[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH v3 17/29] block: Generalise and optimise COR ser
From: |
Benoît Canet |
Subject: |
Re: [Qemu-devel] [PATCH v3 17/29] block: Generalise and optimise COR serialisation |
Date: |
Wed, 22 Jan 2014 21:00:26 +0100 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
Le Friday 17 Jan 2014 à 15:15:07 (+0100), Kevin Wolf a écrit :
> Change the API so that specific requests can be marked serialising. Only
I find the spelling "serialising" instead of "serializing" really odd since QEMU
code is full of z like the word virtualization.
Reviewed-by: Benoit Canet <address@hidden>
> these requests are checked for overlaps then.
>
> This means that during a Copy on Read operation, not all requests
> overlapping other requests are serialised any more, but only those that
> actually overlap with the specific COR request.
>
> Also remove COR from function and variable names because this
> functionality can be useful in other contexts.
>
> Signed-off-by: Kevin Wolf <address@hidden>
> Reviewed-by: Max Reitz <address@hidden>
> ---
> block.c | 48
> ++++++++++++++++++++++++++++-------------------
> include/block/block_int.h | 5 +++--
> 2 files changed, 32 insertions(+), 21 deletions(-)
>
> diff --git a/block.c b/block.c
> index d7156ce..efa8979 100644
> --- a/block.c
> +++ b/block.c
> @@ -2028,6 +2028,10 @@ int bdrv_commit_all(void)
> */
> static void tracked_request_end(BdrvTrackedRequest *req)
> {
> + if (req->serialising) {
> + req->bs->serialising_in_flight--;
> + }
> +
> QLIST_REMOVE(req, list);
> qemu_co_queue_restart_all(&req->wait_queue);
> }
> @@ -2042,10 +2046,11 @@ static void tracked_request_begin(BdrvTrackedRequest
> *req,
> {
> *req = (BdrvTrackedRequest){
> .bs = bs,
> - .offset = offset,
> - .bytes = bytes,
> - .is_write = is_write,
> - .co = qemu_coroutine_self(),
> + .offset = offset,
> + .bytes = bytes,
> + .is_write = is_write,
> + .co = qemu_coroutine_self(),
> + .serialising = false,
> };
>
> qemu_co_queue_init(&req->wait_queue);
> @@ -2053,6 +2058,14 @@ static void tracked_request_begin(BdrvTrackedRequest
> *req,
> QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
> }
>
> +static void mark_request_serialising(BdrvTrackedRequest *req)
> +{
> + if (!req->serialising) {
> + req->bs->serialising_in_flight++;
> + req->serialising = true;
> + }
> +}
> +
> /**
> * Round a region to cluster boundaries
> */
> @@ -2105,26 +2118,31 @@ static bool
> tracked_request_overlaps(BdrvTrackedRequest *req,
> return true;
> }
>
> -static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
> - BdrvTrackedRequest *self, int64_t offset, unsigned int bytes)
> +static void coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
> {
> + BlockDriverState *bs = self->bs;
> BdrvTrackedRequest *req;
> int64_t cluster_offset;
> unsigned int cluster_bytes;
> bool retry;
>
> + if (!bs->serialising_in_flight) {
> + return;
> + }
> +
> /* If we touch the same cluster it counts as an overlap. This guarantees
> * that allocating writes will be serialized and not race with each other
> * for the same cluster. For example, in copy-on-read it ensures that
> the
> * CoR read and write operations are atomic and guest writes cannot
> * interleave between them.
> */
> - round_bytes_to_clusters(bs, offset, bytes, &cluster_offset,
> &cluster_bytes);
> + round_bytes_to_clusters(bs, self->offset, self->bytes,
> + &cluster_offset, &cluster_bytes);
>
> do {
> retry = false;
> QLIST_FOREACH(req, &bs->tracked_requests, list) {
> - if (req == self) {
> + if (req == self || (!req->serialising && !self->serialising)) {
> continue;
> }
> if (tracked_request_overlaps(req, cluster_offset,
> cluster_bytes)) {
> @@ -2743,12 +2761,10 @@ static int coroutine_fn
> bdrv_aligned_preadv(BlockDriverState *bs,
>
> /* Handle Copy on Read and associated serialisation */
> if (flags & BDRV_REQ_COPY_ON_READ) {
> - bs->copy_on_read_in_flight++;
> + mark_request_serialising(req);
> }
>
> - if (bs->copy_on_read_in_flight) {
> - wait_for_overlapping_requests(bs, req, offset, bytes);
> - }
> + wait_serialising_requests(req);
>
> if (flags & BDRV_REQ_COPY_ON_READ) {
> int pnum;
> @@ -2797,10 +2813,6 @@ static int coroutine_fn
> bdrv_aligned_preadv(BlockDriverState *bs,
> }
>
> out:
> - if (flags & BDRV_REQ_COPY_ON_READ) {
> - bs->copy_on_read_in_flight--;
> - }
> -
> return ret;
> }
>
> @@ -2999,9 +3011,7 @@ static int coroutine_fn
> bdrv_aligned_pwritev(BlockDriverState *bs,
> assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
> assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
>
> - if (bs->copy_on_read_in_flight) {
> - wait_for_overlapping_requests(bs, req, offset, bytes);
> - }
> + wait_serialising_requests(req);
>
> ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
>
> diff --git a/include/block/block_int.h b/include/block/block_int.h
> index 97a4d23..d8443df 100644
> --- a/include/block/block_int.h
> +++ b/include/block/block_int.h
> @@ -60,6 +60,7 @@ typedef struct BdrvTrackedRequest {
> int64_t offset;
> unsigned int bytes;
> bool is_write;
> + bool serialising;
> QLIST_ENTRY(BdrvTrackedRequest) list;
> Coroutine *co; /* owner, used for deadlock detection */
> CoQueue wait_queue; /* coroutines blocked on this request */
> @@ -296,8 +297,8 @@ struct BlockDriverState {
> /* Callback before write request is processed */
> NotifierWithReturnList before_write_notifiers;
>
> - /* number of in-flight copy-on-read requests */
> - unsigned int copy_on_read_in_flight;
> + /* number of in-flight serialising requests */
> + unsigned int serialising_in_flight;
>
> /* I/O throttling */
> ThrottleState throttle_state;
> --
> 1.8.1.4
>
>
- Re: [Qemu-devel] [PATCH v3 13/29] block: Introduce bdrv_co_do_pwritev(), (continued)
- [Qemu-devel] [PATCH v3 16/29] block: Make zero-after-EOF work with larger alignment, Kevin Wolf, 2014/01/17
- [Qemu-devel] [PATCH v3 14/29] block: Switch BdrvTrackedRequest to byte granularity, Kevin Wolf, 2014/01/17
- [Qemu-devel] [PATCH v3 15/29] block: Allow waiting for overlapping requests between begin/end, Kevin Wolf, 2014/01/17
- [Qemu-devel] [PATCH v3 17/29] block: Generalise and optimise COR serialisation, Kevin Wolf, 2014/01/17
- Re: [Qemu-devel] [PATCH v3 17/29] block: Generalise and optimise COR serialisation,
Benoît Canet <=
- [Qemu-devel] [PATCH v3 18/29] block: Make overlap range for serialisation dynamic, Kevin Wolf, 2014/01/17
- [Qemu-devel] [PATCH v3 19/29] block: Allow wait_serialising_requests() at any point, Kevin Wolf, 2014/01/17
- [Qemu-devel] [PATCH v3 20/29] block: Align requests in bdrv_co_do_pwritev(), Kevin Wolf, 2014/01/17
- [Qemu-devel] [PATCH v3 21/29] block: Assert serialisation assumptions in pwritev, Kevin Wolf, 2014/01/17