[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH v2 6/8] block: request overlap detection
From: |
Kevin Wolf |
Subject: |
Re: [Qemu-devel] [PATCH v2 6/8] block: request overlap detection |
Date: |
Tue, 22 Nov 2011 17:36:19 +0100 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:8.0) Gecko/20111115 Thunderbird/8.0 |
Am 22.11.2011 16:10, schrieb Stefan Hajnoczi:
> On Tue, Nov 22, 2011 at 3:06 PM, Kevin Wolf <address@hidden> wrote:
>> Am 17.11.2011 14:40, schrieb Stefan Hajnoczi:
>>> Detect overlapping requests and remember to align to cluster boundaries
>>> if the image format uses them. This assumes that allocating I/O is
>>> performed in cluster granularity - which is true for qcow2, qed, etc.
>>>
>>> Signed-off-by: Stefan Hajnoczi <address@hidden>
>>
>>> static void coroutine_fn wait_for_overlapping_requests(BlockDriverState
>>> *bs,
>>> int64_t sector_num, int nb_sectors)
>>> {
>>> BdrvTrackedRequest *req;
>>> + int64_t cluster_sector_num;
>>> + int cluster_nb_sectors;
>>> bool retry;
>>>
>>> + /* If we touch the same cluster it counts as an overlap */
>>> + round_to_clusters(bs, sector_num, nb_sectors,
>>> + &cluster_sector_num, &cluster_nb_sectors);
>>
>> Is this really required? Image formats must be able to deal with two
>> concurrent write requests to the same cluster, and I don't think it
>> makes a difference whether it's a guest write request or a COR one.
>>
>> Or does the queuing protect more than just that a COR never takes
>> precedence over a guest write?
>
> It guarantees that a write request and a copy-on-read request racing
> for the same cluster will be serialized. Either:
> 1. CoR, then write.
> 2. Allocating write, then normal read.
>
> If we don't do this we risk:
> 3. CoR (part 1: read), allocating write, CoR (part 2: write)
>
> The result is that we dropped the write request!
Right, this requirement comes in with the next patch that makes COR
round clusters for optimisation. I think this relationship deserves a
comment here.
Anyway, I merged the series into the block branch (Only to notice that
it would have been better to merge bdrv_co_is_allocated first... Will
review that tomorrow.) If you send another version for this patch to add
a comment, I'll replace it.
Kevin
- [Qemu-devel] [PATCH v2 0/8] block: generic copy-on-read, Stefan Hajnoczi, 2011/11/17
- [Qemu-devel] [PATCH v2 8/8] block: add -drive copy-on-read=on|off, Stefan Hajnoczi, 2011/11/17
- [Qemu-devel] [PATCH v2 3/8] block: add request tracking, Stefan Hajnoczi, 2011/11/17
- [Qemu-devel] [PATCH v2 2/8] coroutine: add qemu_co_queue_restart_all(), Stefan Hajnoczi, 2011/11/17
- [Qemu-devel] [PATCH v2 4/8] block: add bdrv_set_copy_on_read(), Stefan Hajnoczi, 2011/11/17
- [Qemu-devel] [PATCH v2 5/8] block: wait for overlapping requests, Stefan Hajnoczi, 2011/11/17
- [Qemu-devel] [PATCH v2 7/8] block: core copy-on-read logic, Stefan Hajnoczi, 2011/11/17