| Message ID | 20200805181303.7822-3-robert.foley@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | accel/tcg: remove implied BQL from cpu_handle_interrupt/exception path | expand |
On 8/5/20 11:12 AM, Robert Foley wrote: > @@ -299,8 +299,12 @@ void alpha_cpu_do_interrupt(CPUState *cs) > { > AlphaCPU *cpu = ALPHA_CPU(cs); > CPUAlphaState *env = &cpu->env; > - int i = cs->exception_index; > - > + int i; > + bool bql = !qemu_mutex_iothread_locked(); > + if (bql) { > + qemu_mutex_lock_iothread(); > + } Why does this patch for alpha need to check qemu_mutex_iothread_locked and the next patch for arm does not? r~
On Wed, 5 Aug 2020 at 15:18, Richard Henderson <richard.henderson@linaro.org> wrote: > > On 8/5/20 11:12 AM, Robert Foley wrote: > > @@ -299,8 +299,12 @@ void alpha_cpu_do_interrupt(CPUState *cs) > > { > > AlphaCPU *cpu = ALPHA_CPU(cs); > > CPUAlphaState *env = &cpu->env; > > - int i = cs->exception_index; > > - > > + int i; > > + bool bql = !qemu_mutex_iothread_locked(); > > + if (bql) { > > + qemu_mutex_lock_iothread(); > > + } > > Why does this patch for alpha need to check qemu_mutex_iothread_locked and the > next patch for arm does not? > In alpha (and arm) the do_interrupt function can be called separately or by cpu_exec_interrupt. In the case where do_interrupt gets called separately it needs to take the BQL (bql == true). In the case where cpu_exec_interrupt is holding the BQL, and calls do_interrupt, do_interrupt needs to check qemu_mutex_iothread_locked, and in this case not get the lock (bql == false). The next patch for arm, checks qemu_mutex_iothread_locked in its do_interrupt function, but not in its cpu_exec_interrupt function, the same pattern as for alpha. Thanks & Regards, -Rob > > r~
diff --git a/target/alpha/helper.c b/target/alpha/helper.c index 55d7274d94..18169ae1c5 100644 --- a/target/alpha/helper.c +++ b/target/alpha/helper.c @@ -299,8 +299,12 @@ void alpha_cpu_do_interrupt(CPUState *cs) { AlphaCPU *cpu = ALPHA_CPU(cs); CPUAlphaState *env = &cpu->env; - int i = cs->exception_index; - + int i; + bool bql = !qemu_mutex_iothread_locked(); + if (bql) { + qemu_mutex_lock_iothread(); + } + i = cs->exception_index; if (qemu_loglevel_mask(CPU_LOG_INT)) { static int count; const char *name = "<unknown>"; @@ -405,6 +409,9 @@ void alpha_cpu_do_interrupt(CPUState *cs) /* Switch to PALmode. */ env->flags |= ENV_FLAG_PAL_MODE; #endif /* !USER_ONLY */ + if (bql) { + qemu_mutex_unlock_iothread(); + } } bool alpha_cpu_exec_interrupt(CPUState *cs, int interrupt_request) @@ -412,9 +419,11 @@ bool alpha_cpu_exec_interrupt(CPUState *cs, int interrupt_request) AlphaCPU *cpu = ALPHA_CPU(cs); CPUAlphaState *env = &cpu->env; int idx = -1; + qemu_mutex_lock_iothread(); /* We never take interrupts while in PALmode. */ if (env->flags & ENV_FLAG_PAL_MODE) { + qemu_mutex_unlock_iothread(); return false; } @@ -446,8 +455,10 @@ bool alpha_cpu_exec_interrupt(CPUState *cs, int interrupt_request) cs->exception_index = idx; env->error_code = 0; alpha_cpu_do_interrupt(cs); + qemu_mutex_unlock_iothread(); return true; } + qemu_mutex_unlock_iothread(); return false; }
This is part of a series of changes to remove the implied BQL from the common code of cpu_handle_interrupt and cpu_handle_exception. As part of removing the implied BQL from the common code, we are pushing the BQL holding down into the per-arch implementation functions of do_interrupt and cpu_exec_interrupt. The purpose of this set of changes is to set the groundwork so that an arch could move towards removing the BQL from the cpu_handle_interrupt/exception paths. This approach was suggested by Paolo Bonzini. For reference, here are two key posts in the discussion, explaining the reasoning/benefits of this approach. https://lists.gnu.org/archive/html/qemu-devel/2020-07/msg08731.html https://lists.gnu.org/archive/html/qemu-devel/2020-08/msg00044.html Signed-off-by: Robert Foley <robert.foley@linaro.org> --- target/alpha/helper.c | 15 +++++++++++++-- 1 file changed, 13 insertions(+), 2 deletions(-) -- 2.17.1