mm: avoid pages migrated to different zone [1/1]

PD#SWPL-881

Problem:
Amlogic modified code when compatcion from normal case.
And side effect is that pages in normal zone maybe migrated to
highmem zone. Which caused page_address get a NULL value and
kernel panic occurred.

Solution:
Avoid pages migarte to different zone by adding forbid_to_cma
flags.

Verify:
P212

Change-Id: I1d9c6653dc1069562db3c1be3f53a3510a51f0d1
Signed-off-by: tao zeng <tao.zeng@amlogic.com>

diff --git a/include/linux/amlogic/aml_cma.h b/include/linux/amlogic/aml_cma.h
index 947fd0a..4acb96f 100644 (file)
--- a/include/linux/amlogic/aml_cma.h
+++ b/include/linux/amlogic/aml_cma.h
@@ -50,6 +50,7 @@ struct compact_control {
        const int classzone_idx;        /* zone index of a direct compactor */
        struct zone *zone;
        bool contended;                 /* Signal lock or sched contention */
+       bool forbid_to_cma;             /* Forbit to migrate to cma */
 };
 
 static inline bool cma_forbidden_mask(gfp_t gfp_flags)

diff --git a/mm/compaction.c b/mm/compaction.c
index b24f499..2d9ee59 100644 (file)
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -648,6 +648,25 @@ static bool too_many_isolated(struct zone *zone)
 
        return isolated > (inactive + active) / 2;
 }
+#ifdef CONFIG_AMLOGIC_CMA
+static void check_page_to_cma(struct compact_control *cc, struct page *page)
+{
+       struct address_space *mapping;
+
+       if (cc->forbid_to_cma)  /* no need check once it is true */
+               return;
+
+       mapping = page_mapping(page);
+       if ((unsigned long)mapping & PAGE_MAPPING_ANON)
+               mapping = NULL;
+
+       if (PageKsm(page) && !PageSlab(page))
+               cc->forbid_to_cma = true;
+
+       if (mapping && cma_forbidden_mask(mapping_gfp_mask(mapping)))
+               cc->forbid_to_cma = true;
+}
+#endif
 
 /**
  * isolate_migratepages_block() - isolate all migrate-able pages within
@@ -746,6 +765,9 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn,
 
                page = pfn_to_page(low_pfn);
 
+       #ifdef CONFIG_AMLOGIC_CMA
+               check_page_to_cma(cc, page);
+       #endif
                if (!valid_page)
                        valid_page = page;
 
@@ -1098,7 +1120,7 @@ static void isolate_freepages(struct compact_control *cc)
                migrate_type = get_pageblock_migratetype(page);
                if (is_migrate_isolate(migrate_type))
                        continue;
-               if (is_migrate_cma(migrate_type) && cma_alloc_ref())
+               if (is_migrate_cma(migrate_type) && cc->forbid_to_cma)
                        continue;
        #endif /* CONFIG_AMLOGIC_CMA */
                /* Found a block suitable for isolating free pages from. */
@@ -1151,16 +1173,6 @@ static struct page *compaction_alloc(struct page *migratepage,
 {
        struct compact_control *cc = (struct compact_control *)data;
        struct page *freepage;
-#ifdef CONFIG_AMLOGIC_CMA
-       struct address_space *mapping;
-
-       mapping = page_mapping(migratepage);
-       if ((unsigned long)mapping & PAGE_MAPPING_ANON)
-               mapping = NULL;
-
-       if (mapping && !can_use_cma(mapping_gfp_mask(mapping)))
-               return alloc_page(mapping_gfp_mask(mapping) | __GFP_BDEV);
-#endif
 
        /*
         * Isolate free pages if necessary, and if we are not aborting due to
@@ -1558,6 +1570,9 @@ static enum compact_result compact_zone(struct zone *zone, struct compact_contro
 
        migrate_prep_local();
 
+#ifdef CONFIG_AMLOGIC_CMA
+       cc->forbid_to_cma = false;
+#endif
        while ((ret = compact_finished(zone, cc, migratetype)) ==
                                                COMPACT_CONTINUE) {
                int err;