[PATCH 2/2] PAT support for write combining in pci_mmap_page_range

["Cédric Augonnet" <cedric.augonnet@gmail.com>]

Adds support for write-combining in pci_mmap_page_range using PAT6.

Some distinction has to be made between huge pages and normal pages since
the position of the bit encoding PAT depends on that.

Signed-off-by: Cedric Augonnet <cedric.augonnet@ens-lyon.org>
CC: Loic Prylli <loic@myri.com>
CC: Brice Goglin <brice@myri.com>	
---
diff -urN 2.6.23-rc2/arch/i386/pci/i386.c 2.6.23-rc2-pat/arch/i386/pci/i386.c
--- 2.6.23-rc2/arch/i386/pci/i386.c	2007-07-31 10:48:58.000000000 -0400
+++ 2.6.23-rc2-pat/arch/i386/pci/i386.c	2007-08-06 18:19:18.000000000 -0400
@@ -299,13 +299,20 @@
 	 * address on this platform.
 	 */
 	prot = pgprot_val(vma->vm_page_prot);
-	if (boot_cpu_data.x86 > 3)
-		prot |= _PAGE_PCD | _PAGE_PWT;
-	vma->vm_page_prot = __pgprot(prot);
 
-	/* Write-combine setting is ignored, it is changed via the mtrr
-	 * interfaces on this platform.
+	/* Unless PAT is enabling write combining, it is here ignored and
+	 * changed via the mtrr interfaces on this platform.
 	 */
+	if (cpu_has_pat && write_combine) {
+		/* PAT bit is not the same if we have huge pages */
+		vma->vm_page_prot =
+			__pgprot_wc(prot, (vma->vm_flags & VM_HUGETLB));
+	} else {
+		if (boot_cpu_data.x86 > 3)
+			prot |= _PAGE_PCD | _PAGE_PWT;
+		vma->vm_page_prot = __pgprot(prot);
+	}
+
 	if (io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
 			       vma->vm_end - vma->vm_start,
 			       vma->vm_page_prot))
diff -urN 2.6.23-rc2/include/asm-i386/pgtable.h 2.6.23-rc2-pat/include/asm-i386/pgtable.h
--- 2.6.23-rc2/include/asm-i386/pgtable.h	2007-08-06 15:17:12.000000000 -0400
+++ 2.6.23-rc2-pat/include/asm-i386/pgtable.h	2007-08-06 18:33:48.000000000 -0400
@@ -119,6 +119,12 @@
 #define _PAGE_UNUSED2	0x400
 #define _PAGE_UNUSED3	0x800
 
+/* defined for write combining support with PAT enabled */
+#define _PAGE_PAT6		0x088		/* PAT6 for 4K pages */
+#define _PAGE_PAT6_HUGE		0x1008		/* PAT6 for huge pages */
+#define _PAGE_WR_CMB		_PAGE_PAT6
+#define _PAGE_WR_CMB_HUGE	_PAGE_PAT6_HUGE
+
 /* If _PAGE_PRESENT is clear, we use these: */
 #define _PAGE_FILE	0x040	/* nonlinear file mapping, saved PTE; unset:swap */
 #define _PAGE_PROTNONE	0x080	/* if the user mapped it with PROT_NONE;
@@ -171,6 +177,11 @@
 #define PAGE_KERNEL_LARGE	__pgprot(__PAGE_KERNEL_LARGE)
 #define PAGE_KERNEL_LARGE_EXEC	__pgprot(__PAGE_KERNEL_LARGE_EXEC)
 
+#define __pgprot_wc(flag, is_huge_page)				\
+	(__pgprot((is_huge_page) ?				\
+		((flag) & (~_PAGE_PWT)) | _PAGE_WR_CMB_HUGE :	\
+		((flag) & (~_PAGE_PWT)) | _PAGE_WR_CMB) )
+
 /*
  * The i386 can't do page protection for execute, and considers that
  * the same are read. Also, write permissions imply read permissions.
diff -urN 2.6.23-rc2/include/asm-x86_64/pgtable.h 2.6.23-rc2-pat/include/asm-x86_64/pgtable.h
--- 2.6.23-rc2/include/asm-x86_64/pgtable.h	2007-08-06 15:17:15.000000000 -0400
+++ 2.6.23-rc2-pat/include/asm-x86_64/pgtable.h	2007-08-06 18:29:28.000000000 -0400
@@ -164,6 +164,13 @@
 #define _PAGE_GLOBAL	0x100	/* Global TLB entry */
 
 #define _PAGE_PROTNONE	0x080	/* If not present */
+
+/* defined for write combining support with PAT enabled */
+#define _PAGE_PAT6		0x088		/* PAT6 for 4K pages */
+#define _PAGE_PAT6_HUGE		0x1008		/* PAT6 for huge pages */
+#define _PAGE_WR_CMB		_PAGE_PAT6
+#define _PAGE_WR_CMB_HUGE	_PAGE_PAT6_HUGE
+
 #define _PAGE_NX        (_AC(1,UL)<<_PAGE_BIT_NX)
 
 #define _PAGE_TABLE	(_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | _PAGE_ACCESSED | _PAGE_DIRTY)
@@ -196,6 +203,11 @@
 #define __PAGE_KERNEL_LARGE_EXEC \
 	(__PAGE_KERNEL_EXEC | _PAGE_PSE)
 
+#define __pgprot_wc(flag, is_huge_page)				\
+	(__pgprot((is_huge_page) ?				\
+		((flag) & (~_PAGE_PWT)) | _PAGE_WR_CMB_HUGE :	\
+		((flag) & (~_PAGE_PWT)) | _PAGE_WR_CMB))
+
 #define MAKE_GLOBAL(x) __pgprot((x) | _PAGE_GLOBAL)
 
 #define PAGE_KERNEL MAKE_GLOBAL(__PAGE_KERNEL)