Re: has the i915 "black screen" boot issue returned? [BISECTED]

Re: has the i915 "black screen" boot issue returned? [BISECTED]

[Robert P. J. Day]

On Fri, 28 Jan 2011, Chris Wilson wrote:

> On Fri, 28 Jan 2011 08:53:59 -0500 (EST), "Robert P. J. Day" <rpjday@crashcourse.ca> wrote:
> > so the culprit appears to be:
> >
> > b705120e4198315f4ae043de06c62f65e0851fd3 is the first bad commit
> > commit b705120e4198315f4ae043de06c62f65e0851fd3
> > Author: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
> > Date:   Sun Jan 23 18:17:17 2011 +0000
> >
> >     drm/i915: Use consistent mappings for OpRegion between ACPI and i915
> >
> >     The opregion is a shared memory region between ACPI and the graphics
> >     driver. As the ACPI mapping has been changed to cachable in commit
> >     6d5bbf00d251cc73223a71422d69e069dc2e0b8d, mapping the intel opregion
> >     non-cachable now fails. As no bus-master hardware is involved in the
> >     opregion, cachable map should do no harm.
> >
> >     Tested on a Fujitsu Lifebook P8010.
> >
> >     Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
> >     [ickle: convert to acpi_os_ioremap for consistency]
> >     Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> >
> >
> >   thoughts?  once again, the salient output from "lspci -v":
>
> Indeed looks like using ioremap_cache is not as safe as was assumed. Does

  *sigh*.  there was, in fact, an "ioremap_error" message displayed
*very* early in the boot sequence, but it was generated even for
successful boots so i never paid it any mind.  in hindsight, might
have been useful to have mentioned it.

> diff --git a/include/linux/acpi_io.h b/include/linux/acpi_io.h
> index 7180013..42108ab 100644
> --- a/include/linux/acpi_io.h
> +++ b/include/linux/acpi_io.h
> @@ -7,7 +7,7 @@
>  static inline void __iomem *acpi_os_ioremap(acpi_physical_address phys,
>                                             acpi_size size)
>  {
> -       return ioremap_cache(phys, size);
> +       return ioremap_wc(phys, size);
>  }
>
>  int acpi_os_map_generic_address(struct acpi_generic_address *addr);

  ok, i'll make this single change and report back shortly.

rday

-- 

========================================================================
Robert P. J. Day                               Waterloo, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================

Re: has the i915 "black screen" boot issue returned? [BISECTED]

[Robert P. J. Day]

On Fri, 28 Jan 2011, Chris Wilson wrote:

> On Fri, 28 Jan 2011 08:53:59 -0500 (EST), "Robert P. J. Day" <rpjday@crashcourse.ca> wrote:
> > so the culprit appears to be:
> >
> > b705120e4198315f4ae043de06c62f65e0851fd3 is the first bad commit
> > commit b705120e4198315f4ae043de06c62f65e0851fd3
> > Author: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
> > Date:   Sun Jan 23 18:17:17 2011 +0000
> >
> >     drm/i915: Use consistent mappings for OpRegion between ACPI and i915
> >
> >     The opregion is a shared memory region between ACPI and the graphics
> >     driver. As the ACPI mapping has been changed to cachable in commit
> >     6d5bbf00d251cc73223a71422d69e069dc2e0b8d, mapping the intel opregion
> >     non-cachable now fails. As no bus-master hardware is involved in the
> >     opregion, cachable map should do no harm.
> >
> >     Tested on a Fujitsu Lifebook P8010.
> >
> >     Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
> >     [ickle: convert to acpi_os_ioremap for consistency]
> >     Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> >
> >
> >   thoughts?  once again, the salient output from "lspci -v":
>
> Indeed looks like using ioremap_cache is not as safe as was assumed. Does
>
> diff --git a/include/linux/acpi_io.h b/include/linux/acpi_io.h
> index 7180013..42108ab 100644
> --- a/include/linux/acpi_io.h
> +++ b/include/linux/acpi_io.h
> @@ -7,7 +7,7 @@
>  static inline void __iomem *acpi_os_ioremap(acpi_physical_address phys,
>                                             acpi_size size)
>  {
> -       return ioremap_cache(phys, size);
> +       return ioremap_wc(phys, size);
>  }
>
>  int acpi_os_map_generic_address(struct acpi_generic_address *addr);

  that didn't appear to make a difference.  same black screen issue.
BTW, here's an excerpt from /var/log/dmesg for a *successful* boot (i
just rebooted to 8a327f23):

...
[   24.037114] intel ips 0000:00:1f.6: failed to get i915 symbols, graphics turbo disabled
...
[   27.155660] i915 0000:00:02.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16
[   27.155665] i915 0000:00:02.0: setting latency timer to 64
[   27.217407] mtrr: no more MTRRs available
[   27.217411] [drm] MTRR allocation failed.  Graphics performance may suffer.
[   27.217848] ioremap error for 0xb3752000-0xb3755000, requested 0x10, got 0x0
[   27.217933] i915 0000:00:02.0: irq 42 for MSI/MSI-X
[   27.217938] [drm] Supports vblank timestamp caching Rev 1 (10.10.2010).
[   27.217940] [drm] Driver supports precise vblank timestamp query.
...

if any of that is of any use.

> fix your boot issue or do we need to go back to using uncached:
>
> +       return ioremap(phys, size);

  is that the next change you want me to try?

rday

-- 

========================================================================
Robert P. J. Day                               Waterloo, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================

Re: has the i915 "black screen" boot issue returned? [BISECTED]

[Chris Wilson]

On Fri, 28 Jan 2011 09:32:04 -0500 (EST), "Robert P. J. Day" <rpjday@crashcourse.ca> wrote:
> > fix your boot issue or do we need to go back to using uncached:
> >
> > +       return ioremap(phys, size);
> 
>   is that the next change you want me to try?

Yes. (Replacing the current return ioremap_*(phys, size).)
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre

Re: has the i915 "black screen" boot issue returned? [BISECTED]

[Robert P. J. Day]

On Fri, 28 Jan 2011, Chris Wilson wrote:

> On Fri, 28 Jan 2011 09:32:04 -0500 (EST), "Robert P. J. Day" <rpjday@crashcourse.ca> wrote:
> > > fix your boot issue or do we need to go back to using uncached:
> > >
> > > +       return ioremap(phys, size);
> >
> >   is that the next change you want me to try?
>
> Yes. (Replacing the current return ioremap_*(phys, size).)

  sadly, no change -- still black screen.  again, rebooted
successfully under commit 8a327f23.  just to be clear, here's "git
diff":

$ git diff
diff --git a/include/linux/acpi_io.h b/include/linux/acpi_io.h
index 7180013..e035f3c 100644
--- a/include/linux/acpi_io.h
+++ b/include/linux/acpi_io.h
@@ -7,7 +7,7 @@
 static inline void __iomem *acpi_os_ioremap(acpi_physical_address phys,
                                            acpi_size size)
 {
-       return ioremap_cache(phys, size);
+       return ioremap(phys, size);
 }

 int acpi_os_map_generic_address(struct acpi_generic_address *addr);




rday

-- 

========================================================================
Robert P. J. Day                               Waterloo, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================

Re: has the i915 "black screen" boot issue returned? [BISECTED]

[Chris Wilson]

On Fri, 28 Jan 2011 09:51:01 -0500 (EST), "Robert P. J. Day" <rpjday@crashcourse.ca> wrote:
>   sadly, no change -- still black screen.  again, rebooted
> successfully under commit 8a327f23.  just to be clear, here's "git
> diff":
> 
> $ git diff
> diff --git a/include/linux/acpi_io.h b/include/linux/acpi_io.h
> index 7180013..e035f3c 100644
> --- a/include/linux/acpi_io.h
> +++ b/include/linux/acpi_io.h
> @@ -7,7 +7,7 @@
>  static inline void __iomem *acpi_os_ioremap(acpi_physical_address phys,
>                                             acpi_size size)
>  {
> -       return ioremap_cache(phys, size);
> +       return ioremap(phys, size);
>  }

Ok, that implies the new mapping is fine and not the cause of the issue.

Instead you have some OpRegion related regression hidden until till now
because the conflicting mapping disabled it for i915.

That would be easy to test by returning early in intel_opregion_setup():

diff --git a/drivers/gpu/drm/i915/intel_opregion.c
b/drivers/gpu/drm/i915/intel_
index 9efccb9..8c93201 100644
--- a/drivers/gpu/drm/i915/intel_opregion.c
+++ b/drivers/gpu/drm/i915/intel_opregion.c
@@ -470,6 +470,8 @@ int intel_opregion_setup(struct drm_device *dev)
        u32 asls, mboxes;
        int err = 0;
 
+       return -ENOTSUPP;
+

        pci_read_config_dword(dev->pdev, PCI_ASLS, &asls);
        DRM_DEBUG_DRIVER("graphic opregion physical addr: 0x%x\n", asls);
        if (asls == 0) {

-- 
Chris Wilson, Intel Open Source Technology Centre

Re: has the i915 "black screen" boot issue returned? [BISECTED]

[Robert P. J. Day]

On Fri, 28 Jan 2011, Chris Wilson wrote:

> On Fri, 28 Jan 2011 09:51:01 -0500 (EST), "Robert P. J. Day" <rpjday@crashcourse.ca> wrote:
> >   sadly, no change -- still black screen.  again, rebooted
> > successfully under commit 8a327f23.  just to be clear, here's "git
> > diff":
> >
> > $ git diff
> > diff --git a/include/linux/acpi_io.h b/include/linux/acpi_io.h
> > index 7180013..e035f3c 100644
> > --- a/include/linux/acpi_io.h
> > +++ b/include/linux/acpi_io.h
> > @@ -7,7 +7,7 @@
> >  static inline void __iomem *acpi_os_ioremap(acpi_physical_address phys,
> >                                             acpi_size size)
> >  {
> > -       return ioremap_cache(phys, size);
> > +       return ioremap(phys, size);
> >  }
>
> Ok, that implies the new mapping is fine and not the cause of the issue.
>
> Instead you have some OpRegion related regression hidden until till now
> because the conflicting mapping disabled it for i915.
>
> That would be easy to test by returning early in intel_opregion_setup():
>
> diff --git a/drivers/gpu/drm/i915/intel_opregion.c
> b/drivers/gpu/drm/i915/intel_
> index 9efccb9..8c93201 100644
> --- a/drivers/gpu/drm/i915/intel_opregion.c
> +++ b/drivers/gpu/drm/i915/intel_opregion.c
> @@ -470,6 +470,8 @@ int intel_opregion_setup(struct drm_device *dev)
>         u32 asls, mboxes;
>         int err = 0;
>
> +       return -ENOTSUPP;
> +
>
>         pci_read_config_dword(dev->pdev, PCI_ASLS, &asls);
>         DRM_DEBUG_DRIVER("graphic opregion physical addr: 0x%x\n", asls);
>         if (asls == 0) {
>

so you want me to revert to a stock b705120e before doing the above?

rday

-- 

========================================================================
Robert P. J. Day                               Waterloo, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================

Re: has the i915 "black screen" boot issue returned? [BISECTED]

[Rafael J. Wysocki]

On Friday, January 28, 2011, Robert P. J. Day wrote:
> On Fri, 28 Jan 2011, Chris Wilson wrote:
> 
> > On Fri, 28 Jan 2011 09:51:01 -0500 (EST), "Robert P. J. Day" <rpjday@crashcourse.ca> wrote:
> > >   sadly, no change -- still black screen.  again, rebooted
> > > successfully under commit 8a327f23.  just to be clear, here's "git
> > > diff":
> > >
> > > $ git diff
> > > diff --git a/include/linux/acpi_io.h b/include/linux/acpi_io.h
> > > index 7180013..e035f3c 100644
> > > --- a/include/linux/acpi_io.h
> > > +++ b/include/linux/acpi_io.h
> > > @@ -7,7 +7,7 @@
> > >  static inline void __iomem *acpi_os_ioremap(acpi_physical_address phys,
> > >                                             acpi_size size)
> > >  {
> > > -       return ioremap_cache(phys, size);
> > > +       return ioremap(phys, size);
> > >  }
> >
> > Ok, that implies the new mapping is fine and not the cause of the issue.
> >
> > Instead you have some OpRegion related regression hidden until till now
> > because the conflicting mapping disabled it for i915.
> >
> > That would be easy to test by returning early in intel_opregion_setup():
> >
> > diff --git a/drivers/gpu/drm/i915/intel_opregion.c
> > b/drivers/gpu/drm/i915/intel_
> > index 9efccb9..8c93201 100644
> > --- a/drivers/gpu/drm/i915/intel_opregion.c
> > +++ b/drivers/gpu/drm/i915/intel_opregion.c
> > @@ -470,6 +470,8 @@ int intel_opregion_setup(struct drm_device *dev)
> >         u32 asls, mboxes;
> >         int err = 0;
> >
> > +       return -ENOTSUPP;
> > +
> >
> >         pci_read_config_dword(dev->pdev, PCI_ASLS, &asls);
> >         DRM_DEBUG_DRIVER("graphic opregion physical addr: 0x%x\n", asls);
> >         if (asls == 0) {
> >
> 
> so you want me to revert to a stock b705120e before doing the above?

Alternatively, you could take the vanilla Linus' tree and replace
ioremap_cache() with ioremap() in include/linux/acpi_io.h .  Please try that
and see if it makes a difference.

Thanks,
Rafael

Re: has the i915 "black screen" boot issue returned? [BISECTED]

[Robert P. J. Day]

On Fri, 28 Jan 2011, Rafael J. Wysocki wrote:

> On Friday, January 28, 2011, Robert P. J. Day wrote:
> > On Fri, 28 Jan 2011, Chris Wilson wrote:
> >
> > > On Fri, 28 Jan 2011 09:51:01 -0500 (EST), "Robert P. J. Day" <rpjday@crashcourse.ca> wrote:
> > > >   sadly, no change -- still black screen.  again, rebooted
> > > > successfully under commit 8a327f23.  just to be clear, here's "git
> > > > diff":
> > > >
> > > > $ git diff
> > > > diff --git a/include/linux/acpi_io.h b/include/linux/acpi_io.h
> > > > index 7180013..e035f3c 100644
> > > > --- a/include/linux/acpi_io.h
> > > > +++ b/include/linux/acpi_io.h
> > > > @@ -7,7 +7,7 @@
> > > >  static inline void __iomem *acpi_os_ioremap(acpi_physical_address phys,
> > > >                                             acpi_size size)
> > > >  {
> > > > -       return ioremap_cache(phys, size);
> > > > +       return ioremap(phys, size);
> > > >  }
> > >
> > > Ok, that implies the new mapping is fine and not the cause of the issue.
> > >
> > > Instead you have some OpRegion related regression hidden until till now
> > > because the conflicting mapping disabled it for i915.
> > >
> > > That would be easy to test by returning early in intel_opregion_setup():
> > >
> > > diff --git a/drivers/gpu/drm/i915/intel_opregion.c
> > > b/drivers/gpu/drm/i915/intel_
> > > index 9efccb9..8c93201 100644
> > > --- a/drivers/gpu/drm/i915/intel_opregion.c
> > > +++ b/drivers/gpu/drm/i915/intel_opregion.c
> > > @@ -470,6 +470,8 @@ int intel_opregion_setup(struct drm_device *dev)
> > >         u32 asls, mboxes;
> > >         int err = 0;
> > >
> > > +       return -ENOTSUPP;
> > > +
> > >
> > >         pci_read_config_dword(dev->pdev, PCI_ASLS, &asls);
> > >         DRM_DEBUG_DRIVER("graphic opregion physical addr: 0x%x\n", asls);
> > >         if (asls == 0) {
> > >
> >
> > so you want me to revert to a stock b705120e before doing the above?
>
> Alternatively, you could take the vanilla Linus' tree and replace
> ioremap_cache() with ioremap() in include/linux/acpi_io.h .  Please try that
> and see if it makes a difference.

  been there, done that, that's all so ... this morning.  i continued
bisecting and playing and eventually verified the following
workaround for an earlier working version of the kernel:

$ git diff
diff --git a/drivers/gpu/drm/i915/intel_opregion.c b/drivers/gpu/drm/i915/intel_opregion.c
index 64fd644..28adc6d 100644
--- a/drivers/gpu/drm/i915/intel_opregion.c
+++ b/drivers/gpu/drm/i915/intel_opregion.c
@@ -495,6 +495,8 @@ int intel_opregion_setup(struct drm_device *dev)
                opregion->acpi = base + OPREGION_ACPI_OFFSET;
        }

+return 0; // rday
+
        if (mboxes & MBOX_SWSCI) {
                DRM_DEBUG_DRIVER("SWSCI supported\n");
                opregion->swsci = base + OPREGION_SWSCI_OFFSET;


  this was based on a suggestion by chris wilson to deactivate the
"mboxes" tests in that single source file and see what happened.
adding the "return 0" above suddenly gave me a properly booting
kernel.  i applied the same diff to the current kernel and it's
building as we speak.

rday

p.s.  having that "return 0" *after* that test but before the next one
still gave me a black screen.

-- 

========================================================================
Robert P. J. Day                               Waterloo, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================

Re: has the i915 "black screen" boot issue returned? [BISECTED]

[Robert P. J. Day]

On Fri, 28 Jan 2011, Rafael J. Wysocki wrote:

> Alternatively, you could take the vanilla Linus' tree and replace
> ioremap_cache() with ioremap() in include/linux/acpi_io.h .  Please
> try that and see if it makes a difference.

  as a quick followup, i applied the following simple patch:

$ git diff
diff --git a/drivers/gpu/drm/i915/intel_opregion.c b/drivers/gpu/drm/i915/intel_opregion.c
index 64fd644..28adc6d 100644
--- a/drivers/gpu/drm/i915/intel_opregion.c
+++ b/drivers/gpu/drm/i915/intel_opregion.c
@@ -495,6 +495,8 @@ int intel_opregion_setup(struct drm_device *dev)
                opregion->acpi = base + OPREGION_ACPI_OFFSET;
        }

+return 0; // rday
+
        if (mboxes & MBOX_SWSCI) {
                DRM_DEBUG_DRIVER("SWSCI supported\n");
                opregion->swsci = base + OPREGION_SWSCI_OFFSET;


to the latest tree, and it gave me a booting kernel, no black screen
issue.  i will not pretend to understand why that fixes the problem
but it does.

rday

-- 

========================================================================
Robert P. J. Day                               Waterloo, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================