To: linux-mm@xxxxxxxxx, and linux-mips@xxxxxxxxxxxxxx
于 2012年01月20日 13:18, Zumeng Chen 写道:
> This patch fixes two issues as follows:
>
> For some filesystem with fsync == noop_fsync, there is not so much thing
> to do, so sys_msync just passes by for all arches but some CPUs.
>
> For some CPUs with cache aliases(dmesg|grep alias), it maybe has an issue,
> which reported by msync test suites in ltp-full when the memory of memset
> used by msync01 runs into cache alias randomly.
>
> Consider the following scenario used by msync01 in ltp-full:
> fildes = open(TEMPFILE, O_RDWR | O_CREAT, 0666)) < 0);
> .../* initialization fildes by write(fildes); */
> addr = mmap(0, page_sz, PROT_READ | PROT_WRITE, MAP_FILE | MAP_SHARED,
> fildes, 0);
> /* set buf with memset */
> memset(addr + OFFSET_1, 1, BUF_SIZE);
>
> /* msync the addr before using, or MS_SYNC*/
> msync(addr, page_sz, MS_ASYNC)
>
> /* Tries to read fildes */
> lseek(fildes, (off_t) OFFSET_1, SEEK_SET) != (off_t) OFFSET_1) {
> nread = read(fildes, read_buf, sizeof(read_buf));
>
> /* Then test the result */
> if (read_buf[count] != 1) {
>
> The test result is random too for CPUs with cache alias. So in this
> situation, we have to flush the related vma to make sure the read is
> correct.
>
> Signed-off-by: Zumeng Chen <zumeng.chen@xxxxxxxxxxxxx>
> ---
> mm/msync.c | 30 ++++++++++++++++++++++++++++++
> 1 files changed, 30 insertions(+), 0 deletions(-)
>
> diff --git a/mm/msync.c b/mm/msync.c
> index 632df45..0021a7e 100644
> --- a/mm/msync.c
> +++ b/mm/msync.c
> @@ -13,6 +13,14 @@
> #include <linux/file.h>
> #include <linux/syscalls.h>
> #include <linux/sched.h>
> +#include <asm/cacheflush.h>
> +
> +/* Cache aliases should be taken into accounts when msync. */
> +#ifdef cpu_has_dc_aliases
> +#define CPU_HAS_CACHE_ALIAS cpu_has_dc_aliases
> +#else
> +#define CPU_HAS_CACHE_ALIAS 0
> +#endif
>
> /*
> * MS_SYNC syncs the entire file - including mappings.
> @@ -78,6 +86,28 @@ SYSCALL_DEFINE3(msync, unsigned long, start, size_t, len, int, flags)
> }
> file = vma->vm_file;
> start = vma->vm_end;
> +
> + /*
> + * For some filesystems with fsync == noop_fsync, msync just
> + * passes by but some CPUs.
> + * For CPUs with cache alias, msync has to flush the related
> + * vma explicitly to make sure data coherency between memory
> + * and cache, which includes MS_SYNC or MS_ASYNC. That is to
> + * say, cache aliases should not be an async factor, so does
> + * msync on other arches without cache aliases.
> + */
> + if (file && file->f_op && file->f_op->fsync == noop_fsync) {
> + if (CPU_HAS_CACHE_ALIAS)
> + flush_cache_range(vma, vma->vm_start,
> + vma->vm_end);
> + if (start >= end) {
> + error = 0;
> + goto out_unlock;
> + }
> + vma = find_vma(mm, start);
> + continue;
> + }
> +
> if ((flags & MS_SYNC) && file &&
> (vma->vm_flags & VM_SHARED)) {
> get_file(file);
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>