[v2] scsi-disk: Don't enlarge min_io_size to max_io_size
diff mbox series

Message ID 20180327164141.19075-1-famz@redhat.com
State New
Headers show
Series
  • [v2] scsi-disk: Don't enlarge min_io_size to max_io_size
Related show

Commit Message

Fam Zheng March 27, 2018, 4:41 p.m. UTC
Some backends report big max_io_sectors. Making min_io_size the same
value in this case will make it impossible for guest to align memory,
therefore the disk may not be usable at all.

Do not enlarge them when they are zero.

Reported-by: David Gibson <dgibson@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>

---

v2: Leave the values alone if zero. [Paolo]
    At least we can consult block layer for a slightly more sensible
    opt_io_size, but that's for another patch.
---
 hw/scsi/scsi-disk.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

Comments

Daniel Henrique Barboza March 27, 2018, 5:05 p.m. UTC | #1
On 03/27/2018 01:41 PM, Fam Zheng wrote:
> Some backends report big max_io_sectors. Making min_io_size the same
> value in this case will make it impossible for guest to align memory,
> therefore the disk may not be usable at all.
>
> Do not enlarge them when they are zero.
>
> Reported-by: David Gibson <dgibson@redhat.com>
> Signed-off-by: Fam Zheng <famz@redhat.com>
>
> ---
>
> v2: Leave the values alone if zero. [Paolo]
>      At least we can consult block layer for a slightly more sensible
>      opt_io_size, but that's for another patch.
> ---
>   hw/scsi/scsi-disk.c | 10 ++++++----
>   1 file changed, 6 insertions(+), 4 deletions(-)
>
> diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
> index f5ab767ab5..f8ed8cf2b4 100644
> --- a/hw/scsi/scsi-disk.c
> +++ b/hw/scsi/scsi-disk.c
> @@ -714,10 +714,12 @@ static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
>
>                   /* min_io_size and opt_io_size can't be greater than
>                    * max_io_sectors */
> -                min_io_size =
> -                    MIN_NON_ZERO(min_io_size, max_io_sectors);
> -                opt_io_size =
> -                    MIN_NON_ZERO(opt_io_size, max_io_sectors);
> +                if (min_io_size) {
> +                    min_io_size = MIN(min_io_size, max_io_sectors);
> +                }
> +                if (opt_io_size) {
> +                    opt_io_size = MIN(opt_io_size, max_io_sectors);
> +                }
>               }
>               /* required VPD size with unmap support */
>               buflen = 0x40;

Reviewed-by: Daniel Henrique Barboza <danielhb@linux.vnet.ibm.com>
David Gibson March 28, 2018, 2:37 a.m. UTC | #2
On Wed, Mar 28, 2018 at 12:41:41AM +0800, Fam Zheng wrote:
> Some backends report big max_io_sectors. Making min_io_size the same
> value in this case will make it impossible for guest to align memory,
> therefore the disk may not be usable at all.
> 
> Do not enlarge them when they are zero.
> 
> Reported-by: David Gibson <dgibson@redhat.com>
> Signed-off-by: Fam Zheng <famz@redhat.com>

Tested-by: David Gibson <david@gibson.dropbear.id.au>

With this patch applied, I was able to successfully install a ppc64le
guest again.

> 
> ---
> 
> v2: Leave the values alone if zero. [Paolo]
>     At least we can consult block layer for a slightly more sensible
>     opt_io_size, but that's for another patch.
> ---
>  hw/scsi/scsi-disk.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
> index f5ab767ab5..f8ed8cf2b4 100644
> --- a/hw/scsi/scsi-disk.c
> +++ b/hw/scsi/scsi-disk.c
> @@ -714,10 +714,12 @@ static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
>  
>                  /* min_io_size and opt_io_size can't be greater than
>                   * max_io_sectors */
> -                min_io_size =
> -                    MIN_NON_ZERO(min_io_size, max_io_sectors);
> -                opt_io_size =
> -                    MIN_NON_ZERO(opt_io_size, max_io_sectors);
> +                if (min_io_size) {
> +                    min_io_size = MIN(min_io_size, max_io_sectors);
> +                }
> +                if (opt_io_size) {
> +                    opt_io_size = MIN(opt_io_size, max_io_sectors);
> +                }
>              }
>              /* required VPD size with unmap support */
>              buflen = 0x40;
Paolo Bonzini April 5, 2018, 3:54 p.m. UTC | #3
On 27/03/2018 18:41, Fam Zheng wrote:
> Some backends report big max_io_sectors. Making min_io_size the same
> value in this case will make it impossible for guest to align memory,
> therefore the disk may not be usable at all.
> 
> Do not enlarge them when they are zero.
> 
> Reported-by: David Gibson <dgibson@redhat.com>
> Signed-off-by: Fam Zheng <famz@redhat.com>
> 
> ---
> 
> v2: Leave the values alone if zero. [Paolo]
>     At least we can consult block layer for a slightly more sensible
>     opt_io_size, but that's for another patch.
> ---
>  hw/scsi/scsi-disk.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
> index f5ab767ab5..f8ed8cf2b4 100644
> --- a/hw/scsi/scsi-disk.c
> +++ b/hw/scsi/scsi-disk.c
> @@ -714,10 +714,12 @@ static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
>  
>                  /* min_io_size and opt_io_size can't be greater than
>                   * max_io_sectors */
> -                min_io_size =
> -                    MIN_NON_ZERO(min_io_size, max_io_sectors);
> -                opt_io_size =
> -                    MIN_NON_ZERO(opt_io_size, max_io_sectors);
> +                if (min_io_size) {
> +                    min_io_size = MIN(min_io_size, max_io_sectors);
> +                }
> +                if (opt_io_size) {
> +                    opt_io_size = MIN(opt_io_size, max_io_sectors);
> +                }
>              }
>              /* required VPD size with unmap support */
>              buflen = 0x40;
> 

Queued, thanks.

Paolo
David Gibson April 9, 2018, 1:08 a.m. UTC | #4
On Thu, 5 Apr 2018 17:54:00 +0200
Paolo Bonzini <pbonzini@redhat.com> wrote:

> On 27/03/2018 18:41, Fam Zheng wrote:
> > Some backends report big max_io_sectors. Making min_io_size the same
> > value in this case will make it impossible for guest to align memory,
> > therefore the disk may not be usable at all.
> > 
> > Do not enlarge them when they are zero.
> > 
> > Reported-by: David Gibson <dgibson@redhat.com>
> > Signed-off-by: Fam Zheng <famz@redhat.com>
[snip]
> Queued, thanks.
> 
> Paolo

Any ETA on going from queued to actually merged?  This is holding up a
handful of Power bugfixes for 2.12, because I can't complete my usual
testing cycle.
Paolo Bonzini April 9, 2018, 9:28 a.m. UTC | #5
On 09/04/2018 03:08, David Gibson wrote:
> On Thu, 5 Apr 2018 17:54:00 +0200 Paolo Bonzini <pbonzini@redhat.com> wrote:
>> On 27/03/2018 18:41, Fam Zheng wrote:
>>> Some backends report big max_io_sectors. Making min_io_size the same
>>> value in this case will make it impossible for guest to align memory,
>>> therefore the disk may not be usable at all.
>>>
>>> Do not enlarge them when they are zero.
>>
>> Queued, thanks.
>>
>> Paolo
> 
> Any ETA on going from queued to actually merged?  This is holding up a
> handful of Power bugfixes for 2.12, because I can't complete my usual
> testing cycle.

Hi David,

it's already included in my pull request from Friday.

Paolo
David Gibson April 10, 2018, 12:51 a.m. UTC | #6
On Mon, 9 Apr 2018 11:28:16 +0200
Paolo Bonzini <pbonzini@redhat.com> wrote:

> On 09/04/2018 03:08, David Gibson wrote:
>  [...]  
>  [...]  
>  [...]  
>  [...]  
>  [...]  
> 
> Hi David,
> 
> it's already included in my pull request from Friday.

And I see it in master now.  Thanks!

Patch
diff mbox series

diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
index f5ab767ab5..f8ed8cf2b4 100644
--- a/hw/scsi/scsi-disk.c
+++ b/hw/scsi/scsi-disk.c
@@ -714,10 +714,12 @@  static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
 
                 /* min_io_size and opt_io_size can't be greater than
                  * max_io_sectors */
-                min_io_size =
-                    MIN_NON_ZERO(min_io_size, max_io_sectors);
-                opt_io_size =
-                    MIN_NON_ZERO(opt_io_size, max_io_sectors);
+                if (min_io_size) {
+                    min_io_size = MIN(min_io_size, max_io_sectors);
+                }
+                if (opt_io_size) {
+                    opt_io_size = MIN(opt_io_size, max_io_sectors);
+                }
             }
             /* required VPD size with unmap support */
             buflen = 0x40;