On 2018/4/10 23:01, Boris Ostrovsky wrote:
On 04/10/2018 10:31 AM, Jia-Ju Bai wrote:

On 2018/4/10 22:27, Boris Ostrovsky wrote:
On 04/09/2018 11:03 AM, Jia-Ju Bai wrote:
pcistub_probe() is never called in atomic context.
This function is only set as ".probe" in struct pci_driver.

Despite never getting called from atomic context,
pcistub_probe() calls kmalloc() with GFP_ATOMIC,
which does not sleep for allocation.
GFP_ATOMIC is not necessary and can be replaced with GFP_KERNEL,
which can sleep and improve the possibility of sucessful allocation.

This is found by a static analysis tool named DCNS written by myself.
And I also manually check it.

Signed-off-by: Jia-Ju Bai <baijiaju1...@gmail.com>
What about use of GFP_ATOMIC in pcistub_reg_add()?
Thanks for your reply :)
I find pcistub_reg_add() is called by pcistub_quirk_add().
And pcistub_quirk_add() is called in the macro DRIVER_ATTR().
I am not sure whether DRIVER_ATTR() can make the function called in
atomic context,
so I do not analyze it in my tool.
I don't see why it needs to be ATOMIC, it's sysfs access. Can you send a
patch to fix it as well?

Okay, I will send a patch for it soon.
You can have a look :)

Best wishes,
Jia-Ju Bai

Reply via email to