[PATCH 0/3] Added uncached accelerated CPU feature

[PATCH 0/3] Added uncached accelerated CPU feature

[Zhang Le]

This patchset adds uncached accelerated CPU feature and handles write_combine
parameter in pci_mmap_page_range function.

Zhang Le (3):
  MIPS: added cpu_has_uncached_accelerated feature
  MIPS: Loongson 2 has cpu_has_uncached_accelerated feature
  MIPS: handle write_combine in pci_mmap_page_range

 arch/mips/include/asm/cpu-features.h               |    4 ++++
 arch/mips/include/asm/cpu.h                        |    1 +
 .../asm/mach-lemote/cpu-feature-overrides.h        |    1 +
 arch/mips/kernel/cpu-probe.c                       |    4 ++--
 arch/mips/pci/pci.c                                |    8 ++++++--
 5 files changed, 14 insertions(+), 4 deletions(-)

[PATCH 1/3] MIPS: added cpu_has_uncached_accelerated feature

[Zhang Le]

Signed-off-by: Zhang Le <r0bertz@gentoo.org>
---
 arch/mips/include/asm/cpu-features.h |    4 ++++
 arch/mips/include/asm/cpu.h          |    1 +
 arch/mips/kernel/cpu-probe.c         |    4 ++--
 3 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/arch/mips/include/asm/cpu-features.h b/arch/mips/include/asm/cpu-features.h
index c0047f8..63a34d3 100644
--- a/arch/mips/include/asm/cpu-features.h
+++ b/arch/mips/include/asm/cpu-features.h
@@ -220,6 +220,10 @@
 #define cpu_has_inclusive_pcaches	(cpu_data[0].options & MIPS_CPU_INCLUSIVE_CACHES)
 #endif
 
+#ifndef cpu_has_uncached_accelerated
+#define cpu_has_uncached_accelerated	(cpu_data[0].options & MIPS_CPU_UNCACHED_ACCELERATED)
+#endif
+
 #ifndef cpu_dcache_line_size
 #define cpu_dcache_line_size()	cpu_data[0].dcache.linesz
 #endif
diff --git a/arch/mips/include/asm/cpu.h b/arch/mips/include/asm/cpu.h
index 3bdc0e3..846477b 100644
--- a/arch/mips/include/asm/cpu.h
+++ b/arch/mips/include/asm/cpu.h
@@ -265,6 +265,7 @@ enum cpu_type_enum {
 #define MIPS_CPU_VINT		0x00080000 /* CPU supports MIPSR2 vectored interrupts */
 #define MIPS_CPU_VEIC		0x00100000 /* CPU supports MIPSR2 external interrupt controller mode */
 #define MIPS_CPU_ULRI		0x00200000 /* CPU has ULRI feature */
+#define MIPS_CPU_UNCACHED_ACCELERATED	0x00400000 /* CPU has uncached accelerated feature */
 
 /*
  * CPU ASE encodings
diff --git a/arch/mips/kernel/cpu-probe.c b/arch/mips/kernel/cpu-probe.c
index b13b8eb..1910ccf 100644
--- a/arch/mips/kernel/cpu-probe.c
+++ b/arch/mips/kernel/cpu-probe.c
@@ -547,7 +547,7 @@ static inline void cpu_probe_legacy(struct cpuinfo_mips *c, unsigned int cpu)
 		c->options = MIPS_CPU_TLB | MIPS_CPU_4K_CACHE | MIPS_CPU_4KEX |
 		             MIPS_CPU_FPU | MIPS_CPU_32FPR |
 			     MIPS_CPU_COUNTER | MIPS_CPU_WATCH |
-		             MIPS_CPU_LLSC;
+		             MIPS_CPU_LLSC | MIPS_CPU_UNCACHED_ACCELERATED;
 		c->tlbsize = 64;
 		break;
 	case PRID_IMP_R12000:
@@ -576,7 +576,7 @@ static inline void cpu_probe_legacy(struct cpuinfo_mips *c, unsigned int cpu)
 		c->isa_level = MIPS_CPU_ISA_III;
 		c->options = R4K_OPTS |
 			     MIPS_CPU_FPU | MIPS_CPU_LLSC |
-			     MIPS_CPU_32FPR;
+			     MIPS_CPU_32FPR | MIPS_CPU_UNCACHED_ACCELERATED;
 		c->tlbsize = 64;
 		break;
 	}
-- 
1.6.2.3

[PATCH 2/3] MIPS: Loongson 2 has cpu_has_uncached_accelerated feature

[Zhang Le]

Signed-off-by: Zhang Le <r0bertz@gentoo.org>
---
 .../asm/mach-lemote/cpu-feature-overrides.h        |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/arch/mips/include/asm/mach-lemote/cpu-feature-overrides.h b/arch/mips/include/asm/mach-lemote/cpu-feature-overrides.h
index 52ec54b..700dd17 100644
--- a/arch/mips/include/asm/mach-lemote/cpu-feature-overrides.h
+++ b/arch/mips/include/asm/mach-lemote/cpu-feature-overrides.h
@@ -50,6 +50,7 @@
 #define cpu_has_tlb		1
 #define cpu_has_tx39_cache	0
 #define cpu_has_userlocal	0
+#define cpu_has_uncached_accelerated	1
 #define cpu_has_vce		0
 #define cpu_has_vtag_icache	0
 #define cpu_has_watch		1
-- 
1.6.2.3

[PATCH 3/3] MIPS: handle write_combine in pci_mmap_page_range

[Zhang Le]

Signed-off-by: Zhang Le <r0bertz@gentoo.org>
---
 arch/mips/pci/pci.c |    8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
index b0eb9e7..4ca53ef 100644
--- a/arch/mips/pci/pci.c
+++ b/arch/mips/pci/pci.c
@@ -346,10 +346,14 @@ int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
 		return -EINVAL;
 
 	/*
-	 * Ignore write-combine; for now only return uncached mappings.
+	 * For write-combine, return uncached accelerated mappings if CPU
+	 * supports; otherwise, return uncached mappings.
 	 */
 	prot = pgprot_val(vma->vm_page_prot);
-	prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED;
+	if (write_combine && cpu_has_uncached_accelerated)
+		prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED_ACCELERATED;
+	else
+		prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED;
 	vma->vm_page_prot = __pgprot(prot);
 
 	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
-- 
1.6.2.3

Re: [PATCH 3/3] MIPS: handle write_combine in pci_mmap_page_range

[yanh]

在 2009-05-08五的 14:30 +0800，Zhang Le写道：
> Signed-off-by: Zhang Le <r0bertz@gentoo.org>
> ---
>  arch/mips/pci/pci.c |    8 ++++++--
>  1 files changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
> index b0eb9e7..4ca53ef 100644
> --- a/arch/mips/pci/pci.c
> +++ b/arch/mips/pci/pci.c
> @@ -346,10 +346,14 @@ int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
>  		return -EINVAL;
>  
>  	/*
> -	 * Ignore write-combine; for now only return uncached mappings.
> +	 * For write-combine, return uncached accelerated mappings if CPU
> +	 * supports; otherwise, return uncached mappings.
>  	 */
>  	prot = pgprot_val(vma->vm_page_prot);
> -	prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED;
> +	if (write_combine && cpu_has_uncached_accelerated)
> +		prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED_ACCELERATED;
> +	else
> +		prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED;
>  	vma->vm_page_prot = __pgprot(prot);
This should be definietely wrong for MMIOs.
uncache accelleration should only be enabled for addresses which have no
side effect when doing write combine such as video memory.
>  
>  	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,

Re: [PATCH 3/3] MIPS: handle write_combine in pci_mmap_page_range

[Zhang Le]

[-- Attachment #1: Type: text/plain, Size: 2338 bytes --]

On 16:54 Fri 08 May     , yanh wrote:
> 
> 在 2009-05-08五的 14:30 +0800，Zhang Le写道：
> > Signed-off-by: Zhang Le <r0bertz@gentoo.org>
> > ---
> >  arch/mips/pci/pci.c |    8 ++++++--
> >  1 files changed, 6 insertions(+), 2 deletions(-)
> > 
> > diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
> > index b0eb9e7..4ca53ef 100644
> > --- a/arch/mips/pci/pci.c
> > +++ b/arch/mips/pci/pci.c
> > @@ -346,10 +346,14 @@ int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
> >  		return -EINVAL;
> >  
> >  	/*
> > -	 * Ignore write-combine; for now only return uncached mappings.
> > +	 * For write-combine, return uncached accelerated mappings if CPU
> > +	 * supports; otherwise, return uncached mappings.
> >  	 */
> >  	prot = pgprot_val(vma->vm_page_prot);
> > -	prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED;
> > +	if (write_combine && cpu_has_uncached_accelerated)
> > +		prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED_ACCELERATED;
> > +	else
> > +		prot = (prot & ~_CACHE_MASK) | _CACHE_UNCACHED;
> >  	vma->vm_page_prot = __pgprot(prot);
> This should be definietely wrong for MMIOs.
> uncache accelleration should only be enabled for addresses which have no
> side effect when doing write combine such as video memory.

Actually, currently write_combine is true only for prefetchable pci memory:
http://lxr.linux.no/linux+v2.6.29/drivers/pci/pci-sysfs.c#L711

And currently on fuloong 2f box, there is only one prefetchable pci memory
region, which just belongs to video card:

zhangle@2f /sys/devices/pci0000:00 (n32) $ find -name "*wc" -exec ls -l {} \;
-rw------- 1 root root 256M 2009-05-08 03:25 ./0000:00:08.0/resource0_wc

So, are we talking about the same thing? 

Also, I have observed that this region is larger than the actual video ram size
32M. However, libpciaccess only mapped 32M into X's address space:

zhangle@2f ~ (n32) $ sudo cat /proc/$(pidof X)/maps | grep wc$
2b354000-2d354000 rw-s 50000000 00:00 1882 /sys/devices/pci0000:00/0000:00:08.0/resource0_wc

So, I guess this should be just ok, right?

Any other risks?

> >  
> >  	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
> 

-- 
Zhang, Le
Gentoo/Loongson Developer
http://zhangle.is-a-geek.org
0260 C902 B8F8 6506 6586 2B90 BC51 C808 1E4E 2973

[-- Attachment #2: Type: application/pgp-signature, Size: 198 bytes --]