x86: fix deadlock, make pgd_lock irq-safe
lockdep just caught this one:
=================================
[ INFO: inconsistent lock state ]
2.6.24 #38
---------------------------------
inconsistent {in-softirq-W} -> {softirq-on-W} usage.
swapper/1 [HC0[0]:SC0[0]:HE1:SE1] takes:
(pgd_lock){-+..}, at: [<ffffffff8022a9ea>] mm_init+0x1da/0x250
{in-softirq-W} state was registered at:
[<ffffffffffffffff>] 0xffffffffffffffff
irq event stamp: 394559
hardirqs last enabled at (394559): [<ffffffff80267f0a>] get_page_from_freelist+0x30a/0x4c0
hardirqs last disabled at (394558): [<ffffffff80267d25>] get_page_from_freelist+0x125/0x4c0
softirqs last enabled at (393952): [<ffffffff80232f8e>] __do_softirq+0xce/0xe0
softirqs last disabled at (393945): [<ffffffff8020c57c>] call_softirq+0x1c/0x30
other info that might help us debug this:
no locks held by swapper/1.
stack backtrace:
Pid: 1, comm: swapper Not tainted 2.6.24 #38
Call Trace:
[<ffffffff8024e1fb>] print_usage_bug+0x18b/0x190
[<ffffffff8024f55d>] mark_lock+0x53d/0x560
[<ffffffff8024fffa>] __lock_acquire+0x3ca/0xed0
[<ffffffff80250ba8>] lock_acquire+0xa8/0xe0
[<ffffffff8022a9ea>] ? mm_init+0x1da/0x250
[<ffffffff809bcd10>] _spin_lock+0x30/0x70
[<ffffffff8022a9ea>] mm_init+0x1da/0x250
[<ffffffff8022aa99>] mm_alloc+0x39/0x50
[<ffffffff8028b95a>] bprm_mm_init+0x2a/0x1a0
[<ffffffff8028d12b>] do_execve+0x7b/0x220
[<ffffffff80209776>] sys_execve+0x46/0x70
[<ffffffff8020c214>] kernel_execve+0x64/0xd0
[<ffffffff8020901e>] ? _stext+0x1e/0x20
[<ffffffff802090ba>] init_post+0x9a/0xf0
[<ffffffff809bc5f6>] ? trace_hardirqs_on_thunk+0x35/0x3a
[<ffffffff8024f75a>] ? trace_hardirqs_on+0xba/0xd0
[<ffffffff8020c1a8>] ? child_rip+0xa/0x12
[<ffffffff8020bcbc>] ? restore_args+0x0/0x44
[<ffffffff8020c19e>] ? child_rip+0x0/0x12
turns out that pgd_lock has been used on 64-bit x86 in an irq-unsafe
way for almost two years, since commit 8c914cb704a11460e.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index d8ed400..621afb6 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -958,11 +958,12 @@
for (address = start; address <= VMALLOC_END; address += PGDIR_SIZE) {
if (!test_bit(pgd_index(address), insync)) {
const pgd_t *pgd_ref = pgd_offset_k(address);
+ unsigned long flags;
struct page *page;
if (pgd_none(*pgd_ref))
continue;
- spin_lock(&pgd_lock);
+ spin_lock_irqsave(&pgd_lock, flags);
list_for_each_entry(page, &pgd_list, lru) {
pgd_t *pgd;
pgd = (pgd_t *)page_address(page) + pgd_index(address);
@@ -971,7 +972,7 @@
else
BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
}
- spin_unlock(&pgd_lock);
+ spin_unlock_irqrestore(&pgd_lock, flags);
set_bit(pgd_index(address), insync);
}
if (address == start)