El 31 de març de 2012 1:16, Robert Millan <rmh@debian.org> ha escrit: > Somewhat. My latest patch (after fixing other problems [1]) now > passes tst-timer [2], but only *sometimes* (looks like a race). I > isolated my code into a standalone app by renaming the timer_* > functions and found that when run outside libc it nearly always (or > always? can't recall) works, and when run as part of libc it nearly > always fails, but sometimes works. Problem fixed [1]. Now it passes the whole testsuite, including tst-timer5 which was disabled before! \o/ Here's my latest patch. So far only tested with kfreebsd 9.0-1 on i386. I will test older kernels and port it to amd64. In the meantime, would someone be so kind to check if the testsuite problems in ruby, python, etc got better? [1] Grmf, turns out GCC was optimizing my busy wait loop into an infinite loop. Thankfully this taught me to ditch some of my broken cache coherency assumptions... -- Robert Millan
Attachment:
timer.diff
Description: Binary data