On Wed, Dec 18, 2019 at 4:14 PM Josh Triplett <josh@xxxxxxxxxxxxxxxx> wrote:
>
> Er, wrong file. That's the original patch; the attached patch is the
> right one.
This looks correct to me.
If I were to actually commit it, the "split into two waitqueues" would
be a separate patch from the "use wait_event_interruptible_exclusive()
and add "wake_next_reader/writer logic", but for testing purposes the
unified patch was simpler, and your forward port looks good to me.
I ran the original patch for a couple of days, and didn't see any
other issues than the 'make' thing in F30. It was all good with my
self-build make.
But that "ran for a couple of days" wasn't all that stress-full. I did
do the "verify that the thundering herd is gone" test - including that
silly test-case here again:
#include <unistd.h>
int main(int argc, char **argv)
{
int fd[2], counters[2];
pipe(fd);
counters[0] = 0;
counters[1] = -1;
write(fd[1], counters, sizeof(counters));
/* 64 processes */
fork(); fork(); fork(); fork(); fork(); fork();
do {
int i;
read(fd[0], &i, sizeof(i));
if (i < 0)
continue;
counters[0] = i+1;
write(fd[1], counters, (1+(i & 1)) *sizeof(int));
} while (counters[0] < 1000000);
return 0;
}
where you can tweak the numbers - add another fork() or two to create
even more pipe waiters, and maybe change the final count exit value to
match whatever hw performance you have.
Linus
