Implement the lazy preempt for riscv.
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
---
arch/riscv/Kconfig | 1 +
arch/riscv/include/asm/thread_info.h | 7 +++++--
arch/riscv/kernel/asm-offsets.c | 1 +
arch/riscv/kernel/entry.S | 9 +++++++--
4 files changed, 14 insertions(+), 4 deletions(-)
diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
index 7a8134fd7ec9..9f2f1936b1b5 100644
--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig
@@ -105,6 +105,7 @@ config RISCV
select HAVE_PERF_REGS
select HAVE_PERF_USER_STACK_DUMP
select HAVE_POSIX_CPU_TIMERS_TASK_WORK
+ select HAVE_PREEMPT_LAZY
select HAVE_REGS_AND_STACK_ACCESS_API
select HAVE_FUNCTION_ARG_ACCESS_API
select HAVE_STACKPROTECTOR
diff --git a/arch/riscv/include/asm/thread_info.h b/arch/riscv/include/asm/thread_info.h
index 78933ac04995..471915b179a2 100644
--- a/arch/riscv/include/asm/thread_info.h
+++ b/arch/riscv/include/asm/thread_info.h
@@ -56,6 +56,7 @@
struct thread_info {
unsigned long flags; /* low level flags */
int preempt_count; /* 0=>preemptible, <0=>BUG */
+ int preempt_lazy_count; /* 0=>preemptible, <0=>BUG */
/*
* These stack pointers are overwritten on every system call or
* exception. SP is also saved to the stack it can be recovered when
@@ -90,7 +91,7 @@ struct thread_info {
#define TIF_NOTIFY_RESUME 1 /* callback before returning to user */
#define TIF_SIGPENDING 2 /* signal pending */
#define TIF_NEED_RESCHED 3 /* rescheduling necessary */
-#define TIF_RESTORE_SIGMASK 4 /* restore signal mask in do_signal() */
+#define TIF_NEED_RESCHED_LAZY 4 /* lazy rescheduling */
#define TIF_MEMDIE 5 /* is terminating due to OOM killer */
#define TIF_SYSCALL_TRACEPOINT 6 /* syscall tracepoint instrumentation */
#define TIF_SYSCALL_AUDIT 7 /* syscall auditing */
@@ -98,6 +99,7 @@ struct thread_info {
#define TIF_NOTIFY_SIGNAL 9 /* signal notifications exist */
#define TIF_UPROBE 10 /* uprobe breakpoint or singlestep */
#define TIF_32BIT 11 /* compat-mode 32bit process */
+#define TIF_RESTORE_SIGMASK 12 /* restore signal mask in do_signal() */
#define _TIF_SYSCALL_TRACE (1 << TIF_SYSCALL_TRACE)
#define _TIF_NOTIFY_RESUME (1 << TIF_NOTIFY_RESUME)
@@ -108,10 +110,11 @@ struct thread_info {
#define _TIF_SECCOMP (1 << TIF_SECCOMP)
#define _TIF_NOTIFY_SIGNAL (1 << TIF_NOTIFY_SIGNAL)
#define _TIF_UPROBE (1 << TIF_UPROBE)
+#define _TIF_NEED_RESCHED_LAZY (1 << TIF_NEED_RESCHED_LAZY)
#define _TIF_WORK_MASK \
(_TIF_NOTIFY_RESUME | _TIF_SIGPENDING | _TIF_NEED_RESCHED | \
- _TIF_NOTIFY_SIGNAL | _TIF_UPROBE)
+ _TIF_NEED_RESCHED_LAZY | _TIF_NOTIFY_SIGNAL | _TIF_UPROBE)
#define _TIF_SYSCALL_WORK \
(_TIF_SYSCALL_TRACE | _TIF_SYSCALL_TRACEPOINT | _TIF_SYSCALL_AUDIT | \
diff --git a/arch/riscv/kernel/asm-offsets.c b/arch/riscv/kernel/asm-offsets.c
index df9444397908..e38e33822f72 100644
--- a/arch/riscv/kernel/asm-offsets.c
+++ b/arch/riscv/kernel/asm-offsets.c
@@ -35,6 +35,7 @@ void asm_offsets(void)
OFFSET(TASK_THREAD_S11, task_struct, thread.s[11]);
OFFSET(TASK_TI_FLAGS, task_struct, thread_info.flags);
OFFSET(TASK_TI_PREEMPT_COUNT, task_struct, thread_info.preempt_count);
+ OFFSET(TASK_TI_PREEMPT_LAZY_COUNT, task_struct, thread_info.preempt_lazy_count);
OFFSET(TASK_TI_KERNEL_SP, task_struct, thread_info.kernel_sp);
OFFSET(TASK_TI_USER_SP, task_struct, thread_info.user_sp);
diff --git a/arch/riscv/kernel/entry.S b/arch/riscv/kernel/entry.S
index b9eda3fcbd6d..595100a4c2c7 100644
--- a/arch/riscv/kernel/entry.S
+++ b/arch/riscv/kernel/entry.S
@@ -361,9 +361,14 @@ restore_all:
resume_kernel:
REG_L s0, TASK_TI_PREEMPT_COUNT(tp)
bnez s0, restore_all
- REG_L s0, TASK_TI_FLAGS(tp)
- andi s0, s0, _TIF_NEED_RESCHED
+ REG_L s1, TASK_TI_FLAGS(tp)
+ andi s0, s1, _TIF_NEED_RESCHED
+ bnez s0, 1f
+ REG_L s0, TASK_TI_PREEMPT_LAZY_COUNT(tp)
+ bnez s0, restore_all
+ andi s0, s1, _TIF_NEED_RESCHED_LAZY
beqz s0, restore_all
+1:
call preempt_schedule_irq
j restore_all
#endif
--
2.34.1On Thu, Sep 1, 2022 at 2:08 AM Jisheng Zhang <jszhang@kernel.org> wrote:
>
> Implement the lazy preempt for riscv.
>
> Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
> ---
> arch/riscv/Kconfig | 1 +
> arch/riscv/include/asm/thread_info.h | 7 +++++--
> arch/riscv/kernel/asm-offsets.c | 1 +
> arch/riscv/kernel/entry.S | 9 +++++++--
> 4 files changed, 14 insertions(+), 4 deletions(-)
>
> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> index 7a8134fd7ec9..9f2f1936b1b5 100644
> --- a/arch/riscv/Kconfig
> +++ b/arch/riscv/Kconfig
> @@ -105,6 +105,7 @@ config RISCV
> select HAVE_PERF_REGS
> select HAVE_PERF_USER_STACK_DUMP
> select HAVE_POSIX_CPU_TIMERS_TASK_WORK
> + select HAVE_PREEMPT_LAZY
> select HAVE_REGS_AND_STACK_ACCESS_API
> select HAVE_FUNCTION_ARG_ACCESS_API
> select HAVE_STACKPROTECTOR
> diff --git a/arch/riscv/include/asm/thread_info.h b/arch/riscv/include/asm/thread_info.h
> index 78933ac04995..471915b179a2 100644
> --- a/arch/riscv/include/asm/thread_info.h
> +++ b/arch/riscv/include/asm/thread_info.h
> @@ -56,6 +56,7 @@
> struct thread_info {
> unsigned long flags; /* low level flags */
> int preempt_count; /* 0=>preemptible, <0=>BUG */
> + int preempt_lazy_count; /* 0=>preemptible, <0=>BUG */
> /*
> * These stack pointers are overwritten on every system call or
> * exception. SP is also saved to the stack it can be recovered when
> @@ -90,7 +91,7 @@ struct thread_info {
> #define TIF_NOTIFY_RESUME 1 /* callback before returning to user */
> #define TIF_SIGPENDING 2 /* signal pending */
> #define TIF_NEED_RESCHED 3 /* rescheduling necessary */
> -#define TIF_RESTORE_SIGMASK 4 /* restore signal mask in do_signal() */
> +#define TIF_NEED_RESCHED_LAZY 4 /* lazy rescheduling */
> #define TIF_MEMDIE 5 /* is terminating due to OOM killer */
> #define TIF_SYSCALL_TRACEPOINT 6 /* syscall tracepoint instrumentation */
> #define TIF_SYSCALL_AUDIT 7 /* syscall auditing */
> @@ -98,6 +99,7 @@ struct thread_info {
> #define TIF_NOTIFY_SIGNAL 9 /* signal notifications exist */
> #define TIF_UPROBE 10 /* uprobe breakpoint or singlestep */
> #define TIF_32BIT 11 /* compat-mode 32bit process */
> +#define TIF_RESTORE_SIGMASK 12 /* restore signal mask in do_signal() */
>
> #define _TIF_SYSCALL_TRACE (1 << TIF_SYSCALL_TRACE)
> #define _TIF_NOTIFY_RESUME (1 << TIF_NOTIFY_RESUME)
> @@ -108,10 +110,11 @@ struct thread_info {
> #define _TIF_SECCOMP (1 << TIF_SECCOMP)
> #define _TIF_NOTIFY_SIGNAL (1 << TIF_NOTIFY_SIGNAL)
> #define _TIF_UPROBE (1 << TIF_UPROBE)
> +#define _TIF_NEED_RESCHED_LAZY (1 << TIF_NEED_RESCHED_LAZY)
>
> #define _TIF_WORK_MASK \
> (_TIF_NOTIFY_RESUME | _TIF_SIGPENDING | _TIF_NEED_RESCHED | \
> - _TIF_NOTIFY_SIGNAL | _TIF_UPROBE)
> + _TIF_NEED_RESCHED_LAZY | _TIF_NOTIFY_SIGNAL | _TIF_UPROBE)
>
> #define _TIF_SYSCALL_WORK \
> (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_TRACEPOINT | _TIF_SYSCALL_AUDIT | \
> diff --git a/arch/riscv/kernel/asm-offsets.c b/arch/riscv/kernel/asm-offsets.c
> index df9444397908..e38e33822f72 100644
> --- a/arch/riscv/kernel/asm-offsets.c
> +++ b/arch/riscv/kernel/asm-offsets.c
> @@ -35,6 +35,7 @@ void asm_offsets(void)
> OFFSET(TASK_THREAD_S11, task_struct, thread.s[11]);
> OFFSET(TASK_TI_FLAGS, task_struct, thread_info.flags);
> OFFSET(TASK_TI_PREEMPT_COUNT, task_struct, thread_info.preempt_count);
> + OFFSET(TASK_TI_PREEMPT_LAZY_COUNT, task_struct, thread_info.preempt_lazy_count);
> OFFSET(TASK_TI_KERNEL_SP, task_struct, thread_info.kernel_sp);
> OFFSET(TASK_TI_USER_SP, task_struct, thread_info.user_sp);
>
> diff --git a/arch/riscv/kernel/entry.S b/arch/riscv/kernel/entry.S
> index b9eda3fcbd6d..595100a4c2c7 100644
> --- a/arch/riscv/kernel/entry.S
> +++ b/arch/riscv/kernel/entry.S
> @@ -361,9 +361,14 @@ restore_all:
> resume_kernel:
> REG_L s0, TASK_TI_PREEMPT_COUNT(tp)
> bnez s0, restore_all
> - REG_L s0, TASK_TI_FLAGS(tp)
> - andi s0, s0, _TIF_NEED_RESCHED
> + REG_L s1, TASK_TI_FLAGS(tp)
> + andi s0, s1, _TIF_NEED_RESCHED
> + bnez s0, 1f
> + REG_L s0, TASK_TI_PREEMPT_LAZY_COUNT(tp)
> + bnez s0, restore_all
> + andi s0, s1, _TIF_NEED_RESCHED_LAZY
Can you tell me, who increased/decreased the PREEMPT_LAZY_COUNT? And
who set NEED_RESCHED_LAZY?
> beqz s0, restore_all
> +1:
> call preempt_schedule_irq
> j restore_all
> #endif
> --
> 2.34.1
>
>
> --
> kvm-riscv mailing list
> kvm-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/kvm-riscv
--
Best Regards
Guo Ren
On 2022-09-04 23:16:12 [+0800], Guo Ren wrote: > > diff --git a/arch/riscv/kernel/entry.S b/arch/riscv/kernel/entry.S > > index b9eda3fcbd6d..595100a4c2c7 100644 > > --- a/arch/riscv/kernel/entry.S > > +++ b/arch/riscv/kernel/entry.S > > @@ -361,9 +361,14 @@ restore_all: > > resume_kernel: > > REG_L s0, TASK_TI_PREEMPT_COUNT(tp) > > bnez s0, restore_all > > - REG_L s0, TASK_TI_FLAGS(tp) > > - andi s0, s0, _TIF_NEED_RESCHED > > + REG_L s1, TASK_TI_FLAGS(tp) > > + andi s0, s1, _TIF_NEED_RESCHED > > + bnez s0, 1f > > + REG_L s0, TASK_TI_PREEMPT_LAZY_COUNT(tp) > > + bnez s0, restore_all > > + andi s0, s1, _TIF_NEED_RESCHED_LAZY > Can you tell me, who increased/decreased the PREEMPT_LAZY_COUNT? And > who set NEED_RESCHED_LAZY? There is "generic" code in the PREEMPT_RT patch doing that. The counter is incremented/ decremented via preempt_lazy_enable()/disable() and one of the user is migrate_disable()/enable(). Basically if a task is task_is_realtime() then NEED_RESCHED is set for the wakeup. For the remaining states (SCHED_OTHER, …) NEED_RESCHED_LAZY is set for the wakeup. This can be delayed if the task is in a "preempt disable lazy" section (similar to a preempt_disable() section) but a task_is_realtime() can still be scheduled if needed. See details at https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git/plain/patches/sched__Add_support_for_lazy_preemption.patch?h=linux-6.0.y-rt-patches Sebastian
On Mon, Sep 5, 2022 at 2:34 PM Sebastian Andrzej Siewior <bigeasy@linutronix.de> wrote: > > On 2022-09-04 23:16:12 [+0800], Guo Ren wrote: > > > diff --git a/arch/riscv/kernel/entry.S b/arch/riscv/kernel/entry.S > > > index b9eda3fcbd6d..595100a4c2c7 100644 > > > --- a/arch/riscv/kernel/entry.S > > > +++ b/arch/riscv/kernel/entry.S > > > @@ -361,9 +361,14 @@ restore_all: > > > resume_kernel: > > > REG_L s0, TASK_TI_PREEMPT_COUNT(tp) > > > bnez s0, restore_all > > > - REG_L s0, TASK_TI_FLAGS(tp) > > > - andi s0, s0, _TIF_NEED_RESCHED > > > + REG_L s1, TASK_TI_FLAGS(tp) > > > + andi s0, s1, _TIF_NEED_RESCHED > > > + bnez s0, 1f > > > + REG_L s0, TASK_TI_PREEMPT_LAZY_COUNT(tp) > > > + bnez s0, restore_all > > > + andi s0, s1, _TIF_NEED_RESCHED_LAZY > > Can you tell me, who increased/decreased the PREEMPT_LAZY_COUNT? And > > who set NEED_RESCHED_LAZY? > > There is "generic" code in the PREEMPT_RT patch doing that. The counter > is incremented/ decremented via preempt_lazy_enable()/disable() and one > of the user is migrate_disable()/enable(). > Basically if a task is task_is_realtime() then NEED_RESCHED is set for > the wakeup. For the remaining states (SCHED_OTHER, …) NEED_RESCHED_LAZY > is set for the wakeup. This can be delayed if the task is in a "preempt > disable lazy" section (similar to a preempt_disable() section) but a > task_is_realtime() can still be scheduled if needed. Okay, It should be [PATCH RT]. RISC-V would also move to GENERIC_ENTRY [1], so above assembly code would be replaced by generic one, right? [1]: https://lore.kernel.org/linux-riscv/20220904072637.8619-3-guoren@kernel.org/T/#u > See details at > https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git/plain/patches/sched__Add_support_for_lazy_preemption.patch?h=linux-6.0.y-rt-patches > > Sebastian -- Best Regards Guo Ren
On Mon, Sep 05, 2022 at 04:33:54PM +0800, Guo Ren wrote: > On Mon, Sep 5, 2022 at 2:34 PM Sebastian Andrzej Siewior > <bigeasy@linutronix.de> wrote: > > > > On 2022-09-04 23:16:12 [+0800], Guo Ren wrote: > > > > diff --git a/arch/riscv/kernel/entry.S b/arch/riscv/kernel/entry.S > > > > index b9eda3fcbd6d..595100a4c2c7 100644 > > > > --- a/arch/riscv/kernel/entry.S > > > > +++ b/arch/riscv/kernel/entry.S > > > > @@ -361,9 +361,14 @@ restore_all: > > > > resume_kernel: > > > > REG_L s0, TASK_TI_PREEMPT_COUNT(tp) > > > > bnez s0, restore_all > > > > - REG_L s0, TASK_TI_FLAGS(tp) > > > > - andi s0, s0, _TIF_NEED_RESCHED > > > > + REG_L s1, TASK_TI_FLAGS(tp) > > > > + andi s0, s1, _TIF_NEED_RESCHED > > > > + bnez s0, 1f > > > > + REG_L s0, TASK_TI_PREEMPT_LAZY_COUNT(tp) > > > > + bnez s0, restore_all > > > > + andi s0, s1, _TIF_NEED_RESCHED_LAZY > > > Can you tell me, who increased/decreased the PREEMPT_LAZY_COUNT? And > > > who set NEED_RESCHED_LAZY? > > > > There is "generic" code in the PREEMPT_RT patch doing that. The counter > > is incremented/ decremented via preempt_lazy_enable()/disable() and one > > of the user is migrate_disable()/enable(). > > Basically if a task is task_is_realtime() then NEED_RESCHED is set for > > the wakeup. For the remaining states (SCHED_OTHER, …) NEED_RESCHED_LAZY > > is set for the wakeup. This can be delayed if the task is in a "preempt > > disable lazy" section (similar to a preempt_disable() section) but a > > task_is_realtime() can still be scheduled if needed. > Okay, It should be [PATCH RT]. RISC-V would also move to GENERIC_ENTRY As said in the cover letter, this patch is expected to reviewed and maintained in RT tree. If your GENERIC_ENTRY patches are merged, I will send an updated patch. > [1], so above assembly code would be replaced by generic one, right? > > [1]: https://lore.kernel.org/linux-riscv/20220904072637.8619-3-guoren@kernel.org/T/#u > > > > See details at > > https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git/plain/patches/sched__Add_support_for_lazy_preemption.patch?h=linux-6.0.y-rt-patches > > > > > > Sebastian > > > > -- > Best Regards > Guo Ren
On 2022-09-05 16:33:54 [+0800], Guo Ren wrote: > > There is "generic" code in the PREEMPT_RT patch doing that. The counter > > is incremented/ decremented via preempt_lazy_enable()/disable() and one > > of the user is migrate_disable()/enable(). > > Basically if a task is task_is_realtime() then NEED_RESCHED is set for > > the wakeup. For the remaining states (SCHED_OTHER, …) NEED_RESCHED_LAZY > > is set for the wakeup. This can be delayed if the task is in a "preempt > > disable lazy" section (similar to a preempt_disable() section) but a > > task_is_realtime() can still be scheduled if needed. > Okay, It should be [PATCH RT]. RISC-V would also move to GENERIC_ENTRY > [1], so above assembly code would be replaced by generic one, right? correct. Sebastian
On Mon, Sep 5, 2022 at 4:46 PM Sebastian Andrzej Siewior
<bigeasy@linutronix.de> wrote:
>
> On 2022-09-05 16:33:54 [+0800], Guo Ren wrote:
> > > There is "generic" code in the PREEMPT_RT patch doing that. The counter
> > > is incremented/ decremented via preempt_lazy_enable()/disable() and one
> > > of the user is migrate_disable()/enable().
> > > Basically if a task is task_is_realtime() then NEED_RESCHED is set for
> > > the wakeup. For the remaining states (SCHED_OTHER, …) NEED_RESCHED_LAZY
> > > is set for the wakeup. This can be delayed if the task is in a "preempt
> > > disable lazy" section (similar to a preempt_disable() section) but a
> > > task_is_realtime() can still be scheduled if needed.
> > Okay, It should be [PATCH RT]. RISC-V would also move to GENERIC_ENTRY
> > [1], so above assembly code would be replaced by generic one, right?
>
> correct.
Maybe TIF_XXX_RESCHED also could be merged into GENERIC_ENTRY, just
like what you've done in syscall.
struct thread_info {
unsigned long flags;
unsigned long syscall_work; /* SYSCALL_WORK_ flags */
+ unsigned long resched_work; /* RESCHED flags */
Or merge them into one:
struct thread_info {
unsigned long flags;
- unsigned long syscall_work; /* SYSCALL_WORK_ flags */
+ unsigned long ge_flags; /* GENERIC_ENTRY flags */
>
> Sebastian
--
Best Regards
Guo Ren
© 2016 - 2026 Red Hat, Inc.