On 9/20/11 4:24 PM, Kieran Mansley wrote:
On Mon, 2011-09-19 at 10:31 +0300, Matias Mandell wrote:
In the end of event_callback() when it has gone thru all tasks waiting
for a select() it unprotects&protects and starts over from the
beginning. At this point the sock->select_waiting was sometimes zero and
the loop crashed.
In lwip_select() there are protections when adding/removing from list
and increasing/decreasing sock->select_waiting counters but they are not
in sync. How good idea is it to pass a local variable to another task,
there might be context switching in between? I might have a stack
overflow or something going on...
Without looking at the code I'm not quite sure what the problem is. Do
you have a patch to fix it as that would show clearly what you're
describing.
In short, the lwip_select()-function's local variable
struct lwip_select_cb select_cb;
is also used within the event_callback()-function.
I do not get a crash if I:
- exit the event_callback()-loop upon "sock->select_waiting==0", see
attached excerpt.
or
- create static memory for the select_cb:s.
There is not a stack overflow, but I'm not sure how the stack is
saved/loaded upon context switching. I am also experiencing problems
opening new sockets after closing old ones, I have not looked at it but
there seems to be another email discussion about it.
Sincerely, Matias
--
mat!as
/* unlock interrupts with each step */
last_select_cb_ctr = select_cb_ctr;
SYS_ARCH_UNPROTECT(lev);
/* this makes sure interrupt protection time is short */
SYS_ARCH_PROTECT(lev);
if (sock->select_waiting == 0) {
/* noone is waiting for this socket, no need to check
select_cb_list */
SYS_ARCH_UNPROTECT(lev);
return;
}
if (last_select_cb_ctr != select_cb_ctr) {
/* someone has changed select_cb_list, restart at the beginning */
goto again;
}_______________________________________________
lwip-users mailing list
[email protected]
https://lists.nongnu.org/mailman/listinfo/lwip-users
