From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:35268) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1cWkFC-0000Zt-JZ for qemu-devel@nongnu.org; Thu, 26 Jan 2017 08:37:43 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1cWkF7-0002A9-N0 for qemu-devel@nongnu.org; Thu, 26 Jan 2017 08:37:42 -0500 Received: from mail.ispras.ru ([83.149.199.45]:49142) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1cWkF7-00029F-Bi for qemu-devel@nongnu.org; Thu, 26 Jan 2017 08:37:37 -0500 From: "Pavel Dovgalyuk" References: <20170126123411.5412.44769.stgit@PASHA-ISP> <20170126123423.5412.2388.stgit@PASHA-ISP> <06b49612-9faf-5933-d8e9-8645a2fcde4d@redhat.com> In-Reply-To: <06b49612-9faf-5933-d8e9-8645a2fcde4d@redhat.com> Date: Thu, 26 Jan 2017 16:37:40 +0300 Message-ID: <000e01d277d9$5d80dae0$188290a0$@ru> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Content-Language: ru Subject: Re: [Qemu-devel] [PATCH v8 2/9] icount: exit cpu loop on expire List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: 'Paolo Bonzini' , 'Pavel Dovgalyuk' , qemu-devel@nongnu.org Cc: kwolf@redhat.com, peter.maydell@linaro.org, quintela@redhat.com, jasowang@redhat.com, mst@redhat.com, kraxel@redhat.com > From: Paolo Bonzini [mailto:paolo.bonzini@gmail.com] On Behalf Of Paolo Bonzini > On 26/01/2017 13:34, Pavel Dovgalyuk wrote: > > This patch adds check to break cpu loop when icount expires without > > setting the TB_EXIT_ICOUNT_EXPIRED flag. It happens when there is no > > available translated blocks and all instructions were executed. > > In icount replay mode unnecessary tb_find will be called (which may > > cause an exception) and execution will be non-deterministic. > > > > v8: refactored loop exit code and moved it to separate function > > > > Signed-off-by: Pavel Dovgalyuk > > --- > > cpu-exec.c | 24 +++++++++++++++++++----- > > 1 file changed, 19 insertions(+), 5 deletions(-) > > > > diff --git a/cpu-exec.c b/cpu-exec.c > > index fa08c73..f9b8ec8 100644 > > --- a/cpu-exec.c > > +++ b/cpu-exec.c > > @@ -523,9 +523,25 @@ static inline void cpu_handle_interrupt(CPUState *cpu, > > *last_tb = NULL; > > } > > } > > - if (unlikely(atomic_read(&cpu->exit_request) || replay_has_interrupt())) { > > +} > > + > > + > > +static void cpu_check_loop_exit(CPUState *cpu) > > +{ > > + if (unlikely(atomic_read(&cpu->exit_request) > > + /* icount has expired, we need to break the execution loop. > > + This check is needed before tb_find to make execution > > + deterministic - tb_find may cause an exception > > + while translating the code from non-mapped page. */ > > + || (use_icount && ((cpu->icount_extra == 0 > > + && cpu->icount_decr.u16.low == 0) > > + || (int32_t)cpu->icount_decr.u32 < 0)))) { > > Simpler: > > use_icount && > ((int32_t)cpu->icount_decr.u32 < 0 || > cpu->icount_decr.low + cpu->icount_extra == 0) Right. > But I'm not sure that you need to test u32. After all you're not Checking u32 is needed, because sometimes it is less than zero. Consider the code in cpu_loop_exec_tb: int insns_left = cpu->icount_decr.u32; if (insns_left > 0) { It is set to negative in tcg_handle_interrupt. Ten days ago you also offered setting high bits of u32 in gen_icount: + /* Make the next TB exit immediately with TB_EXIT_ICOUNT_EXPIRED. */ + tcg_gen_st16_i32(-1, cpu_env, + -ENV_OFFSET + offsetof(CPUState, icount_decr.u16.high)); > testing tcg_exit_req, which is the equivalent when icount is disabled. Yes, if we want to refactor the whole loop. > > atomic_set(&cpu->exit_request, 0); > > - cpu->exception_index = EXCP_INTERRUPT; > > + /* If there is an exception that wasn't replayed yet, > > + don't change exception_index. */ > > + if (cpu->exception_index == -1) { > > + cpu->exception_index = EXCP_INTERRUPT; > > + } > > cpu_loop_exit(cpu); > > The siglongjmp is effectively the same as exiting the for(;;) loop of > cpu_exec and going back to cpu_handle_exception. So I would just merge > this with cpu_handle_interrupt, which exits a lot with cpu_loop_exit too. Do you want me to create new version of the patch, which changes the loop this way? > All cpu_loop_exit() calls in cpu_handle_interrupt become "return true", > similar to cpu_handle_exception, and then in cpu_exec you have: > > /* if an exception is pending, we execute it here */ > while (!cpu_handle_exception(cpu, &ret)) { > /* if an interrupt is pending, inject it and go back > * to cpu_handle_exception. > */ > while (!cpu_handle_interrupt(cpu, &last_tb)) { > tb = tb_find(cpu, last_tb, tb_exit); > cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc); > /* Try to align the host and virtual clocks > if the guest is in advance */ > align_clocks(&sc, cpu); > } > } > break; It might even work faster without excessive cpu_loop_exit calls. Pavel Dovgalyuk