| Message ID | 20190729223605.7163-1-jsnow@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Revert "ide/ahci: Check for -ECANCELED in aio callbacks" | expand |
Hi John, On 7/30/19 12:36 AM, John Snow wrote: > This reverts commit 0d910cfeaf2076b116b4517166d5deb0fea76394. > > It's not correct to just ignore an error code in a callback; we need to > handle that error and possible report failure to the guest so that they > don't wait indefinitely for an operation that will now never finish. Is this 4.1 material? It looks so. > This ought to help cases reported by Nutanix where iSCSI returns a > legitimate -ECANCELED for certain operations which should be propagated > normally. > > Reported-by: Shaju Abraham <shaju.abraham@nutanix.com> > Signed-off-by: John Snow <jsnow@redhat.com> > --- > hw/ide/ahci.c | 3 --- > hw/ide/core.c | 14 -------------- > 2 files changed, 17 deletions(-) > > diff --git a/hw/ide/ahci.c b/hw/ide/ahci.c > index 00ba422a48..6aaf66534a 100644 > --- a/hw/ide/ahci.c > +++ b/hw/ide/ahci.c > @@ -1023,9 +1023,6 @@ static void ncq_cb(void *opaque, int ret) > IDEState *ide_state = &ncq_tfs->drive->port.ifs[0]; > > ncq_tfs->aiocb = NULL; > - if (ret == -ECANCELED) { > - return; > - } > > if (ret < 0) { > bool is_read = ncq_tfs->cmd == READ_FPDMA_QUEUED; > diff --git a/hw/ide/core.c b/hw/ide/core.c > index 6afadf894f..8e1624f7ce 100644 > --- a/hw/ide/core.c > +++ b/hw/ide/core.c > @@ -722,9 +722,6 @@ static void ide_sector_read_cb(void *opaque, int ret) > s->pio_aiocb = NULL; > s->status &= ~BUSY_STAT; > > - if (ret == -ECANCELED) { > - return; > - } > if (ret != 0) { > if (ide_handle_rw_error(s, -ret, IDE_RETRY_PIO | > IDE_RETRY_READ)) { > @@ -840,10 +837,6 @@ static void ide_dma_cb(void *opaque, int ret) > uint64_t offset; > bool stay_active = false; > > - if (ret == -ECANCELED) { > - return; > - } > - > if (ret == -EINVAL) { > ide_dma_error(s); > return; > @@ -975,10 +968,6 @@ static void ide_sector_write_cb(void *opaque, int ret) > IDEState *s = opaque; > int n; > > - if (ret == -ECANCELED) { > - return; > - } > - > s->pio_aiocb = NULL; > s->status &= ~BUSY_STAT; > > @@ -1058,9 +1047,6 @@ static void ide_flush_cb(void *opaque, int ret) > > s->pio_aiocb = NULL; > > - if (ret == -ECANCELED) { > - return; > - } > if (ret < 0) { > /* XXX: What sector number to set here? */ > if (ide_handle_rw_error(s, -ret, IDE_RETRY_FLUSH)) { >
On 30/07/19 11:54, Philippe Mathieu-Daudé wrote: > Hi John, > > On 7/30/19 12:36 AM, John Snow wrote: >> This reverts commit 0d910cfeaf2076b116b4517166d5deb0fea76394. >> >> It's not correct to just ignore an error code in a callback; we need to >> handle that error and possible report failure to the guest so that they >> don't wait indefinitely for an operation that will now never finish. > > Is this 4.1 material? It looks so. Perhaps could have been last week, but now I suggest Cc qemu-stable and delaying it to 4.2. Paolo >> This ought to help cases reported by Nutanix where iSCSI returns a >> legitimate -ECANCELED for certain operations which should be propagated >> normally. >> >> Reported-by: Shaju Abraham <shaju.abraham@nutanix.com> >> Signed-off-by: John Snow <jsnow@redhat.com> >> --- >> hw/ide/ahci.c | 3 --- >> hw/ide/core.c | 14 -------------- >> 2 files changed, 17 deletions(-) >> >> diff --git a/hw/ide/ahci.c b/hw/ide/ahci.c >> index 00ba422a48..6aaf66534a 100644 >> --- a/hw/ide/ahci.c >> +++ b/hw/ide/ahci.c >> @@ -1023,9 +1023,6 @@ static void ncq_cb(void *opaque, int ret) >> IDEState *ide_state = &ncq_tfs->drive->port.ifs[0]; >> >> ncq_tfs->aiocb = NULL; >> - if (ret == -ECANCELED) { >> - return; >> - } >> >> if (ret < 0) { >> bool is_read = ncq_tfs->cmd == READ_FPDMA_QUEUED; >> diff --git a/hw/ide/core.c b/hw/ide/core.c >> index 6afadf894f..8e1624f7ce 100644 >> --- a/hw/ide/core.c >> +++ b/hw/ide/core.c >> @@ -722,9 +722,6 @@ static void ide_sector_read_cb(void *opaque, int ret) >> s->pio_aiocb = NULL; >> s->status &= ~BUSY_STAT; >> >> - if (ret == -ECANCELED) { >> - return; >> - } >> if (ret != 0) { >> if (ide_handle_rw_error(s, -ret, IDE_RETRY_PIO | >> IDE_RETRY_READ)) { >> @@ -840,10 +837,6 @@ static void ide_dma_cb(void *opaque, int ret) >> uint64_t offset; >> bool stay_active = false; >> >> - if (ret == -ECANCELED) { >> - return; >> - } >> - >> if (ret == -EINVAL) { >> ide_dma_error(s); >> return; >> @@ -975,10 +968,6 @@ static void ide_sector_write_cb(void *opaque, int ret) >> IDEState *s = opaque; >> int n; >> >> - if (ret == -ECANCELED) { >> - return; >> - } >> - >> s->pio_aiocb = NULL; >> s->status &= ~BUSY_STAT; >> >> @@ -1058,9 +1047,6 @@ static void ide_flush_cb(void *opaque, int ret) >> >> s->pio_aiocb = NULL; >> >> - if (ret == -ECANCELED) { >> - return; >> - } >> if (ret < 0) { >> /* XXX: What sector number to set here? */ >> if (ide_handle_rw_error(s, -ret, IDE_RETRY_FLUSH)) { >>
On 7/30/19 1:37 PM, Paolo Bonzini wrote: > On 30/07/19 11:54, Philippe Mathieu-Daudé wrote: >> Hi John, >> >> On 7/30/19 12:36 AM, John Snow wrote: >>> This reverts commit 0d910cfeaf2076b116b4517166d5deb0fea76394. >>> >>> It's not correct to just ignore an error code in a callback; we need to >>> handle that error and possible report failure to the guest so that they >>> don't wait indefinitely for an operation that will now never finish. >> >> Is this 4.1 material? It looks so. > > Perhaps could have been last week, but now I suggest Cc qemu-stable and > delaying it to 4.2. I was wondering because Kevin was preparing a block@ pull request. Anyway this seems fair because the reverted commit is dated from 2014. Thanks, Phil.
On 7/30/19 5:54 AM, Philippe Mathieu-Daudé wrote: > Hi John, > > On 7/30/19 12:36 AM, John Snow wrote: >> This reverts commit 0d910cfeaf2076b116b4517166d5deb0fea76394. >> >> It's not correct to just ignore an error code in a callback; we need to >> handle that error and possible report failure to the guest so that they >> don't wait indefinitely for an operation that will now never finish. > > Is this 4.1 material? It looks so. > Nah. This reverts a commit from 2014. I have no valid argument for why this is critical :) Thanks for checking! --js
On 7/29/19 6:36 PM, John Snow wrote: > This reverts commit 0d910cfeaf2076b116b4517166d5deb0fea76394. > > It's not correct to just ignore an error code in a callback; we need to > handle that error and possible report failure to the guest so that they > don't wait indefinitely for an operation that will now never finish. > > This ought to help cases reported by Nutanix where iSCSI returns a > legitimate -ECANCELED for certain operations which should be propagated > normally. > > Reported-by: Shaju Abraham <shaju.abraham@nutanix.com> > Signed-off-by: John Snow <jsnow@redhat.com> Nobody's yelling, so this is getting staged on my IDE branch, alongside Paolo's dma-helpers fix. Thanks, applied to my IDE tree: https://github.com/jnsnow/qemu/commits/ide https://github.com/jnsnow/qemu.git --js
diff --git a/hw/ide/ahci.c b/hw/ide/ahci.c index 00ba422a48..6aaf66534a 100644 --- a/hw/ide/ahci.c +++ b/hw/ide/ahci.c @@ -1023,9 +1023,6 @@ static void ncq_cb(void *opaque, int ret) IDEState *ide_state = &ncq_tfs->drive->port.ifs[0]; ncq_tfs->aiocb = NULL; - if (ret == -ECANCELED) { - return; - } if (ret < 0) { bool is_read = ncq_tfs->cmd == READ_FPDMA_QUEUED; diff --git a/hw/ide/core.c b/hw/ide/core.c index 6afadf894f..8e1624f7ce 100644 --- a/hw/ide/core.c +++ b/hw/ide/core.c @@ -722,9 +722,6 @@ static void ide_sector_read_cb(void *opaque, int ret) s->pio_aiocb = NULL; s->status &= ~BUSY_STAT; - if (ret == -ECANCELED) { - return; - } if (ret != 0) { if (ide_handle_rw_error(s, -ret, IDE_RETRY_PIO | IDE_RETRY_READ)) { @@ -840,10 +837,6 @@ static void ide_dma_cb(void *opaque, int ret) uint64_t offset; bool stay_active = false; - if (ret == -ECANCELED) { - return; - } - if (ret == -EINVAL) { ide_dma_error(s); return; @@ -975,10 +968,6 @@ static void ide_sector_write_cb(void *opaque, int ret) IDEState *s = opaque; int n; - if (ret == -ECANCELED) { - return; - } - s->pio_aiocb = NULL; s->status &= ~BUSY_STAT; @@ -1058,9 +1047,6 @@ static void ide_flush_cb(void *opaque, int ret) s->pio_aiocb = NULL; - if (ret == -ECANCELED) { - return; - } if (ret < 0) { /* XXX: What sector number to set here? */ if (ide_handle_rw_error(s, -ret, IDE_RETRY_FLUSH)) {
This reverts commit 0d910cfeaf2076b116b4517166d5deb0fea76394. It's not correct to just ignore an error code in a callback; we need to handle that error and possible report failure to the guest so that they don't wait indefinitely for an operation that will now never finish. This ought to help cases reported by Nutanix where iSCSI returns a legitimate -ECANCELED for certain operations which should be propagated normally. Reported-by: Shaju Abraham <shaju.abraham@nutanix.com> Signed-off-by: John Snow <jsnow@redhat.com> --- hw/ide/ahci.c | 3 --- hw/ide/core.c | 14 -------------- 2 files changed, 17 deletions(-)