >> + cached_vm_stacks[i] = NULL; >> + } >> + >> + return 0; >> +} >> #endif >> >> static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node) >> @@ -456,6 +474,11 @@ void __init fork_init(void) >> for (i = 0; i < UCOUNT_COUNTS; i++) { >> init_user_ns.ucount_max[i] = max_threads/2; >> } >> + >> +#ifdef CONFIG_VMAP_STACK >> + cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vmstack_cache", >> + NULL, free_vm_stack_cache); >> +#endif > > If you wrap that in vm_stack_cache_init() and have a stub for !VMAP_STACK > (add an #else to the #ifdef VMAP_STACK section above) then you spare the > ugly ifdef in fork_init() OK. I will. > Thanks, > > tglx
On Sat, Feb 11, 2017 at 6:56 PM, Hoeun Ryu <hoeun.ryu@gmail.com> wrote: > >> On Feb 11, 2017, at 5:31 PM, Thomas Gleixner <tglx@linutronix.de> wrote: >> >>> On Sat, 11 Feb 2017, Hoeun Ryu wrote: >>> #define NR_CACHED_STACKS 2 >>> static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); >>> + >>> +static int free_vm_stack_cache(unsigned int cpu) >>> +{ >>> + struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); >>> + int i; >>> + >>> + for (i = 0; i < NR_CACHED_STACKS; i++) { >>> + struct vm_struct *vm_stack = cached_vm_stacks[i]; >>> + >>> + if (!vm_stack) >>> + continue; >>> + >>> + vfree(vm_stack->addr); >>> + cached_vm_stacks[i] = NULL; >>> + } >>> + >>> + return 0; >>> +} >>> #endif >>> >>> static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node) >>> @@ -456,6 +474,11 @@ void __init fork_init(void) >>> for (i = 0; i < UCOUNT_COUNTS; i++) { >>> init_user_ns.ucount_max[i] = max_threads/2; >>> } >>> + >>> +#ifdef CONFIG_VMAP_STACK >>> + cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vmstack_cache", >>> + NULL, free_vm_stack_cache); >>> +#endif >> >> If you wrap that in vm_stack_cache_init() and have a stub for !VMAP_STACK >> (add an #else to the #ifdef VMAP_STACK section above) then you spare the >> ugly ifdef in fork_init() > > OK. I will. > I sent v5 patch having those stub, (Kbuild error anyway). But, I didn't consider THREAD_SIZE >= PAGE_SIZE. so the blank inlined version of vm_stack_cache_init() should be defined twice, one for THREAD_SIZE < PAGE_SIZE && !CONFIG_VMAP_STACK, one for !CONFIG_VMAP_STACK. Do you really want this? --- # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) # if CONFIG_VMAP_STACK int free_vm_stack_cache() { ... } int vm_stack_cache_init() { cpuhp_setup_state(free_vm_stack_cache); } #else // !CONFIG_VMAP_STACK inline int vm_stack_cache_init() { return 0; } #endif #else // THREAD_SIZE < PAGE_SIZE && !CONFIG_VMAP_STACK ... ... inline int vm_stack_cache_init() { return 0; } #endif >> Thanks, >> >> tglx
diff --git a/kernel/fork.c b/kernel/fork.c index 937ba59..61634d7 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -168,6 +168,24 @@ void __weak arch_release_thread_stack(unsigned long *stack) */ #define NR_CACHED_STACKS 2 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); + +static int free_vm_stack_cache(unsigned int cpu) +{ + struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); + int i; + + for (i = 0; i < NR_CACHED_STACKS; i++) { + struct vm_struct *vm_stack = cached_vm_stacks[i]; + + if (!vm_stack) + continue; + + vfree(vm_stack->addr); + cached_vm_stacks[i] = NULL; + } + + return 0; +} #endif static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node) @@ -456,6 +474,11 @@ void __init fork_init(void) for (i = 0; i < UCOUNT_COUNTS; i++) { init_user_ns.ucount_max[i] = max_threads/2; } + +#ifdef CONFIG_VMAP_STACK + cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vmstack_cache", + NULL, free_vm_stack_cache); +#endif } int __weak arch_dup_task_struct(struct task_struct *dst,
Using virtually mapped stack, kernel stacks are allocated via vmalloc. In the current implementation, two stacks per cpu can be cached when tasks are freed and the cached stacks are used again in task duplications. but the cached stacks may remain unfreed even when cpu are offline. By adding a cpu hotplug callback to free the cached stacks when a cpu goes offline, the pages of the cached stacks are not wasted. Signed-off-by: Hoeun Ryu <hoeun.ryu@gmail.com> --- v4: use CPUHP_BP_PREPARE_DYN state for cpuhp setup fix minor coding style v3: fix misuse of per-cpu api fix location of function definition within CONFIG_VMAP_STACK v2: remove cpuhp callback for `startup`, only `teardown` callback is installed. kernel/fork.c | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+)