summaryrefslogtreecommitdiff
path: root/net/unix/af_unix.c
diff options
context:
space:
mode:
authorLennert Buytenhek <buytenh@org.rmk.(none)>2005-03-21 20:47:35 +0000
committerRussell King <rmk@flint.arm.linux.org.uk>2005-03-21 20:47:35 +0000
commitedfb74968a4f6f6b312652760956f8fd6524924f (patch)
tree0a970e5ff8ac1233694664cd1afe949376d68c71 /net/unix/af_unix.c
parent7882ca4e6934b0ad113f7a80758ca8d818b08daf (diff)
[ARM PATCH] 2571/1: minor time-keeping fixes for ixp2000
Patch from Lennert Buytenhek There are two very minor issues with time-keeping on the ixp2000: 1. It turns out that if T#_CLD (timer reload value) is set to X, it means the timer will spend a full tick on 0, then a full tick on X, and then start counting down towards 0, which then effectively gives the timer a period of X+1 instead of the intended X. Solve by writing X-1 instead of X to T#_CLD. This is relatively harmless, though -- assuming a 50MHz system clock (like on the ENP-2611) and HZ=100, there will be 500000 timer interrupts generated per 500001 real-world jiffies, but that last timer interrupt will call timer_tick twice, taking the system time from 499999 jiffies plus 20000 usec to 500001 jiffies plus 0 usec. Every 500000th jiffy will 'not exist', and every other jiffy will last 10002us instead of 10000us. 2. next_jiffy_time is initialised as '0xffffffff - ticks_per_jiffy', but timer 4 starts ticking from 0xffffffff downwards, which will cause the time offset in the first 1/HZ sec after timer int init to be negative, and the first timer interrupt ever not to call timer_tick(). Neither of these issues can cause clock drift or backwards clock motion, but it's good to fix them nevertheless. Signed-off-by: Lennert Buytenhek Signed-off-by: Deepak Saxena Signed-off-by: Russell King
Diffstat (limited to 'net/unix/af_unix.c')
0 files changed, 0 insertions, 0 deletions