Message ID | 1474289459-15242-16-git-send-email-pbonzini@redhat.com |
---|---|
State | New |
Headers | show |
On Mon, Sep 19, 2016 at 14:50:58 +0200, Paolo Bonzini wrote: > From: Sergey Fedorov <serge.fdrv@gmail.com> > > Use async_safe_run_on_cpu() to make tb_flush() thread safe. This is > possible now that code generation does not happen in the middle of > execution. > > It can happen that multiple threads schedule a safe work to flush the > translation buffer. To keep statistics and debugging output sane, always > check if the translation buffer has already been flushed. > > Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com> > Signed-off-by: Sergey Fedorov <sergey.fedorov@linaro.org> > [AJB: minor re-base fixes] > Signed-off-by: Alex Bennée <alex.bennee@linaro.org> > Message-Id: <1470158864-17651-13-git-send-email-alex.bennee@linaro.org> > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> > --- (snip) > @@ -38,7 +38,7 @@ struct TBContext { > QemuMutex tb_lock; > > /* statistics */ > - int tb_flush_count; > + unsigned tb_flush_count; (snip) > /* flush all the translation blocks */ > -/* XXX: tb_flush is currently not thread safe */ > -void tb_flush(CPUState *cpu) > +static void do_tb_flush(CPUState *cpu, void *data) > { > - if (!tcg_enabled()) { > - return; > + unsigned tb_flush_req = (unsigned) (uintptr_t) data; > + > + tb_lock(); > + > + /* If it's already been done on request of another CPU, > + * just retry. > + */ > + if (atomic_read(&tcg_ctx.tb_ctx.tb_flush_count) != tb_flush_req) { > + goto done; tb_flush_count is always accessed with tb_lock held, right? If so, I don't see a reason to access it with atomic_read/set. (snip) > @@ -1773,7 +1790,8 @@ void dump_exec_info(FILE *f, fprintf_function cpu_fprintf) > qht_statistics_destroy(&hst); > > cpu_fprintf(f, "\nStatistics:\n"); > - cpu_fprintf(f, "TB flush count %d\n", tcg_ctx.tb_ctx.tb_flush_count); > + cpu_fprintf(f, "TB flush count %d\n", > + atomic_read(&tcg_ctx.tb_ctx.tb_flush_count)); s/%d/%u/ would be more appropriate given the type change. E.
On 21/09/2016 18:05, Emilio G. Cota wrote: >> > + tb_lock(); >> > + >> > + /* If it's already been done on request of another CPU, >> > + * just retry. >> > + */ >> > + if (atomic_read(&tcg_ctx.tb_ctx.tb_flush_count) != tb_flush_req) { >> > + goto done; > tb_flush_count is always accessed with tb_lock held, right? If so, I don't > see a reason to access it with atomic_read/set. tb_flush accesses it outside tb_lock. Technically this one you're quoting need not use atomic_read, but others need to. >> + cpu_fprintf(f, "TB flush count %d\n", >> + atomic_read(&tcg_ctx.tb_ctx.tb_flush_count)); > > s/%d/%u/ would be more appropriate given the type change. Ok. Paolo
On Wed, Sep 21, 2016 at 18:19:26 +0200, Paolo Bonzini wrote: > > > On 21/09/2016 18:05, Emilio G. Cota wrote: > >> > + tb_lock(); > >> > + > >> > + /* If it's already been done on request of another CPU, > >> > + * just retry. > >> > + */ > >> > + if (atomic_read(&tcg_ctx.tb_ctx.tb_flush_count) != tb_flush_req) { > >> > + goto done; > > tb_flush_count is always accessed with tb_lock held, right? If so, I don't > > see a reason to access it with atomic_read/set. > > tb_flush accesses it outside tb_lock. Technically this one you're > quoting need not use atomic_read, but others need to. Sorry for being thick, but when does tb_flush not own tb_lock? (I'm assuming we're talking only user-mode, since full-system has for now empty tb_lock/unlock helpers.) E.
On 21/09/2016 19:37, Emilio G. Cota wrote: > On Wed, Sep 21, 2016 at 18:19:26 +0200, Paolo Bonzini wrote: >> >> >> On 21/09/2016 18:05, Emilio G. Cota wrote: >>>>> + tb_lock(); >>>>> + >>>>> + /* If it's already been done on request of another CPU, >>>>> + * just retry. >>>>> + */ >>>>> + if (atomic_read(&tcg_ctx.tb_ctx.tb_flush_count) != tb_flush_req) { >>>>> + goto done; >>> tb_flush_count is always accessed with tb_lock held, right? If so, I don't >>> see a reason to access it with atomic_read/set. >> >> tb_flush accesses it outside tb_lock. Technically this one you're >> quoting need not use atomic_read, but others need to. > > Sorry for being thick, but when does tb_flush not own tb_lock? > (I'm assuming we're talking only user-mode, since full-system has > for now empty tb_lock/unlock helpers.) When called from gdbstub I think it doesn't (and for system-mode in other cases too, so better be ready anyway). Paolo
diff --git a/cpu-exec.c b/cpu-exec.c index b240b9f..a8ff2a1 100644 --- a/cpu-exec.c +++ b/cpu-exec.c @@ -203,20 +203,16 @@ static void cpu_exec_nocache(CPUState *cpu, int max_cycles, TranslationBlock *orig_tb, bool ignore_icount) { TranslationBlock *tb; - bool old_tb_flushed; /* Should never happen. We only end up here when an existing TB is too long. */ if (max_cycles > CF_COUNT_MASK) max_cycles = CF_COUNT_MASK; - old_tb_flushed = cpu->tb_flushed; - cpu->tb_flushed = false; tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags, max_cycles | CF_NOCACHE | (ignore_icount ? CF_IGNORE_ICOUNT : 0)); - tb->orig_tb = cpu->tb_flushed ? NULL : orig_tb; - cpu->tb_flushed |= old_tb_flushed; + tb->orig_tb = orig_tb; /* execute the generated code */ trace_exec_tb_nocache(tb, tb->pc); cpu_tb_exec(cpu, tb); @@ -337,10 +333,7 @@ static inline TranslationBlock *tb_find(CPUState *cpu, tb_lock(); have_tb_lock = true; } - /* Check if translation buffer has been flushed */ - if (cpu->tb_flushed) { - cpu->tb_flushed = false; - } else if (!tb->invalid) { + if (!tb->invalid) { tb_add_jump(last_tb, tb_exit, tb); } } @@ -605,7 +598,6 @@ int cpu_exec(CPUState *cpu) break; } - atomic_mb_set(&cpu->tb_flushed, false); /* reset before first TB lookup */ for(;;) { cpu_handle_interrupt(cpu, &last_tb); tb = tb_find(cpu, last_tb, tb_exit); diff --git a/include/exec/tb-context.h b/include/exec/tb-context.h index dce95d9..c7f17f2 100644 --- a/include/exec/tb-context.h +++ b/include/exec/tb-context.h @@ -38,7 +38,7 @@ struct TBContext { QemuMutex tb_lock; /* statistics */ - int tb_flush_count; + unsigned tb_flush_count; int tb_phys_invalidate_count; }; diff --git a/include/qom/cpu.h b/include/qom/cpu.h index 4092dd9..5dfe74a 100644 --- a/include/qom/cpu.h +++ b/include/qom/cpu.h @@ -253,7 +253,6 @@ struct qemu_work_item; * @crash_occurred: Indicates the OS reported a crash (panic) for this CPU * @tcg_exit_req: Set to force TCG to stop executing linked TBs for this * CPU and return to its top level loop. - * @tb_flushed: Indicates the translation buffer has been flushed. * @singlestep_enabled: Flags for single-stepping. * @icount_extra: Instructions until next timer event. * @icount_decr: Number of cycles left, with interrupt flag in high bit. @@ -306,7 +305,6 @@ struct CPUState { bool unplug; bool crash_occurred; bool exit_request; - bool tb_flushed; uint32_t interrupt_request; int singlestep_enabled; int64_t icount_extra; diff --git a/translate-all.c b/translate-all.c index b6663dc..ab657e7 100644 --- a/translate-all.c +++ b/translate-all.c @@ -832,12 +832,19 @@ static void page_flush_tb(void) } /* flush all the translation blocks */ -/* XXX: tb_flush is currently not thread safe */ -void tb_flush(CPUState *cpu) +static void do_tb_flush(CPUState *cpu, void *data) { - if (!tcg_enabled()) { - return; + unsigned tb_flush_req = (unsigned) (uintptr_t) data; + + tb_lock(); + + /* If it's already been done on request of another CPU, + * just retry. + */ + if (atomic_read(&tcg_ctx.tb_ctx.tb_flush_count) != tb_flush_req) { + goto done; } + #if defined(DEBUG_FLUSH) printf("qemu: flush code_size=%ld nb_tbs=%d avg_tb_size=%ld\n", (unsigned long)(tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer), @@ -856,7 +863,6 @@ void tb_flush(CPUState *cpu) for (i = 0; i < TB_JMP_CACHE_SIZE; ++i) { atomic_set(&cpu->tb_jmp_cache[i], NULL); } - atomic_mb_set(&cpu->tb_flushed, true); } tcg_ctx.tb_ctx.nb_tbs = 0; @@ -866,7 +872,19 @@ void tb_flush(CPUState *cpu) tcg_ctx.code_gen_ptr = tcg_ctx.code_gen_buffer; /* XXX: flush processor icache at this point if cache flush is expensive */ - tcg_ctx.tb_ctx.tb_flush_count++; + atomic_inc(&tcg_ctx.tb_ctx.tb_flush_count); + +done: + tb_unlock(); +} + +void tb_flush(CPUState *cpu) +{ + if (tcg_enabled()) { + uintptr_t tb_flush_req = (uintptr_t) + atomic_read(&tcg_ctx.tb_ctx.tb_flush_count); + async_safe_run_on_cpu(cpu, do_tb_flush, (void *) tb_flush_req); + } } #ifdef DEBUG_TB_CHECK @@ -1173,9 +1191,8 @@ TranslationBlock *tb_gen_code(CPUState *cpu, buffer_overflow: /* flush must be done */ tb_flush(cpu); - /* cannot fail at this point */ - tb = tb_alloc(pc); - assert(tb != NULL); + mmap_unlock(); + cpu_loop_exit(cpu); } gen_code_buf = tcg_ctx.code_gen_ptr; @@ -1773,7 +1790,8 @@ void dump_exec_info(FILE *f, fprintf_function cpu_fprintf) qht_statistics_destroy(&hst); cpu_fprintf(f, "\nStatistics:\n"); - cpu_fprintf(f, "TB flush count %d\n", tcg_ctx.tb_ctx.tb_flush_count); + cpu_fprintf(f, "TB flush count %d\n", + atomic_read(&tcg_ctx.tb_ctx.tb_flush_count)); cpu_fprintf(f, "TB invalidate count %d\n", tcg_ctx.tb_ctx.tb_phys_invalidate_count); cpu_fprintf(f, "TLB flush count %d\n", tlb_flush_count);