Re: [PATCH 2/2] riscv: Enable context tracking

[Greentime Hu]

guoren <guoren@xxxxxxxxxxxxxxxxx> æ 2020å6æ27æ éå äå12:29åéï
>
> Hi Greentime,
>
> On 2020/6/23 9:28 äå, Greentime Hu wrote:
> > This patch implements and enables context tracking for riscv (which is a
> > prerequisite for CONFIG_NO_HZ_FULL support)
> >
> > It adds checking for previous state in the entry that all excepttions and
> > interrupts goes to and calls context_tracking_user_exit() if it comes from
> > user space. It also calls context_tracking_user_enter() if it will return
> > to user space before restore_all.
> >
> > This patch is tested with the dynticks-testing testcase in
> > qemu-system-riscv64 virt machine and Unleashed board.
> > git://git.kernel.org/pub/scm/linux/kernel/git/frederic/dynticks-testing.git
> >
> > We can see the log here. The tick got mostly stopped during the execution
> > of the user loop.
> >
> >                          _-----=> irqs-off
> >                         / _----=> need-resched
> >                        | / _---=> hardirq/softirq
> >                        || / _--=> preempt-depth
> >                        ||| /     delay
> >       TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
> >          | |       |   ||||       |         |
> >     <idle>-0     [001] d..2   604.183512: sched_switch: prev_comm=swapper/1 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=taskset next_pid=273 next_prio=120
> > user_loop-273   [001] d.h1   604.184788: hrtimer_expire_entry: hrtimer=000000002eda5fab function=tick_sched_timer now=604176096300
> > user_loop-273   [001] d.s2   604.184897: workqueue_queue_work: work struct=00000000383402c2 function=vmstat_update workqueue=00000000f36d35d4 req_cpu=1 cpu=1
> > user_loop-273   [001] dns2   604.185039: tick_stop: success=0 dependency=SCHED
> > user_loop-273   [001] dn.1   604.185103: tick_stop: success=0 dependency=SCHED
> > user_loop-273   [001] d..2   604.185154: sched_switch: prev_comm=taskset prev_pid=273 prev_prio=120 prev_state=R+ ==> next_comm=kworker/1:1 next_pid=46 next_prio=120
> >      <...>-46    [001] ....   604.185194: workqueue_execute_start: work struct 00000000383402c2: function vmstat_update
> >      <...>-46    [001] d..2   604.185266: sched_switch: prev_comm=kworker/1:1 prev_pid=46 prev_prio=120 prev_state=I ==> next_comm=taskset next_pid=273 next_prio=120
> > user_loop-273   [001] d.h1   604.188812: hrtimer_expire_entry: hrtimer=000000002eda5fab function=tick_sched_timer now=604180133400
> > user_loop-273   [001] d..1   604.189050: tick_stop: success=1 dependency=NONE
> > user_loop-273   [001] d..2   614.251386: sched_switch: prev_comm=user_loop prev_pid=273 prev_prio=120 prev_state=X ==> next_comm=swapper/1 next_pid=0 next_prio=120
> >     <idle>-0     [001] d..2   614.315391: sched_switch: prev_comm=swapper/1 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=taskset next_pid=276 next_prio=120
> >
> > Signed-off-by: Greentime Hu<greentime.hu@xxxxxxxxxx>
> > ---
> >   arch/riscv/Kconfig        |  1 +
> >   arch/riscv/kernel/entry.S | 23 +++++++++++++++++++++++
> >   2 files changed, 24 insertions(+)
> >
> > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> > index 128192e14ff2..17520e11815b 100644
> > --- a/arch/riscv/Kconfig
> > +++ b/arch/riscv/Kconfig
> > @@ -52,6 +52,7 @@ config RISCV
> >       select HAVE_ARCH_SECCOMP_FILTER
> >       select HAVE_ARCH_TRACEHOOK
> >       select HAVE_ASM_MODVERSIONS
> > +     select HAVE_CONTEXT_TRACKING
> >       select HAVE_COPY_THREAD_TLS
> >       select HAVE_DMA_CONTIGUOUS if MMU
> >       select HAVE_EBPF_JIT if MMU
> > diff --git a/arch/riscv/kernel/entry.S b/arch/riscv/kernel/entry.S
> > index cae7e6d4c7ef..6ed579fc1073 100644
> > --- a/arch/riscv/kernel/entry.S
> > +++ b/arch/riscv/kernel/entry.S
> > @@ -97,6 +97,14 @@ _save_context:
> >       la gp, __global_pointer$
> >   .option pop
> >
> > +#ifdef CONFIG_CONTEXT_TRACKING
> > +     /* If previous state is in user mode, call context_tracking_user_exit. */
> > +     andi a0, s1, SR_SPP
> > +     bnez a0, skip_context_tracking
> > +     call context_tracking_user_exit
>
> It will destroy s1-s5 which are used in entry.S like these:
>
>          REG_S s0, PT_SP(sp)
>          REG_S s1, PT_STATUS(sp)
>          REG_S s2, PT_EPC(sp)
>          REG_S s3, PT_BADADDR(sp)
>          REG_S s4, PT_CAUSE(sp)
>          REG_S s5, PT_TP(sp)
> ...
>          /*
>           * MSB of cause differentiates between
>           * interrupts and exceptions
>           */
>          bge s4, zero, 1f
>
> ...
>
>          /*
>           * Exceptions run with interrupts enabled or disabled depending
> on the
>           * state of SR_PIE in m/sstatus.
>           */
>          andi t0, s1, SR_PIE
>
> ...
>
>          /* Handle syscalls */
>          li t0, EXC_SYSCALL
>          beq s4, t0, handle_syscall
>
> ...
>
> I also solved it in my lockdep patches, please have a look:
> https://lore.kernel.org/linux-riscv/1592323021-98541-4-git-send-email-guoren@xxxxxxxxxx/T/#u
>
>

Hi Guo,

These registers are callee saved registers. I think callee will take care of it.
https://github.com/riscv/riscv-elf-psabi-doc/blob/master/riscv-elf.md#integer-register-convention-