linux-sg2042/arch/arm/mm/copypage-v4wb.c

95 lines
2.8 KiB
C
Raw Normal View History

/*
* linux/arch/arm/mm/copypage-v4wb.c
*
* Copyright (C) 1995-1999 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/init.h>
#include <linux/highmem.h>
/*
* 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
* Dcache aliasing issue. The writes will be forwarded to the write buffer,
* and merged as appropriate.
*
* 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_highpage that does the right thing.
*/
ARM: 8805/2: remove unneeded naked function usage The naked attribute is known to confuse some old gcc versions when function arguments aren't explicitly listed as inline assembly operands despite the gcc documentation. That resulted in commit 9a40ac86152c ("ARM: 6164/1: Add kto and kfrom to input operands list."). Yet that commit has problems of its own by having assembly operand constraints completely wrong. If the generated code has been OK since then, it is due to luck rather than correctness. So this patch also provides proper assembly operand constraints, and removes two instances of redundant register usages in the implementation while at it. Inspection of the generated code with this patch doesn't show any obvious quality degradation either, so not relying on __naked at all will make the code less fragile, and avoid some issues with clang. The only remaining __naked instances (excluding the kprobes test cases) are exynos_pm_power_up_setup(), tc2_pm_power_up_setup() and cci_enable_port_for_self(. But in the first two cases, only the function address is used by the compiler with no chance of inlining it by mistake, and the third case is called from assembly code only. And the fact that no stack is available when the corresponding code is executed does warrant the __naked usage in those cases. Signed-off-by: Nicolas Pitre <nico@linaro.org> Reviewed-by: Stefan Agner <stefan@agner.ch> Tested-by: Stefan Agner <stefan@agner.ch> Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
2018-11-08 00:49:00 +08:00
static void v4wb_copy_user_page(void *kto, const void *kfrom)
{
ARM: 8805/2: remove unneeded naked function usage The naked attribute is known to confuse some old gcc versions when function arguments aren't explicitly listed as inline assembly operands despite the gcc documentation. That resulted in commit 9a40ac86152c ("ARM: 6164/1: Add kto and kfrom to input operands list."). Yet that commit has problems of its own by having assembly operand constraints completely wrong. If the generated code has been OK since then, it is due to luck rather than correctness. So this patch also provides proper assembly operand constraints, and removes two instances of redundant register usages in the implementation while at it. Inspection of the generated code with this patch doesn't show any obvious quality degradation either, so not relying on __naked at all will make the code less fragile, and avoid some issues with clang. The only remaining __naked instances (excluding the kprobes test cases) are exynos_pm_power_up_setup(), tc2_pm_power_up_setup() and cci_enable_port_for_self(. But in the first two cases, only the function address is used by the compiler with no chance of inlining it by mistake, and the third case is called from assembly code only. And the fact that no stack is available when the corresponding code is executed does warrant the __naked usage in those cases. Signed-off-by: Nicolas Pitre <nico@linaro.org> Reviewed-by: Stefan Agner <stefan@agner.ch> Tested-by: Stefan Agner <stefan@agner.ch> Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
2018-11-08 00:49:00 +08:00
int tmp;
asm volatile ("\
ldmia %1!, {r3, r4, ip, lr} @ 4\n\
1: mcr p15, 0, %0, c7, c6, 1 @ 1 invalidate D line\n\
stmia %0!, {r3, r4, ip, lr} @ 4\n\
ldmia %1!, {r3, r4, ip, lr} @ 4+1\n\
stmia %0!, {r3, r4, ip, lr} @ 4\n\
ldmia %1!, {r3, r4, ip, lr} @ 4\n\
mcr p15, 0, %0, c7, c6, 1 @ 1 invalidate D line\n\
stmia %0!, {r3, r4, ip, lr} @ 4\n\
ldmia %1!, {r3, r4, ip, lr} @ 4\n\
subs %2, %2, #1 @ 1\n\
stmia %0!, {r3, r4, ip, lr} @ 4\n\
ldmneia %1!, {r3, r4, ip, lr} @ 4\n\
bne 1b @ 1\n\
ARM: 8805/2: remove unneeded naked function usage The naked attribute is known to confuse some old gcc versions when function arguments aren't explicitly listed as inline assembly operands despite the gcc documentation. That resulted in commit 9a40ac86152c ("ARM: 6164/1: Add kto and kfrom to input operands list."). Yet that commit has problems of its own by having assembly operand constraints completely wrong. If the generated code has been OK since then, it is due to luck rather than correctness. So this patch also provides proper assembly operand constraints, and removes two instances of redundant register usages in the implementation while at it. Inspection of the generated code with this patch doesn't show any obvious quality degradation either, so not relying on __naked at all will make the code less fragile, and avoid some issues with clang. The only remaining __naked instances (excluding the kprobes test cases) are exynos_pm_power_up_setup(), tc2_pm_power_up_setup() and cci_enable_port_for_self(. But in the first two cases, only the function address is used by the compiler with no chance of inlining it by mistake, and the third case is called from assembly code only. And the fact that no stack is available when the corresponding code is executed does warrant the __naked usage in those cases. Signed-off-by: Nicolas Pitre <nico@linaro.org> Reviewed-by: Stefan Agner <stefan@agner.ch> Tested-by: Stefan Agner <stefan@agner.ch> Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
2018-11-08 00:49:00 +08:00
mcr p15, 0, %1, c7, c10, 4 @ 1 drain WB"
: "+&r" (kto), "+&r" (kfrom), "=&r" (tmp)
: "2" (PAGE_SIZE / 64)
: "r3", "r4", "ip", "lr");
}
void v4wb_copy_user_highpage(struct page *to, struct page *from,
unsigned long vaddr, struct vm_area_struct *vma)
{
void *kto, *kfrom;
kto = kmap_atomic(to);
kfrom = kmap_atomic(from);
ARM: Flush user mapping on VIVT processors when copying a page Steven Walter <stevenrwalter@gmail.com> writes: > I've been tracking down an instance of userspace data corruption, > and I believe I have found a window during fork where data can be > lost. The corruption is occurring on an ARMv5 system with VIVT > caches. Here's the scenario in question. Thread A is forking, > Thread B is running in userspace: > > Thread A: flush_cache_mm() (dup_mmap) > Thread B: writes to a page in the above mm > Thread A: pte_wrprotect() the above page (copy_one_pte) > Thread B: writes to the same page again > > During thread B's second write, he'll take a fault and enter the > do_wp_page() case. We'll end up calling copy_page(), which notably > uses the kernel virtual addresses for the old and new pages. This > means that the new page does not necessarily have the data from the > first write. Now there are two conflicting copies of the same > cache-line in dcache. If the userspace cache-line flushes before > the kernel cache-line, we lose the changes made during the first > write. do_wp_page does call flush_dcache_page on the newly-copied > page, but there's still a window where the CPU could flush the > userspace cache-line before then. Resolve this by flushing the user mapping before copying the page on processors with a writeback VIVT cache. Note: this does have a performance impact, and so needs further consideration before being merged - can we optimize out some of the cache flushes if, eg, we know that the page isn't yet mapped? Thread: <e06498070903061426o5875ad13hc6328aa0d3f08ed7@mail.gmail.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
2009-10-05 22:34:22 +08:00
flush_cache_page(vma, vaddr, page_to_pfn(from));
v4wb_copy_user_page(kto, kfrom);
kunmap_atomic(kfrom);
kunmap_atomic(kto);
}
/*
* ARMv4 optimised clear_user_page
*
* Same story as above.
*/
void v4wb_clear_user_highpage(struct page *page, unsigned long vaddr)
{
void *ptr, *kaddr = kmap_atomic(page);
asm volatile("\
mov r1, %2 @ 1\n\
mov r2, #0 @ 1\n\
mov r3, #0 @ 1\n\
mov ip, #0 @ 1\n\
mov lr, #0 @ 1\n\
1: mcr p15, 0, %0, c7, c6, 1 @ 1 invalidate D line\n\
stmia %0!, {r2, r3, ip, lr} @ 4\n\
stmia %0!, {r2, r3, ip, lr} @ 4\n\
mcr p15, 0, %0, c7, c6, 1 @ 1 invalidate D line\n\
stmia %0!, {r2, r3, ip, lr} @ 4\n\
stmia %0!, {r2, r3, ip, lr} @ 4\n\
subs r1, r1, #1 @ 1\n\
bne 1b @ 1\n\
mcr p15, 0, r1, c7, c10, 4 @ 1 drain WB"
: "=r" (ptr)
: "0" (kaddr), "I" (PAGE_SIZE / 64)
: "r1", "r2", "r3", "ip", "lr");
kunmap_atomic(kaddr);
}
struct cpu_user_fns v4wb_user_fns __initdata = {
.cpu_clear_user_highpage = v4wb_clear_user_highpage,
.cpu_copy_user_highpage = v4wb_copy_user_highpage,
};