On Tue, 30 Dec 2014, Sergei Shtylyov wrote:

On 12/30/2014 09:28 PM, Nicholas Mc Guire wrote:

quoted

quoted

quoted

wait_for_completion_timeout does not return negative values so the tests
for <= 0 are not needed and the case differentiation in the error handling
path unnecessary.

quoted

quoted

    I decided to verify your statement and I saw that it seems wrong.
do_wait_for_common() can return -ERESTARTSYS and the return value gets
returned by its callers unchanged.

quoted

the -ERESTARTSYS only can be returned if state matches but
wait_for_completion_timemout passes TASK_UNINTERRUPTIBLE
so signal_pending_state will return 0 and never negativ

quoted

my understanding of the callchain is:
wait_for_completion_timemout with TASK_UNINTERRUPTIBLE
   -> wait_for_common(...TASK_UNINTERRUPTIBLE)
     -> __wait_for_common(...TASK_UNINTERRUPTIBLE)
       -> do_wait_for_common(...TASK_UNINTERRUPTIBLE)
         -> signal_pending_state(TASK_UNINTERRUPTIBLE...)

quoted

static inline int signal_pending_state(long state, struct task_struct *p)
{
         if (!(state & (TASK_INTERRUPTIBLE | TASK_WAKEKILL)))
                 return 0;

   Right. I didn't look into TASK_UNINTERRUPTIBLE thing before sending my mail.

quoted

so wait_for_completion_timemout should return 0 or 1 only

   0 or the remaining time, to be precise.

yup - thanks for the confirmation!

quoted

quoted

quoted

patch was only compile tested x86_64_defconfig + CONFIG_ATH_CARDS=m
CONFIG_ATH10K=m

quoted

quoted

quoted

patch is against linux-next 3.19.0-rc1 -next-20141226

quoted

quoted

    Rather patches. It would have been better to send one patch instead of
4 patches with the same name.

quoted

sorry for that - I had split it into separate patches as it was
in different files - giving them the same name of course was a bit
brain-dead.

   You should have mentioned the modified files in the subject. But IMHO 
it would be better to have just one patch.

resent as a single patch as v2

thx!
hofrat