Message ID | 20221110164807.1306076-2-eesposit@redhat.com |
---|---|
State | New |
Headers | show |
Series | KVM: allow listener to stop all vcpus before | expand |
On 11/10/22 17:48, Emanuele Giuseppe Esposito wrote: > +/* > + * QEMU accel blocker class "Lock to inhibit accelerator ioctls" > + * > + * Copyright (c) 2014 Red Hat Inc. 2022, you can also add an Author line. > +static int accel_in_ioctls(void) Return bool (and return early if ret becomes true). > +void accel_ioctl_inhibit_begin(void) > +{ > + CPUState *cpu; > + > + /* > + * We allow to inhibit only when holding the BQL, so we can identify > + * when an inhibitor wants to issue an ioctl easily. > + */ > + g_assert(qemu_mutex_iothread_locked()); > + > + /* Block further invocations of the ioctls outside the BQL. */ > + CPU_FOREACH(cpu) { > + qemu_lockcnt_lock(&cpu->in_ioctl_lock); > + } > + qemu_lockcnt_lock(&accel_in_ioctl_lock); > + > + /* Keep waiting until there are running ioctls */ > + while (accel_in_ioctls()) { > + /* Reset event to FREE. */ > + qemu_event_reset(&accel_in_ioctl_event); > + > + if (accel_in_ioctls()) { > + > + CPU_FOREACH(cpu) { > + /* exit the ioctl */ > + qemu_cpu_kick(cpu); Only kick if the lockcnt count is > 0? (this is not racy; if it is == 0, it cannot ever become > 0 again while the lock is taken) > diff --git a/include/sysemu/accel-blocker.h b/include/sysemu/accel-blocker.h > new file mode 100644 > index 0000000000..135ebea566 > --- /dev/null > +++ b/include/sysemu/accel-blocker.h > @@ -0,0 +1,45 @@ > +/* > + * Accelerator blocking API, to prevent new ioctls from starting and wait the > + * running ones finish. > + * This mechanism differs from pause/resume_all_vcpus() in that it does not > + * release the BQL. > + * > + * Copyright (c) 2014 Red Hat Inc. 2022, you can also add an Author line here too. > + * This work is licensed under the terms of the GNU GPL, version 2 or later. > + * See the COPYING file in the top-level directory. > + */ > +#ifndef ACCEL_BLOCKER_H > +#define ACCEL_BLOCKER_H > + > +#include "qemu/osdep.h" > +#include "qemu/accel.h" qemu/accel.h not needed? > +#include "sysemu/cpus.h" > + > +extern void accel_blocker_init(void); > + > +/* > + * accel_set_in_ioctl/accel_cpu_set_in_ioctl: > + * Mark when ioctl is about to run or just finished. > + * If @in_ioctl is true, then mark it is beginning. Otherwise marks that it is > + * ending. > + * > + * These functions will block after accel_ioctl_inhibit_begin() is called, > + * preventing new ioctls to run. They will continue only after > + * accel_ioctl_inibith_end(). > + */ > +extern void accel_set_in_ioctl(bool in_ioctl); > +extern void accel_cpu_set_in_ioctl(CPUState *cpu, bool in_ioctl); Why not just extern void accel_ioctl_begin(void); extern void accel_ioctl_end(void); extern void accel_cpu_ioctl_begin(CPUState *cpu); extern void accel_cpu_ioctl_end(CPUState *cpu); ? Otherwise it's very nice. Paolo
Am 11/11/2022 um 11:48 schrieb Paolo Bonzini: > On 11/10/22 17:48, Emanuele Giuseppe Esposito wrote: >> +/* >> + * QEMU accel blocker class > > "Lock to inhibit accelerator ioctls" > >> + * >> + * Copyright (c) 2014 Red Hat Inc. > > 2022, you can also add an Author line. > >> +static int accel_in_ioctls(void) > > Return bool (and return early if ret becomes true). > >> +void accel_ioctl_inhibit_begin(void) >> +{ >> + CPUState *cpu; >> + >> + /* >> + * We allow to inhibit only when holding the BQL, so we can identify >> + * when an inhibitor wants to issue an ioctl easily. >> + */ >> + g_assert(qemu_mutex_iothread_locked()); >> + >> + /* Block further invocations of the ioctls outside the BQL. */ >> + CPU_FOREACH(cpu) { >> + qemu_lockcnt_lock(&cpu->in_ioctl_lock); >> + } >> + qemu_lockcnt_lock(&accel_in_ioctl_lock); >> + >> + /* Keep waiting until there are running ioctls */ >> + while (accel_in_ioctls()) { >> + /* Reset event to FREE. */ >> + qemu_event_reset(&accel_in_ioctl_event); >> + >> + if (accel_in_ioctls()) { >> + >> + CPU_FOREACH(cpu) { >> + /* exit the ioctl */ >> + qemu_cpu_kick(cpu); > > Only kick if the lockcnt count is > 0? (this is not racy; if it is == 0, > it cannot ever become > 0 again while the lock is taken) Better: accel_has_to_wait(void) { CPUState *cpu; bool needs_to_wait = false; CPU_FOREACH(cpu) { if (qemu_lockcnt_count(&cpu->in_ioctl_lock)) { qemu_cpu_kick(cpu); needs_to_wait = true; } } return needs_to_wait || qemu_lockcnt_count(&accel_in_ioctl_lock); } And then the loop becomes: while (true) { qemu_event_reset(&accel_in_ioctl_event); if (accel_has_to_wait()) { qemu_event_wait(&accel_in_ioctl_event); } else { /* No ioctl is running */ return; } } > >> diff --git a/include/sysemu/accel-blocker.h >> b/include/sysemu/accel-blocker.h >> new file mode 100644 >> index 0000000000..135ebea566 >> --- /dev/null >> +++ b/include/sysemu/accel-blocker.h >> @@ -0,0 +1,45 @@ >> +/* >> + * Accelerator blocking API, to prevent new ioctls from starting and >> wait the >> + * running ones finish. >> + * This mechanism differs from pause/resume_all_vcpus() in that it >> does not >> + * release the BQL. >> + * >> + * Copyright (c) 2014 Red Hat Inc. > > 2022, you can also add an Author line here too. > >> + * This work is licensed under the terms of the GNU GPL, version 2 or >> later. >> + * See the COPYING file in the top-level directory. >> + */ >> +#ifndef ACCEL_BLOCKER_H >> +#define ACCEL_BLOCKER_H >> + >> +#include "qemu/osdep.h" >> +#include "qemu/accel.h" > > qemu/accel.h not needed? > >> +#include "sysemu/cpus.h" >> + >> +extern void accel_blocker_init(void); >> + >> +/* >> + * accel_set_in_ioctl/accel_cpu_set_in_ioctl: >> + * Mark when ioctl is about to run or just finished. >> + * If @in_ioctl is true, then mark it is beginning. Otherwise marks >> that it is >> + * ending. >> + * >> + * These functions will block after accel_ioctl_inhibit_begin() is >> called, >> + * preventing new ioctls to run. They will continue only after >> + * accel_ioctl_inibith_end(). >> + */ >> +extern void accel_set_in_ioctl(bool in_ioctl); >> +extern void accel_cpu_set_in_ioctl(CPUState *cpu, bool in_ioctl); > > Why not just > > extern void accel_ioctl_begin(void); > extern void accel_ioctl_end(void); > extern void accel_cpu_ioctl_begin(CPUState *cpu); > extern void accel_cpu_ioctl_end(CPUState *cpu); > > ? Ok, makes sense. Thank you, Emanuele > > Otherwise it's very nice. > > Paolo >
diff --git a/accel/accel-blocker.c b/accel/accel-blocker.c new file mode 100644 index 0000000000..2701a05945 --- /dev/null +++ b/accel/accel-blocker.c @@ -0,0 +1,139 @@ +/* + * QEMU accel blocker class + * + * Copyright (c) 2014 Red Hat Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + */ + +#include "qemu/osdep.h" +#include "qemu/thread.h" +#include "qemu/main-loop.h" +#include "hw/core/cpu.h" +#include "sysemu/accel-blocker.h" + +static QemuLockCnt accel_in_ioctl_lock; +static QemuEvent accel_in_ioctl_event; + +void accel_blocker_init(void) +{ + qemu_lockcnt_init(&accel_in_ioctl_lock); + qemu_event_init(&accel_in_ioctl_event, false); +} + +void accel_set_in_ioctl(bool in_ioctl) +{ + if (likely(qemu_mutex_iothread_locked())) { + return; + } + if (in_ioctl) { + /* block if lock is taken in kvm_ioctl_inhibit_begin() */ + qemu_lockcnt_inc(&accel_in_ioctl_lock); + } else { + qemu_lockcnt_dec(&accel_in_ioctl_lock); + /* change event to SET. If event was BUSY, wake up all waiters */ + qemu_event_set(&accel_in_ioctl_event); + } +} + +void accel_cpu_set_in_ioctl(CPUState *cpu, bool in_ioctl) +{ + if (unlikely(qemu_mutex_iothread_locked())) { + return; + } + if (in_ioctl) { + /* block if lock is taken in kvm_ioctl_inhibit_begin() */ + qemu_lockcnt_inc(&cpu->in_ioctl_lock); + } else { + qemu_lockcnt_dec(&cpu->in_ioctl_lock); + /* change event to SET. If event was BUSY, wake up all waiters */ + qemu_event_set(&accel_in_ioctl_event); + } +} + +static int accel_in_ioctls(void) +{ + CPUState *cpu; + int ret = qemu_lockcnt_count(&accel_in_ioctl_lock); + + CPU_FOREACH(cpu) { + ret += qemu_lockcnt_count(&cpu->in_ioctl_lock); + } + + return ret; +} + +void accel_ioctl_inhibit_begin(void) +{ + CPUState *cpu; + + /* + * We allow to inhibit only when holding the BQL, so we can identify + * when an inhibitor wants to issue an ioctl easily. + */ + g_assert(qemu_mutex_iothread_locked()); + + /* Block further invocations of the ioctls outside the BQL. */ + CPU_FOREACH(cpu) { + qemu_lockcnt_lock(&cpu->in_ioctl_lock); + } + qemu_lockcnt_lock(&accel_in_ioctl_lock); + + /* Keep waiting until there are running ioctls */ + while (accel_in_ioctls()) { + /* Reset event to FREE. */ + qemu_event_reset(&accel_in_ioctl_event); + + if (accel_in_ioctls()) { + + CPU_FOREACH(cpu) { + /* exit the ioctl */ + qemu_cpu_kick(cpu); + } + + /* + * If event is still FREE, and there are ioctls still in progress, + * wait. + * + * If an ioctl finishes before qemu_event_wait(), it will change + * the event state to SET. This will prevent qemu_event_wait() from + * blocking, but it's not a problem because if other ioctls are + * still running (accel_in_ioctls is true) the loop will iterate + * once more and reset the event status to FREE so that it can wait + * properly. + * + * If an ioctls finishes while qemu_event_wait() is blocking, then + * it will be waken up, but also here the while loop makes sure + * to re-enter the wait if there are other running ioctls. + */ + qemu_event_wait(&accel_in_ioctl_event); + } + } +} + +void accel_ioctl_inhibit_end(void) +{ + CPUState *cpu; + + qemu_lockcnt_unlock(&accel_in_ioctl_lock); + CPU_FOREACH(cpu) { + qemu_lockcnt_unlock(&cpu->in_ioctl_lock); + } +} + diff --git a/accel/meson.build b/accel/meson.build index b9a963cf80..a0d49c4f31 100644 --- a/accel/meson.build +++ b/accel/meson.build @@ -1,4 +1,4 @@ -specific_ss.add(files('accel-common.c')) +specific_ss.add(files('accel-common.c', 'accel-blocker.c')) softmmu_ss.add(files('accel-softmmu.c')) user_ss.add(files('accel-user.c')) diff --git a/include/sysemu/accel-blocker.h b/include/sysemu/accel-blocker.h new file mode 100644 index 0000000000..135ebea566 --- /dev/null +++ b/include/sysemu/accel-blocker.h @@ -0,0 +1,45 @@ +/* + * Accelerator blocking API, to prevent new ioctls from starting and wait the + * running ones finish. + * This mechanism differs from pause/resume_all_vcpus() in that it does not + * release the BQL. + * + * Copyright (c) 2014 Red Hat Inc. + * + * This work is licensed under the terms of the GNU GPL, version 2 or later. + * See the COPYING file in the top-level directory. + */ +#ifndef ACCEL_BLOCKER_H +#define ACCEL_BLOCKER_H + +#include "qemu/osdep.h" +#include "qemu/accel.h" +#include "sysemu/cpus.h" + +extern void accel_blocker_init(void); + +/* + * accel_set_in_ioctl/accel_cpu_set_in_ioctl: + * Mark when ioctl is about to run or just finished. + * If @in_ioctl is true, then mark it is beginning. Otherwise marks that it is + * ending. + * + * These functions will block after accel_ioctl_inhibit_begin() is called, + * preventing new ioctls to run. They will continue only after + * accel_ioctl_inibith_end(). + */ +extern void accel_set_in_ioctl(bool in_ioctl); +extern void accel_cpu_set_in_ioctl(CPUState *cpu, bool in_ioctl); + +/* + * accel_ioctl_inhibit_begin/end: start/end critical section + * Between these two calls, no ioctl marked with accel_set_in_ioctl() and + * accel_cpu_set_in_ioctl() is allowed to run. + * + * This allows the caller to access shared data or perform operations without + * worrying of concurrent vcpus accesses. + */ +extern void accel_ioctl_inhibit_begin(void); +extern void accel_ioctl_inhibit_end(void); + +#endif /* ACCEL_BLOCKER_H */
This API allows the accelerators to prevent vcpus from issuing new ioctls while execting a critical section marked with the accel-ioctl_inhibit_begin/end functions. Note that all functions submitting ioctls must mark where the ioctl is being called with accel_{cpu_}set_in_ioctl(). This API requires the caller to always hold the BQL. API documentation is in sysemu/accel-blocker.h Internally, it uses a QemuLockCnt together with a per-CPU QemuLockCnt (to minimize cache line bouncing) to keep avoid that new ioctls run when the critical section starts, and a QemuEvent to wait that all running ioctls finish. Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com> --- accel/accel-blocker.c | 139 +++++++++++++++++++++++++++++++++ accel/meson.build | 2 +- include/sysemu/accel-blocker.h | 45 +++++++++++ 3 files changed, 185 insertions(+), 1 deletion(-) create mode 100644 accel/accel-blocker.c create mode 100644 include/sysemu/accel-blocker.h