Message ID | 1357907697-16790-4-git-send-email-stefanha@redhat.com |
---|---|
State | New |
Headers | show |
Il 11/01/2013 13:34, Stefan Hajnoczi ha scritto: > + iov = &bounce_iov; > + iov_cnt = 1; > + > + if (read) { > + /* Need to copy back from bounce buffer on completion */ > + read_qiov = g_slice_new(QEMUIOVector); > + qemu_iovec_init(read_qiov, iov_cnt); > + qemu_iovec_concat_iov(read_qiov, iov, iov_cnt, 0, qiov.size); This is still wrong, how did you test it? Paolo
On Fri, Jan 11, 2013 at 02:28:17PM +0100, Paolo Bonzini wrote: > Il 11/01/2013 13:34, Stefan Hajnoczi ha scritto: > > + iov = &bounce_iov; > > + iov_cnt = 1; > > + > > + if (read) { > > + /* Need to copy back from bounce buffer on completion */ > > + read_qiov = g_slice_new(QEMUIOVector); > > + qemu_iovec_init(read_qiov, iov_cnt); > > + qemu_iovec_concat_iov(read_qiov, iov, iov_cnt, 0, qiov.size); > > This is still wrong, how did you test it? I sent it before reading the reviews from you and Kevin. Bizarrely Windows saw the files I copied to the disk - even across reboot (no buffer cache involved?). Or is Windows 8 doing tricks like "shutdown" is actually "suspend to disk"? Stefan
Am 11.01.2013 15:40, schrieb Stefan Hajnoczi: > On Fri, Jan 11, 2013 at 02:28:17PM +0100, Paolo Bonzini wrote: >> Il 11/01/2013 13:34, Stefan Hajnoczi ha scritto: >>> + iov = &bounce_iov; >>> + iov_cnt = 1; >>> + >>> + if (read) { >>> + /* Need to copy back from bounce buffer on completion */ >>> + read_qiov = g_slice_new(QEMUIOVector); >>> + qemu_iovec_init(read_qiov, iov_cnt); >>> + qemu_iovec_concat_iov(read_qiov, iov, iov_cnt, 0, qiov.size); >> >> This is still wrong, how did you test it? > > I sent it before reading the reviews from you and Kevin. Bizarrely > Windows saw the files I copied to the disk - even across reboot (no > buffer cache involved?). Or is Windows 8 doing tricks like "shutdown" > is actually "suspend to disk"? As far as I know this is exactly what it's doing. I've seen warnings that you shouldn't mount file systems of a dual-boot Windows 8 if you didn't change its options so that it really shuts down. Kevin
diff --git a/hw/dataplane/virtio-blk.c b/hw/dataplane/virtio-blk.c index a6696b8..88300a6 100644 --- a/hw/dataplane/virtio-blk.c +++ b/hw/dataplane/virtio-blk.c @@ -34,6 +34,8 @@ typedef struct { struct iocb iocb; /* Linux AIO control block */ QEMUIOVector *inhdr; /* iovecs for virtio_blk_inhdr */ unsigned int head; /* vring descriptor index */ + void *bounce_buffer; /* used if guest buffers are unaligned */ + QEMUIOVector *read_qiov; /* for read completion /w bounce buffer */ } VirtIOBlockRequest; struct VirtIOBlockDataPlane { @@ -89,6 +91,15 @@ static void complete_request(struct iocb *iocb, ssize_t ret, void *opaque) trace_virtio_blk_data_plane_complete_request(s, req->head, ret); + if (req->read_qiov) { + assert(req->bounce_buffer); + qemu_iovec_from_buf(req->read_qiov, 0, req->bounce_buffer, len); + qemu_iovec_destroy(req->read_qiov); + g_slice_free(QEMUIOVector, req->read_qiov); + } + + qemu_vfree(req->bounce_buffer); + qemu_iovec_from_buf(req->inhdr, 0, &hdr, sizeof(hdr)); qemu_iovec_destroy(req->inhdr); g_slice_free(QEMUIOVector, req->inhdr); @@ -136,6 +147,29 @@ static int do_rdwr_cmd(VirtIOBlockDataPlane *s, bool read, QEMUIOVector *inhdr) { struct iocb *iocb; + QEMUIOVector qiov; + struct iovec bounce_iov; + void *bounce_buffer = NULL; + QEMUIOVector *read_qiov = NULL; + + qemu_iovec_init_external(&qiov, iov, iov_cnt); + if (!bdrv_qiov_is_aligned(s->blk->conf.bs, &qiov)) { + /* Redirect I/O to aligned bounce buffer */ + bounce_buffer = qemu_blockalign(s->blk->conf.bs, qiov.size); + bounce_iov.iov_base = bounce_buffer; + bounce_iov.iov_len = qiov.size; + iov = &bounce_iov; + iov_cnt = 1; + + if (read) { + /* Need to copy back from bounce buffer on completion */ + read_qiov = g_slice_new(QEMUIOVector); + qemu_iovec_init(read_qiov, iov_cnt); + qemu_iovec_concat_iov(read_qiov, iov, iov_cnt, 0, qiov.size); + } else { + qemu_iovec_to_buf(&qiov, 0, bounce_buffer, qiov.size); + } + } iocb = ioq_rdwr(&s->ioqueue, read, iov, iov_cnt, offset); @@ -143,6 +177,8 @@ static int do_rdwr_cmd(VirtIOBlockDataPlane *s, bool read, VirtIOBlockRequest *req = container_of(iocb, VirtIOBlockRequest, iocb); req->head = head; req->inhdr = inhdr; + req->bounce_buffer = bounce_buffer; + req->read_qiov = read_qiov; return 0; }
O_DIRECT on Linux has alignment requirements on I/O buffers and misaligned requests result in -EINVAL. The Linux virtio_blk guest driver usually submits aligned requests so I forgot to handle misaligned requests. It turns out that virtio-win guest drivers submit misaligned requests. Handle them using a bounce buffer that meets alignment requirements. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> --- hw/dataplane/virtio-blk.c | 36 ++++++++++++++++++++++++++++++++++++ 1 file changed, 36 insertions(+)