<html> <head> <meta content="text/html; charset=windows-1252" http-equiv="Content-Type"> </head> <body text="#000000" bgcolor="#FFFFFF"> Hi Heiko, <div class="moz-cite-prefix">On 12/09/2015 10:51 PM, Heiko Stübner wrote: </div> <blockquote cite="mid:1572876.arr7htU3E0@diego" type="cite"> <pre wrap="">Hi Yakir, Am Mittwoch, 9. Dezember 2015, 11:49:10 schrieb Yakir Yang: </pre> <blockquote type="cite"> <pre wrap="">Thanks a lot for great debugging. On 12/08/2015 11:33 PM, Heiko Stübner wrote: </pre> <blockquote type="cite"> <pre wrap="">Hi Yakir, Am Montag, 7. Dezember 2015, 14:37:19 schrieb Yakir Yang: </pre> <blockquote type="cite"> <pre wrap=""> The Samsung Exynos eDP controller and Rockchip RK3288 eDP controller share the same IP, so a lot of parts can be re-used. I split the common code into bridge directory, then rk3288 and exynos only need to keep some platform code. Cause I can't find the exact IP name of exynos dp controller, so I decide to name dp core driver with "analogix" which I find in rk3288 eDP TRM </pre> </blockquote> <pre wrap=""> [...] </pre> <blockquote type="cite"> <pre wrap="">Changes in v10: - Correct the ROCKCHIP_ANALOGIX_DP indentation in Kconfig to tabs here (Heiko) - Fix the wrong macro value of GRF_EDP_REF_CLK_SEL_INTER_HIWORD_MASK BIT(4) -> BIT(20) - Remove the surplus "plat_data" check. (Heiko) - switch (dp->plat_data && dp->plat_data->dev_type) { + switch (dp->plat_data->dev_type) { - Revert parts of Gustavo Padovan's changes in commit: drm/exynos: do not start enabling DP at bind() phase Add dp phy poweron function in bind time. </pre> </blockquote> <pre wrap=""> The hotplug issue is still present, but I think I found the cause. When the first detect call happens, the display simply is still off. I just did some very basic tracing [0] and it seems the display simply is not enabled when it is supposed to get detected. </pre> </blockquote> <pre wrap=""> Aha, thanks, make a lot of sense. </pre> <blockquote type="cite"> <pre wrap="">And it seems injecting a drm_panel_prepare early for _testing_ [1] really did make the hotplug work on both my jerry and minnie. So I guess we should somehow make sure the panel is actually powered when detection is running. Although I'm not sure yet, how that should look like. </pre> </blockquote> <pre wrap=""> Agree, panel should be powered up before DP controller start to detect hotplug signal. </pre> <blockquote type="cite"> <pre wrap="">Intuition suggests, making drm_panel calls nestable (similar to clk_prepare/unprepare, etc) and simply wrapping the detection code in a prepare-unprepare calls, but I'm not sure if Thierry might have other ideas ;-) </pre> </blockquote> <pre wrap=""> Due to the panel power status would influence the hotplug status, so I think we don't need to unprepared the panel unless in driver enter into suspend time. Things I want: 1. Prepared the panel in driver *bind time* 2. Enable the panel in driver *bridge->enable time* 3. Disable the panel in driver *bridge->disable time* 4. Unprepared the panel in driver*suspend time * 5. Re-prepared the panel in driver *resume time* </pre> </blockquote> <pre wrap="">6. Unprepare the panel in driver at *unbind time* otherwise going that way looks nice. </pre> </blockquote> Yes, <blockquote cite="mid:1572876.arr7htU3E0@diego" type="cite"> <blockquote type="cite"> <blockquote type="cite"> <pre wrap="">Also my "log" below suggests some sort of mismatch between prepare/unprepare calls, as there are a lot more of the prepare-side. </pre> </blockquote> <pre wrap=""> Yes, it's a typo too. I shouldn't place the panel->prepare in connector->get_modes, cause userspace would try to call get_modes once it receive the hotplug event, so there wouldn't have a match between panel prepare/unprepare. Previously, I just want to ensure that panel should be power-up when driver try to read the EDID from panel, so for now must remove the prepare from get_modes time :) </pre> <blockquote type="cite"> <pre wrap="">And the locking issue also seems to be still there [2]. </pre> </blockquote> <pre wrap=""> Hmm, I haven't meet this dead lock on my chromebook (ChromeOS + 4.4-rc3 Kernel) After look at the dead lock trace, I guess this dead lock would happened when hotplug event happened in bridge->disable time, not sure. Would try to find more in trace log and try to reproduce this. </pre> </blockquote> <pre wrap=""> It is not an actual deadlock, but a warning that a deadlock might happen. So you need to have LOCKDEP on. My kernels are currently running with CONFIG_DEBUG_LOCK_ALLOC=y CONFIG_PROVE_LOCKING=y CONFIG_LOCKDEP=y CONFIG_DEBUG_LOCKDEP=y CONFIG_DEBUG_ATOMIC_SLEEP=y and I see this mostly when changing between X11, console and back to X11. </pre> </blockquote> Oops, I have reproduced this issue. I didn't fully understand the potential dead lock, but i guess this is relate the an situation that when hotplug event happened in bridge->disable time. The normal flow would like: IN --> DRM IOCTL 1. Acquire crtc_ww_class_mutex (DRM IOCTL) IN --> analogix_dp_bridge 2. Acquire hpd work lock (Flush hpd work) 3. HPD work already in idle, no need to call drm_helper_hpd_irq_event() OUT <-- analogix_dp_bridge OUT <-- DRM IOCTL The dead lock flow would like: IN --> DRM IOCTL 1. Acquire crtc_ww_class_mutex (DRM IOCTL) IN --> analogix_dp_bridge 2. Acquire hpd work lock (Flush hpd work) IN --> analogix_dp_hotplug IN --> drm_helper_hpd_irq_event 3. Acquire mode_config lock (This lock already have been acquire in previous step 1) ** Dead Lock Now ** Hmm... In other words, dead lock would happened if we call drm_helper_hpd_irq_event in bridge->disbale time. It's wrong to flush the hpd work in bridge->disable time, I guess the original code just want to ensure the delay work must be finish before encoder disabled. For now i think it's better to delete the delay work way to update HPD event, we can take the fast interrupt thread way to update DRM HPD event (like dw_hdmi interrupt code). Would send the fix patch soon. - Yakir <blockquote cite="mid:1572876.arr7htU3E0@diego" type="cite"> <pre wrap=""> Heiko </pre> <blockquote type="cite"> <blockquote type="cite"> <pre wrap="">Heiko [0] [ 2.797383] analogix_dp_reset [ 2.800709] analogix_dp_init_hpd [ 2.803960] analogix_dp_init_video [ 2.807653] rockchip-drm display-subsystem: bound ff970000.dp (ops rockchip_dp_component_ops) [ 2.817176] [drm] Supports vblank timestamp caching Rev 2 (21.10.2013). [ 2.823799] [drm] No driver support for vblank timestamp query. [ 2.829947] analogix_dp_detect [ 2.833015] analogix_dp_get_plug_in_status: hpd status 0 ... [ 2.893425] analogix_dp_get_plug_in_status: hpd status 0 [ 2.893456] rockchip-dp ff970000.dp: failed to get hpd plug status, try to force hpd [ 2.893458] analogix_dp_force_hpd [ 2.893464] analogix_dp_get_plug_in_status: hpd status 112 [ 2.893470] panel_simple_prepare [ 2.952183] rockchip-dp ff970000.dp: EDID data does not include any extensions. [ 2.961727] panel_simple_get_modes [ 3.432154] analogix_dp_detect [ 3.432158] analogix_dp_get_plug_in_status: hpd status 120 [ 3.432160] panel_simple_prepare [ 3.433731] rockchip-dp ff970000.dp: EDID data does not include any extensions. [ 3.443268] panel_simple_get_modes [ 3.444668] panel_simple_prepare [ 3.444755] analogix_dp_reset [ 3.445078] analogix_dp_init_hpd [ 3.445096] panel_simple_disable [ 3.455349] analogix_dp_init_video [ 3.558323] rockchip-dp ff970000.dp: Timeout of video streamclk ok [ 3.558326] rockchip-dp ff970000.dp: unable to config video [ 3.558328] panel_simple_enable [ 3.573915] analogix_dp_detect [ 3.573919] analogix_dp_get_plug_in_status: hpd status 72 [ 3.573921] panel_simple_prepare [1] diff --git a/drivers/gpu/drm/rockchip/analogix_dp-rockchip.c b/drivers/gpu/drm/rockchip/analogix_dp-rockchip.c index 3990951..0c2dca5 100644 --- a/drivers/gpu/drm/rockchip/analogix_dp-rockchip.c +++ b/drivers/gpu/drm/rockchip/analogix_dp-rockchip.c @@ -399,6 +399,8 @@ static int rockchip_dp_probe(struct platform_device *pdev)> dp->plat_data.panel = panel; +drm_panel_prepare(dp->plat_data.panel); + /* * We just use the drvdata until driver run into component * add function, and then we would set drvdata to null, so [2] [ 11.971277] panel_simple_get_modes [ OK ] Started LSB: X display manager for KDE. [ OK ] Started LSB: Speech Dispatcher. [ 12.007120] panel_simple_disable [ 12.012323] [ 12.013820] ====================================================== [ 12.019993] [ INFO: possible circular locking dependency detected ] [ 12.026250] 4.4.0-rc3+ #2755 Not tainted [ 12.030165] ------------------------------------------------------- [ 12.036417] Xorg/793 is trying to acquire lock: [ 12.040855] ((&dp->hotplug_work)){+.+...}[ 12.040870] [ 12.040870] but task is already holding lock: [ 12.040871] (crtc_ww_class_mutex){+.+.+.}, at: [<c0383600>] drm_modeset_lock+0x84/0x104 [ 12.040881] [ 12.040881] which lock already depends on the new lock. [ 12.040881] [ 12.040882] [ 12.040882] the existing dependency chain (in reverse order) is: [ 12.040883] [ 12.040883] -> #2 (crtc_ww_class_mutex){+.+.+.}: [ 12.040887] [<c0756374>] mutex_lock_nested+0x78/0x41c [ 12.040893] [<c038365c>] drm_modeset_lock+0xe0/0x104 [ 12.040896] [<c0377888>] drm_mode_getconnector+0x168/0x38c [ 12.040902] [<c036ae58>] drm_ioctl+0x274/0x408 [ 12.040907] [<c017cbc8>] do_vfs_ioctl+0x670/0x750 [ 12.040911] [<c017cd04>] SyS_ioctl+0x5c/0x84 [ 12.040914] [<c000ff80>] ret_fast_syscall+0x0/0x1c [ 12.040920] [ 12.040920] -> #1 (&dev->mode_config.mutex){+.+.+.}: [ 12.040924] [<c0756374>] mutex_lock_nested+0x78/0x41c [ 12.040928] [<c035b3b4>] drm_helper_hpd_irq_event+0x40/0x150 [ 12.040934] [<c038e0b0>] analogix_dp_hotplug+0x24/0x28 [ 12.040938] [<c00450c4>] process_one_work+0x328/0x668 [ 12.040942] [<c0046334>] worker_thread+0x2cc/0x41c [ 12.040945] [<c004bbe8>] kthread+0xf4/0x10c [ 12.040950] [<c0010010>] ret_from_fork+0x14/0x24 [ 12.040954] [ 12.040954] -> #0 ((&dp->hotplug_work)){+.+...}: [ 12.040958] [<c007f4ac>] lock_acquire+0x178/0x218 [ 12.040963] [<c00437d0>] flush_work+0x4c/0x22c [ 12.040966] [<c038e45c>] analogix_dp_bridge_disable+0x74/0xf0 [ 12.040970] [<c0385ef4>] drm_bridge_disable+0x34/0x38 [ 12.040973] [<c0359044>] drm_crtc_helper_set_mode+0x200/0x424 [ 12.040977] [<c0359a38>] drm_crtc_helper_set_config+0x6c0/0x988 [ 12.040981] [<c0373ba8>] drm_mode_set_config_internal+0x60/0xdc [ 12.040984] [<c0378938>] drm_mode_setcrtc+0x3cc/0x474 [ 12.040988] [<c036ae58>] drm_ioctl+0x274/0x408 [ 12.040991] [<c017cbc8>] do_vfs_ioctl+0x670/0x750 [ 12.040994] [<c017cd04>] SyS_ioctl+0x5c/0x84 [ 12.040997] [<c000ff80>] ret_fast_syscall+0x0/0x1c [ 12.041001] [ 12.041001] other info that might help us debug this: [ 12.041001] [ 12.041002] Chain exists of: [ 12.041002] (&dp->hotplug_work) --> &dev->mode_config.mutex --> crtc_ww_class_mutex [ 12.041007] [ 12.041008] Possible unsafe locking scenario: [ 12.041008] [ 12.041009] CPU0 CPU1 [ 12.041010] ---- ---- [ 12.041011] lock(crtc_ww_class_mutex); [ 12.041013] lock(&dev->mode_config.mutex); [ 12.041016] lock(crtc_ww_class_mutex); [ 12.041018] lock((&dp->hotplug_work)); [ 12.041020] [ 12.041020] *** DEADLOCK *** [ 12.041020] [ 12.041023] 3 locks held by Xorg/793: [ 12.041023] #0: (&dev->mode_config.mutex){+.+.+.}, at: [<c0383d48>] drm_modeset_lock_all+0x50/0xd8 [ 12.041030] #1: (crtc_ww_class_acquire){+.+.+.}, at: [<c0383d58>] drm_modeset_lock_all+0x60/0xd8 [ 12.041037] #2: (crtc_ww_class_mutex){+.+.+.}, at: [<c0383600>] drm_modeset_lock+0x84/0x104 [ 12.041044] [ 12.041044] stack backtrace: [ 12.041048] CPU: 3 PID: 793 Comm: Xorg Not tainted 4.4.0-rc3+ #2755 [ 12.041049] Hardware name: Rockchip (Device Tree) [ 12.041059] [<c0019914>] (unwind_backtrace) from [<c0014bcc>] (show_stack+0x20/0x24) [ 12.041066] [<c0014bcc>] (show_stack) from [<c02c4484>] (dump_stack+0x84/0xb8) [ 12.041072] [<c02c4484>] (dump_stack) from [<c007a2b8>] (print_circular_bug+0x278/0x2cc) [ 12.041078] [<c007a2b8>] (print_circular_bug) from [<c007e850>] (__lock_acquire+0x14a0/0x1aec) [ 12.041082] [<c007e850>] (__lock_acquire) from [<c007f4ac>] (lock_acquire+0x178/0x218) [ 12.041086] [<c007f4ac>] (lock_acquire) from [<c00437d0>] (flush_work+0x4c/0x22c) [ 12.041091] [<c00437d0>] (flush_work) from [<c038e45c>] (analogix_dp_bridge_disable+0x74/0xf0) [ 12.041097] [<c038e45c>] (analogix_dp_bridge_disable) from [<c0385ef4>] (drm_bridge_disable+0x34/0x38) [ 12.041102] [<c0385ef4>] (drm_bridge_disable) from [<c0359044>] (drm_crtc_helper_set_mode+0x200/0x424) [ 12.041108] [<c0359044>] (drm_crtc_helper_set_mode) from [<c0359a38>] (drm_crtc_helper_set_config+0x6c0/0x988) [ 12.041114] [<c0359a38>] (drm_crtc_helper_set_config) from [<c0373ba8>] (drm_mode_set_config_internal+0x60/0xdc) [ 12.041119] [<c0373ba8>] (drm_mode_set_config_internal) from [<c0378938>] (drm_mode_setcrtc+0x3cc/0x474) [ 12.041124] [<c0378938>] (drm_mode_setcrtc) from [<c036ae58>] (drm_ioctl+0x274/0x408) [ 12.041129] [<c036ae58>] (drm_ioctl) from [<c017cbc8>] (do_vfs_ioctl+0x670/0x750) [ 12.041134] [<c017cbc8>] (do_vfs_ioctl) from [<c017cd04>] (SyS_ioctl+0x5c/0x84) [ 12.041138] [<c017cd04>] (SyS_ioctl) from [<c000ff80>] (ret_fast_syscall+0x0/0x1c) [ 12.041210] panel_simple_unprepare [ 12.041359] panel_simple_prepare [ 12.101638] analogix_dp_reset [ 12.101879] analogix_dp_init_hpd [ 12.101897] panel_simple_disable [ 12.108543] analogix_dp_irq_handler: irq-type 0 [ 12.110452] rockchip-dp ff970000.dp: Link Training Clock Recovery success [ 12.111900] rockchip-dp ff970000.dp: Link Training success! [ 12.113384] analogix_dp_init_video [ 12.215825] rockchip-dp ff970000.dp: Timeout of video streamclk ok [ 12.215829] rockchip-dp ff970000.dp: unable to config video [ 12.215832] panel_simple_enable ,at: [<c0043784>] flush_work+0x0/0x22c </pre> </blockquote> </blockquote> <pre wrap=""> </pre> </blockquote> </body> </html>