Search
SailfishOS Open Build Service
>
Projects
>
nemo
:
devel:hw
:
maguro
>
systemd
> _service:tar_git:systemd-187-ignore-omap-vsync.patch
Log In
Username
Password
Cancel
Overview
Repositories
Revisions
Requests
Users
Advanced
Attributes
Meta
File _service:tar_git:systemd-187-ignore-omap-vsync.patch of Package systemd
Description: Ignore VSYNC change uevents from omapfb, as they happen at each vsync and thus keep waking up udevd (including rules processing), upstart, and other parts of the system 60 times a second. Author: Martin Pitt <martin.pitt@ubuntu.com> Forwarded: Heck, don't ever show this embarassing hack to anyone. Except if that anyone is an omapfb author, then please don't stop showing this to them until they fix the driver :-) Bug-Ubuntu: https://launchpad.net/bugs/1234743 Index: systemd/src/libudev/libudev-monitor.c =================================================================== --- systemd.orig/src/libudev/libudev-monitor.c 2013-02-20 02:32:41.900839006 +0100 +++ systemd/src/libudev/libudev-monitor.c 2013-10-17 11:38:56.216374114 +0200 @@ -198,6 +198,9 @@ (*i)++; } +/* return uint32_t for the first 4 characters in s */ +#define str2word(s) ((s[0] << 24) | (s[1] << 16) | (s[2] << 8) | s[3]) + /** * udev_monitor_filter_update: * @udev_monitor: monitor @@ -216,6 +219,11 @@ int err; if (udev_list_get_entry(&udev_monitor->filter_subsystem_list) == NULL && +#ifdef __arm__ + /* we always need a filter chain for kernel monitors to get rid + * of the omapfb VSYNC uevents */ + udev_monitor->snl.nl.nl_groups != UDEV_MONITOR_KERNEL && +#endif udev_list_get_entry(&udev_monitor->filter_tag_list) == NULL) return 0; @@ -224,8 +232,45 @@ /* load magic in A */ bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_ABS, offsetof(struct udev_monitor_netlink_header, magic)); +#ifndef __arm__ /* jump if magic matches */ bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, UDEV_MONITOR_MAGIC, 1, 0); +#else + /* jump over our whole omapfb/VSYNC check if magic matches */ + bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, UDEV_MONITOR_MAGIC, 10, 0); + + /* filter out omapfb change VSYNC uevents (LP #1234743); they look like +change@/devices/platform/omapfb\x00ACTION=change\x00DEVPATH=/devices/platform/omapfb\x00SUBSYSTEM=platform\x00VSYNC=1737775695800\x00DRIVER=omapfb\x00MODALIAS=platform:omapfb\x00SEQNUM=8909\x00 + + Matching on the VSYNC attribute is a bit finicky, as its exact + position in the packet might not be reliable. The real VSYNC uevents + have packets of ~ 170 bytes length (varying with the string length of + VSYNC= and SEQNUM), while a simple synthetic + "echo change | tee /sys/devices/platform/omapfb/uevent" only has 150. + So we instead just compare the length of the packet and use 160 as a divider. + As we don't generally care about that device at all, this logic does + not need to be rock solid, though -- if we would just ignore all + change events from that device nothing should break. + */ + + /* if the event is shorter than 160 bytes, it can't be omapfb/VSYNC, jump to "pass" */ + bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_LEN, 0); + bpf_jmp(ins, &i, BPF_JMP|BPF_JGE|BPF_K, 160, 0, 7); + + /* check if it is a change event; if not, jump to "pass" */ + bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_ABS, 0); + bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, str2word("chan"), 0, 5); + + /* check if it is from omapfb */ + bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_ABS, 24); + bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, str2word("/oma"), 0, 3); + bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_ABS, 28); + bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, str2word("pfb\0"), 0, 1); + + /* we found the omapfb change uevent, reject it */ + bpf_stmt(ins, &i, BPF_RET|BPF_K, 0); +#endif /* omapfb filter hack on __arm__ */ + /* wrong magic, pass packet */ bpf_stmt(ins, &i, BPF_RET|BPF_K, 0xffffffff);
