Commit 13ece886d99cd668483113f7238e419d5331af26

Authored by Andrea Arcangeli
Committed by Linus Torvalds
1 parent ce83d2174e

thp: transparent hugepage config choice

Allow to choose between the always|madvise default for page faults and
khugepaged at config time.  madvise guarantees zero risk of higher memory
footprint for applications (applications using madvise(MADV_HUGEPAGE)
won't risk to use any more memory by backing their virtual regions with
hugepages).

Initially set the default to N and don't depend on EMBEDDED.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Showing 2 changed files with 30 additions and 2 deletions Side-by-side Diff

... ... @@ -303,9 +303,8 @@
303 303 See Documentation/nommu-mmap.txt for more information.
304 304  
305 305 config TRANSPARENT_HUGEPAGE
306   - bool "Transparent Hugepage Support" if EMBEDDED
  306 + bool "Transparent Hugepage Support"
307 307 depends on X86 && MMU
308   - default y
309 308 help
310 309 Transparent Hugepages allows the kernel to use huge pages and
311 310 huge tlb transparently to the applications whenever possible.
... ... @@ -315,6 +314,30 @@
315 314 up the pagetable walking.
316 315  
317 316 If memory constrained on embedded, you may want to say N.
  317 +
  318 +choice
  319 + prompt "Transparent Hugepage Support sysfs defaults"
  320 + depends on TRANSPARENT_HUGEPAGE
  321 + default TRANSPARENT_HUGEPAGE_ALWAYS
  322 + help
  323 + Selects the sysfs defaults for Transparent Hugepage Support.
  324 +
  325 + config TRANSPARENT_HUGEPAGE_ALWAYS
  326 + bool "always"
  327 + help
  328 + Enabling Transparent Hugepage always, can increase the
  329 + memory footprint of applications without a guaranteed
  330 + benefit but it will work automatically for all applications.
  331 +
  332 + config TRANSPARENT_HUGEPAGE_MADVISE
  333 + bool "madvise"
  334 + help
  335 + Enabling Transparent Hugepage madvise, will only provide a
  336 + performance improvement benefit to the applications using
  337 + madvise(MADV_HUGEPAGE) but it won't risk to increase the
  338 + memory footprint of applications without a guaranteed
  339 + benefit.
  340 +endchoice
318 341  
319 342 #
320 343 # UP and nommu archs use km based percpu allocator
... ... @@ -27,7 +27,12 @@
27 27 * allocations.
28 28 */
29 29 unsigned long transparent_hugepage_flags __read_mostly =
  30 +#ifdef CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS
30 31 (1<<TRANSPARENT_HUGEPAGE_FLAG)|
  32 +#endif
  33 +#ifdef CONFIG_TRANSPARENT_HUGEPAGE_MADVISE
  34 + (1<<TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG)|
  35 +#endif
31 36 (1<<TRANSPARENT_HUGEPAGE_DEFRAG_FLAG)|
32 37 (1<<TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG);
33 38