Hi.
(The following (with minor differences in the text) was sent to linux-kernel
Thursday. I have received no response and the item is still on Teds list
of todo (www version anyway) so I'll risk a limb to the Quake monster :)
by sending this to you.)
Some time ago there was a thread about subject and a patch was posted
(by davej?). It was rejected because of the vmlist_modify_{un}lock
mess (AFAIR) and nothing has been done since. The patch below tries
to remake the attempt to move mm->rss inside the page_table_lock in mm/*.
I noticed that mm->rss is also modified in fs/{exec.c,binfmt_aout.c,
binfmt_elf.c} without the lock being held. Am I missing something or
are these buggy too? I have no idea of what assumptions can be made
in these code paths so I have not tried to produce patches for them.
Patch against 240-test11. Please comment.
diff -uar linux-240-t11/mm/memory.c linux/mm/memory.c
--- linux-240-t11/mm/memory.c Wed Nov 22 22:41:45 2000
+++ linux/mm/memory.c Thu Nov 23 21:45:58 2000
@@ -369,7 +369,6 @@
address = (address + PGDIR_SIZE) & PGDIR_MASK;
dir++;
} while (address && (address < end));
- spin_unlock(&mm->page_table_lock);
/*
* Update rss for the mm_struct (not necessarily current->mm)
* Notice that rss is an unsigned long.
@@ -378,6 +377,7 @@
mm->rss -= freed;
else
mm->rss = 0;
+ spin_unlock(&mm->page_table_lock);
}
@@ -1074,7 +1074,9 @@
flush_icache_page(vma, page);
}
+ spin_lock(&mm->page_table_lock);
mm->rss++;
+ spin_unlock(&mm->page_table_lock);
pte = mk_pte(page, vma->vm_page_prot);
@@ -1113,7 +1115,9 @@
return -1;
clear_user_highpage(page, addr);
entry = pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
+ spin_lock(&mm->page_table_lock);
mm->rss++;
+ spin_unlock(&mm->page_table_lock);
flush_page_to_ram(page);
}
set_pte(page_table, entry);
@@ -1152,7 +1156,9 @@
return 0;
if (new_page == NOPAGE_OOM)
return -1;
+ spin_lock(&mm->page_table_lock);
++mm->rss;
+ spin_unlock(&mm->page_table_lock);
/*
* This silly early PAGE_DIRTY setting removes a race
* due to the bad i386 page protection. But it's valid
diff -uar linux-240-t11/mm/mmap.c linux/mm/mmap.c
--- linux-240-t11/mm/mmap.c Wed Nov 22 22:41:45 2000
+++ linux/mm/mmap.c Thu Nov 23 21:45:58 2000
@@ -889,8 +889,8 @@
spin_lock(&mm->page_table_lock);
mpnt = mm->mmap;
mm->mmap = mm->mmap_avl = mm->mmap_cache = NULL;
- spin_unlock(&mm->page_table_lock);
mm->rss = 0;
+ spin_unlock(&mm->page_table_lock);
mm->total_vm = 0;
mm->locked_vm = 0;
while (mpnt) {
diff -uar linux-240-t11/mm/swapfile.c linux/mm/swapfile.c
--- linux-240-t11/mm/swapfile.c Sat Nov 4 23:27:17 2000
+++ linux/mm/swapfile.c Thu Nov 23 21:45:58 2000
@@ -231,7 +231,9 @@
set_pte(dir, pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
swap_free(entry);
get_page(page);
+ spin_lock(&vma->vm_mm->page_table_lock);
++vma->vm_mm->rss;
+ spin_unlock(&vma->vm_mm->page_table_lock);
}
static inline void unuse_pmd(struct vm_area_struct * vma, pmd_t *dir,
diff -uar linux-240-t11/mm/vmscan.c linux/mm/vmscan.c
--- linux-240-t11/mm/vmscan.c Wed Nov 22 22:41:45 2000
+++ linux/mm/vmscan.c Thu Nov 23 21:45:58 2000
@@ -95,7 +95,9 @@
set_pte(page_table, swp_entry_to_pte(entry));
drop_pte:
UnlockPage(page);
+ spin_lock(&mm->page_table_lock);
mm->rss--;
+ spin_unlock(&mm->page_table_lock);
flush_tlb_page(vma, address);
deactivate_page(page);
page_cache_release(page);
--
Regards,
Rasmus([email protected])
Gates' Law: Every 18 months, the speed of software halves
-- Anonymous