summaryrefslogtreecommitdiffstats
path: root/lib/libssl/src
diff options
context:
space:
mode:
authorgluk <gluk@openbsd.org>2001-04-22 19:20:35 +0000
committergluk <gluk@openbsd.org>2001-04-22 19:20:35 +0000
commit2a2de9773441135363eb16073710c11e3117ea06 (patch)
treef6da495ffe4cb8aa3fd57edb280cde6836d4c58a /lib/libssl/src
parentpermit compilation without I586_CPU or I686_CPU; armin@wolfermann.org (diff)
downloadwireguard-openbsd-2a2de9773441135363eb16073710c11e3117ea06.tar.xz
wireguard-openbsd-2a2de9773441135363eb16073710c11e3117ea06.zip
Remove -march=i{56}86 optimization because of compiler bug. This bug
results in system lockup, which many people report for 2.8 and -current when they doing a big network transfer. This problem affect only custom kernels in which only one cpu type enabled (option I586_CPU or I686_CPU). When lockup occur I can't switch between virtual wscons terminals. System continue respond to pings and forward ip packets. It is possible to enter into ddb. DDB show that several processes in runnable state, but it seems that task switching not occur. More and more processes becomes runnable. Stack of curproc looks like: > _end(e99d8fac, e0101dcc, 4, e0635a00, e99d8f80) at 0xe99d8f78 > _end(e99d8fa0, e028a62e, e99d8fac, 0, 0) at 0xe99d8f78 > ddb> Sometimes 'boot sync' cleanly unmount all file systems. I reproduce this bug by transfering two big files from ftp simultaneously. It seems that at least one process must perform a network transfer and two or more processes must fight for the processor. The following PRs probably a result of this problem: 1504, 1716, 1751, 1771, 1780. deraadt@ ok.
Diffstat (limited to 'lib/libssl/src')
0 files changed, 0 insertions, 0 deletions