Patchwork [2/2] qxl: change rom size to 8192

login
register
mail settings
Submitter Gerd Hoffmann
Date Jan. 22, 2013, 10:09 a.m.
Message ID <1358849355-28132-3-git-send-email-kraxel@redhat.com>
Download mbox | patch
Permalink /patch/214514/
State New
Headers show

Comments

Gerd Hoffmann - Jan. 22, 2013, 10:09 a.m.
From: Alon Levy <alevy@redhat.com>

This is a simpler solution to 869981, where migration breaks since qxl's
rom bar size has changed. Instead of ignoring fields in QXLRom, which is what has
actually changed, we remove some of the modes, a mechanism already
accounted for by the guest. The modes left allow for portrait and
landscape only modes, corresponding to orientations 0 and 1.
Orientations 2 and 3 are dropped.

Added assert so that rom size will fit the future QXLRom increases via
spice-protocol changes.

This patch has been tested with 6.1.0.10015. With the newer 6.1.0.10016
there are problems with both "(flipped)" modes prior to the patch, and
the patch loses the ability to set "Portrait" modes. But this is a
separate bug to be fixed in the driver, and besides the patch doesn't
affect the new arbitrary mode setting functionality.

Signed-off-by: Alon Levy <alevy@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
---
 hw/qxl.c |   13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
Cole Robinson - Feb. 6, 2013, 10:52 p.m.
On 01/22/2013 05:09 AM, Gerd Hoffmann wrote:
> From: Alon Levy <alevy@redhat.com>
> 
> This is a simpler solution to 869981, where migration breaks since qxl's
> rom bar size has changed. Instead of ignoring fields in QXLRom, which is what has
> actually changed, we remove some of the modes, a mechanism already
> accounted for by the guest. The modes left allow for portrait and
> landscape only modes, corresponding to orientations 0 and 1.
> Orientations 2 and 3 are dropped.
> 
> Added assert so that rom size will fit the future QXLRom increases via
> spice-protocol changes.
> 
> This patch has been tested with 6.1.0.10015. With the newer 6.1.0.10016
> there are problems with both "(flipped)" modes prior to the patch, and
> the patch loses the ability to set "Portrait" modes. But this is a
> separate bug to be fixed in the driver, and besides the patch doesn't
> affect the new arbitrary mode setting functionality.
> 
> Signed-off-by: Alon Levy <alevy@redhat.com>
> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> ---
>  hw/qxl.c |   13 +++++++------
>  1 file changed, 7 insertions(+), 6 deletions(-)
> 
> diff --git a/hw/qxl.c b/hw/qxl.c
> index 0d81816..a125e29 100644
> --- a/hw/qxl.c
> +++ b/hw/qxl.c
> @@ -80,9 +80,7 @@
>  
>  #define QXL_MODE_EX(x_res, y_res)                 \
>      QXL_MODE_16_32(x_res, y_res, 0),              \
> -    QXL_MODE_16_32(y_res, x_res, 1),              \
> -    QXL_MODE_16_32(x_res, y_res, 2),              \
> -    QXL_MODE_16_32(y_res, x_res, 3)
> +    QXL_MODE_16_32(x_res, y_res, 1)
>  
>  static QXLMode qxl_modes[] = {
>      QXL_MODE_EX(640, 480),
> @@ -306,10 +304,13 @@ static inline uint32_t msb_mask(uint32_t val)
>  
>  static ram_addr_t qxl_rom_size(void)
>  {
> -    uint32_t rom_size = sizeof(QXLRom) + sizeof(QXLModes) + sizeof(qxl_modes);
> +    uint32_t required_rom_size = sizeof(QXLRom) + sizeof(QXLModes) +
> +                                 sizeof(qxl_modes);
> +    uint32_t rom_size = 8192; /* two pages */
>  
> -    rom_size = MAX(rom_size, TARGET_PAGE_SIZE);
> -    rom_size = msb_mask(rom_size * 2 - 1);
> +    required_rom_size = MAX(required_rom_size, TARGET_PAGE_SIZE);
> +    required_rom_size = msb_mask(required_rom_size * 2 - 1);
> +    assert(required_rom_size <= rom_size);
>      return rom_size;
>  }
>  
> 

This breaks migration for me with -M pc-1.0 at least.

Before this commit:

./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice port=5905
-monitor stdio
> migrate "exec: cat > foo.img"
> quit

After this commit:

./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice port=5905
-monitor stdio -incoming "exec: cat foo.img"
qemu: warning: error while loading state for instance 0x0 of device 'ram'
cat: write error: Broken pipe
load of migration failed

Thanks,
Cole
Cole Robinson - Feb. 6, 2013, 11:44 p.m.
On 02/06/2013 05:52 PM, Cole Robinson wrote:
> On 01/22/2013 05:09 AM, Gerd Hoffmann wrote:
>> From: Alon Levy <alevy@redhat.com>
>>
>> This is a simpler solution to 869981, where migration breaks since qxl's
>> rom bar size has changed. Instead of ignoring fields in QXLRom, which is what has
>> actually changed, we remove some of the modes, a mechanism already
>> accounted for by the guest. The modes left allow for portrait and
>> landscape only modes, corresponding to orientations 0 and 1.
>> Orientations 2 and 3 are dropped.
>>
>> Added assert so that rom size will fit the future QXLRom increases via
>> spice-protocol changes.
>>
>> This patch has been tested with 6.1.0.10015. With the newer 6.1.0.10016
>> there are problems with both "(flipped)" modes prior to the patch, and
>> the patch loses the ability to set "Portrait" modes. But this is a
>> separate bug to be fixed in the driver, and besides the patch doesn't
>> affect the new arbitrary mode setting functionality.
>>
>> Signed-off-by: Alon Levy <alevy@redhat.com>
>> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
>> ---
>>  hw/qxl.c |   13 +++++++------
>>  1 file changed, 7 insertions(+), 6 deletions(-)
>>
>> diff --git a/hw/qxl.c b/hw/qxl.c
>> index 0d81816..a125e29 100644
>> --- a/hw/qxl.c
>> +++ b/hw/qxl.c
>> @@ -80,9 +80,7 @@
>>  
>>  #define QXL_MODE_EX(x_res, y_res)                 \
>>      QXL_MODE_16_32(x_res, y_res, 0),              \
>> -    QXL_MODE_16_32(y_res, x_res, 1),              \
>> -    QXL_MODE_16_32(x_res, y_res, 2),              \
>> -    QXL_MODE_16_32(y_res, x_res, 3)
>> +    QXL_MODE_16_32(x_res, y_res, 1)
>>  
>>  static QXLMode qxl_modes[] = {
>>      QXL_MODE_EX(640, 480),
>> @@ -306,10 +304,13 @@ static inline uint32_t msb_mask(uint32_t val)
>>  
>>  static ram_addr_t qxl_rom_size(void)
>>  {
>> -    uint32_t rom_size = sizeof(QXLRom) + sizeof(QXLModes) + sizeof(qxl_modes);
>> +    uint32_t required_rom_size = sizeof(QXLRom) + sizeof(QXLModes) +
>> +                                 sizeof(qxl_modes);
>> +    uint32_t rom_size = 8192; /* two pages */
>>  
>> -    rom_size = MAX(rom_size, TARGET_PAGE_SIZE);
>> -    rom_size = msb_mask(rom_size * 2 - 1);
>> +    required_rom_size = MAX(required_rom_size, TARGET_PAGE_SIZE);
>> +    required_rom_size = msb_mask(required_rom_size * 2 - 1);
>> +    assert(required_rom_size <= rom_size);
>>      return rom_size;
>>  }
>>  
>>
> 
> This breaks migration for me with -M pc-1.0 at least.
> 
> Before this commit:
> 
> ./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice port=5905
> -monitor stdio
>> migrate "exec: cat > foo.img"
>> quit
> 
> After this commit:
> 
> ./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice port=5905
> -monitor stdio -incoming "exec: cat foo.img"
> qemu: warning: error while loading state for instance 0x0 of device 'ram'
> cat: write error: Broken pipe
> load of migration failed
> 

Sorry, I realize that may seem a bit artificial, so just to clarify a bit. If
I start on v1.3.1, then cherry-pick this commit, it breaks migration using the
commands above. I also tried v1.3.1 to git head, reverting the seabios rebase
that caused other migration issues, and got the same error (is there any way
to get better error reporting out of migration failures?).

Thanks,
Cole
Alon Levy - Feb. 7, 2013, 12:31 p.m.
> On 02/06/2013 05:52 PM, Cole Robinson wrote:
> > On 01/22/2013 05:09 AM, Gerd Hoffmann wrote:
> >> From: Alon Levy <alevy@redhat.com>
> >>
> >> This is a simpler solution to 869981, where migration breaks since
> >> qxl's
> >> rom bar size has changed. Instead of ignoring fields in QXLRom,
> >> which is what has
> >> actually changed, we remove some of the modes, a mechanism already
> >> accounted for by the guest. The modes left allow for portrait and
> >> landscape only modes, corresponding to orientations 0 and 1.
> >> Orientations 2 and 3 are dropped.
> >>
> >> Added assert so that rom size will fit the future QXLRom increases
> >> via
> >> spice-protocol changes.
> >>
> >> This patch has been tested with 6.1.0.10015. With the newer
> >> 6.1.0.10016
> >> there are problems with both "(flipped)" modes prior to the patch,
> >> and
> >> the patch loses the ability to set "Portrait" modes. But this is a
> >> separate bug to be fixed in the driver, and besides the patch
> >> doesn't
> >> affect the new arbitrary mode setting functionality.
> >>
> >> Signed-off-by: Alon Levy <alevy@redhat.com>
> >> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> >> ---
> >>  hw/qxl.c |   13 +++++++------
> >>  1 file changed, 7 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/hw/qxl.c b/hw/qxl.c
> >> index 0d81816..a125e29 100644
> >> --- a/hw/qxl.c
> >> +++ b/hw/qxl.c
> >> @@ -80,9 +80,7 @@
> >>  
> >>  #define QXL_MODE_EX(x_res, y_res)                 \
> >>      QXL_MODE_16_32(x_res, y_res, 0),              \
> >> -    QXL_MODE_16_32(y_res, x_res, 1),              \
> >> -    QXL_MODE_16_32(x_res, y_res, 2),              \
> >> -    QXL_MODE_16_32(y_res, x_res, 3)
> >> +    QXL_MODE_16_32(x_res, y_res, 1)
> >>  
> >>  static QXLMode qxl_modes[] = {
> >>      QXL_MODE_EX(640, 480),
> >> @@ -306,10 +304,13 @@ static inline uint32_t msb_mask(uint32_t
> >> val)
> >>  
> >>  static ram_addr_t qxl_rom_size(void)
> >>  {
> >> -    uint32_t rom_size = sizeof(QXLRom) + sizeof(QXLModes) +
> >> sizeof(qxl_modes);
> >> +    uint32_t required_rom_size = sizeof(QXLRom) +
> >> sizeof(QXLModes) +
> >> +                                 sizeof(qxl_modes);
> >> +    uint32_t rom_size = 8192; /* two pages */
> >>  
> >> -    rom_size = MAX(rom_size, TARGET_PAGE_SIZE);
> >> -    rom_size = msb_mask(rom_size * 2 - 1);
> >> +    required_rom_size = MAX(required_rom_size, TARGET_PAGE_SIZE);
> >> +    required_rom_size = msb_mask(required_rom_size * 2 - 1);
> >> +    assert(required_rom_size <= rom_size);
> >>      return rom_size;
> >>  }
> >>  
> >>
> > 
> > This breaks migration for me with -M pc-1.0 at least.
> > 
> > Before this commit:
> > 
> > ./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice
> > port=5905
> > -monitor stdio
> >> migrate "exec: cat > foo.img"
> >> quit
> > 
> > After this commit:
> > 
> > ./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice
> > port=5905
> > -monitor stdio -incoming "exec: cat foo.img"
> > qemu: warning: error while loading state for instance 0x0 of device
> > 'ram'
> > cat: write error: Broken pipe
> > load of migration failed
> > 
> 
> Sorry, I realize that may seem a bit artificial, so just to clarify a
> bit. If
> I start on v1.3.1, then cherry-pick this commit, it breaks migration
> using the
> commands above. I also tried v1.3.1 to git head, reverting the
> seabios rebase
> that caused other migration issues, and got the same error (is there
> any way
> to get better error reporting out of migration failures?).

Migration error reporting sucks. Did you create foo.img with pre-patched qemu and then try to read it with post patched, or both are using the same version?

> 
> Thanks,
> Cole
>
Cole Robinson - Feb. 7, 2013, 3:04 p.m.
On 02/07/2013 07:31 AM, Alon Levy wrote:
>> On 02/06/2013 05:52 PM, Cole Robinson wrote:
>>> On 01/22/2013 05:09 AM, Gerd Hoffmann wrote:
>>>> From: Alon Levy <alevy@redhat.com>
>>>>
>>>> This is a simpler solution to 869981, where migration breaks since
>>>> qxl's
>>>> rom bar size has changed. Instead of ignoring fields in QXLRom,
>>>> which is what has
>>>> actually changed, we remove some of the modes, a mechanism already
>>>> accounted for by the guest. The modes left allow for portrait and
>>>> landscape only modes, corresponding to orientations 0 and 1.
>>>> Orientations 2 and 3 are dropped.
>>>>
>>>> Added assert so that rom size will fit the future QXLRom increases
>>>> via
>>>> spice-protocol changes.
>>>>
>>>> This patch has been tested with 6.1.0.10015. With the newer
>>>> 6.1.0.10016
>>>> there are problems with both "(flipped)" modes prior to the patch,
>>>> and
>>>> the patch loses the ability to set "Portrait" modes. But this is a
>>>> separate bug to be fixed in the driver, and besides the patch
>>>> doesn't
>>>> affect the new arbitrary mode setting functionality.
>>>>
>>>> Signed-off-by: Alon Levy <alevy@redhat.com>
>>>> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
>>>> ---
>>>>  hw/qxl.c |   13 +++++++------
>>>>  1 file changed, 7 insertions(+), 6 deletions(-)
>>>>
>>>> diff --git a/hw/qxl.c b/hw/qxl.c
>>>> index 0d81816..a125e29 100644
>>>> --- a/hw/qxl.c
>>>> +++ b/hw/qxl.c
>>>> @@ -80,9 +80,7 @@
>>>>  
>>>>  #define QXL_MODE_EX(x_res, y_res)                 \
>>>>      QXL_MODE_16_32(x_res, y_res, 0),              \
>>>> -    QXL_MODE_16_32(y_res, x_res, 1),              \
>>>> -    QXL_MODE_16_32(x_res, y_res, 2),              \
>>>> -    QXL_MODE_16_32(y_res, x_res, 3)
>>>> +    QXL_MODE_16_32(x_res, y_res, 1)
>>>>  
>>>>  static QXLMode qxl_modes[] = {
>>>>      QXL_MODE_EX(640, 480),
>>>> @@ -306,10 +304,13 @@ static inline uint32_t msb_mask(uint32_t
>>>> val)
>>>>  
>>>>  static ram_addr_t qxl_rom_size(void)
>>>>  {
>>>> -    uint32_t rom_size = sizeof(QXLRom) + sizeof(QXLModes) +
>>>> sizeof(qxl_modes);
>>>> +    uint32_t required_rom_size = sizeof(QXLRom) +
>>>> sizeof(QXLModes) +
>>>> +                                 sizeof(qxl_modes);
>>>> +    uint32_t rom_size = 8192; /* two pages */
>>>>  
>>>> -    rom_size = MAX(rom_size, TARGET_PAGE_SIZE);
>>>> -    rom_size = msb_mask(rom_size * 2 - 1);
>>>> +    required_rom_size = MAX(required_rom_size, TARGET_PAGE_SIZE);
>>>> +    required_rom_size = msb_mask(required_rom_size * 2 - 1);
>>>> +    assert(required_rom_size <= rom_size);
>>>>      return rom_size;
>>>>  }
>>>>  
>>>>
>>>
>>> This breaks migration for me with -M pc-1.0 at least.
>>>
>>> Before this commit:
>>>
>>> ./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice
>>> port=5905
>>> -monitor stdio
>>>> migrate "exec: cat > foo.img"
>>>> quit
>>>
>>> After this commit:
>>>
>>> ./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice
>>> port=5905
>>> -monitor stdio -incoming "exec: cat foo.img"
>>> qemu: warning: error while loading state for instance 0x0 of device
>>> 'ram'
>>> cat: write error: Broken pipe
>>> load of migration failed
>>>
>>
>> Sorry, I realize that may seem a bit artificial, so just to clarify a
>> bit. If
>> I start on v1.3.1, then cherry-pick this commit, it breaks migration
>> using the
>> commands above. I also tried v1.3.1 to git head, reverting the
>> seabios rebase
>> that caused other migration issues, and got the same error (is there
>> any way
>> to get better error reporting out of migration failures?).
> 
> Migration error reporting sucks. Did you create foo.img with pre-patched qemu and then try to read it with post patched, or both are using the same version?

I tried:

foo.img generated with stock qemu 1.3.1, migration fails when cherry picking
the patch.

foo.img generated with stock qemu 1.3.1, migration fails with git head, still
fails after reverting the seabios 1.7.2 update which supposedly causes other
migrate issues (commit 3588185b8396eb97fd9efd41c2b97775465f67c4)

- Cole
Alon Levy - Feb. 7, 2013, 4:10 p.m.
> On 02/07/2013 07:31 AM, Alon Levy wrote:
> >> On 02/06/2013 05:52 PM, Cole Robinson wrote:
> >>> On 01/22/2013 05:09 AM, Gerd Hoffmann wrote:
> >>>> From: Alon Levy <alevy@redhat.com>
> >>>>
> >>>> This is a simpler solution to 869981, where migration breaks
> >>>> since
> >>>> qxl's
> >>>> rom bar size has changed. Instead of ignoring fields in QXLRom,
> >>>> which is what has
> >>>> actually changed, we remove some of the modes, a mechanism
> >>>> already
> >>>> accounted for by the guest. The modes left allow for portrait
> >>>> and
> >>>> landscape only modes, corresponding to orientations 0 and 1.
> >>>> Orientations 2 and 3 are dropped.
> >>>>
> >>>> Added assert so that rom size will fit the future QXLRom
> >>>> increases
> >>>> via
> >>>> spice-protocol changes.
> >>>>
> >>>> This patch has been tested with 6.1.0.10015. With the newer
> >>>> 6.1.0.10016
> >>>> there are problems with both "(flipped)" modes prior to the
> >>>> patch,
> >>>> and
> >>>> the patch loses the ability to set "Portrait" modes. But this is
> >>>> a
> >>>> separate bug to be fixed in the driver, and besides the patch
> >>>> doesn't
> >>>> affect the new arbitrary mode setting functionality.
> >>>>
> >>>> Signed-off-by: Alon Levy <alevy@redhat.com>
> >>>> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> >>>> ---
> >>>>  hw/qxl.c |   13 +++++++------
> >>>>  1 file changed, 7 insertions(+), 6 deletions(-)
> >>>>
> >>>> diff --git a/hw/qxl.c b/hw/qxl.c
> >>>> index 0d81816..a125e29 100644
> >>>> --- a/hw/qxl.c
> >>>> +++ b/hw/qxl.c
> >>>> @@ -80,9 +80,7 @@
> >>>>  
> >>>>  #define QXL_MODE_EX(x_res, y_res)                 \
> >>>>      QXL_MODE_16_32(x_res, y_res, 0),              \
> >>>> -    QXL_MODE_16_32(y_res, x_res, 1),              \
> >>>> -    QXL_MODE_16_32(x_res, y_res, 2),              \
> >>>> -    QXL_MODE_16_32(y_res, x_res, 3)
> >>>> +    QXL_MODE_16_32(x_res, y_res, 1)
> >>>>  
> >>>>  static QXLMode qxl_modes[] = {
> >>>>      QXL_MODE_EX(640, 480),
> >>>> @@ -306,10 +304,13 @@ static inline uint32_t msb_mask(uint32_t
> >>>> val)
> >>>>  
> >>>>  static ram_addr_t qxl_rom_size(void)
> >>>>  {
> >>>> -    uint32_t rom_size = sizeof(QXLRom) + sizeof(QXLModes) +
> >>>> sizeof(qxl_modes);
> >>>> +    uint32_t required_rom_size = sizeof(QXLRom) +
> >>>> sizeof(QXLModes) +
> >>>> +                                 sizeof(qxl_modes);
> >>>> +    uint32_t rom_size = 8192; /* two pages */
> >>>>  
> >>>> -    rom_size = MAX(rom_size, TARGET_PAGE_SIZE);
> >>>> -    rom_size = msb_mask(rom_size * 2 - 1);
> >>>> +    required_rom_size = MAX(required_rom_size,
> >>>> TARGET_PAGE_SIZE);
> >>>> +    required_rom_size = msb_mask(required_rom_size * 2 - 1);
> >>>> +    assert(required_rom_size <= rom_size);
> >>>>      return rom_size;
> >>>>  }
> >>>>  
> >>>>
> >>>
> >>> This breaks migration for me with -M pc-1.0 at least.
> >>>
> >>> Before this commit:
> >>>
> >>> ./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice
> >>> port=5905
> >>> -monitor stdio
> >>>> migrate "exec: cat > foo.img"
> >>>> quit
> >>>
> >>> After this commit:
> >>>
> >>> ./x86_64-softmmu/qemu-system-x86_64 -M pc-1.0 -vga qxl -spice
> >>> port=5905
> >>> -monitor stdio -incoming "exec: cat foo.img"
> >>> qemu: warning: error while loading state for instance 0x0 of
> >>> device
> >>> 'ram'
> >>> cat: write error: Broken pipe
> >>> load of migration failed
> >>>
> >>
> >> Sorry, I realize that may seem a bit artificial, so just to
> >> clarify a
> >> bit. If
> >> I start on v1.3.1, then cherry-pick this commit, it breaks
> >> migration
> >> using the
> >> commands above. I also tried v1.3.1 to git head, reverting the
> >> seabios rebase
> >> that caused other migration issues, and got the same error (is
> >> there
> >> any way
> >> to get better error reporting out of migration failures?).
> > 
> > Migration error reporting sucks. Did you create foo.img with
> > pre-patched qemu and then try to read it with post patched, or
> > both are using the same version?
> 
> I tried:
> 
> foo.img generated with stock qemu 1.3.1, migration fails when cherry
> picking
> the patch.
> 
> foo.img generated with stock qemu 1.3.1, migration fails with git
> head, still
> fails after reverting the seabios 1.7.2 update which supposedly
> causes other
> migrate issues (commit 3588185b8396eb97fd9efd41c2b97775465f67c4)

So this is by design - that patch changes the rom size for revision 4 (which is the default for qemu 1.3.1) from 16384 bytes to 8192. This breaks migration from previously created revision 4 images... But it unbreaks migration from other revisions. Different things that could be done: do a special migration hook to copy from src 16384 byte rom bar to dst 8192 byte rom bar. Gerd, what do you think?

> 
> - Cole
> 
> 
>
Gerd Hoffmann - Feb. 8, 2013, 9:36 a.m.
Hi,

>> foo.img generated with stock qemu 1.3.1, migration fails with git
>> head, still
>> fails after reverting the seabios 1.7.2 update which supposedly
>> causes other
>> migrate issues (commit 3588185b8396eb97fd9efd41c2b97775465f67c4)
> 
> So this is by design - that patch changes the rom size for revision 4 (which is the default for qemu 1.3.1) from 16384 bytes to 8192. This breaks migration from previously created revision 4 images... But it unbreaks migration from other revisions. Different things that could be done: do a special migration hook to copy from src 16384 byte rom bar to dst 8192 byte rom bar. Gerd, what do you think?

So the change making the rpm 16k was before 1.3, the fix thereafter, and
now we have migration 1.2 -> 1.3 broken (maybe masked by -M 1.2
defaulting to qxl rev 3), 1.3 -> 1.4 broken and 1.2 -> 1.4 working, correct?

So a compat property changing the rom size to 16k for pc-1.3 should do
the trick I think.

cheers,
  Gerd
Cole Robinson - Feb. 8, 2013, 5:11 p.m.
On 02/08/2013 04:36 AM, Gerd Hoffmann wrote:
>   Hi,
> 
>>> foo.img generated with stock qemu 1.3.1, migration fails with git
>>> head, still
>>> fails after reverting the seabios 1.7.2 update which supposedly
>>> causes other
>>> migrate issues (commit 3588185b8396eb97fd9efd41c2b97775465f67c4)
>>
>> So this is by design - that patch changes the rom size for revision 4 (which is the default for qemu 1.3.1) from 16384 bytes to 8192. This breaks migration from previously created revision 4 images... But it unbreaks migration from other revisions. Different things that could be done: do a special migration hook to copy from src 16384 byte rom bar to dst 8192 byte rom bar. Gerd, what do you think?
> 
> So the change making the rpm 16k was before 1.3, the fix thereafter, and
> now we have migration 1.2 -> 1.3 broken (maybe masked by -M 1.2
> defaulting to qxl rev 3), 1.3 -> 1.4 broken and 1.2 -> 1.4 working, correct?
> 
> So a compat property changing the rom size to 16k for pc-1.3 should do
> the trick I think.
> 

The example commands I gave were using -M pc-1.0, I can't tell if you're
solution factors that in.

Thanks,
Cole
Cole Robinson - Feb. 19, 2013, 10:15 p.m.
On 02/08/2013 04:36 AM, Gerd Hoffmann wrote:
>   Hi,
> 
>>> foo.img generated with stock qemu 1.3.1, migration fails with git
>>> head, still
>>> fails after reverting the seabios 1.7.2 update which supposedly
>>> causes other
>>> migrate issues (commit 3588185b8396eb97fd9efd41c2b97775465f67c4)
>>
>> So this is by design - that patch changes the rom size for revision 4 (which is the default for qemu 1.3.1) from 16384 bytes to 8192. This breaks migration from previously created revision 4 images... But it unbreaks migration from other revisions. Different things that could be done: do a special migration hook to copy from src 16384 byte rom bar to dst 8192 byte rom bar. Gerd, what do you think?
> 
> So the change making the rpm 16k was before 1.3, the fix thereafter, and
> now we have migration 1.2 -> 1.3 broken (maybe masked by -M 1.2
> defaulting to qxl rev 3), 1.3 -> 1.4 broken and 1.2 -> 1.4 working, correct?
> 
> So a compat property changing the rom size to 16k for pc-1.3 should do
> the trick I think.
> 

I took a stab at this, unfortunately the situation isn't that simple. qxl
rom_size is actually dependent on the spice version we build against:

stock qemu 1.2 built on F17 against spice-0.10.1:
sizeof(QXLRom)=72 rom_size=8192

stock qemu 1.2 built on F18 against spice-0.12.2:
sizeof(QXLRom)=1160 rom_size=16834

Maybe a more complex solution like Alon proposed up thread is needed.

Thanks,
Cole

Patch

diff --git a/hw/qxl.c b/hw/qxl.c
index 0d81816..a125e29 100644
--- a/hw/qxl.c
+++ b/hw/qxl.c
@@ -80,9 +80,7 @@ 
 
 #define QXL_MODE_EX(x_res, y_res)                 \
     QXL_MODE_16_32(x_res, y_res, 0),              \
-    QXL_MODE_16_32(y_res, x_res, 1),              \
-    QXL_MODE_16_32(x_res, y_res, 2),              \
-    QXL_MODE_16_32(y_res, x_res, 3)
+    QXL_MODE_16_32(x_res, y_res, 1)
 
 static QXLMode qxl_modes[] = {
     QXL_MODE_EX(640, 480),
@@ -306,10 +304,13 @@  static inline uint32_t msb_mask(uint32_t val)
 
 static ram_addr_t qxl_rom_size(void)
 {
-    uint32_t rom_size = sizeof(QXLRom) + sizeof(QXLModes) + sizeof(qxl_modes);
+    uint32_t required_rom_size = sizeof(QXLRom) + sizeof(QXLModes) +
+                                 sizeof(qxl_modes);
+    uint32_t rom_size = 8192; /* two pages */
 
-    rom_size = MAX(rom_size, TARGET_PAGE_SIZE);
-    rom_size = msb_mask(rom_size * 2 - 1);
+    required_rom_size = MAX(required_rom_size, TARGET_PAGE_SIZE);
+    required_rom_size = msb_mask(required_rom_size * 2 - 1);
+    assert(required_rom_size <= rom_size);
     return rom_size;
 }