| Message ID | 20201128160141.1003903-6-npiggin@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | shoot lazy tlbs | expand |
On Sat, Nov 28, 2020 at 8:02 AM Nicholas Piggin <npiggin@gmail.com> wrote: > > NOMMU systems could easily go without this and save a bit of code > and the refcount atomics, because their mm switch is a no-op. I > haven't flipped them over because haven't audited all arch code to > convert over to using the _lazy_tlb refcounting. > > Signed-off-by: Nicholas Piggin <npiggin@gmail.com> > --- > arch/Kconfig | 11 +++++++ > include/linux/sched/mm.h | 13 ++++++-- > kernel/sched/core.c | 68 +++++++++++++++++++++++++++++----------- > kernel/sched/sched.h | 4 ++- > 4 files changed, 75 insertions(+), 21 deletions(-) > > diff --git a/arch/Kconfig b/arch/Kconfig > index 56b6ccc0e32d..596bf589d74b 100644 > --- a/arch/Kconfig > +++ b/arch/Kconfig > @@ -430,6 +430,17 @@ config ARCH_WANT_IRQS_OFF_ACTIVATE_MM > irqs disabled over activate_mm. Architectures that do IPI based TLB > shootdowns should enable this. > > +# Should make this depend on MMU, because there is little use for lazy mm switching > +# with NOMMU. Must audit NOMMU architecture code for lazy mm refcounting first. > +config MMU_LAZY_TLB > + def_bool y > + help > + Enable "lazy TLB" mmu context switching for kernel threads. > + > +config MMU_LAZY_TLB_REFCOUNT > + def_bool y > + depends on MMU_LAZY_TLB > + This could use some documentation as to what "no" means. > config ARCH_HAVE_NMI_SAFE_CMPXCHG > bool > > diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h > index 7157c0f6fef8..bd0f27402d4b 100644 > --- a/include/linux/sched/mm.h > +++ b/include/linux/sched/mm.h > @@ -51,12 +51,21 @@ static inline void mmdrop(struct mm_struct *mm) > /* Helpers for lazy TLB mm refcounting */ > static inline void mmgrab_lazy_tlb(struct mm_struct *mm) > { > - mmgrab(mm); > + if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) > + mmgrab(mm); > } > > static inline void mmdrop_lazy_tlb(struct mm_struct *mm) > { > - mmdrop(mm); > + if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) { > + mmdrop(mm); > + } else { > + /* > + * mmdrop_lazy_tlb must provide a full memory barrier, see the > + * membarrier comment finish_task_switch. "membarrier comment in finish_task_switch()", perhaps? > + */ > + smp_mb(); > + } > } > > /** > diff --git a/kernel/sched/core.c b/kernel/sched/core.c > index e372b613d514..3b79c6cc3a37 100644 > --- a/kernel/sched/core.c > +++ b/kernel/sched/core.c > @@ -3579,7 +3579,7 @@ static struct rq *finish_task_switch(struct task_struct *prev) > __releases(rq->lock) > { > struct rq *rq = this_rq(); > - struct mm_struct *mm = rq->prev_mm; > + struct mm_struct *mm = NULL; > long prev_state; > > /* > @@ -3598,7 +3598,10 @@ static struct rq *finish_task_switch(struct task_struct *prev) > current->comm, current->pid, preempt_count())) > preempt_count_set(FORK_PREEMPT_COUNT); > > - rq->prev_mm = NULL; > +#ifdef CONFIG_MMU_LAZY_TLB_REFCOUNT > + mm = rq->prev_lazy_mm; > + rq->prev_lazy_mm = NULL; > +#endif > > /* > * A task struct has one reference for the use as "current". > @@ -3630,6 +3633,8 @@ static struct rq *finish_task_switch(struct task_struct *prev) > * rq->curr, before returning to userspace, for > * {PRIVATE,GLOBAL}_EXPEDITED. This is implicitly provided by > * mmdrop_lazy_tlb(). > + * > + * This same issue applies to other places that mmdrop_lazy_tlb(). > */ > if (mm) > mmdrop_lazy_tlb(mm); > @@ -3719,22 +3724,10 @@ asmlinkage __visible void schedule_tail(struct task_struct *prev) > calculate_sigpending(); > } > > -/* > - * context_switch - switch to the new MM and the new thread's register state. > - */ > -static __always_inline struct rq * > -context_switch(struct rq *rq, struct task_struct *prev, > - struct task_struct *next, struct rq_flags *rf) > +static __always_inline void > +context_switch_mm(struct rq *rq, struct task_struct *prev, > + struct task_struct *next) > { > - prepare_task_switch(rq, prev, next); > - > - /* > - * For paravirt, this is coupled with an exit in switch_to to > - * combine the page table reload and the switch backend into > - * one hypercall. > - */ > - arch_start_context_switch(prev); > - > /* > * kernel -> kernel lazy + transfer active > * user -> kernel lazy + mmgrab_lazy_tlb() active > @@ -3765,11 +3758,50 @@ context_switch(struct rq *rq, struct task_struct *prev, > if (!prev->mm) { // from kernel > exit_lazy_tlb(prev->active_mm, next); > > +#ifdef CONFIG_MMU_LAZY_TLB_REFCOUNT > /* will mmdrop_lazy_tlb() in finish_task_switch(). */ > - rq->prev_mm = prev->active_mm; > + rq->prev_lazy_mm = prev->active_mm; > prev->active_mm = NULL; > +#else > + /* See membarrier comment in finish_task_switch(). */ > + smp_mb(); > +#endif > } > } > +} > + Comment here describing what this does, please. > +static __always_inline void > +context_switch_mm_nolazy(struct rq *rq, struct task_struct *prev, > + struct task_struct *next) > +{ > + if (!next->mm) > + next->active_mm = &init_mm; > + membarrier_switch_mm(rq, prev->active_mm, next->active_mm); > + switch_mm_irqs_off(prev->active_mm, next->active_mm, next); > + if (!prev->mm) > + prev->active_mm = NULL; > +} > + > +/* > + * context_switch - switch to the new MM and the new thread's register state. > + */ > +static __always_inline struct rq * > +context_switch(struct rq *rq, struct task_struct *prev, > + struct task_struct *next, struct rq_flags *rf) > +{ > + prepare_task_switch(rq, prev, next); > + > + /* > + * For paravirt, this is coupled with an exit in switch_to to > + * combine the page table reload and the switch backend into > + * one hypercall. > + */ > + arch_start_context_switch(prev); > + > + if (IS_ENABLED(CONFIG_MMU_LAZY_TLB)) > + context_switch_mm(rq, prev, next); > + else > + context_switch_mm_nolazy(rq, prev, next); > > rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); > > diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h > index df80bfcea92e..3b72aec5a2f2 100644 > --- a/kernel/sched/sched.h > +++ b/kernel/sched/sched.h > @@ -950,7 +950,9 @@ struct rq { > struct task_struct *idle; > struct task_struct *stop; > unsigned long next_balance; > - struct mm_struct *prev_mm; > +#ifdef CONFIG_MMU_LAZY_TLB_REFCOUNT > + struct mm_struct *prev_lazy_mm; > +#endif > > unsigned int clock_update_flags; > u64 clock; > -- > 2.23.0 >
Excerpts from Andy Lutomirski's message of November 29, 2020 10:36 am: > On Sat, Nov 28, 2020 at 8:02 AM Nicholas Piggin <npiggin@gmail.com> wrote: >> >> NOMMU systems could easily go without this and save a bit of code >> and the refcount atomics, because their mm switch is a no-op. I >> haven't flipped them over because haven't audited all arch code to >> convert over to using the _lazy_tlb refcounting. >> >> Signed-off-by: Nicholas Piggin <npiggin@gmail.com> >> --- >> arch/Kconfig | 11 +++++++ >> include/linux/sched/mm.h | 13 ++++++-- >> kernel/sched/core.c | 68 +++++++++++++++++++++++++++++----------- >> kernel/sched/sched.h | 4 ++- >> 4 files changed, 75 insertions(+), 21 deletions(-) >> >> diff --git a/arch/Kconfig b/arch/Kconfig >> index 56b6ccc0e32d..596bf589d74b 100644 >> --- a/arch/Kconfig >> +++ b/arch/Kconfig >> @@ -430,6 +430,17 @@ config ARCH_WANT_IRQS_OFF_ACTIVATE_MM >> irqs disabled over activate_mm. Architectures that do IPI based TLB >> shootdowns should enable this. >> >> +# Should make this depend on MMU, because there is little use for lazy mm switching >> +# with NOMMU. Must audit NOMMU architecture code for lazy mm refcounting first. >> +config MMU_LAZY_TLB >> + def_bool y >> + help >> + Enable "lazy TLB" mmu context switching for kernel threads. >> + >> +config MMU_LAZY_TLB_REFCOUNT >> + def_bool y >> + depends on MMU_LAZY_TLB >> + > > This could use some documentation as to what "no" means. Sure I can add a bit more. > >> config ARCH_HAVE_NMI_SAFE_CMPXCHG >> bool >> >> diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h >> index 7157c0f6fef8..bd0f27402d4b 100644 >> --- a/include/linux/sched/mm.h >> +++ b/include/linux/sched/mm.h >> @@ -51,12 +51,21 @@ static inline void mmdrop(struct mm_struct *mm) >> /* Helpers for lazy TLB mm refcounting */ >> static inline void mmgrab_lazy_tlb(struct mm_struct *mm) >> { >> - mmgrab(mm); >> + if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) >> + mmgrab(mm); >> } >> >> static inline void mmdrop_lazy_tlb(struct mm_struct *mm) >> { >> - mmdrop(mm); >> + if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) { >> + mmdrop(mm); >> + } else { >> + /* >> + * mmdrop_lazy_tlb must provide a full memory barrier, see the >> + * membarrier comment finish_task_switch. > > "membarrier comment in finish_task_switch()", perhaps? Sure. Thanks, Nick
diff --git a/arch/Kconfig b/arch/Kconfig index 56b6ccc0e32d..596bf589d74b 100644 --- a/arch/Kconfig +++ b/arch/Kconfig @@ -430,6 +430,17 @@ config ARCH_WANT_IRQS_OFF_ACTIVATE_MM irqs disabled over activate_mm. Architectures that do IPI based TLB shootdowns should enable this. +# Should make this depend on MMU, because there is little use for lazy mm switching +# with NOMMU. Must audit NOMMU architecture code for lazy mm refcounting first. +config MMU_LAZY_TLB + def_bool y + help + Enable "lazy TLB" mmu context switching for kernel threads. + +config MMU_LAZY_TLB_REFCOUNT + def_bool y + depends on MMU_LAZY_TLB + config ARCH_HAVE_NMI_SAFE_CMPXCHG bool diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h index 7157c0f6fef8..bd0f27402d4b 100644 --- a/include/linux/sched/mm.h +++ b/include/linux/sched/mm.h @@ -51,12 +51,21 @@ static inline void mmdrop(struct mm_struct *mm) /* Helpers for lazy TLB mm refcounting */ static inline void mmgrab_lazy_tlb(struct mm_struct *mm) { - mmgrab(mm); + if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) + mmgrab(mm); } static inline void mmdrop_lazy_tlb(struct mm_struct *mm) { - mmdrop(mm); + if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) { + mmdrop(mm); + } else { + /* + * mmdrop_lazy_tlb must provide a full memory barrier, see the + * membarrier comment finish_task_switch. + */ + smp_mb(); + } } /** diff --git a/kernel/sched/core.c b/kernel/sched/core.c index e372b613d514..3b79c6cc3a37 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -3579,7 +3579,7 @@ static struct rq *finish_task_switch(struct task_struct *prev) __releases(rq->lock) { struct rq *rq = this_rq(); - struct mm_struct *mm = rq->prev_mm; + struct mm_struct *mm = NULL; long prev_state; /* @@ -3598,7 +3598,10 @@ static struct rq *finish_task_switch(struct task_struct *prev) current->comm, current->pid, preempt_count())) preempt_count_set(FORK_PREEMPT_COUNT); - rq->prev_mm = NULL; +#ifdef CONFIG_MMU_LAZY_TLB_REFCOUNT + mm = rq->prev_lazy_mm; + rq->prev_lazy_mm = NULL; +#endif /* * A task struct has one reference for the use as "current". @@ -3630,6 +3633,8 @@ static struct rq *finish_task_switch(struct task_struct *prev) * rq->curr, before returning to userspace, for * {PRIVATE,GLOBAL}_EXPEDITED. This is implicitly provided by * mmdrop_lazy_tlb(). + * + * This same issue applies to other places that mmdrop_lazy_tlb(). */ if (mm) mmdrop_lazy_tlb(mm); @@ -3719,22 +3724,10 @@ asmlinkage __visible void schedule_tail(struct task_struct *prev) calculate_sigpending(); } -/* - * context_switch - switch to the new MM and the new thread's register state. - */ -static __always_inline struct rq * -context_switch(struct rq *rq, struct task_struct *prev, - struct task_struct *next, struct rq_flags *rf) +static __always_inline void +context_switch_mm(struct rq *rq, struct task_struct *prev, + struct task_struct *next) { - prepare_task_switch(rq, prev, next); - - /* - * For paravirt, this is coupled with an exit in switch_to to - * combine the page table reload and the switch backend into - * one hypercall. - */ - arch_start_context_switch(prev); - /* * kernel -> kernel lazy + transfer active * user -> kernel lazy + mmgrab_lazy_tlb() active @@ -3765,11 +3758,50 @@ context_switch(struct rq *rq, struct task_struct *prev, if (!prev->mm) { // from kernel exit_lazy_tlb(prev->active_mm, next); +#ifdef CONFIG_MMU_LAZY_TLB_REFCOUNT /* will mmdrop_lazy_tlb() in finish_task_switch(). */ - rq->prev_mm = prev->active_mm; + rq->prev_lazy_mm = prev->active_mm; prev->active_mm = NULL; +#else + /* See membarrier comment in finish_task_switch(). */ + smp_mb(); +#endif } } +} + +static __always_inline void +context_switch_mm_nolazy(struct rq *rq, struct task_struct *prev, + struct task_struct *next) +{ + if (!next->mm) + next->active_mm = &init_mm; + membarrier_switch_mm(rq, prev->active_mm, next->active_mm); + switch_mm_irqs_off(prev->active_mm, next->active_mm, next); + if (!prev->mm) + prev->active_mm = NULL; +} + +/* + * context_switch - switch to the new MM and the new thread's register state. + */ +static __always_inline struct rq * +context_switch(struct rq *rq, struct task_struct *prev, + struct task_struct *next, struct rq_flags *rf) +{ + prepare_task_switch(rq, prev, next); + + /* + * For paravirt, this is coupled with an exit in switch_to to + * combine the page table reload and the switch backend into + * one hypercall. + */ + arch_start_context_switch(prev); + + if (IS_ENABLED(CONFIG_MMU_LAZY_TLB)) + context_switch_mm(rq, prev, next); + else + context_switch_mm_nolazy(rq, prev, next); rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index df80bfcea92e..3b72aec5a2f2 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -950,7 +950,9 @@ struct rq { struct task_struct *idle; struct task_struct *stop; unsigned long next_balance; - struct mm_struct *prev_mm; +#ifdef CONFIG_MMU_LAZY_TLB_REFCOUNT + struct mm_struct *prev_lazy_mm; +#endif unsigned int clock_update_flags; u64 clock;
NOMMU systems could easily go without this and save a bit of code and the refcount atomics, because their mm switch is a no-op. I haven't flipped them over because haven't audited all arch code to convert over to using the _lazy_tlb refcounting. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> --- arch/Kconfig | 11 +++++++ include/linux/sched/mm.h | 13 ++++++-- kernel/sched/core.c | 68 +++++++++++++++++++++++++++++----------- kernel/sched/sched.h | 4 ++- 4 files changed, 75 insertions(+), 21 deletions(-)