in_interrupt() semantics are confusing and wrong for most users
as it also returns true when bh is disabled. Thus we open coded
a proper check for interrupts in __sanitizer_cov_trace_pc()
with a lengthy explanatory comment.
Use the new in_task() predicate instead.
Signed-off-by: Dmitry Vyukov <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Kefeng Wang <[email protected]>
Cc: James Morse <[email protected]>
Cc: Alexander Popov <[email protected]>
Cc: Andrey Konovalov <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
---
kernel/kcov.c | 9 +--------
1 file changed, 1 insertion(+), 8 deletions(-)
diff --git a/kernel/kcov.c b/kernel/kcov.c
index 85e5546cd791..cd771993f96f 100644
--- a/kernel/kcov.c
+++ b/kernel/kcov.c
@@ -60,15 +60,8 @@ void notrace __sanitizer_cov_trace_pc(void)
/*
* We are interested in code coverage as a function of a syscall inputs,
* so we ignore code executed in interrupts.
- * The checks for whether we are in an interrupt are open-coded, because
- * 1. We can't use in_interrupt() here, since it also returns true
- * when we are inside local_bh_disable() section.
- * 2. We don't want to use (in_irq() | in_serving_softirq() | in_nmi()),
- * since that leads to slower generated code (three separate tests,
- * one for each of the flags).
*/
- if (!t || (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_OFFSET
- | NMI_MASK)))
+ if (!t || !in_task())
return;
mode = READ_ONCE(t->kcov_mode);
if (mode == KCOV_MODE_TRACE) {
--
2.12.1.500.gab5fba24ee-goog
On March 21, 2017 5:10 PM Dmitry Vyukov wrote:
>
> @@ -60,15 +60,8 @@ void notrace __sanitizer_cov_trace_pc(void)
> /*
> * We are interested in code coverage as a function of a syscall inputs,
> * so we ignore code executed in interrupts.
> - * The checks for whether we are in an interrupt are open-coded, because
> - * 1. We can't use in_interrupt() here, since it also returns true
> - * when we are inside local_bh_disable() section.
> - * 2. We don't want to use (in_irq() | in_serving_softirq() | in_nmi()),
> - * since that leads to slower generated code (three separate tests,
> - * one for each of the flags).
> */
> - if (!t || (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_OFFSET
> - | NMI_MASK)))
> + if (!t || !in_task())
> return;
Nit: can we get the current task check cut off?
On Tue, Mar 21, 2017 at 10:28 AM, Hillf Danton <[email protected]> wrote:
>
> On March 21, 2017 5:10 PM Dmitry Vyukov wrote:
>>
>> @@ -60,15 +60,8 @@ void notrace __sanitizer_cov_trace_pc(void)
>> /*
>> * We are interested in code coverage as a function of a syscall inputs,
>> * so we ignore code executed in interrupts.
>> - * The checks for whether we are in an interrupt are open-coded, because
>> - * 1. We can't use in_interrupt() here, since it also returns true
>> - * when we are inside local_bh_disable() section.
>> - * 2. We don't want to use (in_irq() | in_serving_softirq() | in_nmi()),
>> - * since that leads to slower generated code (three separate tests,
>> - * one for each of the flags).
>> */
>> - if (!t || (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_OFFSET
>> - | NMI_MASK)))
>> + if (!t || !in_task())
>> return;
>
> Nit: can we get the current task check cut off?
Humm... good question.
I don't remember why exactly I added it. I guess something was
crashing during boot. Note that this call is inserted into almost all
kernel code. But probably that was before I disabled instrumentation
of some early boot code for other reasons (with KCOV_INSTRUMENT := n
in Makefile), because now I can boot kernel in qemu without this
check. But I am still not sure about real hardware/arm/etc.
Does anybody know if current can ever (including early boot) return
invalid pointer?