s390/test_unwind: fix CALL_ON_STACK tests

[ Upstream commit f22b9c219a798e1bf11110a3d2733d883e6da059 ]

The CALL_ON_STACK tests use the no_dat stack to switch to a different
stack for unwinding tests. If an interrupt or machine check happens
while using that stack, and previously being on the async stack, the
interrupt / machine check entry code (SWITCH_ASYNC) will assume that
the previous context did not use the async stack and happily use the
async stack again.

This will lead to stack corruption of the previous context.

To solve this disable both interrupts and machine checks before
switching to the no_dat stack.

Fixes: 7868249fbbc8 ("s390/test_unwind: add CALL_ON_STACK tests")
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
This commit is contained in:
Heiko Carstens 2020-12-13 18:05:56 +01:00 committed by Greg Kroah-Hartman
parent 0633094ec7
commit 1856405862

View File

@ -205,12 +205,15 @@ static noinline int unwindme_func3(struct unwindme *u)
/* This function must appear in the backtrace. */
static noinline int unwindme_func2(struct unwindme *u)
{
unsigned long flags;
int rc;
if (u->flags & UWM_SWITCH_STACK) {
preempt_disable();
local_irq_save(flags);
local_mcck_disable();
rc = CALL_ON_STACK(unwindme_func3, S390_lowcore.nodat_stack, 1, u);
preempt_enable();
local_mcck_enable();
local_irq_restore(flags);
return rc;
} else {
return unwindme_func3(u);