From: Igor Mammedov <imammedo@redhat.com>
To: Paolo Bonzini <pbonzini@redhat.com>
Cc: qemu-devel@nongnu.org, peterx@redhat.com, richard.henderson@linaro.org
Subject: Re: [PATCH 09/18] accel: use store_release/load_acquire for cross-thread exit_request
Date: Mon, 1 Sep 2025 15:13:43 +0200 [thread overview]
Message-ID: <20250901151343.76fa22f6@fedora> (raw)
In-Reply-To: <20250829153115.1590048-5-pbonzini@redhat.com>
On Fri, 29 Aug 2025 17:31:06 +0200
Paolo Bonzini <pbonzini@redhat.com> wrote:
> Reads and writes cpu->exit_request do not use a load-acquire/store-release
> pair right now, but this means that cpu_exit() may not write cpu->exit_request
> after any flags that are read by the vCPU thread.
>
> Probably everything is protected one way or the other by the BQL, because
> cpu->exit_request leads to the slow path, where the CPU thread often takes
> the BQL (for example, to go to sleep by waiting on the BQL-protected
> cpu->halt_cond); but it's not clear, so use load-acquire/store-release
> consistently.
>
> Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
> Reviewed-by: Peter Xu <peterx@redhat.com>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
nit below:
> ---
> accel/kvm/kvm-all.c | 19 +++++++++----------
> accel/tcg/cpu-exec.c | 7 +++++--
> accel/tcg/tcg-accel-ops-rr.c | 2 +-
> hw/core/cpu-common.c | 3 ++-
> target/i386/nvmm/nvmm-all.c | 5 ++---
> target/i386/whpx/whpx-all.c | 3 ++-
> 6 files changed, 21 insertions(+), 18 deletions(-)
>
> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
> index f36dfe33492..bd9e5e3886d 100644
> --- a/accel/kvm/kvm-all.c
> +++ b/accel/kvm/kvm-all.c
> @@ -3029,10 +3029,6 @@ static void kvm_eat_signals(CPUState *cpu)
>
> if (kvm_immediate_exit) {
> qatomic_set(&cpu->kvm_run->immediate_exit, 0);
> - /* Write kvm_run->immediate_exit before the cpu->exit_request
> - * write in kvm_cpu_exec.
> - */
> - smp_wmb();
> return;
> }
>
> @@ -3187,7 +3183,8 @@ int kvm_cpu_exec(CPUState *cpu)
> }
>
> kvm_arch_pre_run(cpu, run);
> - if (qatomic_read(&cpu->exit_request)) {
> + /* Corresponding store-release is in cpu_exit. */
> + if (qatomic_load_acquire(&cpu->exit_request)) {
> trace_kvm_interrupt_exit_request();
> /*
> * KVM requires us to reenter the kernel after IO exits to complete
> @@ -3197,13 +3194,15 @@ int kvm_cpu_exec(CPUState *cpu)
> kvm_cpu_kick_self();
> }
>
> - /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit.
> - * Matching barrier in kvm_eat_signals.
> - */
> - smp_rmb();
> -
> run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
>
> + /*
> + * After writing cpu->exit_request, cpu_exit() sends a signal that writes
> + * kvm->run->immediate_exit. The signal is already happening after the
> + * write to cpu->exit_request so, if KVM read kvm->run->immediate_exit
> + * as true, cpu->exit_request will always read as true.
> + */
> +
> attrs = kvm_arch_post_run(cpu, run);
>
> #ifdef KVM_HAVE_MCE_INJECTION
> diff --git a/accel/tcg/cpu-exec.c b/accel/tcg/cpu-exec.c
> index 508d2d2d9e2..f838535d111 100644
> --- a/accel/tcg/cpu-exec.c
> +++ b/accel/tcg/cpu-exec.c
> @@ -851,8 +851,11 @@ static inline bool cpu_handle_interrupt(CPUState *cpu,
> }
> #endif /* !CONFIG_USER_ONLY */
>
> - /* Finally, check if we need to exit to the main loop. */
> - if (unlikely(qatomic_read(&cpu->exit_request)) || icount_exit_request(cpu)) {
> + /*
> + * Finally, check if we need to exit to the main loop.
> + * The corresponding store-release is in cpu_exit.
> + */
> + if (unlikely(qatomic_load_acquire(&cpu->exit_request)) || icount_exit_request(cpu)) {
> qatomic_set(&cpu->exit_request, 0);
> if (cpu->exception_index == -1) {
> cpu->exception_index = EXCP_INTERRUPT;
> diff --git a/accel/tcg/tcg-accel-ops-rr.c b/accel/tcg/tcg-accel-ops-rr.c
> index 6eec5c9eee9..1e551e92d6d 100644
> --- a/accel/tcg/tcg-accel-ops-rr.c
> +++ b/accel/tcg/tcg-accel-ops-rr.c
> @@ -242,7 +242,7 @@ static void *rr_cpu_thread_fn(void *arg)
> cpu = first_cpu;
> }
>
should we have here a similar comment as in the previous hunk?
> - while (cpu && cpu_work_list_empty(cpu) && !cpu->exit_request) {
> + while (cpu && cpu_work_list_empty(cpu) && !qatomic_load_acquire(&cpu->exit_request)) {
> /* Store rr_current_cpu before evaluating cpu_can_run(). */
> qatomic_set_mb(&rr_current_cpu, cpu);
>
> diff --git a/hw/core/cpu-common.c b/hw/core/cpu-common.c
> index 9ea1f3764a8..ca00accd162 100644
> --- a/hw/core/cpu-common.c
> +++ b/hw/core/cpu-common.c
> @@ -74,7 +74,8 @@ void cpu_reset_interrupt(CPUState *cpu, int mask)
>
> void cpu_exit(CPUState *cpu)
> {
> - qatomic_set(&cpu->exit_request, 1);
> + /* Ensure cpu_exec will see the reason why the exit request was set. */
> + qatomic_store_release(&cpu->exit_request, true);
> /* Ensure cpu_exec will see the exit request after TCG has exited. */
> smp_wmb();
> qatomic_set(&cpu->neg.icount_decr.u16.high, -1);
> diff --git a/target/i386/nvmm/nvmm-all.c b/target/i386/nvmm/nvmm-all.c
> index e1151b04c6e..10bd51d9b59 100644
> --- a/target/i386/nvmm/nvmm-all.c
> +++ b/target/i386/nvmm/nvmm-all.c
> @@ -743,7 +743,8 @@ nvmm_vcpu_loop(CPUState *cpu)
>
> nvmm_vcpu_pre_run(cpu);
>
> - if (qatomic_read(&cpu->exit_request)) {
> + /* Corresponding store-release is in cpu_exit. */
> + if (qatomic_load_acquire(&cpu->exit_request)) {
> #if NVMM_USER_VERSION >= 2
> nvmm_vcpu_stop(vcpu);
> #else
> @@ -751,8 +752,6 @@ nvmm_vcpu_loop(CPUState *cpu)
> #endif
> }
>
> - /* Read exit_request before the kernel reads the immediate exit flag */
> - smp_rmb();
> ret = nvmm_vcpu_run(mach, vcpu);
> if (ret == -1) {
> error_report("NVMM: Failed to exec a virtual processor,"
> diff --git a/target/i386/whpx/whpx-all.c b/target/i386/whpx/whpx-all.c
> index c09a0a64f22..2106c29c3a0 100644
> --- a/target/i386/whpx/whpx-all.c
> +++ b/target/i386/whpx/whpx-all.c
> @@ -1714,7 +1714,8 @@ static int whpx_vcpu_run(CPUState *cpu)
> if (exclusive_step_mode == WHPX_STEP_NONE) {
> whpx_vcpu_pre_run(cpu);
>
> - if (qatomic_read(&cpu->exit_request)) {
> + /* Corresponding store-release is in cpu_exit. */
> + if (qatomic_load_acquire(&cpu->exit_request)) {
> whpx_vcpu_kick(cpu);
> }
> }
next prev parent reply other threads:[~2025-09-01 13:14 UTC|newest]
Thread overview: 58+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-08-29 15:28 [PATCH v2 00/18] accel, cpus: clean up cpu->exit_request Paolo Bonzini
2025-08-29 15:28 ` [PATCH 01/18] target/ppc: limit cpu_interrupt_exittb to system emulation Paolo Bonzini
2025-08-29 21:32 ` Richard Henderson
2025-09-01 13:03 ` Igor Mammedov
2025-08-29 15:28 ` [PATCH 02/18] target/sparc: limit cpu_check_irqs " Paolo Bonzini
2025-08-29 21:32 ` Richard Henderson
2025-09-01 11:20 ` Philippe Mathieu-Daudé
2025-08-29 15:28 ` [PATCH 03/18] target/i386: limit a20 " Paolo Bonzini
2025-08-29 21:32 ` Richard Henderson
2025-09-01 7:44 ` Philippe Mathieu-Daudé
2025-09-01 13:05 ` Igor Mammedov
2025-08-29 15:28 ` [PATCH 04/18] target-arm: remove uses of cpu_interrupt() for user-mode emulation Paolo Bonzini
2025-08-29 22:18 ` Richard Henderson
2025-08-29 15:31 ` [PATCH 05/18] user-exec: remove cpu_interrupt() stub Paolo Bonzini
2025-08-29 21:33 ` Richard Henderson
2025-09-01 11:22 ` Philippe Mathieu-Daudé
2025-08-29 15:31 ` [PATCH 06/18] treewide: clear bits of cs->interrupt_request with cpu_reset_interrupt() Paolo Bonzini
2025-08-29 21:34 ` Richard Henderson
2025-09-01 11:23 ` Philippe Mathieu-Daudé
2025-09-01 11:35 ` Philippe Mathieu-Daudé
2025-09-01 12:59 ` Igor Mammedov
2025-08-29 15:31 ` [PATCH 07/18] cpu-common: use atomic access for interrupt_request Paolo Bonzini
2025-08-29 21:38 ` Richard Henderson
2025-09-01 8:00 ` Philippe Mathieu-Daudé
2025-09-01 13:01 ` Igor Mammedov
2025-08-29 15:31 ` [PATCH 08/18] cpus: document that qemu_cpu_kick() can be used for BQL-less operation Paolo Bonzini
2025-08-29 21:39 ` Richard Henderson
2025-08-29 15:31 ` [PATCH 09/18] accel: use store_release/load_acquire for cross-thread exit_request Paolo Bonzini
2025-09-01 13:13 ` Igor Mammedov [this message]
2025-08-29 15:31 ` [PATCH 10/18] accel: use atomic accesses for exit_request Paolo Bonzini
2025-09-01 13:15 ` Igor Mammedov
2025-08-29 15:31 ` [PATCH 11/18] accel/tcg: create a thread-kick function for TCG Paolo Bonzini
2025-08-29 21:41 ` Richard Henderson
2025-09-01 8:03 ` Philippe Mathieu-Daudé
2025-09-01 13:19 ` Igor Mammedov
2025-08-29 15:31 ` [PATCH 12/18] accel/tcg: inline cpu_exit() Paolo Bonzini
2025-08-29 21:41 ` Richard Henderson
2025-09-01 13:21 ` Igor Mammedov
2025-08-29 15:31 ` [PATCH 13/18] cpus: remove TCG-ism from cpu_exit() Paolo Bonzini
2025-09-01 13:22 ` Igor Mammedov
2025-08-29 15:31 ` [PATCH 14/18] cpus: properly kick CPUs out of inner execution loop Paolo Bonzini
2025-08-29 21:56 ` Richard Henderson
2025-08-29 23:05 ` Paolo Bonzini
2025-08-30 3:04 ` Richard Henderson
2025-08-29 15:31 ` [PATCH 15/18] bsd-user, linux-user: introduce qemu_wait_io_event Paolo Bonzini
2025-08-29 22:11 ` Richard Henderson
2025-09-01 11:32 ` Philippe Mathieu-Daudé
2025-08-29 15:31 ` [PATCH 16/18] cpus: clear exit_request in qemu_wait_io_event Paolo Bonzini
2025-08-29 22:11 ` Richard Henderson
2025-08-29 22:19 ` Paolo Bonzini
2025-09-01 13:32 ` Igor Mammedov
2025-08-29 15:31 ` [PATCH 17/18] accel: make all calls to qemu_wait_io_event look the same Paolo Bonzini
2025-08-29 22:16 ` Richard Henderson
2025-09-01 11:29 ` Philippe Mathieu-Daudé
2025-09-01 11:31 ` Philippe Mathieu-Daudé
2025-09-01 13:32 ` Igor Mammedov
2025-08-29 15:31 ` [PATCH 18/18] tcg/user: do not set exit_request gratuitously Paolo Bonzini
2025-09-01 13:33 ` Igor Mammedov
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20250901151343.76fa22f6@fedora \
--to=imammedo@redhat.com \
--cc=pbonzini@redhat.com \
--cc=peterx@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).