2005-04-17 02:20:36 +04:00
|
|
|
/*
|
|
|
|
* linux/arch/arm/mm/flush.c
|
|
|
|
*
|
|
|
|
* Copyright (C) 1995-2002 Russell King
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or modify
|
|
|
|
* it under the terms of the GNU General Public License version 2 as
|
|
|
|
* published by the Free Software Foundation.
|
|
|
|
*/
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/mm.h>
|
|
|
|
#include <linux/pagemap.h>
|
2010-12-15 23:14:45 +03:00
|
|
|
#include <linux/highmem.h>
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
#include <asm/cacheflush.h>
|
2008-08-10 21:10:19 +04:00
|
|
|
#include <asm/cachetype.h>
|
2010-03-30 00:46:02 +04:00
|
|
|
#include <asm/highmem.h>
|
2009-11-05 16:29:36 +03:00
|
|
|
#include <asm/smp_plat.h>
|
2005-05-10 20:31:43 +04:00
|
|
|
#include <asm/tlbflush.h>
|
2013-05-17 15:33:28 +04:00
|
|
|
#include <linux/hugetlb.h>
|
2005-05-10 20:31:43 +04:00
|
|
|
|
2006-08-21 20:06:38 +04:00
|
|
|
#include "mm.h"
|
|
|
|
|
ARM: move heavy barrier support out of line
The existing memory barrier macro causes a significant amount of code
to be inserted inline at every call site. For example, in
gpio_set_irq_type(), we have this for mb():
c0344c08: f57ff04e dsb st
c0344c0c: e59f8190 ldr r8, [pc, #400] ; c0344da4 <gpio_set_irq_type+0x230>
c0344c10: e3590004 cmp r9, #4
c0344c14: e5983014 ldr r3, [r8, #20]
c0344c18: 0a000054 beq c0344d70 <gpio_set_irq_type+0x1fc>
c0344c1c: e3530000 cmp r3, #0
c0344c20: 0a000004 beq c0344c38 <gpio_set_irq_type+0xc4>
c0344c24: e50b2030 str r2, [fp, #-48] ; 0xffffffd0
c0344c28: e50bc034 str ip, [fp, #-52] ; 0xffffffcc
c0344c2c: e12fff33 blx r3
c0344c30: e51bc034 ldr ip, [fp, #-52] ; 0xffffffcc
c0344c34: e51b2030 ldr r2, [fp, #-48] ; 0xffffffd0
c0344c38: e5963004 ldr r3, [r6, #4]
Moving the outer_cache_sync() call out of line reduces the impact of
the barrier:
c0344968: f57ff04e dsb st
c034496c: e35a0004 cmp sl, #4
c0344970: e50b2030 str r2, [fp, #-48] ; 0xffffffd0
c0344974: 0a000044 beq c0344a8c <gpio_set_irq_type+0x1b8>
c0344978: ebf363dd bl c001d8f4 <arm_heavy_mb>
c034497c: e5953004 ldr r3, [r5, #4]
This should reduce the cache footprint of this code. Overall, this
results in a reduction of around 20K in the kernel size:
text data bss dec hex filename
10773970 667392 10369656 21811018 14ccf4a ../build/imx6/vmlinux-old
10754219 667392 10369656 21791267 14c8223 ../build/imx6/vmlinux-new
Another advantage to this approach is that we can finally resolve the
issue of SoCs which have their own memory barrier requirements within
multiplatform kernels (such as OMAP.) Here, the bus interconnects
need additional handling to ensure that writes become visible in the
correct order (eg, between dma_map() operations, writes to DMA
coherent memory, and MMIO accesses.)
Acked-by: Tony Lindgren <tony@atomide.com>
Acked-by: Richard Woodruff <r-woodruff2@ti.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2015-06-02 01:44:46 +03:00
|
|
|
#ifdef CONFIG_ARM_HEAVY_MB
|
2015-06-03 15:10:16 +03:00
|
|
|
void (*soc_mb)(void);
|
|
|
|
|
ARM: move heavy barrier support out of line
The existing memory barrier macro causes a significant amount of code
to be inserted inline at every call site. For example, in
gpio_set_irq_type(), we have this for mb():
c0344c08: f57ff04e dsb st
c0344c0c: e59f8190 ldr r8, [pc, #400] ; c0344da4 <gpio_set_irq_type+0x230>
c0344c10: e3590004 cmp r9, #4
c0344c14: e5983014 ldr r3, [r8, #20]
c0344c18: 0a000054 beq c0344d70 <gpio_set_irq_type+0x1fc>
c0344c1c: e3530000 cmp r3, #0
c0344c20: 0a000004 beq c0344c38 <gpio_set_irq_type+0xc4>
c0344c24: e50b2030 str r2, [fp, #-48] ; 0xffffffd0
c0344c28: e50bc034 str ip, [fp, #-52] ; 0xffffffcc
c0344c2c: e12fff33 blx r3
c0344c30: e51bc034 ldr ip, [fp, #-52] ; 0xffffffcc
c0344c34: e51b2030 ldr r2, [fp, #-48] ; 0xffffffd0
c0344c38: e5963004 ldr r3, [r6, #4]
Moving the outer_cache_sync() call out of line reduces the impact of
the barrier:
c0344968: f57ff04e dsb st
c034496c: e35a0004 cmp sl, #4
c0344970: e50b2030 str r2, [fp, #-48] ; 0xffffffd0
c0344974: 0a000044 beq c0344a8c <gpio_set_irq_type+0x1b8>
c0344978: ebf363dd bl c001d8f4 <arm_heavy_mb>
c034497c: e5953004 ldr r3, [r5, #4]
This should reduce the cache footprint of this code. Overall, this
results in a reduction of around 20K in the kernel size:
text data bss dec hex filename
10773970 667392 10369656 21811018 14ccf4a ../build/imx6/vmlinux-old
10754219 667392 10369656 21791267 14c8223 ../build/imx6/vmlinux-new
Another advantage to this approach is that we can finally resolve the
issue of SoCs which have their own memory barrier requirements within
multiplatform kernels (such as OMAP.) Here, the bus interconnects
need additional handling to ensure that writes become visible in the
correct order (eg, between dma_map() operations, writes to DMA
coherent memory, and MMIO accesses.)
Acked-by: Tony Lindgren <tony@atomide.com>
Acked-by: Richard Woodruff <r-woodruff2@ti.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2015-06-02 01:44:46 +03:00
|
|
|
void arm_heavy_mb(void)
|
|
|
|
{
|
|
|
|
#ifdef CONFIG_OUTER_CACHE_SYNC
|
|
|
|
if (outer_cache.sync)
|
|
|
|
outer_cache.sync();
|
|
|
|
#endif
|
2015-06-03 15:10:16 +03:00
|
|
|
if (soc_mb)
|
|
|
|
soc_mb();
|
ARM: move heavy barrier support out of line
The existing memory barrier macro causes a significant amount of code
to be inserted inline at every call site. For example, in
gpio_set_irq_type(), we have this for mb():
c0344c08: f57ff04e dsb st
c0344c0c: e59f8190 ldr r8, [pc, #400] ; c0344da4 <gpio_set_irq_type+0x230>
c0344c10: e3590004 cmp r9, #4
c0344c14: e5983014 ldr r3, [r8, #20]
c0344c18: 0a000054 beq c0344d70 <gpio_set_irq_type+0x1fc>
c0344c1c: e3530000 cmp r3, #0
c0344c20: 0a000004 beq c0344c38 <gpio_set_irq_type+0xc4>
c0344c24: e50b2030 str r2, [fp, #-48] ; 0xffffffd0
c0344c28: e50bc034 str ip, [fp, #-52] ; 0xffffffcc
c0344c2c: e12fff33 blx r3
c0344c30: e51bc034 ldr ip, [fp, #-52] ; 0xffffffcc
c0344c34: e51b2030 ldr r2, [fp, #-48] ; 0xffffffd0
c0344c38: e5963004 ldr r3, [r6, #4]
Moving the outer_cache_sync() call out of line reduces the impact of
the barrier:
c0344968: f57ff04e dsb st
c034496c: e35a0004 cmp sl, #4
c0344970: e50b2030 str r2, [fp, #-48] ; 0xffffffd0
c0344974: 0a000044 beq c0344a8c <gpio_set_irq_type+0x1b8>
c0344978: ebf363dd bl c001d8f4 <arm_heavy_mb>
c034497c: e5953004 ldr r3, [r5, #4]
This should reduce the cache footprint of this code. Overall, this
results in a reduction of around 20K in the kernel size:
text data bss dec hex filename
10773970 667392 10369656 21811018 14ccf4a ../build/imx6/vmlinux-old
10754219 667392 10369656 21791267 14c8223 ../build/imx6/vmlinux-new
Another advantage to this approach is that we can finally resolve the
issue of SoCs which have their own memory barrier requirements within
multiplatform kernels (such as OMAP.) Here, the bus interconnects
need additional handling to ensure that writes become visible in the
correct order (eg, between dma_map() operations, writes to DMA
coherent memory, and MMIO accesses.)
Acked-by: Tony Lindgren <tony@atomide.com>
Acked-by: Richard Woodruff <r-woodruff2@ti.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2015-06-02 01:44:46 +03:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(arm_heavy_mb);
|
|
|
|
#endif
|
|
|
|
|
2005-05-10 20:31:43 +04:00
|
|
|
#ifdef CONFIG_CPU_CACHE_VIPT
|
2005-09-08 18:32:23 +04:00
|
|
|
|
2005-09-30 19:07:04 +04:00
|
|
|
static void flush_pfn_alias(unsigned long pfn, unsigned long vaddr)
|
|
|
|
{
|
2011-07-02 17:46:27 +04:00
|
|
|
unsigned long to = FLUSH_ALIAS_START + (CACHE_COLOUR(vaddr) << PAGE_SHIFT);
|
2006-03-11 01:26:47 +03:00
|
|
|
const int zero = 0;
|
2005-09-30 19:07:04 +04:00
|
|
|
|
2011-07-02 18:20:44 +04:00
|
|
|
set_top_pte(to, pfn_pte(pfn, PAGE_KERNEL));
|
2005-09-30 19:07:04 +04:00
|
|
|
|
|
|
|
asm( "mcrr p15, 0, %1, %0, c14\n"
|
2009-10-25 01:36:36 +04:00
|
|
|
" mcr p15, 0, %2, c7, c10, 4"
|
2005-09-30 19:07:04 +04:00
|
|
|
:
|
2014-11-29 04:54:27 +03:00
|
|
|
: "r" (to), "r" (to + PAGE_SIZE - 1), "r" (zero)
|
2005-09-30 19:07:04 +04:00
|
|
|
: "cc");
|
|
|
|
}
|
|
|
|
|
2010-09-13 19:19:41 +04:00
|
|
|
static void flush_icache_alias(unsigned long pfn, unsigned long vaddr, unsigned long len)
|
|
|
|
{
|
2011-07-02 18:20:44 +04:00
|
|
|
unsigned long va = FLUSH_ALIAS_START + (CACHE_COLOUR(vaddr) << PAGE_SHIFT);
|
2010-09-13 19:19:41 +04:00
|
|
|
unsigned long offset = vaddr & (PAGE_SIZE - 1);
|
|
|
|
unsigned long to;
|
|
|
|
|
2011-07-02 18:20:44 +04:00
|
|
|
set_top_pte(va, pfn_pte(pfn, PAGE_KERNEL));
|
|
|
|
to = va + offset;
|
2010-09-13 19:19:41 +04:00
|
|
|
flush_icache_range(to, to + len);
|
|
|
|
}
|
|
|
|
|
2005-09-08 18:32:23 +04:00
|
|
|
void flush_cache_mm(struct mm_struct *mm)
|
|
|
|
{
|
|
|
|
if (cache_is_vivt()) {
|
2009-10-25 13:40:02 +03:00
|
|
|
vivt_flush_cache_mm(mm);
|
2005-09-08 18:32:23 +04:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cache_is_vipt_aliasing()) {
|
|
|
|
asm( "mcr p15, 0, %0, c7, c14, 0\n"
|
2009-10-25 01:36:36 +04:00
|
|
|
" mcr p15, 0, %0, c7, c10, 4"
|
2005-09-08 18:32:23 +04:00
|
|
|
:
|
|
|
|
: "r" (0)
|
|
|
|
: "cc");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void flush_cache_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
|
|
|
|
{
|
|
|
|
if (cache_is_vivt()) {
|
2009-10-25 13:40:02 +03:00
|
|
|
vivt_flush_cache_range(vma, start, end);
|
2005-09-08 18:32:23 +04:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cache_is_vipt_aliasing()) {
|
|
|
|
asm( "mcr p15, 0, %0, c7, c14, 0\n"
|
2009-10-25 01:36:36 +04:00
|
|
|
" mcr p15, 0, %0, c7, c10, 4"
|
2005-09-08 18:32:23 +04:00
|
|
|
:
|
|
|
|
: "r" (0)
|
|
|
|
: "cc");
|
|
|
|
}
|
2009-10-25 16:35:13 +03:00
|
|
|
|
2009-10-25 17:12:27 +03:00
|
|
|
if (vma->vm_flags & VM_EXEC)
|
2009-10-25 16:35:13 +03:00
|
|
|
__flush_icache_all();
|
2005-09-08 18:32:23 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
void flush_cache_page(struct vm_area_struct *vma, unsigned long user_addr, unsigned long pfn)
|
|
|
|
{
|
|
|
|
if (cache_is_vivt()) {
|
2009-10-25 13:40:02 +03:00
|
|
|
vivt_flush_cache_page(vma, user_addr, pfn);
|
2005-09-08 18:32:23 +04:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2009-10-25 01:58:40 +04:00
|
|
|
if (cache_is_vipt_aliasing()) {
|
2005-09-08 18:32:23 +04:00
|
|
|
flush_pfn_alias(pfn, user_addr);
|
2009-10-25 01:58:40 +04:00
|
|
|
__flush_icache_all();
|
|
|
|
}
|
2009-10-25 16:35:13 +03:00
|
|
|
|
|
|
|
if (vma->vm_flags & VM_EXEC && icache_is_vivt_asid_tagged())
|
|
|
|
__flush_icache_all();
|
2005-09-08 18:32:23 +04:00
|
|
|
}
|
2010-09-13 19:19:41 +04:00
|
|
|
|
2009-11-05 16:29:36 +03:00
|
|
|
#else
|
2010-09-13 19:19:41 +04:00
|
|
|
#define flush_pfn_alias(pfn,vaddr) do { } while (0)
|
|
|
|
#define flush_icache_alias(pfn,vaddr,len) do { } while (0)
|
2009-11-05 16:29:36 +03:00
|
|
|
#endif
|
2006-09-02 21:43:20 +04:00
|
|
|
|
2014-04-29 07:20:52 +04:00
|
|
|
#define FLAG_PA_IS_EXEC 1
|
|
|
|
#define FLAG_PA_CORE_IN_MM 2
|
|
|
|
|
2009-11-05 16:29:36 +03:00
|
|
|
static void flush_ptrace_access_other(void *args)
|
|
|
|
{
|
|
|
|
__flush_icache_all();
|
|
|
|
}
|
|
|
|
|
2014-04-29 07:20:52 +04:00
|
|
|
static inline
|
|
|
|
void __flush_ptrace_access(struct page *page, unsigned long uaddr, void *kaddr,
|
|
|
|
unsigned long len, unsigned int flags)
|
2006-09-02 21:43:20 +04:00
|
|
|
{
|
|
|
|
if (cache_is_vivt()) {
|
2014-04-29 07:20:52 +04:00
|
|
|
if (flags & FLAG_PA_CORE_IN_MM) {
|
2009-11-05 16:29:36 +03:00
|
|
|
unsigned long addr = (unsigned long)kaddr;
|
|
|
|
__cpuc_coherent_kern_range(addr, addr + len);
|
|
|
|
}
|
2006-09-02 21:43:20 +04:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cache_is_vipt_aliasing()) {
|
|
|
|
flush_pfn_alias(page_to_pfn(page), uaddr);
|
2009-10-25 01:58:40 +04:00
|
|
|
__flush_icache_all();
|
2006-09-02 21:43:20 +04:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2010-09-13 19:19:41 +04:00
|
|
|
/* VIPT non-aliasing D-cache */
|
2014-04-29 07:20:52 +04:00
|
|
|
if (flags & FLAG_PA_IS_EXEC) {
|
2006-09-02 21:43:20 +04:00
|
|
|
unsigned long addr = (unsigned long)kaddr;
|
2010-09-13 19:19:41 +04:00
|
|
|
if (icache_is_vipt_aliasing())
|
|
|
|
flush_icache_alias(page_to_pfn(page), uaddr, len);
|
|
|
|
else
|
|
|
|
__cpuc_coherent_kern_range(addr, addr + len);
|
2009-11-05 16:29:36 +03:00
|
|
|
if (cache_ops_need_broadcast())
|
|
|
|
smp_call_function(flush_ptrace_access_other,
|
|
|
|
NULL, 1);
|
2006-09-02 21:43:20 +04:00
|
|
|
}
|
|
|
|
}
|
2009-11-05 16:29:36 +03:00
|
|
|
|
2014-04-29 07:20:52 +04:00
|
|
|
static
|
|
|
|
void flush_ptrace_access(struct vm_area_struct *vma, struct page *page,
|
|
|
|
unsigned long uaddr, void *kaddr, unsigned long len)
|
|
|
|
{
|
|
|
|
unsigned int flags = 0;
|
|
|
|
if (cpumask_test_cpu(smp_processor_id(), mm_cpumask(vma->vm_mm)))
|
|
|
|
flags |= FLAG_PA_CORE_IN_MM;
|
|
|
|
if (vma->vm_flags & VM_EXEC)
|
|
|
|
flags |= FLAG_PA_IS_EXEC;
|
|
|
|
__flush_ptrace_access(page, uaddr, kaddr, len, flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
void flush_uprobe_xol_access(struct page *page, unsigned long uaddr,
|
|
|
|
void *kaddr, unsigned long len)
|
|
|
|
{
|
|
|
|
unsigned int flags = FLAG_PA_CORE_IN_MM|FLAG_PA_IS_EXEC;
|
|
|
|
|
|
|
|
__flush_ptrace_access(page, uaddr, kaddr, len, flags);
|
|
|
|
}
|
|
|
|
|
2009-11-05 16:29:36 +03:00
|
|
|
/*
|
|
|
|
* Copy user data from/to a page which is mapped into a different
|
|
|
|
* processes address space. Really, we want to allow our "user
|
|
|
|
* space" model to handle this.
|
|
|
|
*
|
|
|
|
* Note that this code needs to run on the current CPU.
|
|
|
|
*/
|
|
|
|
void copy_to_user_page(struct vm_area_struct *vma, struct page *page,
|
|
|
|
unsigned long uaddr, void *dst, const void *src,
|
|
|
|
unsigned long len)
|
|
|
|
{
|
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
preempt_disable();
|
2005-05-10 20:31:43 +04:00
|
|
|
#endif
|
2009-11-05 16:29:36 +03:00
|
|
|
memcpy(dst, src, len);
|
|
|
|
flush_ptrace_access(vma, page, uaddr, dst, len);
|
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
preempt_enable();
|
|
|
|
#endif
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2005-06-20 12:51:03 +04:00
|
|
|
void __flush_dcache_page(struct address_space *mapping, struct page *page)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
|
|
|
/*
|
|
|
|
* Writeback any data associated with the kernel mapping of this
|
|
|
|
* page. This ensures that data in the physical page is mutually
|
|
|
|
* coherent with the kernels mapping.
|
|
|
|
*/
|
2010-03-30 00:46:02 +04:00
|
|
|
if (!PageHighMem(page)) {
|
2013-05-17 15:33:28 +04:00
|
|
|
size_t page_size = PAGE_SIZE << compound_order(page);
|
|
|
|
__cpuc_flush_dcache_area(page_address(page), page_size);
|
2010-03-30 00:46:02 +04:00
|
|
|
} else {
|
2013-05-17 15:33:28 +04:00
|
|
|
unsigned long i;
|
2013-04-05 06:16:14 +04:00
|
|
|
if (cache_is_vipt_nonaliasing()) {
|
2013-05-17 15:33:28 +04:00
|
|
|
for (i = 0; i < (1 << compound_order(page)); i++) {
|
2013-12-16 20:25:52 +04:00
|
|
|
void *addr = kmap_atomic(page + i);
|
2013-04-05 06:16:14 +04:00
|
|
|
__cpuc_flush_dcache_area(addr, PAGE_SIZE);
|
2013-05-17 15:33:28 +04:00
|
|
|
kunmap_atomic(addr);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
for (i = 0; i < (1 << compound_order(page)); i++) {
|
2013-12-16 20:25:52 +04:00
|
|
|
void *addr = kmap_high_get(page + i);
|
2013-05-17 15:33:28 +04:00
|
|
|
if (addr) {
|
|
|
|
__cpuc_flush_dcache_area(addr, PAGE_SIZE);
|
2013-12-16 20:25:52 +04:00
|
|
|
kunmap_high(page + i);
|
2013-05-17 15:33:28 +04:00
|
|
|
}
|
2013-04-05 06:16:14 +04:00
|
|
|
}
|
2010-03-30 00:46:02 +04:00
|
|
|
}
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
/*
|
2005-06-20 12:51:03 +04:00
|
|
|
* If this is a page cache page, and we have an aliasing VIPT cache,
|
|
|
|
* we only need to do one flush - which would be at the relevant
|
2005-05-10 20:31:43 +04:00
|
|
|
* userspace colour, which is congruent with page->index.
|
|
|
|
*/
|
2009-10-25 02:05:34 +04:00
|
|
|
if (mapping && cache_is_vipt_aliasing())
|
2005-06-20 12:51:03 +04:00
|
|
|
flush_pfn_alias(page_to_pfn(page),
|
|
|
|
page->index << PAGE_CACHE_SHIFT);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void __flush_dcache_aliases(struct address_space *mapping, struct page *page)
|
|
|
|
{
|
|
|
|
struct mm_struct *mm = current->active_mm;
|
|
|
|
struct vm_area_struct *mpnt;
|
|
|
|
pgoff_t pgoff;
|
2005-05-10 20:31:43 +04:00
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
/*
|
|
|
|
* There are possible user space mappings of this page:
|
|
|
|
* - VIVT cache: we need to also write back and invalidate all user
|
|
|
|
* data in the current VM view associated with this page.
|
|
|
|
* - aliasing VIPT: we only need to find one mapping of this page.
|
|
|
|
*/
|
|
|
|
pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
|
|
|
|
|
|
|
|
flush_dcache_mmap_lock(mapping);
|
2012-10-09 03:31:25 +04:00
|
|
|
vma_interval_tree_foreach(mpnt, &mapping->i_mmap, pgoff, pgoff) {
|
2005-04-17 02:20:36 +04:00
|
|
|
unsigned long offset;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If this VMA is not in our MM, we can ignore it.
|
|
|
|
*/
|
|
|
|
if (mpnt->vm_mm != mm)
|
|
|
|
continue;
|
|
|
|
if (!(mpnt->vm_flags & VM_MAYSHARE))
|
|
|
|
continue;
|
|
|
|
offset = (pgoff - mpnt->vm_pgoff) << PAGE_SHIFT;
|
|
|
|
flush_cache_page(mpnt, mpnt->vm_start + offset, page_to_pfn(page));
|
|
|
|
}
|
|
|
|
flush_dcache_mmap_unlock(mapping);
|
|
|
|
}
|
|
|
|
|
2010-09-13 18:58:06 +04:00
|
|
|
#if __LINUX_ARM_ARCH__ >= 6
|
|
|
|
void __sync_icache_dcache(pte_t pteval)
|
|
|
|
{
|
|
|
|
unsigned long pfn;
|
|
|
|
struct page *page;
|
|
|
|
struct address_space *mapping;
|
|
|
|
|
|
|
|
if (cache_is_vipt_nonaliasing() && !pte_exec(pteval))
|
|
|
|
/* only flush non-aliasing VIPT caches for exec mappings */
|
|
|
|
return;
|
|
|
|
pfn = pte_pfn(pteval);
|
|
|
|
if (!pfn_valid(pfn))
|
|
|
|
return;
|
|
|
|
|
|
|
|
page = pfn_to_page(pfn);
|
|
|
|
if (cache_is_vipt_aliasing())
|
|
|
|
mapping = page_mapping(page);
|
|
|
|
else
|
|
|
|
mapping = NULL;
|
|
|
|
|
|
|
|
if (!test_and_set_bit(PG_dcache_clean, &page->flags))
|
|
|
|
__flush_dcache_page(mapping, page);
|
2011-05-16 18:41:15 +04:00
|
|
|
|
|
|
|
if (pte_exec(pteval))
|
2010-09-13 18:58:06 +04:00
|
|
|
__flush_icache_all();
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
/*
|
|
|
|
* Ensure cache coherency between kernel mapping and userspace mapping
|
|
|
|
* of this page.
|
|
|
|
*
|
|
|
|
* We have three cases to consider:
|
|
|
|
* - VIPT non-aliasing cache: fully coherent so nothing required.
|
|
|
|
* - VIVT: fully aliasing, so we need to handle every alias in our
|
|
|
|
* current VM view.
|
|
|
|
* - VIPT aliasing: need to handle one alias in our current VM view.
|
|
|
|
*
|
|
|
|
* If we need to handle aliasing:
|
|
|
|
* If the page only exists in the page cache and there are no user
|
|
|
|
* space mappings, we can be lazy and remember that we may have dirty
|
|
|
|
* kernel cache lines for later. Otherwise, we assume we have
|
|
|
|
* aliasing mappings.
|
2005-11-30 19:02:54 +03:00
|
|
|
*
|
2011-05-16 14:25:21 +04:00
|
|
|
* Note that we disable the lazy flush for SMP configurations where
|
|
|
|
* the cache maintenance operations are not automatically broadcasted.
|
2005-04-17 02:20:36 +04:00
|
|
|
*/
|
|
|
|
void flush_dcache_page(struct page *page)
|
|
|
|
{
|
2009-10-25 13:23:04 +03:00
|
|
|
struct address_space *mapping;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The zero page is never written to, so never has any dirty
|
|
|
|
* cache lines, and therefore never needs to be flushed.
|
|
|
|
*/
|
|
|
|
if (page == ZERO_PAGE(0))
|
|
|
|
return;
|
|
|
|
|
|
|
|
mapping = page_mapping(page);
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2010-09-13 18:58:37 +04:00
|
|
|
if (!cache_ops_need_broadcast() &&
|
ARM: 7746/1: mm: lazy cache flushing on non-mapped pages
Currently flush_dcache_page() thinks pages as non-mapped if
mapping_mapped(mapping) return false. This approach is very
coase:
- mmap on part of file may cause all pages backed on
the file being thought as mmaped
- file-backed pages aren't mapped into user space actually
if the memory mmaped on the file isn't accessed
This patch uses page_mapped() to decide if the page has been
mapped.
From the attached test code, I find there is much performance
improvement(>25%) when accessing page caches via read under this
situations, so memcpy benefits a lot from not flushing cache
under this situation.
No. read time without the patch No. read time with the patch
================================================================
No. 0, time 22615636 us No. 0, time 22014717 us
No. 1, time 4387851 us No. 1, time 3113184 us
No. 2, time 4276535 us No. 2, time 3005244 us
No. 3, time 4259821 us No. 3, time 3001565 us
No. 4, time 4263811 us No. 4, time 3002748 us
No. 5, time 4258486 us No. 5, time 3004104 us
No. 6, time 4253009 us No. 6, time 3002188 us
No. 7, time 4262809 us No. 7, time 2998196 us
No. 8, time 4264525 us No. 8, time 3007255 us
No. 9, time 4267795 us No. 9, time 3005094 us
1), No.0. is to read the file from storage device, and others are
to read the file from page caches basically.
2), file size is 512M, and is on ext4 over usb mass storage.
3), the test is done on Pandaboard.
unsigned int sum = 0;
unsigned long sum_val = 0;
static unsigned long tv_diff(struct timeval *tv1, struct timeval *tv2)
{
return (tv2->tv_sec - tv1->tv_sec) * 1000000 +
(tv2->tv_usec - tv1->tv_usec);
}
int main(int argc, char *argv[])
{
char *mbuf, fbuf;
int fd;
int i;
unsigned long page_size, size;
struct stat stat;
struct timeval t1, t2;
unsigned char *rbuf = malloc(32 * page_size);
if (!rbuf) {
printf(" %sn", "malloc failed");
exit(-1);
}
page_size = getpagesize();
fd = open(argv[1], O_RDWR);
assert(fd >= 0);
fstat(fd, &stat);
size = stat.st_size;
printf("%s: file %s, size %lu, page size %lun",
argv[0],
argv[1], size, page_size);
gettimeofday(&t1, NULL);
mbuf = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
if (!mbuf) {
printf(" %sn", "mmap failed");
exit(-1);
}
for (i = 0 ; i < size ; i += (page_size * 32)) {
int rcnt;
lseek(fd, i, SEEK_SET);
rcnt = read(fd, rbuf, page_size * 32);
if (rcnt != page_size * 32) {
printf("%s: read faildn", __func__);
exit(-1);
}
}
free(rbuf);
munmap(mbuf, size);
gettimeofday(&t2, NULL);
printf("tread mmaped time: %luusn", tv_diff(&t1, &t2));
close(fd);
}
Cc: Michel Lespinasse <walken@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2013-06-05 05:44:00 +04:00
|
|
|
mapping && !page_mapped(page))
|
2010-09-13 18:57:36 +04:00
|
|
|
clear_bit(PG_dcache_clean, &page->flags);
|
2010-09-13 18:58:37 +04:00
|
|
|
else {
|
2005-04-17 02:20:36 +04:00
|
|
|
__flush_dcache_page(mapping, page);
|
2005-06-20 12:51:03 +04:00
|
|
|
if (mapping && cache_is_vivt())
|
|
|
|
__flush_dcache_aliases(mapping, page);
|
2008-06-13 13:28:36 +04:00
|
|
|
else if (mapping)
|
|
|
|
__flush_icache_all();
|
2010-09-13 18:57:36 +04:00
|
|
|
set_bit(PG_dcache_clean, &page->flags);
|
2005-06-20 12:51:03 +04:00
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(flush_dcache_page);
|
2013-06-11 00:10:12 +04:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Ensure cache coherency for the kernel mapping of this page. We can
|
|
|
|
* assume that the page is pinned via kmap.
|
|
|
|
*
|
|
|
|
* If the page only exists in the page cache and there are no user
|
|
|
|
* space mappings, this is a no-op since the page was already marked
|
|
|
|
* dirty at creation. Otherwise, we need to flush the dirty kernel
|
|
|
|
* cache lines directly.
|
|
|
|
*/
|
|
|
|
void flush_kernel_dcache_page(struct page *page)
|
|
|
|
{
|
|
|
|
if (cache_is_vivt() || cache_is_vipt_aliasing()) {
|
|
|
|
struct address_space *mapping;
|
|
|
|
|
|
|
|
mapping = page_mapping(page);
|
|
|
|
|
|
|
|
if (!mapping || mapping_mapped(mapping)) {
|
|
|
|
void *addr;
|
|
|
|
|
|
|
|
addr = page_address(page);
|
|
|
|
/*
|
|
|
|
* kmap_atomic() doesn't set the page virtual
|
|
|
|
* address for highmem pages, and
|
|
|
|
* kunmap_atomic() takes care of cache
|
|
|
|
* flushing already.
|
|
|
|
*/
|
|
|
|
if (!IS_ENABLED(CONFIG_HIGHMEM) || addr)
|
|
|
|
__cpuc_flush_dcache_area(addr, PAGE_SIZE);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(flush_kernel_dcache_page);
|
2006-12-31 02:17:40 +03:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Flush an anonymous page so that users of get_user_pages()
|
|
|
|
* can safely access the data. The expected sequence is:
|
|
|
|
*
|
|
|
|
* get_user_pages()
|
|
|
|
* -> flush_anon_page
|
|
|
|
* memcpy() to/from page
|
|
|
|
* if written to page, flush_dcache_page()
|
|
|
|
*/
|
|
|
|
void __flush_anon_page(struct vm_area_struct *vma, struct page *page, unsigned long vmaddr)
|
|
|
|
{
|
|
|
|
unsigned long pfn;
|
|
|
|
|
|
|
|
/* VIPT non-aliasing caches need do nothing */
|
|
|
|
if (cache_is_vipt_nonaliasing())
|
|
|
|
return;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Write back and invalidate userspace mapping.
|
|
|
|
*/
|
|
|
|
pfn = page_to_pfn(page);
|
|
|
|
if (cache_is_vivt()) {
|
|
|
|
flush_cache_page(vma, vmaddr, pfn);
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* For aliasing VIPT, we can flush an alias of the
|
|
|
|
* userspace address only.
|
|
|
|
*/
|
|
|
|
flush_pfn_alias(pfn, vmaddr);
|
2009-10-25 01:58:40 +04:00
|
|
|
__flush_icache_all();
|
2006-12-31 02:17:40 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Invalidate kernel mapping. No data should be contained
|
|
|
|
* in this mapping of the page. FIXME: this is overkill
|
|
|
|
* since we actually ask for a write-back and invalidate.
|
|
|
|
*/
|
2009-11-26 15:56:21 +03:00
|
|
|
__cpuc_flush_dcache_area(page_address(page), PAGE_SIZE);
|
2006-12-31 02:17:40 +03:00
|
|
|
}
|
2014-10-10 02:29:20 +04:00
|
|
|
|
|
|
|
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
|
|
|
|
#ifdef CONFIG_HAVE_RCU_TABLE_FREE
|
|
|
|
void pmdp_splitting_flush(struct vm_area_struct *vma, unsigned long address,
|
|
|
|
pmd_t *pmdp)
|
|
|
|
{
|
|
|
|
pmd_t pmd = pmd_mksplitting(*pmdp);
|
|
|
|
VM_BUG_ON(address & ~PMD_MASK);
|
|
|
|
set_pmd_at(vma->vm_mm, address, pmdp, pmd);
|
|
|
|
|
|
|
|
/* dummy IPI to serialise against fast_gup */
|
|
|
|
kick_all_cpus_sync();
|
|
|
|
}
|
|
|
|
#endif /* CONFIG_HAVE_RCU_TABLE_FREE */
|
|
|
|
#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
|