From: David Woodhouse <dwmw2@infradead.org>
I have just spotted a bug in the 'infinite' case there though -- if they've
asked for a timeout longer than MAX_INT64_SECONDS, then the 'else' case
should be setting timeout to -1 instead of to MAX_SCHEDULE_TIMEOUT. I must
have missed that when I fixed the 24-day bug. Let's do it like this...
Signed-off-by: Andrew Morton <akpm@osdl.org>
---
fs/select.c | 14 +++++++++++---
1 files changed, 11 insertions(+), 3 deletions(-)
diff -puN fs/select.c~pselect-ppoll-system-calls-fix fs/select.c
--- 25/fs/select.c~pselect-ppoll-system-calls-fix Sun Jun 26 16:33:09 2005
+++ 25-akpm/fs/select.c Sun Jun 26 16:37:44 2005
@@ -672,11 +672,19 @@ asmlinkage long sys_ppoll(struct pollfd
if (copy_from_user(&ts, tsp, sizeof(ts)))
return -EFAULT;
- if (ts.tv_sec < MAX_INT64_SECONDS) {
+#if BITS_PER_LONG > 32
+ /*
+ * Only 64-bit platforms can overflow an int64_t, and if we let
+ * GCC optimise this out for itself, it whinges that the
+ * comparison is always false, and makes people unhappy
+ */
+ if (ts.tv_sec >= MAX_INT64_SECONDS)
+ timeout = -1; /* infinite */
+ else
+#endif
+ {
timeout = ROUND_UP(ts.tv_sec, 1000000000/HZ);
timeout += ts.tv_sec * HZ;
- } else {
- timeout = MAX_SCHEDULE_TIMEOUT;
}
}
_