a.k.a. mm-use-pagevec-to-rotate-reclaimable-page-fix-2.patch rotate_reclaimable_page() is not necessarily called with IRQ disabled: it must do so when calling the helpfully commented pagevec_move_tail().
Hmm, if pagevec_move_tail() is assuming IRQ disabled, why should it bother with irqsave/irqrestore variants of spin_lock? Because we like to see them on lru_lock? But vmscan.c already has one bare spin_lock(). Signed-off-by: Hugh Dickins <[EMAIL PROTECTED]> --- mm/swap.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) --- 2.6.23-rc7-mm1/mm/swap.c 2007-09-24 11:05:55.000000000 +0100 +++ linux/mm/swap.c 2007-09-24 13:08:12.000000000 +0100 @@ -102,7 +102,6 @@ static void pagevec_move_tail(struct pag int i; int pgmoved = 0; struct zone *zone = NULL; - unsigned long uninitialized_var(flags); for (i = 0; i < pagevec_count(pvec); i++) { struct page *page = pvec->pages[i]; @@ -110,9 +109,9 @@ static void pagevec_move_tail(struct pag if (pagezone != zone) { if (zone) - spin_unlock_irqrestore(&zone->lru_lock, flags); + spin_unlock(&zone->lru_lock); zone = pagezone; - spin_lock_irqsave(&zone->lru_lock, flags); + spin_lock(&zone->lru_lock); } if (PageLRU(page) && !PageActive(page)) { list_move_tail(&page->lru, &zone->inactive_list); @@ -120,7 +119,7 @@ static void pagevec_move_tail(struct pag } } if (zone) - spin_unlock_irqrestore(&zone->lru_lock, flags); + spin_unlock(&zone->lru_lock); __count_vm_events(PGROTATED, pgmoved); release_pages(pvec->pages, pvec->nr, pvec->cold); pagevec_reinit(pvec); @@ -150,6 +149,7 @@ void move_tail_pages() int rotate_reclaimable_page(struct page *page) { struct pagevec *pvec; + unsigned long flags; if (PageLocked(page)) return 1; @@ -162,9 +162,11 @@ int rotate_reclaimable_page(struct page if (PageLRU(page) && !PageActive(page)) { page_cache_get(page); + local_irq_save(flags); pvec = &__get_cpu_var(rotate_pvecs); if (!pagevec_add(pvec, page)) pagevec_move_tail(pvec); + local_irq_restore(flags); } if (!test_clear_page_writeback(page)) BUG(); - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/