....
 

Guardian Digital Inc. > InfoCenter > Mailing List Archives > Linux Kernel

Linux Kernel Mailing List Archive

From: Andrea Arcangeli (andrea@suse.de)
Date: Wed Dec 15 2004 - 14:03:43 EST


This was my original code in GA/SP1:

#ifndef CONFIG_DISCONTIGMEM
        /* this check is unreliable with numa enabled */
        BUG_ON(!pfn_valid(page_to_pfn(new_page)));
#endif
        pageable = !PageReserved(new_page);
        as = !!new_page->mapping;

        BUG_ON(!pageable && as);

        pageable &= as;

        /* ->nopage cannot return swapcache */
        BUG_ON(PageSwapCache(new_page));
        /* ->nopage cannot return anonymous pages */
        BUG_ON(PageAnon(new_page));

[..]

                /* a reserved vma cannot have pageable pages in it */
                BUG_ON(vma->vm_flags & VM_RESERVED);
[..]
        if (likely(pageable))
                        page_add_rmap(new_page, vma, address, anon);

Hugh rejected all the above robustness bugcheck for no apparent good
reason.

In the process he seems that Hugh introduced a bug that didn't exist in
my original code, can you try this?

From: Andrea Arcangeli <andrea@suse.de>
Subject: don't track pages not belonging to a mapping out of ->nopage

Signed-off-by: Andrea Arcangeli <andrea@suse.de>

--- sl9.2/mm/memory.c.~1~ 2004-12-15 14:20:54.000000000 +0100
+++ sl9.2/mm/memory.c 2004-12-15 19:59:59.806978776 +0100
@@ -427,7 +427,8 @@ static void zap_pte_range(struct mmu_gat
                         if (pte_young(pte) && !PageAnon(page))
                                 mark_page_accessed(page);
                         tlb->freed++;
- page_remove_rmap(page);
+ if (page->mapping)
+ page_remove_rmap(page);
                         tlb_remove_page(tlb, page);
                         continue;
                 }
@@ -1495,7 +1496,7 @@ do_no_page(struct mm_struct *mm, struct
         pte_t entry;
         int sequence = 0;
         int ret = VM_FAULT_MINOR;
- int anon = 0;
+ int anon, pageable, as;
 
         if (!vma->vm_ops || !vma->vm_ops->nopage)
                 return do_anonymous_page(mm, vma, page_table,
@@ -1517,9 +1518,26 @@ retry:
         if (new_page == NOPAGE_OOM)
                 return VM_FAULT_OOM;
 
+#ifndef CONFIG_DISCONTIGMEM
+ /* this check is unreliable with numa enabled */
+ BUG_ON(!pfn_valid(page_to_pfn(new_page)));
+#endif
+ pageable = !PageReserved(new_page);
+ as = !!new_page->mapping;
+
+ BUG_ON(!pageable && as);
+
+ pageable &= as;
+
+ /* ->nopage cannot return swapcache */
+ BUG_ON(PageSwapCache(new_page));
+ /* ->nopage cannot return anonymous pages */
+ BUG_ON(PageAnon(new_page));
+
         /*
          * Should we do an early C-O-W break?
          */
+ anon = 0;
         if (write_access && !(vma->vm_flags & VM_SHARED)) {
                 struct page *page;
 
@@ -1532,6 +1550,10 @@ retry:
                 page_cache_release(new_page);
                 new_page = page;
                 anon = 1;
+ pageable = 1; /* not really necessary but cleaner */
+
+ /* a reserved vma cannot have pageable pages in it */
+ BUG_ON(vma->vm_flags & VM_RESERVED);
         }
 
         spin_lock(&mm->page_table_lock);
@@ -1571,7 +1593,7 @@ retry:
                 if (anon) {
                         lru_cache_add_active(new_page);
                         page_add_anon_rmap(new_page, vma, address);
- } else
+ } else if (pageable)
                         page_add_file_rmap(new_page);
                 pte_unmap(page_table);
         } else {
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



[ About Guardian Digital ] - [ Press Center ] - [ Contact Us ] - [ System Activation ] - [ Reseller Info ] - [ Online Store ] - [ Site Map ]
Copyright (c) 2000 - 2004 Guardian Digital, Inc. Linux Lockbox and EnGarde are Trademarks of Guardian Digital, Inc.