qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Fabiano Rosas <farosas@suse.de>
To: Hao Xiang <hao.xiang@bytedance.com>,
	peter.maydell@linaro.org, quintela@redhat.com, peterx@redhat.com,
	marcandre.lureau@redhat.com, bryan.zhang@bytedance.com,
	qemu-devel@nongnu.org
Cc: Hao Xiang <hao.xiang@bytedance.com>
Subject: Re: [PATCH v2 11/20] util/dsa: Implement DSA task asynchronous submission and wait for completion.
Date: Wed, 13 Dec 2023 11:01:00 -0300	[thread overview]
Message-ID: <87o7euxkv7.fsf@suse.de> (raw)
In-Reply-To: <20231114054032.1192027-12-hao.xiang@bytedance.com>

Hao Xiang <hao.xiang@bytedance.com> writes:

> * Add a DSA task completion callback.
> * DSA completion thread will call the tasks's completion callback
> on every task/batch task completion.
> * DSA submission path to wait for completion.
> * Implement CPU fallback if DSA is not able to complete the task.
>
> Signed-off-by: Hao Xiang <hao.xiang@bytedance.com>
> Signed-off-by: Bryan Zhang <bryan.zhang@bytedance.com>
> ---
>  include/qemu/dsa.h |  14 +++++
>  util/dsa.c         | 153 ++++++++++++++++++++++++++++++++++++++++++++-
>  2 files changed, 164 insertions(+), 3 deletions(-)
>
> diff --git a/include/qemu/dsa.h b/include/qemu/dsa.h
> index b10e7b8fb7..3f8ee07004 100644
> --- a/include/qemu/dsa.h
> +++ b/include/qemu/dsa.h
> @@ -65,6 +65,20 @@ void buffer_zero_batch_task_init(struct buffer_zero_batch_task *task,
>   */
>  void buffer_zero_batch_task_destroy(struct buffer_zero_batch_task *task);
>  
> +/**
> + * @brief Performs buffer zero comparison on a DSA batch task asynchronously.
> + *
> + * @param batch_task A pointer to the batch task.
> + * @param buf An array of memory buffers.
> + * @param count The number of buffers in the array.
> + * @param len The buffer length.
> + *
> + * @return Zero if successful, otherwise non-zero.
> + */
> +int
> +buffer_is_zero_dsa_batch_async(struct buffer_zero_batch_task *batch_task,
> +                               const void **buf, size_t count, size_t len);
> +
>  /**
>   * @brief Initializes DSA devices.
>   *
> diff --git a/util/dsa.c b/util/dsa.c
> index 3cc017b8a0..06c6fbf2ca 100644
> --- a/util/dsa.c
> +++ b/util/dsa.c
> @@ -470,6 +470,41 @@ poll_completion(struct dsa_completion_record *completion,
>      return 0;
>  }
>  
> +/**
> + * @brief Use CPU to complete a single zero page checking task.
> + *
> + * @param task A pointer to the task.
> + */
> +static void
> +task_cpu_fallback(struct buffer_zero_batch_task *task)
> +{
> +    assert(task->task_type == DSA_TASK);
> +
> +    struct dsa_completion_record *completion = &task->completions[0];
> +    const uint8_t *buf;
> +    size_t len;
> +
> +    if (completion->status == DSA_COMP_SUCCESS) {
> +        return;
> +    }
> +
> +    /*
> +     * DSA was able to partially complete the operation. Check the
> +     * result. If we already know this is not a zero page, we can
> +     * return now.
> +     */
> +    if (completion->bytes_completed != 0 && completion->result != 0) {
> +        task->results[0] = false;
> +        return;
> +    }
> +
> +    /* Let's fallback to use CPU to complete it. */
> +    buf = (const uint8_t *)task->descriptors[0].src_addr;
> +    len = task->descriptors[0].xfer_size;
> +    task->results[0] = buffer_is_zero(buf + completion->bytes_completed,
> +                                      len - completion->bytes_completed);
> +}
> +
>  /**
>   * @brief Complete a single DSA task in the batch task.
>   *
> @@ -548,6 +583,62 @@ poll_batch_task_completion(struct buffer_zero_batch_task *batch_task)
>      }
>  }
>  
> +/**
> + * @brief Use CPU to complete the zero page checking batch task.
> + *
> + * @param batch_task A pointer to the batch task.
> + */
> +static void
> +batch_task_cpu_fallback(struct buffer_zero_batch_task *batch_task)
> +{
> +    assert(batch_task->task_type == DSA_BATCH_TASK);
> +
> +    struct dsa_completion_record *batch_completion =
> +        &batch_task->batch_completion;
> +    struct dsa_completion_record *completion;
> +    uint8_t status;
> +    const uint8_t *buf;
> +    size_t len;
> +    bool *results = batch_task->results;
> +    uint32_t count = batch_task->batch_descriptor.desc_count;
> +
> +    // DSA is able to complete the entire batch task.
> +    if (batch_completion->status == DSA_COMP_SUCCESS) {
> +        assert(count == batch_completion->bytes_completed);
> +        return;
> +    }
> +
> +    /*
> +     * DSA encounters some error and is not able to complete
> +     * the entire batch task. Use CPU fallback.
> +     */
> +    for (int i = 0; i < count; i++) {
> +        completion = &batch_task->completions[i];
> +        status = completion->status;
> +        if (status == DSA_COMP_SUCCESS) {
> +            continue;
> +        }
> +        assert(status == DSA_COMP_PAGE_FAULT_NOBOF);
> +
> +        /*
> +         * DSA was able to partially complete the operation. Check the
> +         * result. If we already know this is not a zero page, we can
> +         * return now.
> +         */
> +        if (completion->bytes_completed != 0 && completion->result != 0) {
> +            results[i] = false;
> +            continue;
> +        }
> +
> +        /* Let's fallback to use CPU to complete it. */
> +        buf = (uint8_t *)batch_task->descriptors[i].src_addr;
> +        len = batch_task->descriptors[i].xfer_size;
> +        results[i] =
> +            buffer_is_zero(buf + completion->bytes_completed,
> +                           len - completion->bytes_completed);

Here the same thing is happening as in other patches, the batch task
operation is just a repeat of the task operation n times. So this whole
inner code here could be nicely replaced by task_cpu_fallback() with
some adjustment of the function arguments. That makes intuitive sense
and removes code duplication.

> +    }
> +}
> +
>  /**
>   * @brief Handles an asynchronous DSA batch task completion.
>   *
> @@ -825,7 +916,6 @@ buffer_zero_batch_task_set(struct buffer_zero_batch_task *batch_task,
>   *
>   * @return int Zero if successful, otherwise an appropriate error code.
>   */
> -__attribute__((unused))
>  static int
>  buffer_zero_dsa_async(struct buffer_zero_batch_task *task,
>                        const void *buf, size_t len)
> @@ -844,7 +934,6 @@ buffer_zero_dsa_async(struct buffer_zero_batch_task *task,
>   * @param count The number of buffers.
>   * @param len The buffer length.
>   */
> -__attribute__((unused))
>  static int
>  buffer_zero_dsa_batch_async(struct buffer_zero_batch_task *batch_task,
>                              const void **buf, size_t count, size_t len)
> @@ -876,13 +965,29 @@ buffer_zero_dsa_completion(void *context)
>   *
>   * @param batch_task A pointer to the buffer zero comparison batch task.
>   */
> -__attribute__((unused))
>  static void
>  buffer_zero_dsa_wait(struct buffer_zero_batch_task *batch_task)
>  {
>      qemu_sem_wait(&batch_task->sem_task_complete);
>  }
>  
> +/**
> + * @brief Use CPU to complete the zero page checking task if DSA
> + *        is not able to complete it.
> + *
> + * @param batch_task A pointer to the batch task.
> + */
> +static void
> +buffer_zero_cpu_fallback(struct buffer_zero_batch_task *batch_task)
> +{
> +    if (batch_task->task_type == DSA_TASK) {
> +        task_cpu_fallback(batch_task);
> +    } else {
> +        assert(batch_task->task_type == DSA_BATCH_TASK);
> +        batch_task_cpu_fallback(batch_task);
> +    }
> +}
> +
>  /**
>   * @brief Check if DSA is running.
>   *
> @@ -956,6 +1061,41 @@ void dsa_cleanup(void)
>      dsa_device_group_cleanup(&dsa_group);
>  }
>  
> +/**
> + * @brief Performs buffer zero comparison on a DSA batch task asynchronously.
> + *
> + * @param batch_task A pointer to the batch task.
> + * @param buf An array of memory buffers.
> + * @param count The number of buffers in the array.
> + * @param len The buffer length.
> + *
> + * @return Zero if successful, otherwise non-zero.
> + */
> +int
> +buffer_is_zero_dsa_batch_async(struct buffer_zero_batch_task *batch_task,
> +                               const void **buf, size_t count, size_t len)
> +{
> +    if (count <= 0 || count > batch_task->batch_size) {
> +        return -1;
> +    }
> +
> +    assert(batch_task != 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);
> +    } else {
> +        buffer_zero_dsa_batch_async(batch_task, buf, count, len);
> +    }
> +
> +    buffer_zero_dsa_wait(batch_task);
> +    buffer_zero_cpu_fallback(batch_task);
> +
> +    return 0;
> +}
> +
>  #else
>  
>  void buffer_zero_batch_task_init(struct buffer_zero_batch_task *task,
> @@ -981,5 +1121,12 @@ void dsa_stop(void) {}
>  
>  void dsa_cleanup(void) {}
>  
> +int
> +buffer_is_zero_dsa_batch_async(struct buffer_zero_batch_task *batch_task,
> +                               const void **buf, size_t count, size_t len)
> +{
> +    exit(1);
> +}
> +
>  #endif


  reply	other threads:[~2023-12-13 14:02 UTC|newest]

Thread overview: 51+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-11-14  5:40 [PATCH v2 00/20] Use Intel DSA accelerator to offload zero page checking in multifd live migration Hao Xiang
2023-11-14  5:40 ` [PATCH v2 01/20] multifd: Add capability to enable/disable zero_page Hao Xiang
2023-11-16 15:15   ` Fabiano Rosas
2023-11-14  5:40 ` [PATCH v2 02/20] multifd: Support for zero pages transmission Hao Xiang
2023-11-14  5:40 ` [PATCH v2 03/20] multifd: Zero " Hao Xiang
2023-12-18  2:43   ` Wang, Lei
2023-11-14  5:40 ` [PATCH v2 04/20] So we use multifd to transmit zero pages Hao Xiang
2023-11-16 15:14   ` Fabiano Rosas
2024-01-23  4:28     ` [External] " Hao Xiang
2024-01-25 21:55       ` Hao Xiang
2024-01-25 23:14         ` Fabiano Rosas
2024-01-25 23:46           ` Hao Xiang
2023-11-14  5:40 ` [PATCH v2 05/20] meson: Introduce new instruction set enqcmd to the build system Hao Xiang
2023-12-11 15:41   ` Fabiano Rosas
2023-12-16  0:26     ` [External] " Hao Xiang
2023-11-14  5:40 ` [PATCH v2 06/20] util/dsa: Add dependency idxd Hao Xiang
2023-11-14  5:40 ` [PATCH v2 07/20] util/dsa: Implement DSA device start and stop logic Hao Xiang
2023-12-11 21:28   ` Fabiano Rosas
2023-12-19  6:41     ` [External] " Hao Xiang
2023-12-19 13:18       ` Fabiano Rosas
2023-12-27  6:00         ` Hao Xiang
2023-11-14  5:40 ` [PATCH v2 08/20] util/dsa: Implement DSA task enqueue and dequeue Hao Xiang
2023-12-12 16:10   ` Fabiano Rosas
2023-12-27  0:07     ` [External] " Hao Xiang
2023-11-14  5:40 ` [PATCH v2 09/20] util/dsa: Implement DSA task asynchronous completion thread model Hao Xiang
2023-12-12 19:36   ` Fabiano Rosas
2023-12-18  3:11   ` Wang, Lei
2023-12-18 18:57     ` [External] " Hao Xiang
2023-12-19  1:33       ` Wang, Lei
2023-12-19  5:12         ` Hao Xiang
2023-11-14  5:40 ` [PATCH v2 10/20] util/dsa: Implement zero page checking in DSA task Hao Xiang
2023-11-14  5:40 ` [PATCH v2 11/20] util/dsa: Implement DSA task asynchronous submission and wait for completion Hao Xiang
2023-12-13 14:01   ` Fabiano Rosas [this message]
2023-12-27  6:26     ` [External] " Hao Xiang
2023-11-14  5:40 ` [PATCH v2 12/20] migration/multifd: Add new migration option for multifd DSA offloading Hao Xiang
2023-12-11 19:44   ` Fabiano Rosas
2023-12-18 18:34     ` [External] " Hao Xiang
2023-12-18  3:12   ` Wang, Lei
2023-11-14  5:40 ` [PATCH v2 13/20] migration/multifd: Prepare to introduce DSA acceleration on the multifd path Hao Xiang
2023-12-18  3:20   ` Wang, Lei
2023-11-14  5:40 ` [PATCH v2 14/20] migration/multifd: Enable DSA offloading in multifd sender path Hao Xiang
2023-11-14  5:40 ` [PATCH v2 15/20] migration/multifd: Add test hook to set normal page ratio Hao Xiang
2023-11-14  5:40 ` [PATCH v2 16/20] migration/multifd: Enable set normal page ratio test hook in multifd Hao Xiang
2023-11-14  5:40 ` [PATCH v2 17/20] migration/multifd: Add migration option set packet size Hao Xiang
2023-11-14  5:40 ` [PATCH v2 18/20] migration/multifd: Enable set packet size migration option Hao Xiang
2023-12-13 17:33   ` Fabiano Rosas
2024-01-03 20:04     ` [External] " Hao Xiang
2023-11-14  5:40 ` [PATCH v2 19/20] util/dsa: Add unit test coverage for Intel DSA task submission and completion Hao Xiang
2023-11-14  5:40 ` [PATCH v2 20/20] migration/multifd: Add integration tests for multifd with Intel DSA offloading Hao Xiang
2023-11-15 17:43 ` [PATCH v2 00/20] Use Intel DSA accelerator to offload zero page checking in multifd live migration Elena Ufimtseva
2023-11-15 19:37   ` [External] " Hao Xiang

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=87o7euxkv7.fsf@suse.de \
    --to=farosas@suse.de \
    --cc=bryan.zhang@bytedance.com \
    --cc=hao.xiang@bytedance.com \
    --cc=marcandre.lureau@redhat.com \
    --cc=peter.maydell@linaro.org \
    --cc=peterx@redhat.com \
    --cc=qemu-devel@nongnu.org \
    --cc=quintela@redhat.com \
    /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).