drm/i915: protect force_wake_(get|put) with the gt_lock
Hi Daniel,
ok here is the sysprof result (see below). No cpu/gpu intensive apps are running, just the idling desktop. CPU Temperature is low and frequency is throttled down to 800 MHz like it should be. The System is responsive and behaves normal but top shows full cpu load. Honestly I cannot see the reason why X and gkrellm should eat up all the gpu since they clearly don't hog the cpu. Maybe the kernel reports wrong cpu stats? Then again the load goes down when I kill X! Hope this gives you a clue.
If you need more info, please let me know.
Yours,
Nic
[Everything] 0.00% 100.00% [/usr/bin/X] 0.00% 29.26% No map [/usr/bin/X] 0.00% 12.95% select 0.00% 11.79%
- - kernel - - 0.00% 11.79% _raw_spin_unlock_irqrestore 2.91% 2.91% unix_poll 1.60% 1.60% fget_light 1.16% 1.16% sock_poll 0.73% 0.73% do_select 0.58% 0.58% fput 0.44% 0.44% __pollwait 0.44% 0.44% get_work_gcwq 0.44% 0.44% get_page_from_freelist 0.44% 0.44% evdev_poll 0.29% 0.29% _raw_spin_lock_irqsave 0.29% 0.29% finish_task_switch 0.29% 0.29% copy_user_generic_unrolled 0.29% 0.29% n_tty_poll 0.15% 0.15% start_flush_work 0.15% 0.15% tty_poll 0.15% 0.15% datagram_poll 0.15% 0.15% get_signal_to_deliver 0.15% 0.15% next_zones_zonelist 0.15% 0.15% __alloc_pages_nodemask 0.15% 0.15% core_sys_select 0.15% 0.15% sys_rt_sigreturn 0.15% 0.15% _cond_resched 0.15% 0.15% save_i387_xstate 0.15% 0.15% timerfd_poll 0.15% 0.15% hrtimer_init 0.15% 0.15% In file /usr/lib64/xorg/modules/drivers/intel_drv.so 0.73% 0.87%
- - kernel - - 0.00% 0.15% retint_signal 0.15% 0.15% __errno_location 0.15% 0.15% WakeupHandler 0.15% 0.15% ioctl 0.00% 3.35% In file /usr/lib64/xorg/modules/drivers/intel_drv.so 2.62% 2.62% In file /usr/lib64/libpixman-1.so.0.23.9 2.04% 2.04% __read 0.15% 1.46% In file /usr/bin/Xorg 1.46% 1.46% In file /lib64/libc-2.13.so 1.02% 1.16% In file /usr/lib64/xorg/modules/input/synaptics_drv.so 0.58% 0.87% fbPixmapToRegion 0.29% 0.44% fbBlt 0.29% 0.29% ChangeGC 0.29% 0.29% ReadRequestFromClient 0.15% 0.15% mmap 0.00% 0.15% ProcQueryPointer 0.15% 0.15% In file [heap] 0.00% 0.15% fbPolyline32 0.15% 0.15% In file /usr/lib64/xorg/modules/input/evdev_drv.so 0.15% 0.15% dixLookupDrawable 0.15% 0.15% ChangeWindowAttributes 0.15% 0.15% GetExtensionEntry 0.15% 0.15% ValidatePicture 0.15% 0.15% dixLookupResourceByClass 0.15% 0.15% timerfd_settime 0.15% 0.15% IsMaster 0.15% 0.15% malloc 0.15% 0.15% dixLookupResourceByType 0.15% 0.15% ConfigureWindow 0.15% 0.15% [gkrellm] 0.00% 26.49% read 0.15% 12.66%
- - kernel - - 0.00% 12.52% format_decode 3.35% 3.35% vsnprintf 2.04% 2.04% string.clone.2 1.46% 1.46% number.clone.1 1.16% 1.16% seq_printf 0.58% 0.58% e1000e_update_stats 0.58% 0.58% show_stat 0.29% 0.29% put_dec_trunc 0.29% 0.29% strnlen 0.29% 0.29% kstat_irqs 0.29% 0.29% part_round_stats 0.15% 0.15% all_vm_events 0.15% 0.15% dev_seq_printf_stats 0.15% 0.15% acpi_ds_exec_end_op 0.15% 0.15% put_dec_full 0.15% 0.15% kref_put 0.15% 0.15% get_device 0.15% 0.15% diskstats_show 0.15% 0.15% seq_read 0.15% 0.15% irq_to_desc 0.15% 0.15% vmstat_show 0.15% 0.15% vmstat_next 0.15% 0.15% system_call_after_swapgs 0.15% 0.15% jiffies_to_msecs 0.15% 0.15% sys_read 0.15% 0.15% __isoc99_vsscanf 0.00% 1.31% No map [gkrellm] 0.00% 1.16% In file /usr/lib64/libcairo.so.2.11000.2 0.87% 0.87% In file /usr/lib64/libpango-1.0.so.0.2904.0 0.58% 0.58% In file /usr/lib64/libgdk-x11-2.0.so.0.2400.7 0.44% 0.44% In file /lib64/libc-2.13.so 0.29% 0.44% g_type_check_instance_is_a 0.44% 0.44% pango_default_break 0.44% 0.44% In file /usr/lib64/libpangocairo-1.0.so.0.2904.0 0.29% 0.29% g_type_instance_get_private 0.29% 0.29% In file /usr/lib64/libglib-2.0.so.0.3000.1 0.15% 0.29% _XUpdateGCCache 0.29% 0.29% _IO_file_seekoff 0.29% 0.29% __open64 0.00% 0.29% g_markup_parse_context_parse 0.29% 0.29% gdk_gc_get_type 0.29% 0.29% In file /usr/bin/gkrellm 0.29% 0.29% In file /usr/lib64/libfreetype.so.6.7.2 0.29% 0.29% update_host 0.15% 0.15% In file /usr/lib64/gkrellm2/plugins/cpufreq.so 0.15% 0.15% g_unichar_type 0.15% 0.15% g_type_fundamental 0.15% 0.15% pango_renderer_set_matrix 0.15% 0.15% g_object_ref 0.15% 0.15% gdk_draw_drawable 0.00% 0.15% In file /lib64/libpthread-2.13.so 0.15% 0.15% free 0.15% 0.15% poll 0.00% 0.15% g_type_create_instance 0.15% 0.15% g_object_unref 0.15% 0.15% _XFlushGCCache 0.15% 0.15% g_pointer_bit_unlock 0.15% 0.15% cairo_close_path 0.15% 0.15% xcb_writev 0.15% 0.15% pango_layout_line_get_extents 0.15% 0.15% gdk_pango_renderer_get_type 0.15% 0.15% pango_renderer_deactivate 0.15% 0.15% __pthread_mutex_lock 0.15% 0.15% In file /usr/lib64/libX11.so.6.3.0 0.15% 0.15% pango_ot_buffer_output 0.15% 0.15% g_type_class_peek_static 0.15% 0.15% cairo_pattern_destroy 0.15% 0.15% cairo_destroy 0.15% 0.15% g_unichar_get_script 0.15% 0.15% pango_script_iter_next 0.15% 0.15% open 0.00% 0.15% pango_renderer_set_color 0.15% 0.15% g_malloc_n 0.15% 0.15% XSetClipMask 0.15% 0.15% gkrellm_update_krell 0.15% 0.15% finite 0.15% 0.15% g_object_newv 0.15% 0.15% lseek64 0.00% 0.15% g_datalist_id_get_data 0.15% 0.15% [kicker [kdeinit]] 0.00% 8.88% [/opt/apps/firefox/bin/firefox] 0.00% 6.40% [sysprof] 0.00% 5.97% [top] 0.00% 5.68% [/opt/java/jdk/jre/bin/java] 0.00% 5.24% [konsole [kdeinit]] 0.00% 5.09% [kcpuload] 0.00% 2.47% [/opt/apps/claws-mail/bin/claws-mail] 0.00% 1.75% [kwin [kdeinit] -session 10d66f6e7a000128602678900000074690000_1320681] 0.00% 0.87% [kworker/u:2] 0.00% 0.44% [konsole [kdeinit] -session 10d66f6e7a000132061162800000028890007_1320] 0.00% 0.44% [konqueror [kdeinit] -session 10d66f6e7a000132061213500000034570005_13] 0.00% 0.29% [kworker/0:0] 0.00% 0.29% [kwrapper] 0.00% 0.15% [kworker/1:0] 0.00% 0.15% [/usr/bin/wpa_cli] 0.00% 0.15%
-----Ursprüngliche Nachricht-----
Von: "Daniel Vetter" <[email protected]>
Gesendet: Nov 7, 2011 5:56:38 PM
An: "Nicolas Kalkhof" <[email protected]>
Betreff: Re: [Intel-gfx] [PATCH] drm/i915: protect force_wake_(get|put) with the gt_lock
Quoted 19 lines · Nicolas Kalkhof
Quoted 8 lines
Hi Daniel,
confirmed. Linus's linux-next.git shows the same behaviour. Until now I've used a patched 3.1-rc-6+ Kernel from Dave Airlie's branch
(git://people.freedesktop.org/~airlied/linux' from Oct 26th.). dmesg/syslog shows nothing special, neither doesi915_error_state.Any advice?
If the behaviour-difference is clear, you could try to bisect this. The alternative is to use your favourite system profiler and see where the cycles get wasted (either sysprof or perf). I suggest you'll try whatever approach your more familiar with first and then switch over to the other if the first one doesn't yield any clear results. If the difference is really clear, I'd start with the bisect (aside: don't restrict the bisect to any subdir, the change causing the regression could equally likely be somewhere in the core kernel).
Quoted 5 lines
Yours, Daniel
--
Daniel Vetter
Mail: [email protected]
Mobile: +41 (0)79 365 57 48
___________________________________________________________
SMS schreiben mit WEB.DE FreeMail - einfach, schnell und kostenguenstig. Jetzt gleich testen! http://f.web.de/?mc=021192
_______________________________________________
Intel-gfx mailing list [email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 7 lines · Nicolas Kalkhof
ok here is the sysprof result (see below). No cpu/gpu intensive apps are running, just the idling desktop. CPU Temperature is low and frequency is throttled down to 800 MHz like it should be. The System is responsive and behaves normal but top shows full cpu load. Honestly I cannot see the reason why X and gkrellm should eat up all the gpu since they clearly don't hog the cpu. Maybe the kernel reports wrong cpu stats? Then again the load goes down when I kill X! Hope this gives you a clue.
On a quick look it seems that at least X is busy-spinning in the select loop and gkrellm seems to be eqally busy doing not much. Have you tried what happens when you kill gkrellm?
Otherwise I think it's time for a bit of bisecting. -Daniel
Signature · 3 lines
Daniel Vetter Mail: [email protected] Mobile: +41 (0)79 365 57 48