Re: [External] Re: [PATCH v3 13/20] migration/multifd: Prepare to introduce DSA acceleration on the multifd path.

[Hao Xiang]

On Sun, Jan 14, 2024 at 10:46 PM Shivam Kumar <shivam.kumar1@nutanix.com> wrote:
>
>
>
> > On 04-Jan-2024, at 6:14 AM, Hao Xiang <hao.xiang@bytedance.com> wrote:
> >
> > 1. Refactor multifd_send_thread function.
> > 2. Implement buffer_is_zero_use_cpu to handle CPU based zero page
> > checking.
> > 3. Introduce the batch task structure in MultiFDSendParams.
> >
> > Signed-off-by: Hao Xiang <hao.xiang@bytedance.com>
> > ---
> > include/qemu/dsa.h  | 43 +++++++++++++++++++++++--
> > migration/multifd.c | 77 ++++++++++++++++++++++++++++++++++++---------
> > migration/multifd.h |  2 ++
> > util/dsa.c          | 51 +++++++++++++++++++++++++-----
> > 4 files changed, 148 insertions(+), 25 deletions(-)
> >
> > diff --git a/include/qemu/dsa.h b/include/qemu/dsa.h
> > index e002652879..fe7772107a 100644
> > --- a/include/qemu/dsa.h
> > +++ b/include/qemu/dsa.h
> > @@ -2,6 +2,7 @@
> > #define QEMU_DSA_H
> >
> > #include "qemu/error-report.h"
> > +#include "exec/cpu-common.h"
> > #include "qemu/thread.h"
> > #include "qemu/queue.h"
> >
> > @@ -42,6 +43,20 @@ typedef struct dsa_batch_task {
> >     QSIMPLEQ_ENTRY(dsa_batch_task) entry;
> > } dsa_batch_task;
> >
> > +#endif
> > +
> > +struct batch_task {
> > +    /* Address of each pages in pages */
> > +    ram_addr_t *addr;
> > +    /* Zero page checking results */
> > +    bool *results;
> > +#ifdef CONFIG_DSA_OPT
> > +    struct dsa_batch_task *dsa_batch;
> > +#endif
> > +};
> > +
> > +#ifdef CONFIG_DSA_OPT
> > +
> > /**
> >  * @brief Initializes DSA devices.
> >  *
> > @@ -74,7 +89,7 @@ void dsa_cleanup(void);
> > bool dsa_is_running(void);
> >
> > /**
> > - * @brief Initializes a buffer zero batch task.
> > + * @brief Initializes a buffer zero DSA batch task.
> >  *
> >  * @param task A pointer to the batch task to initialize.
> >  * @param results A pointer to an array of zero page checking results.
> > @@ -102,7 +117,7 @@ void buffer_zero_batch_task_destroy(struct 
> > dsa_batch_task *task);
> >  * @return Zero if successful, otherwise non-zero.
> >  */
> > int
> > -buffer_is_zero_dsa_batch_async(struct dsa_batch_task *batch_task,
> > +buffer_is_zero_dsa_batch_async(struct batch_task *batch_task,
> >                                const void **buf, size_t count, size_t len);
> >
> > #else
> > @@ -128,6 +143,30 @@ static inline void dsa_stop(void) {}
> >
> > static inline void dsa_cleanup(void) {}
> >
> > +static inline int
> > +buffer_is_zero_dsa_batch_async(struct batch_task *batch_task,
> > +                               const void **buf, size_t count, size_t len)
> > +{
> > +    exit(1);
> > +}
> > +
> > #endif
> >
> > +/**
> > + * @brief Initializes a general buffer zero batch task.
> > + *
> > + * @param task A pointer to the general batch task to initialize.
> > + * @param batch_size The number of zero page checking tasks in the batch.
> > + */
> > +void
> > +batch_task_init(struct batch_task *task, int batch_size);
> > +
> > +/**
> > + * @brief Destroys a general buffer zero batch task.
> > + *
> > + * @param task A pointer to the general batch task to destroy.
> > + */
> > +void
> > +batch_task_destroy(struct batch_task *task);
> > +
> > #endif
> > diff --git a/migration/multifd.c b/migration/multifd.c
> > index eece85569f..e7c549b93e 100644
> > --- a/migration/multifd.c
> > +++ b/migration/multifd.c
> > @@ -14,6 +14,8 @@
> > #include "qemu/cutils.h"
> > #include "qemu/rcu.h"
> > #include "qemu/cutils.h"
> > +#include "qemu/dsa.h"
> > +#include "qemu/memalign.h"
> > #include "exec/target_page.h"
> > #include "sysemu/sysemu.h"
> > #include "exec/ramblock.h"
> > @@ -574,6 +576,8 @@ void multifd_save_cleanup(void)
> >         p->name = NULL;
> >         multifd_pages_clear(p->pages);
> >         p->pages = NULL;
> > +        batch_task_destroy(p->batch_task);
> > +        p->batch_task = NULL;
> >         p->packet_len = 0;
> >         g_free(p->packet);
> >         p->packet = NULL;
> > @@ -678,13 +682,66 @@ int multifd_send_sync_main(QEMUFile *f)
> >     return 0;
> > }
> >
> > +static void set_page(MultiFDSendParams *p, bool zero_page, uint64_t offset)
> > +{
> > +    RAMBlock *rb = p->pages->block;
> > +    if (zero_page) {
> > +        p->zero[p->zero_num] = offset;
> > +        p->zero_num++;
> > +        ram_release_page(rb->idstr, offset);
> > +    } else {
> > +        p->normal[p->normal_num] = offset;
> > +        p->normal_num++;
> > +    }
> > +}
> > +
> > +static void buffer_is_zero_use_cpu(MultiFDSendParams *p)
> > +{
> > +    const void **buf = (const void **)p->batch_task->addr;
> > +    assert(!migrate_use_main_zero_page());
> > +
> > +    for (int i = 0; i < p->pages->num; i++) {
> > +        p->batch_task->results[i] = buffer_is_zero(buf[i], p->page_size);
> > +    }
> > +}
> > +
> > +static void set_normal_pages(MultiFDSendParams *p)
> > +{
> > +    for (int i = 0; i < p->pages->num; i++) {
> > +        p->batch_task->results[i] = false;
> > +    }
> > +}
> Please correct me if I am wrong but set_normal_pages will not be a part of 
> the final patch, right? They are there for testing out the performance 
> against different zero page ration scenarios. If it’s so, can we isolate 
> these parts into a separate patch.

set_normal_pages is used for performance testing and testing only. It
won't introduce any "incorrect" behavior and I would love to see it
being part of the upstream code. But the argument that testing change
should remain private is always correct. So I am totally OK with
isolating the parts into a separate patch.

> > +
> > +static void multifd_zero_page_check(MultiFDSendParams *p)
> > +{
> > +    /* older qemu don't understand zero page on multifd channel */
> > +    bool use_multifd_zero_page = !migrate_use_main_zero_page();
> > +
> > +    RAMBlock *rb = p->pages->block;
> > +
> > +    for (int i = 0; i < p->pages->num; i++) {
> > +        p->batch_task->addr[i] = (ram_addr_t)(rb->host + 
> > p->pages->offset[i]);
> > +    }
> > +
> > +    if (use_multifd_zero_page) {
> > +        buffer_is_zero_use_cpu(p);
> > +    } else {
> > +        /* No zero page checking. All pages are normal pages. */
> > +        set_normal_pages(p);
> > +    }
> > +
> > +    for (int i = 0; i < p->pages->num; i++) {
> > +        uint64_t offset = p->pages->offset[i];
> > +        bool zero_page = p->batch_task->results[i];
> > +        set_page(p, zero_page, offset);
> > +    }
> > +}
> > +
> > static void *multifd_send_thread(void *opaque)
> > {
> >     MultiFDSendParams *p = opaque;
> >     MigrationThread *thread = NULL;
> >     Error *local_err = NULL;
> > -    /* qemu older than 8.2 don't understand zero page on multifd channel */
> > -    bool use_multifd_zero_page = !migrate_use_main_zero_page();
> >     int ret = 0;
> >     bool use_zero_copy_send = migrate_zero_copy_send();
> >
> > @@ -710,7 +767,6 @@ static void *multifd_send_thread(void *opaque)
> >         qemu_mutex_lock(&p->mutex);
> >
> >         if (p->pending_job) {
> > -            RAMBlock *rb = p->pages->block;
> >             uint64_t packet_num = p->packet_num;
> >             uint32_t flags;
> >
> > @@ -723,18 +779,7 @@ static void *multifd_send_thread(void *opaque)
> >                 p->iovs_num = 1;
> >             }
> >
> > -            for (int i = 0; i < p->pages->num; i++) {
> > -                uint64_t offset = p->pages->offset[i];
> > -                if (use_multifd_zero_page &&
> > -                    buffer_is_zero(rb->host + offset, p->page_size)) {
> > -                    p->zero[p->zero_num] = offset;
> > -                    p->zero_num++;
> > -                    ram_release_page(rb->idstr, offset);
> > -                } else {
> > -                    p->normal[p->normal_num] = offset;
> > -                    p->normal_num++;
> > -                }
> > -            }
> > +            multifd_zero_page_check(p);
> >
> >             if (p->normal_num) {
> >                 ret = multifd_send_state->ops->send_prepare(p, &local_err);
> > @@ -975,6 +1020,8 @@ int multifd_save_setup(Error **errp)
> >         p->pending_job = 0;
> >         p->id = i;
> >         p->pages = multifd_pages_init(page_count);
> > +        p->batch_task = g_malloc0(sizeof(struct batch_task));
> > +        batch_task_init(p->batch_task, page_count);
> >         p->packet_len = sizeof(MultiFDPacket_t)
> >                       + sizeof(uint64_t) * page_count;
> >         p->packet = g_malloc0(p->packet_len);
> > diff --git a/migration/multifd.h b/migration/multifd.h
> > index 13762900d4..97b5f888a7 100644
> > --- a/migration/multifd.h
> > +++ b/migration/multifd.h
> > @@ -119,6 +119,8 @@ typedef struct {
> >      * pending_job != 0 -> multifd_channel can use it.
> >      */
> >     MultiFDPages_t *pages;
> > +    /* Zero page checking batch task */
> > +    struct batch_task *batch_task;
> >
> >     /* thread local variables. No locking required */
> >
> > diff --git a/util/dsa.c b/util/dsa.c
> > index 5a2bf33651..f6224a27d4 100644
> > --- a/util/dsa.c
> > +++ b/util/dsa.c
> > @@ -802,7 +802,7 @@ buffer_zero_task_init_int(struct dsa_hw_desc 
> > *descriptor,
> > }
> >
> > /**
> > - * @brief Initializes a buffer zero batch task.
> > + * @brief Initializes a buffer zero DSA batch task.
> >  *
> >  * @param task A pointer to the batch task to initialize.
> >  * @param results A pointer to an array of zero page checking results.
> > @@ -1107,29 +1107,64 @@ void dsa_cleanup(void)
> >  * @return Zero if successful, otherwise non-zero.
> >  */
> > int
> > -buffer_is_zero_dsa_batch_async(struct dsa_batch_task *batch_task,
> > +buffer_is_zero_dsa_batch_async(struct batch_task *batch_task,
> >                                const void **buf, size_t count, size_t len)
> > {
> > -    if (count <= 0 || count > batch_task->batch_size) {
> > +    struct dsa_batch_task *dsa_batch = batch_task->dsa_batch;
> > +
> > +    if (count <= 0 || count > dsa_batch->batch_size) {
> >         return -1;
> >     }
> >
> > -    assert(batch_task != NULL);
> > +    assert(dsa_batch != NULL);
> >     assert(len != 0);
> >     assert(buf != NULL);
> >
> >     if (count == 1) {
> >         /* DSA doesn't take batch operation with only 1 task. */
> > -        buffer_zero_dsa_async(batch_task, buf[0], len);
> > +        buffer_zero_dsa_async(dsa_batch, buf[0], len);
> >     } else {
> > -        buffer_zero_dsa_batch_async(batch_task, buf, count, len);
> > +        buffer_zero_dsa_batch_async(dsa_batch, buf, count, len);
> >     }
> >
> > -    buffer_zero_dsa_wait(batch_task);
> > -    buffer_zero_cpu_fallback(batch_task);
> > +    buffer_zero_dsa_wait(dsa_batch);
> > +    buffer_zero_cpu_fallback(dsa_batch);
> >
> >     return 0;
> > }
> >
> > #endif
> >
> > +/**
> > + * @brief Initializes a general buffer zero batch task.
> > + *
> > + * @param task A pointer to the general batch task to initialize.
> > + * @param batch_size The number of zero page checking tasks in the batch.
> > + */
> > +void
> > +batch_task_init(struct batch_task *task, int batch_size)
> > +{
> > +    task->addr = g_new0(ram_addr_t, batch_size);
> > +    task->results = g_new0(bool, batch_size);
> > +#ifdef CONFIG_DSA_OPT
> > +    task->dsa_batch = qemu_memalign(64, sizeof(struct dsa_batch_task));
> > +    buffer_zero_batch_task_init(task->dsa_batch, task->results, 
> > batch_size);
> > +#endif
> > +}
> > +
> > +/**
> > + * @brief Destroys a general buffer zero batch task.
> > + *
> > + * @param task A pointer to the general batch task to destroy.
> > + */
> > +void
> > +batch_task_destroy(struct batch_task *task)
> > +{
> > +    g_free(task->addr);
> > +    g_free(task->results);
> > +#ifdef CONFIG_DSA_OPT
> > +    buffer_zero_batch_task_destroy(task->dsa_batch);
> > +    qemu_vfree(task->dsa_batch);
> > +#endif
> > +}
> > +
> > --
> > 2.30.2
> >
> >
> >
>