Re: [PATCH] checkpatch: Add warnings for use of mdelay()
From: Jia-Ju Bai
Date: Sat Jul 07 2018 - 08:38:40 EST
On 2018/7/6 13:49, Julia Lawall wrote:
On Thu, 5 Jul 2018, Dan Carpenter wrote:
Neither Smatch nor Coccinelle do a good job tracking when you're in
atomic context. I've wanted to add this to Smatch but even then it
would be to warn that "We're holding a spinlock so we can't sleep".
It's trickier to say for sure when you're not holding a lock...
Jia-Ju Bai is working on this. The tool is available on github. It's
still being improved, though, so perhaps it's not yet ready for eg 0-day
inclusion. He can give more details.
Thanks for Julia's recommendation :)
I am doing the similar work with Julia, from the beginning of this year.
We develop two new LLVM-based tools to find two problems in the Linux
kernel:
(1) Sleeping in atomic context. The tool is named DSAC.
(2) Using non-sleep function calls in non-atomic context. The tool is
named DCNS.
We handle two common examples of atomic context:
(1) Holding a spinlock.
(2) In an interrupt handler.
DSAC and DCNS can basically work now, and some of the defects found by
them have been confirmed and fixed in the Linux kernel.
But these tools are still being improved.
In fact, I encounter a hard problem when writing the tools, namely how
to accurately and completely handle function pointer calls.
I have handled the function pointer in form of data structure field, but
I do not find a good way to handle the function pointer that is used as
a function argument.
Can someone give me good advice?
We also have made slides introducing DSAC and DCNS tools.
If you are interested in our work, I can send you the slides :)
Best wishes,
Jia-Ju Bai