diff options
| author | Luden <luden@ghostmail.com> | 2016-02-13 23:44:14 +0100 |
|---|---|---|
| committer | Ziyan <jaraidaniel@gmail.com> | 2016-05-01 23:35:55 +0200 |
| commit | ab26843c057773f42f5b46e4e4a519b39707253e (patch) | |
| tree | d90a6f0111048a797cf217becfc2d3b70bb6e6b8 /include | |
| parent | f819ad93dea3adee5f4a7ea87e1f6631aea83d44 (diff) | |
| download | kernel_samsung_tuna-ab26843c057773f42f5b46e4e4a519b39707253e.zip kernel_samsung_tuna-ab26843c057773f42f5b46e4e4a519b39707253e.tar.gz kernel_samsung_tuna-ab26843c057773f42f5b46e4e4a519b39707253e.tar.bz2 | |
Retry CMA allocations.
It looks like Linux pages migration code was never designed to be
deterministic or synchronous, there are multiple race conditions
between different parts of the code that make CMA allocation in one
step very likely to fail, especially for large memory ranges that
we need for Ducati.
Therefore, changing the allocation code to perform multiple allocation
attempts. To further increase the chances of the allocation to succeed
and to make things faster, the results of the previous allocation
attempts are kept - that is, all pages that are already isolated
stay isolated, so that retries are only for those pages that failed
isolation or migration in previous steps.
Additionally, there's a small delay between the steps so that the
chances of the other code to free the pages we need are higher.
Diffstat (limited to 'include')
| -rw-r--r-- | include/linux/ksm.h | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/include/linux/ksm.h b/include/linux/ksm.h index 3af13d3..d20721e 100644 --- a/include/linux/ksm.h +++ b/include/linux/ksm.h @@ -91,7 +91,7 @@ int rmap_walk_ksm(struct page *page, int (*rmap_one)(struct page *, void ksm_migrate_page(struct page *newpage, struct page *oldpage); void ksm_start_migration(void); void ksm_finalize_migration(unsigned long start_pfn, unsigned long nr_pages); -void ksm_abort_migration(void); +void ksm_stop_migration(void); #else /* !CONFIG_KSM */ |