linux_dsm_epyc7002/arch/arm/mm/flush.c

424 lines
11 KiB
C
Raw Normal View History

/*
* 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>
#include <linux/highmem.h>
#include <asm/cacheflush.h>
#include <asm/cachetype.h>
ARM: 6007/1: fix highmem with VIPT cache and DMA The VIVT cache of a highmem page is always flushed before the page is unmapped. This cache flush is explicit through flush_cache_kmaps() in flush_all_zero_pkmaps(), or through __cpuc_flush_dcache_area() in kunmap_atomic(). There is also an implicit flush of those highmem pages that were part of a process that just terminated making those pages free as the whole VIVT cache has to be flushed on every task switch. Hence unmapped highmem pages need no cache maintenance in that case. However unmapped pages may still be cached with a VIPT cache because the cache is tagged with physical addresses. There is no need for a whole cache flush during task switching for that reason, and despite the explicit cache flushes in flush_all_zero_pkmaps() and kunmap_atomic(), some highmem pages that were mapped in user space end up still cached even when they become unmapped. So, we do have to perform cache maintenance on those unmapped highmem pages in the context of DMA when using a VIPT cache. Unfortunately, it is not possible to perform that cache maintenance using physical addresses as all the L1 cache maintenance coprocessor functions accept virtual addresses only. Therefore we have no choice but to set up a temporary virtual mapping for that purpose. And of course the explicit cache flushing when unmapping a highmem page on a system with a VIPT cache now can go, which should increase performance. While at it, because the code in __flush_dcache_page() has to be modified anyway, let's also make sure the mapped highmem pages are pinned with kmap_high_get() for the duration of the cache maintenance operation. Because kunmap() does unmap highmem pages lazily, it was reported by Gary King <GKing@nvidia.com> that those pages ended up being unmapped during cache maintenance on SMP causing segmentation faults. Signed-off-by: Nicolas Pitre <nico@marvell.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-03-30 03:46:02 +07:00
#include <asm/highmem.h>
#include <asm/smp_plat.h>
#include <asm/tlbflush.h>
#include <linux/hugetlb.h>
#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 05:44:46 +07:00
#ifdef CONFIG_ARM_HEAVY_MB
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 05:44:46 +07:00
void arm_heavy_mb(void)
{
#ifdef CONFIG_OUTER_CACHE_SYNC
if (outer_cache.sync)
outer_cache.sync();
#endif
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 05:44:46 +07:00
}
EXPORT_SYMBOL(arm_heavy_mb);
#endif
#ifdef CONFIG_CPU_CACHE_VIPT
static void flush_pfn_alias(unsigned long pfn, unsigned long vaddr)
{
unsigned long to = FLUSH_ALIAS_START + (CACHE_COLOUR(vaddr) << PAGE_SHIFT);
const int zero = 0;
set_top_pte(to, pfn_pte(pfn, PAGE_KERNEL));
asm( "mcrr p15, 0, %1, %0, c14\n"
" mcr p15, 0, %2, c7, c10, 4"
:
: "r" (to), "r" (to + PAGE_SIZE - 1), "r" (zero)
: "cc");
}
static void flush_icache_alias(unsigned long pfn, unsigned long vaddr, unsigned long len)
{
unsigned long va = FLUSH_ALIAS_START + (CACHE_COLOUR(vaddr) << PAGE_SHIFT);
unsigned long offset = vaddr & (PAGE_SIZE - 1);
unsigned long to;
set_top_pte(va, pfn_pte(pfn, PAGE_KERNEL));
to = va + offset;
flush_icache_range(to, to + len);
}
void flush_cache_mm(struct mm_struct *mm)
{
if (cache_is_vivt()) {
vivt_flush_cache_mm(mm);
return;
}
if (cache_is_vipt_aliasing()) {
asm( "mcr p15, 0, %0, c7, c14, 0\n"
" mcr p15, 0, %0, c7, c10, 4"
:
: "r" (0)
: "cc");
}
}
void flush_cache_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
{
if (cache_is_vivt()) {
vivt_flush_cache_range(vma, start, end);
return;
}
if (cache_is_vipt_aliasing()) {
asm( "mcr p15, 0, %0, c7, c14, 0\n"
" mcr p15, 0, %0, c7, c10, 4"
:
: "r" (0)
: "cc");
}
if (vma->vm_flags & VM_EXEC)
__flush_icache_all();
}
void flush_cache_page(struct vm_area_struct *vma, unsigned long user_addr, unsigned long pfn)
{
if (cache_is_vivt()) {
vivt_flush_cache_page(vma, user_addr, pfn);
return;
}
if (cache_is_vipt_aliasing()) {
flush_pfn_alias(pfn, user_addr);
__flush_icache_all();
}
if (vma->vm_flags & VM_EXEC && icache_is_vivt_asid_tagged())
__flush_icache_all();
}
#else
#define flush_pfn_alias(pfn,vaddr) do { } while (0)
#define flush_icache_alias(pfn,vaddr,len) do { } while (0)
#endif
[ARM] 3762/1: Fix ptrace cache coherency bug for ARM1136 VIPT nonaliasing Harvard caches Patch from George G. Davis Resolve ARM1136 VIPT non-aliasing cache coherency issues observed when using ptrace to set breakpoints and cleanup copy_{to,from}_user_page() while we're here as requested by Russell King because "it's also far too heavy on non-v6 CPUs". NOTES: 1. Only access_process_vm() calls copy_{to,from}_user_page(). 2. access_process_vm() calls get_user_pages() to pin down the "page". 3. get_user_pages() calls flush_dcache_page(page) which ensures cache coherency between kernel and userspace mappings of "page". However flush_dcache_page(page) may not invalidate I-Cache over this range for all cases, specifically, I-Cache is not invalidated for the VIPT non-aliasing case. So memory is consistent between kernel and user space mappings of "page" but I-Cache may still be hot over this range. IOW, we don't have to worry about flush_cache_page() before memcpy(). 4. Now, for the copy_to_user_page() case, after memcpy(), we must flush the caches so memory is consistent with kernel cache entries and invalidate the I-Cache if this mm region is executable. We don't need to do anything after memcpy() for the copy_from_user_page() case since kernel cache entries will be invalidated via the same process above if we access "page" again. The flush_ptrace_access() function (borrowed from SPARC64 implementation) is added to handle cache flushing after memcpy() for the copy_to_user_page() case. Signed-off-by: George G. Davis <gdavis@mvista.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2006-09-03 00:43:20 +07:00
2014-04-29 10:20:52 +07:00
#define FLAG_PA_IS_EXEC 1
#define FLAG_PA_CORE_IN_MM 2
static void flush_ptrace_access_other(void *args)
{
__flush_icache_all();
}
2014-04-29 10:20:52 +07:00
static inline
void __flush_ptrace_access(struct page *page, unsigned long uaddr, void *kaddr,
unsigned long len, unsigned int flags)
[ARM] 3762/1: Fix ptrace cache coherency bug for ARM1136 VIPT nonaliasing Harvard caches Patch from George G. Davis Resolve ARM1136 VIPT non-aliasing cache coherency issues observed when using ptrace to set breakpoints and cleanup copy_{to,from}_user_page() while we're here as requested by Russell King because "it's also far too heavy on non-v6 CPUs". NOTES: 1. Only access_process_vm() calls copy_{to,from}_user_page(). 2. access_process_vm() calls get_user_pages() to pin down the "page". 3. get_user_pages() calls flush_dcache_page(page) which ensures cache coherency between kernel and userspace mappings of "page". However flush_dcache_page(page) may not invalidate I-Cache over this range for all cases, specifically, I-Cache is not invalidated for the VIPT non-aliasing case. So memory is consistent between kernel and user space mappings of "page" but I-Cache may still be hot over this range. IOW, we don't have to worry about flush_cache_page() before memcpy(). 4. Now, for the copy_to_user_page() case, after memcpy(), we must flush the caches so memory is consistent with kernel cache entries and invalidate the I-Cache if this mm region is executable. We don't need to do anything after memcpy() for the copy_from_user_page() case since kernel cache entries will be invalidated via the same process above if we access "page" again. The flush_ptrace_access() function (borrowed from SPARC64 implementation) is added to handle cache flushing after memcpy() for the copy_to_user_page() case. Signed-off-by: George G. Davis <gdavis@mvista.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2006-09-03 00:43:20 +07:00
{
if (cache_is_vivt()) {
2014-04-29 10:20:52 +07:00
if (flags & FLAG_PA_CORE_IN_MM) {
unsigned long addr = (unsigned long)kaddr;
__cpuc_coherent_kern_range(addr, addr + len);
}
[ARM] 3762/1: Fix ptrace cache coherency bug for ARM1136 VIPT nonaliasing Harvard caches Patch from George G. Davis Resolve ARM1136 VIPT non-aliasing cache coherency issues observed when using ptrace to set breakpoints and cleanup copy_{to,from}_user_page() while we're here as requested by Russell King because "it's also far too heavy on non-v6 CPUs". NOTES: 1. Only access_process_vm() calls copy_{to,from}_user_page(). 2. access_process_vm() calls get_user_pages() to pin down the "page". 3. get_user_pages() calls flush_dcache_page(page) which ensures cache coherency between kernel and userspace mappings of "page". However flush_dcache_page(page) may not invalidate I-Cache over this range for all cases, specifically, I-Cache is not invalidated for the VIPT non-aliasing case. So memory is consistent between kernel and user space mappings of "page" but I-Cache may still be hot over this range. IOW, we don't have to worry about flush_cache_page() before memcpy(). 4. Now, for the copy_to_user_page() case, after memcpy(), we must flush the caches so memory is consistent with kernel cache entries and invalidate the I-Cache if this mm region is executable. We don't need to do anything after memcpy() for the copy_from_user_page() case since kernel cache entries will be invalidated via the same process above if we access "page" again. The flush_ptrace_access() function (borrowed from SPARC64 implementation) is added to handle cache flushing after memcpy() for the copy_to_user_page() case. Signed-off-by: George G. Davis <gdavis@mvista.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2006-09-03 00:43:20 +07:00
return;
}
if (cache_is_vipt_aliasing()) {
flush_pfn_alias(page_to_pfn(page), uaddr);
__flush_icache_all();
[ARM] 3762/1: Fix ptrace cache coherency bug for ARM1136 VIPT nonaliasing Harvard caches Patch from George G. Davis Resolve ARM1136 VIPT non-aliasing cache coherency issues observed when using ptrace to set breakpoints and cleanup copy_{to,from}_user_page() while we're here as requested by Russell King because "it's also far too heavy on non-v6 CPUs". NOTES: 1. Only access_process_vm() calls copy_{to,from}_user_page(). 2. access_process_vm() calls get_user_pages() to pin down the "page". 3. get_user_pages() calls flush_dcache_page(page) which ensures cache coherency between kernel and userspace mappings of "page". However flush_dcache_page(page) may not invalidate I-Cache over this range for all cases, specifically, I-Cache is not invalidated for the VIPT non-aliasing case. So memory is consistent between kernel and user space mappings of "page" but I-Cache may still be hot over this range. IOW, we don't have to worry about flush_cache_page() before memcpy(). 4. Now, for the copy_to_user_page() case, after memcpy(), we must flush the caches so memory is consistent with kernel cache entries and invalidate the I-Cache if this mm region is executable. We don't need to do anything after memcpy() for the copy_from_user_page() case since kernel cache entries will be invalidated via the same process above if we access "page" again. The flush_ptrace_access() function (borrowed from SPARC64 implementation) is added to handle cache flushing after memcpy() for the copy_to_user_page() case. Signed-off-by: George G. Davis <gdavis@mvista.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2006-09-03 00:43:20 +07:00
return;
}
/* VIPT non-aliasing D-cache */
2014-04-29 10:20:52 +07:00
if (flags & FLAG_PA_IS_EXEC) {
[ARM] 3762/1: Fix ptrace cache coherency bug for ARM1136 VIPT nonaliasing Harvard caches Patch from George G. Davis Resolve ARM1136 VIPT non-aliasing cache coherency issues observed when using ptrace to set breakpoints and cleanup copy_{to,from}_user_page() while we're here as requested by Russell King because "it's also far too heavy on non-v6 CPUs". NOTES: 1. Only access_process_vm() calls copy_{to,from}_user_page(). 2. access_process_vm() calls get_user_pages() to pin down the "page". 3. get_user_pages() calls flush_dcache_page(page) which ensures cache coherency between kernel and userspace mappings of "page". However flush_dcache_page(page) may not invalidate I-Cache over this range for all cases, specifically, I-Cache is not invalidated for the VIPT non-aliasing case. So memory is consistent between kernel and user space mappings of "page" but I-Cache may still be hot over this range. IOW, we don't have to worry about flush_cache_page() before memcpy(). 4. Now, for the copy_to_user_page() case, after memcpy(), we must flush the caches so memory is consistent with kernel cache entries and invalidate the I-Cache if this mm region is executable. We don't need to do anything after memcpy() for the copy_from_user_page() case since kernel cache entries will be invalidated via the same process above if we access "page" again. The flush_ptrace_access() function (borrowed from SPARC64 implementation) is added to handle cache flushing after memcpy() for the copy_to_user_page() case. Signed-off-by: George G. Davis <gdavis@mvista.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2006-09-03 00:43:20 +07:00
unsigned long addr = (unsigned long)kaddr;
if (icache_is_vipt_aliasing())
flush_icache_alias(page_to_pfn(page), uaddr, len);
else
__cpuc_coherent_kern_range(addr, addr + len);
if (cache_ops_need_broadcast())
smp_call_function(flush_ptrace_access_other,
NULL, 1);
[ARM] 3762/1: Fix ptrace cache coherency bug for ARM1136 VIPT nonaliasing Harvard caches Patch from George G. Davis Resolve ARM1136 VIPT non-aliasing cache coherency issues observed when using ptrace to set breakpoints and cleanup copy_{to,from}_user_page() while we're here as requested by Russell King because "it's also far too heavy on non-v6 CPUs". NOTES: 1. Only access_process_vm() calls copy_{to,from}_user_page(). 2. access_process_vm() calls get_user_pages() to pin down the "page". 3. get_user_pages() calls flush_dcache_page(page) which ensures cache coherency between kernel and userspace mappings of "page". However flush_dcache_page(page) may not invalidate I-Cache over this range for all cases, specifically, I-Cache is not invalidated for the VIPT non-aliasing case. So memory is consistent between kernel and user space mappings of "page" but I-Cache may still be hot over this range. IOW, we don't have to worry about flush_cache_page() before memcpy(). 4. Now, for the copy_to_user_page() case, after memcpy(), we must flush the caches so memory is consistent with kernel cache entries and invalidate the I-Cache if this mm region is executable. We don't need to do anything after memcpy() for the copy_from_user_page() case since kernel cache entries will be invalidated via the same process above if we access "page" again. The flush_ptrace_access() function (borrowed from SPARC64 implementation) is added to handle cache flushing after memcpy() for the copy_to_user_page() case. Signed-off-by: George G. Davis <gdavis@mvista.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2006-09-03 00:43:20 +07:00
}
}
2014-04-29 10:20:52 +07:00
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);
}
/*
* 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();
#endif
memcpy(dst, src, len);
flush_ptrace_access(vma, page, uaddr, dst, len);
#ifdef CONFIG_SMP
preempt_enable();
#endif
}
void __flush_dcache_page(struct address_space *mapping, struct page *page)
{
/*
* 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.
*/
ARM: 6007/1: fix highmem with VIPT cache and DMA The VIVT cache of a highmem page is always flushed before the page is unmapped. This cache flush is explicit through flush_cache_kmaps() in flush_all_zero_pkmaps(), or through __cpuc_flush_dcache_area() in kunmap_atomic(). There is also an implicit flush of those highmem pages that were part of a process that just terminated making those pages free as the whole VIVT cache has to be flushed on every task switch. Hence unmapped highmem pages need no cache maintenance in that case. However unmapped pages may still be cached with a VIPT cache because the cache is tagged with physical addresses. There is no need for a whole cache flush during task switching for that reason, and despite the explicit cache flushes in flush_all_zero_pkmaps() and kunmap_atomic(), some highmem pages that were mapped in user space end up still cached even when they become unmapped. So, we do have to perform cache maintenance on those unmapped highmem pages in the context of DMA when using a VIPT cache. Unfortunately, it is not possible to perform that cache maintenance using physical addresses as all the L1 cache maintenance coprocessor functions accept virtual addresses only. Therefore we have no choice but to set up a temporary virtual mapping for that purpose. And of course the explicit cache flushing when unmapping a highmem page on a system with a VIPT cache now can go, which should increase performance. While at it, because the code in __flush_dcache_page() has to be modified anyway, let's also make sure the mapped highmem pages are pinned with kmap_high_get() for the duration of the cache maintenance operation. Because kunmap() does unmap highmem pages lazily, it was reported by Gary King <GKing@nvidia.com> that those pages ended up being unmapped during cache maintenance on SMP causing segmentation faults. Signed-off-by: Nicolas Pitre <nico@marvell.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-03-30 03:46:02 +07:00
if (!PageHighMem(page)) {
size_t page_size = PAGE_SIZE << compound_order(page);
__cpuc_flush_dcache_area(page_address(page), page_size);
ARM: 6007/1: fix highmem with VIPT cache and DMA The VIVT cache of a highmem page is always flushed before the page is unmapped. This cache flush is explicit through flush_cache_kmaps() in flush_all_zero_pkmaps(), or through __cpuc_flush_dcache_area() in kunmap_atomic(). There is also an implicit flush of those highmem pages that were part of a process that just terminated making those pages free as the whole VIVT cache has to be flushed on every task switch. Hence unmapped highmem pages need no cache maintenance in that case. However unmapped pages may still be cached with a VIPT cache because the cache is tagged with physical addresses. There is no need for a whole cache flush during task switching for that reason, and despite the explicit cache flushes in flush_all_zero_pkmaps() and kunmap_atomic(), some highmem pages that were mapped in user space end up still cached even when they become unmapped. So, we do have to perform cache maintenance on those unmapped highmem pages in the context of DMA when using a VIPT cache. Unfortunately, it is not possible to perform that cache maintenance using physical addresses as all the L1 cache maintenance coprocessor functions accept virtual addresses only. Therefore we have no choice but to set up a temporary virtual mapping for that purpose. And of course the explicit cache flushing when unmapping a highmem page on a system with a VIPT cache now can go, which should increase performance. While at it, because the code in __flush_dcache_page() has to be modified anyway, let's also make sure the mapped highmem pages are pinned with kmap_high_get() for the duration of the cache maintenance operation. Because kunmap() does unmap highmem pages lazily, it was reported by Gary King <GKing@nvidia.com> that those pages ended up being unmapped during cache maintenance on SMP causing segmentation faults. Signed-off-by: Nicolas Pitre <nico@marvell.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-03-30 03:46:02 +07:00
} else {
unsigned long i;
if (cache_is_vipt_nonaliasing()) {
for (i = 0; i < (1 << compound_order(page)); i++) {
void *addr = kmap_atomic(page + i);
__cpuc_flush_dcache_area(addr, PAGE_SIZE);
kunmap_atomic(addr);
}
} else {
for (i = 0; i < (1 << compound_order(page)); i++) {
void *addr = kmap_high_get(page + i);
if (addr) {
__cpuc_flush_dcache_area(addr, PAGE_SIZE);
kunmap_high(page + i);
}
}
ARM: 6007/1: fix highmem with VIPT cache and DMA The VIVT cache of a highmem page is always flushed before the page is unmapped. This cache flush is explicit through flush_cache_kmaps() in flush_all_zero_pkmaps(), or through __cpuc_flush_dcache_area() in kunmap_atomic(). There is also an implicit flush of those highmem pages that were part of a process that just terminated making those pages free as the whole VIVT cache has to be flushed on every task switch. Hence unmapped highmem pages need no cache maintenance in that case. However unmapped pages may still be cached with a VIPT cache because the cache is tagged with physical addresses. There is no need for a whole cache flush during task switching for that reason, and despite the explicit cache flushes in flush_all_zero_pkmaps() and kunmap_atomic(), some highmem pages that were mapped in user space end up still cached even when they become unmapped. So, we do have to perform cache maintenance on those unmapped highmem pages in the context of DMA when using a VIPT cache. Unfortunately, it is not possible to perform that cache maintenance using physical addresses as all the L1 cache maintenance coprocessor functions accept virtual addresses only. Therefore we have no choice but to set up a temporary virtual mapping for that purpose. And of course the explicit cache flushing when unmapping a highmem page on a system with a VIPT cache now can go, which should increase performance. While at it, because the code in __flush_dcache_page() has to be modified anyway, let's also make sure the mapped highmem pages are pinned with kmap_high_get() for the duration of the cache maintenance operation. Because kunmap() does unmap highmem pages lazily, it was reported by Gary King <GKing@nvidia.com> that those pages ended up being unmapped during cache maintenance on SMP causing segmentation faults. Signed-off-by: Nicolas Pitre <nico@marvell.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2010-03-30 03:46:02 +07:00
}
}
/*
* 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
* userspace colour, which is congruent with page->index.
*/
if (mapping && cache_is_vipt_aliasing())
flush_pfn_alias(page_to_pfn(page),
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
page->index << PAGE_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;
/*
* 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.
*/
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
pgoff = page->index;
flush_dcache_mmap_lock(mapping);
vma_interval_tree_foreach(mpnt, &mapping->i_mmap, pgoff, pgoff) {
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);
}
#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);
if (pte_exec(pteval))
__flush_icache_all();
}
#endif
/*
* 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.
*
* Note that we disable the lazy flush for SMP configurations where
* the cache maintenance operations are not automatically broadcasted.
*/
void flush_dcache_page(struct page *page)
{
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;
if (!cache_ops_need_broadcast() && cache_is_vipt_nonaliasing()) {
if (test_bit(PG_dcache_clean, &page->flags))
clear_bit(PG_dcache_clean, &page->flags);
return;
}
mapping = page_mapping(page);
if (!cache_ops_need_broadcast() &&
mapping && !page_mapcount(page))
clear_bit(PG_dcache_clean, &page->flags);
else {
__flush_dcache_page(mapping, page);
if (mapping && cache_is_vivt())
__flush_dcache_aliases(mapping, page);
else if (mapping)
__flush_icache_all();
set_bit(PG_dcache_clean, &page->flags);
}
}
EXPORT_SYMBOL(flush_dcache_page);
/*
* 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);
/*
* 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);
__flush_icache_all();
}
/*
* 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.
*/
__cpuc_flush_dcache_area(page_address(page), PAGE_SIZE);
}