lkml.org 
[lkml]   [2015]   [Apr]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 4/4] mm: make every pte dirty on do_swap_page
Hello Andrew,

On Thu, Apr 09, 2015 at 01:59:39PM -0700, Andrew Morton wrote:
> On Thu, 9 Apr 2015 08:50:25 +0900 Minchan Kim <minchan@kernel.org> wrote:
>
> > Bump.
>
> I'm getting the feeling that MADV_FREE is out of control.
>
> Below is the overall rollup of
>
> mm-support-madvisemadv_free.patch
> mm-support-madvisemadv_free-fix.patch
> mm-support-madvisemadv_free-fix-2.patch
> mm-dont-split-thp-page-when-syscall-is-called.patch
> mm-dont-split-thp-page-when-syscall-is-called-fix.patch
> mm-dont-split-thp-page-when-syscall-is-called-fix-2.patch
> mm-free-swp_entry-in-madvise_free.patch
> mm-move-lazy-free-pages-to-inactive-list.patch
> mm-move-lazy-free-pages-to-inactive-list-fix.patch
> mm-move-lazy-free-pages-to-inactive-list-fix-fix.patch
> mm-move-lazy-free-pages-to-inactive-list-fix-fix-fix.patch
> mm-make-every-pte-dirty-on-do_swap_page.patch
>
>
> It's pretty large and has its sticky little paws in all sorts of places.
>
>
> The feature would need to be pretty darn useful to justify a mainline
> merge. Has any such usefulness been demonstrated?

Jemalloc has used MADV_FREE instead of MADV_DONTNEED for a long time
in MADV_FREE supporting OSes(FreeBSD, Solaris, Darwin, Windows).
It used MADV_DONTNEED on only Linux because there was no the feature.

========================== &< ===========================

jemalloc:

/*
* Methods for purging unused pages differ between operating systems.
*
* madvise(..., MADV_DONTNEED) : On Linux, this immediately discards pages,
* such that new pages will be demand-zeroed if
* the address region is later touched.
* madvise(..., MADV_FREE) : On FreeBSD and Darwin, this marks pages as being
* unused, such that they will be discarded rather
* than swapped out.
*/
...

bool
pages_purge(void *addr, size_t length)
{
bool unzeroed;

#ifdef _WIN32
VirtualAlloc(addr, length, MEM_RESET, PAGE_READWRITE);
unzeroed = true;
#elif defined(JEMALLOC_HAVE_MADVISE)
# ifdef JEMALLOC_PURGE_MADVISE_DONTNEED
# define JEMALLOC_MADV_PURGE MADV_DONTNEED
# define JEMALLOC_MADV_ZEROS true
# elif defined(JEMALLOC_PURGE_MADVISE_FREE)
# define JEMALLOC_MADV_PURGE MADV_FREE
# define JEMALLOC_MADV_ZEROS false
# else
# error "No madvise(2) flag defined for purging unused dirty pages."
# endif
int err = madvise(addr, length, JEMALLOC_MADV_PURGE);
unzeroed = (!JEMALLOC_MADV_ZEROS || err != 0);
# undef JEMALLOC_MADV_PURGE
# undef JEMALLOC_MADV_ZEROS
#else
/* Last resort no-op. */
unzeroed = true;
#endif
return (unzeroed);
}


Tcmalloc is same page.

========================== &< ===========================

// MADV_FREE is specifically designed for use by malloc(), but only
// FreeBSD supports it; in linux we fall back to the somewhat inferior
// MADV_DONTNEED.
#if !defined(MADV_FREE) && defined(MADV_DONTNEED)
# define MADV_FREE MADV_DONTNEED
#endif

..

bool TCMalloc_SystemRelease(void* start, size_t length) {
#ifdef MADV_FREE
if (FLAGS_malloc_devmem_start) {
// It's not safe to use MADV_FREE/MADV_DONTNEED if we've been
// mapping /dev/mem for heap memory.
return false;
}
if (FLAGS_malloc_disable_memory_release) return false;
if (pagesize == 0) pagesize = getpagesize();
const size_t pagemask = pagesize - 1;

size_t new_start = reinterpret_cast<size_t>(start);
size_t end = new_start + length;
size_t new_end = end;

// Round up the starting address and round down the ending address
// to be page aligned:
new_start = (new_start + pagesize - 1) & ~pagemask;
new_end = new_end & ~pagemask;

ASSERT((new_start & pagemask) == 0);
ASSERT((new_end & pagemask) == 0);
ASSERT(new_start >= reinterpret_cast<size_t>(start));
ASSERT(new_end <= end);

if (new_end > new_start) {
int result;
do {
result = madvise(reinterpret_cast<char*>(new_start),
new_end - new_start, MADV_FREE);
} while (result == -1 && errno == EAGAIN);

return result != -1;
}
#endif
return false;
}

glibc want it, too.
https://sourceware.org/ml/libc-alpha/2015-02/msg00197.html


--
Kind regards,
Minchan Kim


\
 
 \ /
  Last update: 2015-04-10 02:41    [W:0.070 / U:2.804 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site