[ARM] copypage: provide our own copy_user_highpage()

We used to override the copy_user_page() function.  However, this
is not only inefficient, it also causes additional complexity for
highmem support, since we convert from a struct page to a kernel
direct mapped address and back to a struct page again.

Moreover, with highmem support, we end up pointlessly setting up
kmap entries for pages which we're going to remap.  So, push the
kmapping down into the copypage implementation files where it's
required.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

9 files changed, 147 insertions, 74 deletions

diff --git a/arch/arm/mm/copypage-feroceon.c b/arch/arm/mm/copypage-feroceon.c
index c8347670ab00..edd71686b8df 100644
--- a/arch/arm/mm/copypage-feroceon.c
+++ b/arch/arm/mm/copypage-feroceon.c
@@ -7,15 +7,14 @@
  * it under the terms of the GNU General Public License version 2 as
  * published by the Free Software Foundation.
  *
- * This handles copy_user_page and clear_user_page on Feroceon
+ * This handles copy_user_highpage and clear_user_page on Feroceon
  * more optimally than the generic implementations.
  */
 #include <linux/init.h>
+#include <linux/highmem.h>
 
-#include <asm/page.h>
-
-void __attribute__((naked))
-feroceon_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
+static void __attribute__((naked))
+feroceon_copy_user_page(void *kto, const void *kfrom)
 {
         asm("\
         stmfd   sp!, {r4-r9, lr}                \n\
@@ -68,6 +67,18 @@ feroceon_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
         : "I" (PAGE_SIZE));
 }
 
+void feroceon_copy_user_highpage(struct page *to, struct page *from,
+        unsigned long vaddr)
+{
+        void *kto, *kfrom;
+
+        kto = kmap_atomic(to, KM_USER0);
+        kfrom = kmap_atomic(from, KM_USER1);
+        feroceon_copy_user_page(kto, kfrom);
+        kunmap_atomic(kfrom, KM_USER1);
+        kunmap_atomic(kto, KM_USER0);
+}
+
 void __attribute__((naked))
 feroceon_clear_user_page(void *kaddr, unsigned long vaddr)
 {
@@ -95,6 +106,6 @@ feroceon_clear_user_page(void *kaddr, unsigned long vaddr)
 
 struct cpu_user_fns feroceon_user_fns __initdata = {
         .cpu_clear_user_page    = feroceon_clear_user_page,
-        .cpu_copy_user_page     = feroceon_copy_user_page,
+        .cpu_copy_user_highpage = feroceon_copy_user_highpage,
 };
 
diff --git a/arch/arm/mm/copypage-v3.c b/arch/arm/mm/copypage-v3.c
index 184911089e6c..52df8f04d3f7 100644
--- a/arch/arm/mm/copypage-v3.c
+++ b/arch/arm/mm/copypage-v3.c
@@ -8,16 +8,15 @@
  * published by the Free Software Foundation.
  */
 #include <linux/init.h>
-
-#include <asm/page.h>
+#include <linux/highmem.h>
 
 /*
- * ARMv3 optimised copy_user_page
+ * ARMv3 optimised copy_user_highpage
  *
  * FIXME: do we need to handle cache stuff...
  */
-void __attribute__((naked))
-v3_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
+static void __attribute__((naked))
+v3_copy_user_page(void *kto, const void *kfrom)
 {
         asm("\n\
         stmfd   sp!, {r4, lr}                   @       2\n\
@@ -38,6 +37,18 @@ v3_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
         : "r" (kfrom), "r" (kto), "I" (PAGE_SIZE / 64));
 }
 
+void v3_copy_user_highpage(struct page *to, struct page *from,
+        unsigned long vaddr)
+{
+        void *kto, *kfrom;
+
+        kto = kmap_atomic(to, KM_USER0);
+        kfrom = kmap_atomic(from, KM_USER1);
+        v3_copy_user_page(kto, kfrom);
+        kunmap_atomic(kfrom, KM_USER1);
+        kunmap_atomic(kto, KM_USER0);
+}
+
 /*
  * ARMv3 optimised clear_user_page
  *
@@ -65,5 +76,5 @@ void __attribute__((naked)) v3_clear_user_page(void *kaddr, unsigned long vaddr)
 
 struct cpu_user_fns v3_user_fns __initdata = {
         .cpu_clear_user_page    = v3_clear_user_page,
-        .cpu_copy_user_page     = v3_copy_user_page,
+        .cpu_copy_user_highpage = v3_copy_user_highpage,
 };
diff --git a/arch/arm/mm/copypage-v4mc.c b/arch/arm/mm/copypage-v4mc.c
index 8d33e2549344..a7dc838fee76 100644
--- a/arch/arm/mm/copypage-v4mc.c
+++ b/arch/arm/mm/copypage-v4mc.c
@@ -15,8 +15,8 @@
  */
 #include <linux/init.h>
 #include <linux/mm.h>
+#include <linux/highmem.h>
 
-#include <asm/page.h>
 #include <asm/pgtable.h>
 #include <asm/tlbflush.h>
 #include <asm/cacheflush.h>
@@ -33,7 +33,7 @@
 static DEFINE_SPINLOCK(minicache_lock);
 
 /*
- * ARMv4 mini-dcache optimised copy_user_page
+ * ARMv4 mini-dcache optimised copy_user_highpage
  *
  * We flush the destination cache lines just before we write the data into the
  * corresponding address.  Since the Dcache is read-allocate, this removes the
@@ -42,7 +42,7 @@ static DEFINE_SPINLOCK(minicache_lock);
  *
  * Note: We rely on all ARMv4 processors implementing the "invalidate D line"
  * instruction.  If your processor does not supply this, you have to write your
- * own copy_user_page that does the right thing.
+ * own copy_user_highpage that does the right thing.
  */
 static void __attribute__((naked))
 mc_copy_user_page(void *from, void *to)
@@ -68,21 +68,24 @@ mc_copy_user_page(void *from, void *to)
         : "r" (from), "r" (to), "I" (PAGE_SIZE / 64));
 }
 
-void v4_mc_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
+void v4_mc_copy_user_highpage(struct page *from, struct page *to,
+        unsigned long vaddr)
 {
-        struct page *page = virt_to_page(kfrom);
+        void *kto = kmap_atomic(to, KM_USER1);
 
-        if (test_and_clear_bit(PG_dcache_dirty, &page->flags))
-                __flush_dcache_page(page_mapping(page), page);
+        if (test_and_clear_bit(PG_dcache_dirty, &from->flags))
+                __flush_dcache_page(page_mapping(from), from);
 
         spin_lock(&minicache_lock);
 
-        set_pte_ext(TOP_PTE(0xffff8000), pfn_pte(__pa(kfrom) >> PAGE_SHIFT, minicache_pgprot), 0);
+        set_pte_ext(TOP_PTE(0xffff8000), pfn_pte(page_to_pfn(from), minicache_pgprot), 0);
         flush_tlb_kernel_page(0xffff8000);
 
         mc_copy_user_page((void *)0xffff8000, kto);
 
         spin_unlock(&minicache_lock);
+
+        kunmap_atomic(kto, KM_USER1);
 }
 
 /*
@@ -113,5 +116,5 @@ v4_mc_clear_user_page(void *kaddr, unsigned long vaddr)
 
 struct cpu_user_fns v4_mc_user_fns __initdata = {
         .cpu_clear_user_page    = v4_mc_clear_user_page, 
-        .cpu_copy_user_page     = v4_mc_copy_user_page,
+        .cpu_copy_user_highpage = v4_mc_copy_user_highpage,
 };
diff --git a/arch/arm/mm/copypage-v4wb.c b/arch/arm/mm/copypage-v4wb.c
index 230210822961..186a68a794a9 100644
--- a/arch/arm/mm/copypage-v4wb.c
+++ b/arch/arm/mm/copypage-v4wb.c
@@ -8,11 +8,10 @@
  * published by the Free Software Foundation.
  */
 #include <linux/init.h>
-
-#include <asm/page.h>
+#include <linux/highmem.h>
 
 /*
- * ARMv4 optimised copy_user_page
+ * ARMv4 optimised copy_user_highpage
  *
  * We flush the destination cache lines just before we write the data into the
  * corresponding address.  Since the Dcache is read-allocate, this removes the
@@ -21,10 +20,10 @@
  *
  * Note: We rely on all ARMv4 processors implementing the "invalidate D line"
  * instruction.  If your processor does not supply this, you have to write your
- * own copy_user_page that does the right thing.
+ * own copy_user_highpage that does the right thing.
  */
-void __attribute__((naked))
-v4wb_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
+static void __attribute__((naked))
+v4wb_copy_user_page(void *kto, const void *kfrom)
 {
         asm("\
         stmfd   sp!, {r4, lr}                   @ 2\n\
@@ -48,6 +47,18 @@ v4wb_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
         : "I" (PAGE_SIZE / 64));
 }
 
+void v4wb_copy_user_highpage(struct page *to, struct page *from,
+        unsigned long vaddr)
+{
+        void *kto, *kfrom;
+
+        kto = kmap_atomic(to, KM_USER0);
+        kfrom = kmap_atomic(from, KM_USER1);
+        v4wb_copy_user_page(kto, kfrom);
+        kunmap_atomic(kfrom, KM_USER1);
+        kunmap_atomic(kto, KM_USER0);
+}
+
 /*
  * ARMv4 optimised clear_user_page
  *
@@ -79,5 +90,5 @@ v4wb_clear_user_page(void *kaddr, unsigned long vaddr)
 
 struct cpu_user_fns v4wb_user_fns __initdata = {
         .cpu_clear_user_page    = v4wb_clear_user_page,
-        .cpu_copy_user_page     = v4wb_copy_user_page,
+        .cpu_copy_user_highpage = v4wb_copy_user_highpage,
 };
diff --git a/arch/arm/mm/copypage-v4wt.c b/arch/arm/mm/copypage-v4wt.c
index d8ef39503ff0..86c2cfdbde03 100644
--- a/arch/arm/mm/copypage-v4wt.c
+++ b/arch/arm/mm/copypage-v4wt.c
@@ -11,18 +11,17 @@
  *  the only supported cache operation.
  */
 #include <linux/init.h>
-
-#include <asm/page.h>
+#include <linux/highmem.h>
 
 /*
- * ARMv4 optimised copy_user_page
+ * ARMv4 optimised copy_user_highpage
  *
  * Since we have writethrough caches, we don't have to worry about
  * dirty data in the cache.  However, we do have to ensure that
  * subsequent reads are up to date.
  */
-void __attribute__((naked))
-v4wt_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
+static void __attribute__((naked))
+v4wt_copy_user_page(void *kto, const void *kfrom)
 {
         asm("\
         stmfd   sp!, {r4, lr}                   @ 2\n\
@@ -44,6 +43,18 @@ v4wt_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
         : "I" (PAGE_SIZE / 64));
 }
 
+void v4wt_copy_user_highpage(struct page *to, struct page *from,
+        unsigned long vaddr)
+{
+        void *kto, *kfrom;
+
+        kto = kmap_atomic(to, KM_USER0);
+        kfrom = kmap_atomic(from, KM_USER1);
+        v4wt_copy_user_page(kto, kfrom);
+        kunmap_atomic(kfrom, KM_USER1);
+        kunmap_atomic(kto, KM_USER0);
+}
+
 /*
  * ARMv4 optimised clear_user_page
  *
@@ -73,5 +84,5 @@ v4wt_clear_user_page(void *kaddr, unsigned long vaddr)
 
 struct cpu_user_fns v4wt_user_fns __initdata = {
         .cpu_clear_user_page    = v4wt_clear_user_page,
-        .cpu_copy_user_page     = v4wt_copy_user_page,
+        .cpu_copy_user_highpage = v4wt_copy_user_highpage,
 };
diff --git a/arch/arm/mm/copypage-v6.c b/arch/arm/mm/copypage-v6.c
index 0e21c0767580..2ea75d0f5048 100644
--- a/arch/arm/mm/copypage-v6.c
+++ b/arch/arm/mm/copypage-v6.c
@@ -10,8 +10,8 @@
 #include <linux/init.h>
 #include <linux/spinlock.h>
 #include <linux/mm.h>
+#include <linux/highmem.h>
 
-#include <asm/page.h>
 #include <asm/pgtable.h>
 #include <asm/shmparam.h>
 #include <asm/tlbflush.h>
@@ -33,9 +33,16 @@ static DEFINE_SPINLOCK(v6_lock);
  * Copy the user page.  No aliasing to deal with so we can just
  * attack the kernel's existing mapping of these pages.
  */
-static void v6_copy_user_page_nonaliasing(void *kto, const void *kfrom, unsigned long vaddr)
+static void v6_copy_user_highpage_nonaliasing(struct page *to,
+        struct page *from, unsigned long vaddr)
 {
+        void *kto, *kfrom;
+
+        kfrom = kmap_atomic(from, KM_USER0);
+        kto = kmap_atomic(to, KM_USER1);
         copy_page(kto, kfrom);
+        kunmap_atomic(kto, KM_USER1);
+        kunmap_atomic(kfrom, KM_USER0);
 }
 
 /*
@@ -48,26 +55,32 @@ static void v6_clear_user_page_nonaliasing(void *kaddr, unsigned long vaddr)
 }
 
 /*
- * Copy the page, taking account of the cache colour.
+ * Discard data in the kernel mapping for the new page.
+ * FIXME: needs this MCRR to be supported.
  */
-static void v6_copy_user_page_aliasing(void *kto, const void *kfrom, unsigned long vaddr)
+static void discard_old_kernel_data(void *kto)
 {
-        unsigned int offset = CACHE_COLOUR(vaddr);
-        unsigned long from, to;
-        struct page *page = virt_to_page(kfrom);
-
-        if (test_and_clear_bit(PG_dcache_dirty, &page->flags))
-                __flush_dcache_page(page_mapping(page), page);
-
-        /*
-         * Discard data in the kernel mapping for the new page.
-         * FIXME: needs this MCRR to be supported.
-         */
         __asm__("mcrr   p15, 0, %1, %0, c6      @ 0xec401f06"
            :
            : "r" (kto),
              "r" ((unsigned long)kto + PAGE_SIZE - L1_CACHE_BYTES)
            : "cc");
+}
+
+/*
+ * Copy the page, taking account of the cache colour.
+ */
+static void v6_copy_user_highpage_aliasing(struct page *to,
+        struct page *from, unsigned long vaddr)
+{
+        unsigned int offset = CACHE_COLOUR(vaddr);
+        unsigned long kfrom, kto;
+
+        if (test_and_clear_bit(PG_dcache_dirty, &from->flags))
+                __flush_dcache_page(page_mapping(from), from);
+
+        /* FIXME: not highmem safe */
+        discard_old_kernel_data(page_address(to));
 
         /*
          * Now copy the page using the same cache colour as the
@@ -75,16 +88,16 @@ static void v6_copy_user_page_aliasing(void *kto, const void *kfrom, unsigned lo
          */
         spin_lock(&v6_lock);
 
-        set_pte_ext(TOP_PTE(from_address) + offset, pfn_pte(__pa(kfrom) >> PAGE_SHIFT, PAGE_KERNEL), 0);
-        set_pte_ext(TOP_PTE(to_address) + offset, pfn_pte(__pa(kto) >> PAGE_SHIFT, PAGE_KERNEL), 0);
+        set_pte_ext(TOP_PTE(from_address) + offset, pfn_pte(page_to_pfn(from), PAGE_KERNEL), 0);
+        set_pte_ext(TOP_PTE(to_address) + offset, pfn_pte(page_to_pfn(to), PAGE_KERNEL), 0);
 
-        from = from_address + (offset << PAGE_SHIFT);
-        to   = to_address + (offset << PAGE_SHIFT);
+        kfrom = from_address + (offset << PAGE_SHIFT);
+        kto   = to_address + (offset << PAGE_SHIFT);
 
-        flush_tlb_kernel_page(from);
-        flush_tlb_kernel_page(to);
+        flush_tlb_kernel_page(kfrom);
+        flush_tlb_kernel_page(kto);
 
-        copy_page((void *)to, (void *)from);
+        copy_page((void *)kto, (void *)kfrom);
 
         spin_unlock(&v6_lock);
 }
@@ -124,14 +137,14 @@ static void v6_clear_user_page_aliasing(void *kaddr, unsigned long vaddr)
 
 struct cpu_user_fns v6_user_fns __initdata = {
         .cpu_clear_user_page    = v6_clear_user_page_nonaliasing,
-        .cpu_copy_user_page     = v6_copy_user_page_nonaliasing,
+        .cpu_copy_user_highpage = v6_copy_user_highpage_nonaliasing,
 };
 
 static int __init v6_userpage_init(void)
 {
         if (cache_is_vipt_aliasing()) {
                 cpu_user.cpu_clear_user_page = v6_clear_user_page_aliasing;
-                cpu_user.cpu_copy_user_page = v6_copy_user_page_aliasing;
+                cpu_user.cpu_copy_user_highpage = v6_copy_user_highpage_aliasing;
         }
 
         return 0;
diff --git a/arch/arm/mm/copypage-xsc3.c b/arch/arm/mm/copypage-xsc3.c
index 51ed502e5777..caa697ccd8db 100644
--- a/arch/arm/mm/copypage-xsc3.c
+++ b/arch/arm/mm/copypage-xsc3.c
@@ -11,8 +11,7 @@
  * Author: Matt Gilbert (matthew.m.gilbert@intel.com)
  */
 #include <linux/init.h>
-
-#include <asm/page.h>
+#include <linux/highmem.h>
 
 /*
  * General note:
@@ -21,18 +20,17 @@
  */
 
 /*
- * XSC3 optimised copy_user_page
+ * XSC3 optimised copy_user_highpage
  *  r0 = destination
  *  r1 = source
- *  r2 = virtual user address of ultimate destination page
  *
  * The source page may have some clean entries in the cache already, but we
  * can safely ignore them - break_cow() will flush them out of the cache
  * if we eventually end up using our copied page.
  *
  */
-void __attribute__((naked))
-xsc3_mc_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
+static void __attribute__((naked))
+xsc3_mc_copy_user_page(void *kto, const void *kfrom)
 {
         asm("\
         stmfd   sp!, {r4, r5, lr}               \n\
@@ -72,6 +70,18 @@ xsc3_mc_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
         : "I" (PAGE_SIZE / 64 - 1));
 }
 
+void xsc3_mc_copy_user_highpage(struct page *to, struct page *from,
+        unsigned long vaddr)
+{
+        void *kto, *kfrom;
+
+        kto = kmap_atomic(to, KM_USER0);
+        kfrom = kmap_atomic(from, KM_USER1);
+        xsc3_mc_copy_user_page(kto, kfrom);
+        kunmap_atomic(kfrom, KM_USER1);
+        kunmap_atomic(kto, KM_USER0);
+}
+
 /*
  * XScale optimised clear_user_page
  *  r0 = destination
@@ -98,5 +108,5 @@ xsc3_mc_clear_user_page(void *kaddr, unsigned long vaddr)
 
 struct cpu_user_fns xsc3_mc_user_fns __initdata = {
         .cpu_clear_user_page    = xsc3_mc_clear_user_page,
-        .cpu_copy_user_page     = xsc3_mc_copy_user_page,
+        .cpu_copy_user_highpage = xsc3_mc_copy_user_highpage,
 };
diff --git a/arch/arm/mm/copypage-xscale.c b/arch/arm/mm/copypage-xscale.c
index bad49331bbf9..01bafafce181 100644
--- a/arch/arm/mm/copypage-xscale.c
+++ b/arch/arm/mm/copypage-xscale.c
@@ -15,8 +15,8 @@
  */
 #include <linux/init.h>
 #include <linux/mm.h>
+#include <linux/highmem.h>
 
-#include <asm/page.h>
 #include <asm/pgtable.h>
 #include <asm/tlbflush.h>
 #include <asm/cacheflush.h>
@@ -35,7 +35,7 @@
 static DEFINE_SPINLOCK(minicache_lock);
 
 /*
- * XScale mini-dcache optimised copy_user_page
+ * XScale mini-dcache optimised copy_user_highpage
  *
  * We flush the destination cache lines just before we write the data into the
  * corresponding address.  Since the Dcache is read-allocate, this removes the
@@ -90,21 +90,24 @@ mc_copy_user_page(void *from, void *to)
         : "r" (from), "r" (to), "I" (PAGE_SIZE / 64 - 1));
 }
 
-void xscale_mc_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
+void xscale_mc_copy_user_highpage(struct page *to, struct page *from,
+        unsigned long vaddr)
 {
-        struct page *page = virt_to_page(kfrom);
+        void *kto = kmap_atomic(to, KM_USER1);
 
-        if (test_and_clear_bit(PG_dcache_dirty, &page->flags))
-                __flush_dcache_page(page_mapping(page), page);
+        if (test_and_clear_bit(PG_dcache_dirty, &from->flags))
+                __flush_dcache_page(page_mapping(from), from);
 
         spin_lock(&minicache_lock);
 
-        set_pte_ext(TOP_PTE(COPYPAGE_MINICACHE), pfn_pte(__pa(kfrom) >> PAGE_SHIFT, minicache_pgprot), 0);
+        set_pte_ext(TOP_PTE(COPYPAGE_MINICACHE), pfn_pte(page_to_pfn(from), minicache_pgprot), 0);
         flush_tlb_kernel_page(COPYPAGE_MINICACHE);
 
         mc_copy_user_page((void *)COPYPAGE_MINICACHE, kto);
 
         spin_unlock(&minicache_lock);
+
+        kunmap_atomic(kto, KM_USER1);
 }
 
 /*
@@ -133,5 +136,5 @@ xscale_mc_clear_user_page(void *kaddr, unsigned long vaddr)
 
 struct cpu_user_fns xscale_mc_user_fns __initdata = {
         .cpu_clear_user_page    = xscale_mc_clear_user_page, 
-        .cpu_copy_user_page     = xscale_mc_copy_user_page,
+        .cpu_copy_user_highpage = xscale_mc_copy_user_highpage,
 };
diff --git a/arch/arm/mm/proc-syms.c b/arch/arm/mm/proc-syms.c
index 2b5ba396e3a6..b9743e6416c4 100644
--- a/arch/arm/mm/proc-syms.c
+++ b/arch/arm/mm/proc-syms.c
@@ -34,7 +34,7 @@ EXPORT_SYMBOL(cpu_cache);
 #ifdef CONFIG_MMU
 #ifndef MULTI_USER
 EXPORT_SYMBOL(__cpu_clear_user_page);
-EXPORT_SYMBOL(__cpu_copy_user_page);
+EXPORT_SYMBOL(__cpu_copy_user_highpage);
 #else
 EXPORT_SYMBOL(cpu_user);
 #endif