[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [RFC v3 09/13] cpus.c: introduce simple callback suppor
From: |
Paolo Bonzini |
Subject: |
Re: [Qemu-devel] [RFC v3 09/13] cpus.c: introduce simple callback support |
Date: |
Fri, 10 Jul 2015 12:24:12 +0200 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.0.1 |
On 10/07/2015 11:47, alvise rigo wrote:
> I tried to use it, but it would then create a deadlock at a very early
> stage of the stress test.
> The problem is likely related to the fact that flush_queued_work
> happens with the global mutex locked.
Let's fix that and move the global mutex inside the callbacks. I can
take a look.
Paolo
> As Frederick suggested, we can use the newly introduced
> flush_queued_safe_work for this.
>
> Regards,
> alvise
>
> On Fri, Jul 10, 2015 at 11:36 AM, Paolo Bonzini <address@hidden> wrote:
>>
>>
>> On 10/07/2015 10:23, Alvise Rigo wrote:
>>> In order to perfom "lazy" TLB invalidation requests, introduce a
>>> queue of callbacks at every vCPU disposal that will be fired just
>>> before entering the next TB.
>>>
>>> Suggested-by: Jani Kokkonen <address@hidden>
>>> Suggested-by: Claudio Fontana <address@hidden>
>>> Signed-off-by: Alvise Rigo <address@hidden>
>>
>> Why is async_run_on_cpu not enough?
>>
>> Paolo
>>
>>> ---
>>> cpus.c | 34 ++++++++++++++++++++++++++++++++++
>>> exec.c | 1 +
>>> include/qom/cpu.h | 20 ++++++++++++++++++++
>>> 3 files changed, 55 insertions(+)
>>>
>>> diff --git a/cpus.c b/cpus.c
>>> index f4d938e..b9f0329 100644
>>> --- a/cpus.c
>>> +++ b/cpus.c
>>> @@ -1421,6 +1421,7 @@ static int tcg_cpu_exec(CPUArchState *env)
>>> cpu->icount_extra = count;
>>> }
>>> qemu_mutex_unlock_iothread();
>>> + cpu_exit_callbacks_call_all(cpu);
>>> ret = cpu_exec(env);
>>> cpu->tcg_executing = 0;
>>>
>>> @@ -1469,6 +1470,39 @@ static void tcg_exec_all(CPUState *cpu)
>>> cpu->exit_request = 0;
>>> }
>>>
>>> +void cpu_exit_callback_add(CPUState *cpu, CPUExitCallback callback,
>>> + void *opaque)
>>> +{
>>> + CPUExitCB *cb;
>>> +
>>> + cb = g_malloc(sizeof(*cb));
>>> + cb->callback = callback;
>>> + cb->opaque = opaque;
>>> +
>>> + qemu_mutex_lock(&cpu->exit_cbs.mutex);
>>> + QTAILQ_INSERT_TAIL(&cpu->exit_cbs.exit_callbacks, cb, entry);
>>> + qemu_mutex_unlock(&cpu->exit_cbs.mutex);
>>> +}
>>> +
>>> +void cpu_exit_callbacks_call_all(CPUState *cpu)
>>> +{
>>> + CPUExitCB *cb, *next;
>>> +
>>> + if (QTAILQ_EMPTY(&cpu->exit_cbs.exit_callbacks)) {
>>> + return;
>>> + }
>>> +
>>> + QTAILQ_FOREACH_SAFE(cb, &cpu->exit_cbs.exit_callbacks, entry, next) {
>>> + cb->callback(cpu, cb->opaque);
>>> +
>>> + /* one-shot callbacks, remove it after using it */
>>> + qemu_mutex_lock(&cpu->exit_cbs.mutex);
>>> + QTAILQ_REMOVE(&cpu->exit_cbs.exit_callbacks, cb, entry);
>>> + g_free(cb);
>>> + qemu_mutex_unlock(&cpu->exit_cbs.mutex);
>>> + }
>>> +}
>>> +
>>> void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
>>> {
>>> /* XXX: implement xxx_cpu_list for targets that still miss it */
>>> diff --git a/exec.c b/exec.c
>>> index 51958ed..322f2c6 100644
>>> --- a/exec.c
>>> +++ b/exec.c
>>> @@ -531,6 +531,7 @@ void cpu_exec_init(CPUArchState *env)
>>> cpu->numa_node = 0;
>>> QTAILQ_INIT(&cpu->breakpoints);
>>> QTAILQ_INIT(&cpu->watchpoints);
>>> + QTAILQ_INIT(&cpu->exit_cbs.exit_callbacks);
>>> #ifndef CONFIG_USER_ONLY
>>> cpu->as = &address_space_memory;
>>> cpu->thread_id = qemu_get_thread_id();
>>> diff --git a/include/qom/cpu.h b/include/qom/cpu.h
>>> index 8d121b3..0ec020b 100644
>>> --- a/include/qom/cpu.h
>>> +++ b/include/qom/cpu.h
>>> @@ -201,6 +201,24 @@ typedef struct CPUWatchpoint {
>>> QTAILQ_ENTRY(CPUWatchpoint) entry;
>>> } CPUWatchpoint;
>>>
>>> +/* vCPU exit callbacks */
>>> +typedef void (*CPUExitCallback)(CPUState *cpu, void *opaque);
>>> +struct CPUExitCBs {
>>> + QemuMutex mutex;
>>> + QTAILQ_HEAD(exit_callbacks_head, CPUExitCB) exit_callbacks;
>>> +};
>>> +
>>> +typedef struct CPUExitCB {
>>> + CPUExitCallback callback;
>>> + void *opaque;
>>> +
>>> + QTAILQ_ENTRY(CPUExitCB) entry;
>>> +} CPUExitCB;
>>> +
>>> +void cpu_exit_callback_add(CPUState *cpu, CPUExitCallback callback,
>>> + void *opaque);
>>> +void cpu_exit_callbacks_call_all(CPUState *cpu);
>>> +
>>> /* Rendezvous support */
>>> #define TCG_RDV_POLLING_PERIOD 10
>>> typedef struct CpuExitRendezvous {
>>> @@ -305,6 +323,8 @@ struct CPUState {
>>>
>>> void *opaque;
>>>
>>> + /* One-shot callbacks for stopping requests. */
>>> + struct CPUExitCBs exit_cbs;
>>> volatile int pending_rdv;
>>>
>>> /* In order to avoid passing too many arguments to the MMIO helpers,
>>>
>
>
- Re: [Qemu-devel] [RFC v3 08/13] exec.c: introduce a simple rendezvous support, (continued)
[Qemu-devel] [RFC v3 11/13] Introduce exit_flush_req and tcg_excl_access_lock, Alvise Rigo, 2015/07/10
[Qemu-devel] [RFC v3 10/13] Simple TLB flush wrap to use as exit callback, Alvise Rigo, 2015/07/10
[Qemu-devel] [RFC v3 13/13] softmmu_template.h: move to multithreading, Alvise Rigo, 2015/07/10
[Qemu-devel] [RFC v3 12/13] softmmu_llsc_template.h: move to multithreading, Alvise Rigo, 2015/07/10
Re: [Qemu-devel] [RFC v3 00/13] Slow-path for atomic instruction translation, Mark Burton, 2015/07/10