xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Zhang Le]

Hi, all,

I have tried xorg-server-1.5.2 on loongson 2f recently.
But I found it doesn't work.
It's mainly because of this change:
http://www.x.org/wiki/PciReworkProposal

In short:
"Rather than duplicating the efforts of kernel developers, X.org needs to use the
interfaces provided by the kernel as much as possible."

I have read some code of libpciaccess, the new library utilizing kernel function
to access pci bus. It will try to mmap this file:
/sys/bus/pci/devices/0000:0x:xx.x/resource0
(replace x with any digit appropriate)
Note there is a 0 at the end of the file name. This file's permission is 600.

However, I found on my loongson system, there is only 
/sys/bus/pci/devices/0000:0x:xx.x/resource
Note there is no 0 at the end.

Then I tried to read kernel code. I found it seems that for mips linux to have
this file, HAVE_PCI_MMAP must be defined. However, it is currently not defined.

Since I am not familiar with PCI, yet.
So could someone please shed some light on this?
Why HAVE_PCI_MMAP is not defined?

Thanks in advance!

Zhang, Le

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Arnaud Patard]

Zhang Le <r0bertz@gentoo.org> writes:

> Hi, all,

Hi,

> Then I tried to read kernel code. I found it seems that for mips linux to have
> this file, HAVE_PCI_MMAP must be defined. However, it is currently not defined.
>
> Since I am not familiar with PCI, yet.
> So could someone please shed some light on this?
> Why HAVE_PCI_MMAP is not defined?

HAVE_PCI_MMAP must be defined when you have a pci_mmap_page_range()
function (see Documentation/filesystems/sysfs-pci.txt) and we don't have
a pci_mmap_page_range() on mips.

Hope that helps you.

Arnaud

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Ralf Baechle]

On Fri, Dec 05, 2008 at 06:59:35PM +0100, Arnaud Patard wrote:

Arnaud,

> > Then I tried to read kernel code. I found it seems that for mips linux to have
> > this file, HAVE_PCI_MMAP must be defined. However, it is currently not defined.
> >
> > Since I am not familiar with PCI, yet.
> > So could someone please shed some light on this?
> > Why HAVE_PCI_MMAP is not defined?
> 
> HAVE_PCI_MMAP must be defined when you have a pci_mmap_page_range()
> function (see Documentation/filesystems/sysfs-pci.txt) and we don't have
> a pci_mmap_page_range() on mips.

Correct - but if the code is not present the kernel does not use an
alternative implementation.  I just looked into implementing that and it
seems nothing that I could do quickly to still get it into 2.6.28, sorry ...

  Ralf

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Ralf Baechle]

On Fri, Dec 05, 2008 at 11:43:42PM +0800, Zhang Le wrote:

> I have tried xorg-server-1.5.2 on loongson 2f recently.
> But I found it doesn't work.
> It's mainly because of this change:
> http://www.x.org/wiki/PciReworkProposal
> 
> In short:
> "Rather than duplicating the efforts of kernel developers, X.org needs to use the
> interfaces provided by the kernel as much as possible."
> 
> I have read some code of libpciaccess, the new library utilizing kernel function
> to access pci bus. It will try to mmap this file:
> /sys/bus/pci/devices/0000:0x:xx.x/resource0
> (replace x with any digit appropriate)
> Note there is a 0 at the end of the file name. This file's permission is 600.
> 
> However, I found on my loongson system, there is only 
> /sys/bus/pci/devices/0000:0x:xx.x/resource
> Note there is no 0 at the end.
> 
> Then I tried to read kernel code. I found it seems that for mips linux to have
> this file, HAVE_PCI_MMAP must be defined. However, it is currently not defined.
> 
> Since I am not familiar with PCI, yet.
> So could someone please shed some light on this?
> Why HAVE_PCI_MMAP is not defined?

Here is a quick'n'dirty solution which I've not tested beyond just
compiling.  It should work but performance will be bad.  Either way, I'm
interested in a test report with X.

  Ralf

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

diff --git a/arch/mips/include/asm/pci.h b/arch/mips/include/asm/pci.h
index 5510c53..053e463 100644
--- a/arch/mips/include/asm/pci.h
+++ b/arch/mips/include/asm/pci.h
@@ -79,6 +79,11 @@ static inline void pcibios_penalize_isa_irq(int irq, int active)
 	/* We don't do dynamic PCI IRQ allocation */
 }
 
+#define HAVE_PCI_MMAP
+
+extern int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
+	enum pci_mmap_state mmap_state, int write_combine);
+
 /*
  * Dynamic DMA mapping stuff.
  * MIPS has everything mapped statically.
diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
index a377e9d..9233193 100644
--- a/arch/mips/pci/pci.c
+++ b/arch/mips/pci/pci.c
@@ -354,6 +354,22 @@ EXPORT_SYMBOL(PCIBIOS_MIN_IO);
 EXPORT_SYMBOL(PCIBIOS_MIN_MEM);
 #endif
 
+int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
+			enum pci_mmap_state mmap_state, int write_combine)
+{
+	unsigned long prot;
+
+	/*
+	 * Ignore write-combine; for now only return uncached mappings.
+	 */
+	prot = pgprot_val(vma->vm_page_prot);
+	prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED;
+	vma->vm_page_prot = __pgprot(prot);
+
+	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
+		vma->vm_end - vma->vm_start, vma->vm_page_prot);
+}
+
 char * (*pcibios_plat_setup)(char *str) __devinitdata;
 
 char *__devinit pcibios_setup(char *str)

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Zhang Le]

On 10:20 Sat 06 Dec     , Ralf Baechle wrote:
> On Fri, Dec 05, 2008 at 11:43:42PM +0800, Zhang Le wrote:
> 
> > I have tried xorg-server-1.5.2 on loongson 2f recently.
> > But I found it doesn't work.
> > It's mainly because of this change:
> > http://www.x.org/wiki/PciReworkProposal
> > 
> > In short:
> > "Rather than duplicating the efforts of kernel developers, X.org needs to use the
> > interfaces provided by the kernel as much as possible."
> > 
> > I have read some code of libpciaccess, the new library utilizing kernel function
> > to access pci bus. It will try to mmap this file:
> > /sys/bus/pci/devices/0000:0x:xx.x/resource0
> > (replace x with any digit appropriate)
> > Note there is a 0 at the end of the file name. This file's permission is 600.
> > 
> > However, I found on my loongson system, there is only 
> > /sys/bus/pci/devices/0000:0x:xx.x/resource
> > Note there is no 0 at the end.
> > 
> > Then I tried to read kernel code. I found it seems that for mips linux to have
> > this file, HAVE_PCI_MMAP must be defined. However, it is currently not defined.
> > 
> > Since I am not familiar with PCI, yet.
> > So could someone please shed some light on this?
> > Why HAVE_PCI_MMAP is not defined?
> 
> Here is a quick'n'dirty solution which I've not tested beyond just
> compiling.  It should work but performance will be bad.  Either way, I'm
> interested in a test report with X.

Thanks, Ralf.
I will test it.

> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
> 
> diff --git a/arch/mips/include/asm/pci.h b/arch/mips/include/asm/pci.h
> index 5510c53..053e463 100644
> --- a/arch/mips/include/asm/pci.h
> +++ b/arch/mips/include/asm/pci.h
> @@ -79,6 +79,11 @@ static inline void pcibios_penalize_isa_irq(int irq, int active)
>  	/* We don't do dynamic PCI IRQ allocation */
>  }
>  
> +#define HAVE_PCI_MMAP
> +
> +extern int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
> +	enum pci_mmap_state mmap_state, int write_combine);
> +
>  /*
>   * Dynamic DMA mapping stuff.
>   * MIPS has everything mapped statically.
> diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
> index a377e9d..9233193 100644
> --- a/arch/mips/pci/pci.c
> +++ b/arch/mips/pci/pci.c
> @@ -354,6 +354,22 @@ EXPORT_SYMBOL(PCIBIOS_MIN_IO);
>  EXPORT_SYMBOL(PCIBIOS_MIN_MEM);
>  #endif
>  
> +int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
> +			enum pci_mmap_state mmap_state, int write_combine)
> +{
> +	unsigned long prot;
> +
> +	/*
> +	 * Ignore write-combine; for now only return uncached mappings.
> +	 */
> +	prot = pgprot_val(vma->vm_page_prot);
> +	prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED;
> +	vma->vm_page_prot = __pgprot(prot);
> +
> +	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
> +		vma->vm_end - vma->vm_start, vma->vm_page_prot);
> +}
> +
>  char * (*pcibios_plat_setup)(char *str) __devinitdata;
>  
>  char *__devinit pcibios_setup(char *str)
> 

Zhang, Le

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Zhang Le]

On 00:47 Sun 07 Dec     , Zhang Le wrote:
> On 10:20 Sat 06 Dec     , Ralf Baechle wrote:
> > On Fri, Dec 05, 2008 at 11:43:42PM +0800, Zhang Le wrote:
> > > Then I tried to read kernel code. I found it seems that for mips linux to have
> > > this file, HAVE_PCI_MMAP must be defined. However, it is currently not defined.
> > > 
> > > Since I am not familiar with PCI, yet.
> > > So could someone please shed some light on this?
> > > Why HAVE_PCI_MMAP is not defined?
> > 
> > Here is a quick'n'dirty solution which I've not tested beyond just
> > compiling.  It should work but performance will be bad.  Either way, I'm
> > interested in a test report with X.

Hi, Ralf,

I have build a new kernel with this patch.
Now I have the resource{0,1} file with permission 600.
However, X is not working yet, since there is an undefined reference in sis's
X driver.

I will try to fix it and report back.

Zhang, Le

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Zhang Le]

On 01:22 Wed 10 Dec     , Zhang Le wrote:
> On 00:47 Sun 07 Dec     , Zhang Le wrote:
> > On 10:20 Sat 06 Dec     , Ralf Baechle wrote:
> > > On Fri, Dec 05, 2008 at 11:43:42PM +0800, Zhang Le wrote:
> > > > Then I tried to read kernel code. I found it seems that for mips linux to have
> > > > this file, HAVE_PCI_MMAP must be defined. However, it is currently not defined.
> > > > 
> > > > Since I am not familiar with PCI, yet.
> > > > So could someone please shed some light on this?
> > > > Why HAVE_PCI_MMAP is not defined?
> > > 
> > > Here is a quick'n'dirty solution which I've not tested beyond just
> > > compiling.  It should work but performance will be bad.  Either way, I'm
> > > interested in a test report with X.
> 
> Hi, Ralf,
> 
> I have build a new kernel with this patch.
> Now I have the resource{0,1} file with permission 600.
> However, X is not working yet, since there is an undefined reference in sis's
> X driver.

It is working now! Just forget to load "vbe" module.
Thanks!

Zhang, Le

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Ralf Baechle]

On Wed, Dec 10, 2008 at 01:39:08AM +0800, Zhang Le wrote:

> > I have build a new kernel with this patch.
> > Now I have the resource{0,1} file with permission 600.
> > However, X is not working yet, since there is an undefined reference in sis's
> > X driver.
> 
> It is working now! Just forget to load "vbe" module.
> Thanks!

Wonderful, thanks for testing!

Now, my patch doesn't try to handle all the caching modes.  Any mapping
will be uncached which is guaranteed to deliver the lowest possible
performance.  Registers should be mapped uncached but a framebuffer
will usually work best if mapped in uncached accelerated mode.  On
most MIPS implementations the effect of mapping something multiple times
in different caching modes is undefined so that needs to be avoided
which will require some bookkeeping etc.  Too much for 2.6.28 but this
simple solution seems harmless enough to be pushed for 2.6.28 and
-stable.

  Ralf

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Arnaud Patard]

Ralf Baechle <ralf@linux-mips.org> writes:

Hi,

> Here is a quick'n'dirty solution which I've not tested beyond just
> compiling.  It should work but performance will be bad.  Either way, I'm
> interested in a test report with X.
>
>   Ralf
>
> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
>
> diff --git a/arch/mips/include/asm/pci.h b/arch/mips/include/asm/pci.h
> index 5510c53..053e463 100644
> --- a/arch/mips/include/asm/pci.h
> +++ b/arch/mips/include/asm/pci.h
> @@ -79,6 +79,11 @@ static inline void pcibios_penalize_isa_irq(int irq, int active)
>  	/* We don't do dynamic PCI IRQ allocation */
>  }
>  
> +#define HAVE_PCI_MMAP
> +
> +extern int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
> +	enum pci_mmap_state mmap_state, int write_combine);
> +
>  /*
>   * Dynamic DMA mapping stuff.
>   * MIPS has everything mapped statically.
> diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
> index a377e9d..9233193 100644
> --- a/arch/mips/pci/pci.c
> +++ b/arch/mips/pci/pci.c
> @@ -354,6 +354,22 @@ EXPORT_SYMBOL(PCIBIOS_MIN_IO);
>  EXPORT_SYMBOL(PCIBIOS_MIN_MEM);
>  #endif
>  
> +int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
> +			enum pci_mmap_state mmap_state, int write_combine)
> +{
> +	unsigned long prot;
> +
> +	/*
> +	 * Ignore write-combine; for now only return uncached mappings.
> +	 */
> +	prot = pgprot_val(vma->vm_page_prot);
> +	prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED;

What about things like _CACHE_UNCACHED_ACCELERATED ? Is there a way to
use this flag ?


Arnaud

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Ralf Baechle]

On Tue, Dec 09, 2008 at 07:17:13PM +0100, Arnaud Patard wrote:

> What about things like _CACHE_UNCACHED_ACCELERATED ? Is there a way to
> use this flag ?

That's possible but will require some additional care in the code.
Multiple mappings using different cache modes need to be avoided and
also not all processors support _CACHE_UNCACHED_ACCELERATED.

If you know your software is playing nice and your CPU supports
_CACHE_UNCACHED_ACCELERATED, you can hack the mmap function to use
_CACHE_UNCACHED_ACCELERATED if write_combine is set.  Just for now
and for something which I'm planning to push for 2.6.28 I don't want
anything that's more than trivial.

  Ralf

Re: xorg-server-1.5.2 doesn't work because of missing sysfs pci resource files

[Arnaud Patard]

Ralf Baechle <ralf@linux-mips.org> writes:

Hi,

> On Tue, Dec 09, 2008 at 07:17:13PM +0100, Arnaud Patard wrote:
>
>> What about things like _CACHE_UNCACHED_ACCELERATED ? Is there a way to
>> use this flag ?
>
> That's possible but will require some additional care in the code.
> Multiple mappings using different cache modes need to be avoided and
> also not all processors support _CACHE_UNCACHED_ACCELERATED.

well, for that, one can always define a Kconfig entry for that. It would
be usefull for using _CACHE_UNCACHED_ACCELERATED in fb_pgprotect in
asm/fb.h too.

>
> If you know your software is playing nice and your CPU supports
> _CACHE_UNCACHED_ACCELERATED, you can hack the mmap function to use
> _CACHE_UNCACHED_ACCELERATED if write_combine is set.  Just for now
> and for something which I'm planning to push for 2.6.28 I don't want
> anything that's more than trivial.

Ok, no problem. I'm only asking if it will be possible to use it. For
now, I don't have the pci_mmap_page_range requirement on my systems.


Thanks,
Arnaud