Bert Miemietz wrote:
> Came upon a strange thing with ddi_get_lbolt and cv_timedwait.
> Following a code fragment to show the issue:
>
> mutex_enter(&ptr->mutex);
> while (ptr->busy != 0) {
> clock_t until;
> (void) ddi_get_lbolt(); /* just call function, seems not to work at first
> call */
> until = ddi_get_lbolt() + drv_usectohz(tmo);
> if (cv_timedwait(&ptr->cv, &ptr->mutex, until) == -1) {
> mutex_exit(&ptr->mutex);
> err = EAGAIN;
> goto somewhere_else;
> }
> }
> mutex_exit(&mstr_ptr->vv_mutex);
>
> In theory the line "(void) ddi_get_lbolt();" should be completely unnecessary.
> But in case I omit this line and ddi_get_lbolt() has never been called before
> in the driver cv_timedwait() will not return when timeout is reached without
> the condition being signaled. Inserting that (in theory useless) line makes
> everything work fine.
>
> Did I understand or do something wrong or is it really a bug?
> I'm a little bit unsure if I have seen this on all platforms. Last time I was
> getting
> into this trap was on Solaris x86 / single cpu machine.
>
I suspect a bug in your code. Are you holding the ptr->mutex when doing
the cv_signal?
What is the value of tmo?
Also, be aware that mutex deadlocks may occur -- cv_timedwait() drops,
and reacquires the lock. If it can't reacquire the lock, then badness
will ensue, and it may appear hung when it is merely blocked.
-- Garrett
>
>
> This message posted from opensolaris.org
> _______________________________________________
> opensolaris-code mailing list
> opensolaris-code@opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
>
_______________________________________________
opensolaris-code mailing list
opensolaris-code@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
