- 24 Jan, 2014 31 commits
-
-
Ben Widawsky authored
This statenment became false here: commit 4fc688ce Author: Jesse Barnes <jbarnes@virtuousgeek.org> Date: Fri Nov 2 11:14:01 2012 -0700 drm/i915: protect RPS/RC6 related accesses (including PCU) with a new mutex Signed-off-by:
Ben Widawsky <ben@bwidawsk.net> Signed-off-by:
Daniel Vetter <daniel.vetter@ffwll.ch>
-
Jesse Barnes authored
Now that we have DDI support, we can check these all the time. Signed-off-by:
Jesse Barnes <jbarnes@virtuousgeek.org> Signed-off-by:
-
Jesse Barnes authored
Read out and calculate the port and pixel clocks on DDI configs as well. This means we have to grab the DP divider values and look at the port mapping to figure out which clock select reg to read out. v2: do the work from ddi_get_config (Ville) v3: check WRPLL reference clock (Ville) add additional SPLL freqs (Ville) clean up port/crtc clock calc (Ville) fix up crtc_clock conditionals (Ville) drop superfluous dp_get_m_n from get_config (Ville) Signed-off-by:Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by:
-
Damien Lespiau authored
We need a bit more flexibility here in the future, bits get shuffled around. v2: more descriptive commit message (Jani Nikula) Reviewed-by:
Jani Nikula <jani.nikula@intel.com> Signed-off-by:
Damien Lespiau <damien.lespiau@intel.com> Signed-off-by:
-
Damien Lespiau authored
So it's easier to compare what we program with the documentation, not having to jump at all. Signed-off-by:
-
Damien Lespiau authored
Also, move that computation outside of the for loop that tries 5 times, this value doesn't change between tries. Signed-off-by:
-
Damien Lespiau authored
A tiny clean-up to allow better code separation between platforms. v2: Fix comment placement (put in in i9xx_get_aux_clock_divider()) and nuke the outdated PCH eDP comment (Jani Nikula) Signed-off-by: -
Jani Nikula authored
Since commit ee1452d7 Author: Jani Nikula <jani.nikula@intel.com> Date: Fri Sep 20 15:05:30 2013 +0300 drm/i915: assume all GM45 Acer laptops use inverted backlight PWM failed and was later reverted in commit be505f64 Author: Alexander van Heukelum <heukelum@fastmail.fm> Date: Sat Dec 28 21:00:39 2013 +0100 Revert "drm/i915: assume all GM45 Acer laptops use inverted backlight PWM" fix the individual broken machine instead. Note to backporters: http://patchwork.freedesktop.org/patch/17837/ is the patch you want for 3.13 and older. Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=54171 Reference: http://mid.gmane.org/DUB115-W7628C7C710EA51AA110CD4A5000@phx.gbl CC: stable@vger.kernel.org Signed-off-by:
-
Jani Nikula authored
It's unused, and nowadays specifying unknown parameters no longer prevents modules from being loaded. Signed-off-by:
-
Todd Previte authored
For HSW+ platforms, enable the 5.4Ghz (HBR2) link rate for devices that support it. The sink device must report that is supports Displayport 1.2 and the HBR2 bit rate in the DPCD in order to use HBR2. Signed-off-by:
Todd Previte <tprevite@gmail.com> Signed-off-by:
-
Ville Syrjälä authored
Group the sprite register writes a bit tighter. We want to write the registers atomically, and so doing the base address/offset artihmetic within the critical section is pointless when it can all be done beforehand. Reviewed-by:
-
Daniel Vetter authored
Currently we're doing the reset handling a bit late, and we're doing it both in the driver load code and on resume. This makes it unusable for e.g. resetting the panel power sequence state like Paulo wants to. Instead of adding yet another single-use callback shuffle things around: - Output handling code is responsible to reset/init all state on its own at driver load time. - We call the reset functions much earlier, before we start using any of the modeset code. Compared to Paulo's new ->resume callback the only difference in placement is that ->reset is still called without dev->struct_mutex held. Which is imo a feature. v2: Rebase on top of the now merge dinq. Cc: Paulo Zanoni <przanoni@gmail.com> Reviewed-by:
Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by:
-
Paulo Zanoni authored
Because we already do the wait in software: see ironlake_wait_backlight_on and ironlake_edp_wait_backlight_off. For the "backlight on" delay, even BSpec says we need to program 0x1 to PP_ON_DELAYS 12:0. For the "backlight off" delay, if we don't do the same thing, when we call ironlake_wait_panel_off we'll end up waiting for the it again. On my machine the off delay is 200ms, so we save this amount of time whenever we disable the panel (e.g, suspend). Signed-off-by:
-
Ville Syrjälä authored
I forgot to set new_config and new_enabled appropriately in the load detect code. Fix it up. v2: Handle the other error path in intel_get_load_detect_pipe() too (Imre) Signed-off-by:
Imre Deak <imre.deak@intel.com> Signed-off-by:
-
Ville Syrjälä authored
Not sure anyone cares about this information. I suppose most people would just look at /proc/interrupts instead. Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Ville Syrjälä authored
irq_received is used as a boolean in i965_irq_handler(), so make it bool. This also makes i965_irq_handler() closer to i915_irq_handler(). Signed-off-by:
-
Ville Syrjälä authored
Add intel_hpd_irq_uninstall() which will cancel the hotplug re-enable timer. Also s/i915_reenable_hotplug_timer_func/intel_hpd_irq_reenable/ Suggested-by:
-
Paulo Zanoni authored
Function ironlake_wait_panel_off should just wait for the power off delay, while function ironlake_wait_panel_power_cycle should wait for the panel cycle (that's required after we turn the panel off, before we enable it again). The problem is that, currently, ironlake_wait_panel_off is waiting not just for the panel to be off, but also for the power cycle delay and the backlight off delay. This function relies on the PP_STATUS bits 3:0, which are not documented and not supposed to be used. A quick analysis of the values we get while waiting quickly shows that power off is reached while bits 3:0 are still 0x1, and the time it takes to become 0x0 is the power cycle delay. On my system with backlight off delay of 200ms, power down delay of 50ms and power cycle delay of 500ms, this is what I get: - Start waiting with value 0x80000008, timestamp 6.429364. - Jumps to 0xa0000003, timestamp 6.431360 (time waited: 0.001996) - Jumps to 0xa0000002, timestamp 6.631277 (time waited: 0.201913) - Jumps to 0x08000001, timestamp 6.681258 (time waited: 0.251894) - Jumps to 0x00000000, timestamp 7.192012 (time waited: 0.762648) As you can see, ironlake_wait_panel_off is sleeping 760ms instead of the expected 50ms: the first 200ms matches the backlight off delay (which we should already have waited for!), then the 50ms for the real panel off delay, then the 500ms for the panel power cycle. This patch makes is look just at bits 31 and 29:28, which will ignore the panel power cycle. And just to be clear: this saves 500ms on my system every time we disable the panel. But we can still save 200ms more (the backlight off delay) on the next patches. Signed-off-by:
Jesse Barnes <jbarnes@virtuougseek.org> Signed-off-by:
-
Paulo Zanoni authored
I like how the macros are nicely column-aligned, so we can properly compare what each macro waits for, but a column full of zeroes doesn't really help anything: it just makes the lines bigger, and they're already way past 80 columns. I imagine this column was used in the past, but IMHO now we can get rid of it. Signed-off-by:
-
Daniel Vetter authored
They now also work on vlv, which has the regs somewhere else. And daring a glance into the looking glass it seems like this functionality will continue to work the same for the next few hardware platforms. So it's better to just remove that misleading prefix and have a bit shorter code for better readability. The only exceptions are the panel/backlight functions shared with intel_ddi.c, those get an intel_ prefix. While at it make the vdd_on/off functions static. And one straggler was missing the edp_ in the name, so make everything neatly OCD. Cc: Paulo Zanoni <paulo.r.zanoni@intel.com> Cc: Jani Nikula <jani.nikula@intel.com> Reviewed-by:
-
Paulo Zanoni authored
The eDP spec defines some points where after you do action A, you have to wait some time before action B. The thing is that in our driver action B does not happen exactly after action A, but we still use msleep() calls directly. What this patch does is that we record the timestamp of when action A happened, then, just before action B, we look at how much time has passed and only sleep the remaining amount needed. With this change, I am able to save about 5-20ms (out of the total 200ms) of the backlight_off delay and completely skip the 1ms backlight_on delay. The 600ms vdd_off delay doesn't happen during normal usage anymore due to a previous patch. v2: - Rename ironlake_wait_jiffies_delay to intel_wait_until_after and move it to intel_display.c - Fix the msleep call: diff is in jiffies v3: - Use "tmp_jiffies" so we don't need to worry about the value of "jiffies" advancing while we're doing the math. v4: - Rename function again. - Move function to i915_drv.h. - Store last_power_cycle at edp_panel_off too. - Use msecs_to_jiffies_timeout, then replace the msleep with an open-coded version that avoids the extra +1 jiffy. - Try to add units to every variable name so we don't confuse jiffies with milliseconds. Signed-off-by: -
Paulo Zanoni authored
Our driver has two different ways of waiting for panel power sequencing delays. One of these ways is through ironlake_wait_panel_status, which implicitly uses the values written to our registers. The other way is through the functions that call intel_wait_until_after, and on this case we do direct msleep() calls on the intel_dp->xxx_delay variables. Function intel_dp_init_panel_power_sequencer is responsible for initializing the _delay variables and deciding which values we need to write to the registers, but it does not write these values to the registers. Only at intel_dp_init_panel_power_sequencer_registers we actually do this write. Then problem is that when we call intel_dp_i2c_init, we will get some I2C calls, which will trigger a VDD enable, which will make use of the panel power sequencing registers and the _delay variables, so we need to have both ready by this time. Today, when this happens, the _delay variables are zero (because they were not computed) and the panel power sequence registers contain whatever values were written by the BIOS (which are usually correct). What this patch does is to make sure that function intel_dp_init_panel_power_sequencer is called earlier, so by the time we call intel_dp_i2c_init, the _delay variables will already be initialized. The actual registers won't contain their final values, but at least they will contain the values set by the BIOS. The good side is that we were reading the values, but were not using them for anything (because we were just skipping the msleep(0) calls), so this "fix" shouldn't fix any real existing bugs. I was only able to identify the problem because I added some debug code to check how much time time we were saving with my previous patch. Regression introduced by: commit ed92f0b2 Author: Paulo Zanoni <paulo.r.zanoni@intel.com> Date: Wed Jun 12 17:27:24 2013 -0300 drm/i915: extract intel_edp_init_connector v2: - Rewrite commit message. Reviewed-by: -
Daniel Vetter authored
This was forgotten in commit 565ee389 Author: Jani Nikula <jani.nikula@intel.com> Date: Wed Nov 13 12:56:29 2013 +0200 drm/i915: do not save/restore backlight registers in KMS Since the confusion was likely due to the duplicated definition for this pci config register, let's unify that, too. Cc: Jani Nikula <jani.nikula@intel.com> Reviewed-by:
-
Imre Deak authored
Atm, we don't print these events for all platforms and for VLV/G4X we also print them for DP AUX completion events which is unnecessary spam. Fix both issues. Signed-off-by:
-
Damien Lespiau authored
Those are two distinct concepts. Just use a comment to remind us to remove that W/A at some point. Signed-off-by:
-
Ville Syrjälä authored
crtc->new_config is only relevant for pipes that are going to be active post-modeset. Set the pointer to NULL for all pipes that are going to be disabled. This is done to help catch bugs where some piece of code would go looking at crtc->new_config even if the data there is stale. v2: Clear new_config in disable_crtc_nofb() too (Imre) Suggested-by:
-
Ville Syrjälä authored
If the first modeset operation fails, we will attempt to restore the previous configuration that we read out from the hardware. But as we don't yet reconstruct the framebuffer information, we end up calling the modeset code with an enabled crtc but with fb==NULL. This will lead to an oops within the modeset code. Check for NULL fb when restoring the configuration, and instead of oopsing simply disable the pipe. Signed-off-by:
-
Ville Syrjälä authored
On VLV we need to compute the new cdclk before we've updated the current state. The code achieved that in a somewhat complex way. Now that we have new_enabled and new_config, we can simplify the code quite a bit. Signed-off-by:
-
Ville Syrjälä authored
Add a new_config pointer to intel_crtc which will point to the new pipe config for said crtc while intel_crtc.config will still contain the old config during first parts of the modeset operation. This is a step towards having the entire new state available during the compute phase, so that we can make accurate decisions about global resource usage. Signed-off-by:
-
Ville Syrjälä authored
Add 'new_enabled' to intel_crtc and precompute it alongside new_encoder and new_crtc. This will allow making decisions about shared resources that are affected by the set of active pipes, before we've clobbered anything for real. Signed-off-by:
-
Daniel Vetter authored
This reverts commit 446f2545. I've left the masking in the pageflip code since that seems to be some useful piece of preemptive robustness. Iirc I've merged this patch under the assumption that the BIOS leaves some random gunk in the lower bits and gets unhappy if we trample on them. We have quite a few case like this, so this made sense. Now I've just learned that there's actual hardware features bits in the low 12 bits, and the kernel needs to preserve them to allow a userspace blob to do its job. Given Dave Airlie's clear stance on userspace blob drivers I've quickly chatted with him and he doesn't seem too happy. So let's revert this. If there are indeed bits that we must preserve in this range then we can ressurrect this patch, but with proper documentation for those bits supplied. And we probably also need to think a bit about interactions with our driver. Cc: Armin Reese <armin.c.reese@intel.com> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> Cc: Dave Airlie <airlied@linux.ie> Reviewed-by:
-
- 23 Jan, 2014 1 commit
-
-
Todd Previte authored
Add new definitions for hotplug live status bits for VLV2 since they're in reverse order from the gen4x ones. Changelog: - Restored gen4 bit definitions - Added new definitions for VLV2 - Added platform check for IS_VALLEYVIEW() in dp_detect to use the correct bit defintions - Replaced a lost trailing brace for the added switch() Signed-off-by:
-
- 22 Jan, 2014 8 commits
-
-
Kenneth Graunke authored
Nothing's changed here; we just need to bump the generation check. Signed-off-by:
Kenneth Graunke <kenneth@whitecape.org> Signed-off-by:
-
Chris Wilson authored
Since an old pageflip will keep its scanout buffer object pinned until it has executed its unpin task on the common workqueue, we can clog up our GGTT with stale pinned objects. As we cannot flush those workqueues without dropping our locks, we have to resort to falling back to userspace and telling them to repeat the operation in order to have a chance to run our workqueues and free up the required memory. If we fail, then we are forced to report ENOSPC back to userspace causing the operation to fail and best-case scenario is that it introduces temporary corruption. Signed-off-by:
Jon Bloomfield <jon.bloomfield@intel.com> Signed-off-by:
-
Chris Wilson authored
On older generations (gen2, gen3) the GPU requires fences for many operations, such as blits. The display hardware also requires fences for scanouts and this leads to a situation where an arbitrary number of fences may be pinned by old scanouts following a pageflip but before we have executed the unpin workqueue. This is unpredictable by userspace and leads to random EDEADLK when submitting an otherwise benign execbuffer. However, we can detect when we have an outstanding flip and so cause userspace to wait upon their completion before finally declaring that the system is starved of fences. This is really no worse than forcing the GPU to stall waiting for older execbuffer to retire and release their fences before we can reallocate them for the next execbuffer. v2: move the test for a pending fb unpin to a common routine for later reuse during eviction Reported-and-tested-by: dimon@gmx.net Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=73696Signed-off-by:
-
Imre Deak authored
Atm after a failed link training we disable the DP port. This can happen during a modeset-enable or a DP link re-establishment. The latter can be a problem and we shouldn't disable the DP port, see the previous patch for the reasoning. In the former case the right thing would be to disable the DP port, but also the rest of the pipe. As a stop-gap solution leave the DP port enabled in both cases. It is an improvement on its own (avoiding HW lock ups) and the proper solution for the first case requires a bigger change, so let's keep that on the TODO list. v2: - fix explanation of change impact (Chris) Suggested-by:
Daniel Vetter <daniel@ffwll.ch> Signed-off-by:
-
Imre Deak authored
Currently if the DP link is lost (either because of a hot unplug, or failed link status check) we disable the DP port, but leave the rest of the pipe running. This is incompatible with the modeset disabling sequence of some platforms/configurations. At least this is the case for DP ports on the CPU as opposed to PCH. Atm we'll also get a warning when we do a modeset disable after the above link lost event, since we expect the DP port to be enabled at this point (see the bugzilla ticket for the related dmesg). Note that with this patch we'll still end up disabling the port, thanks to the HPD uevent and subsequent modeset disable. See also the next patch fixing the other half of this issue. Solution suggested by Ville. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70570Signed-off-by:
-
Ville Syrjälä authored
My 855gm doesn't register the intel backlight but it still ends up calling the backlight code to enable/disable the backlight via the LVDS code. This leads to some WARNs due to backlight.max being 0. Let's have intel_panel_enable_backlight() and intel_panel_disable_backlight() check whether there's a backlight present or not. Also move the backlight.present check from asle_set_backlight() into intel_panel_set_backlight() for some extra symmetry. Signed-off-by:
-
Imre Deak authored
Fix typo possibly leading to timed out DP aux transactions on ports C,D. Introduced in: Commmit 4aeebd74 Author: Daniel Vetter <daniel.vetter@ffwll.ch> Date: Thu Oct 31 09:53:36 2013 +0100 Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=72210 Signed off-by: Imre Deak <imre.deak@intel.com> Signed-off-by:
-
Ben Widawsky authored
We need to defer the free request until the object/vma is capable of being freed - or else we have a problem when we try to destroy the context. The exact same issue is described and fixed here: commit e2078043 Author: Ben Widawsky <ben@bwidawsk.net> Date: Fri Dec 6 14:11:22 2013 -0800 drm/i915: Defer request freeing I had this fix previously, but decided not to keep it for some reason I can no longer remember. gem_reset_stats is a really good test at hitting the problem. For the inquisitive: [ 170.516392] ------------[ cut here ]------------ [ 170.517227] WARNING: CPU: 1 PID: 105 at drivers/gpu/drm/drm_mm.c:578 drm_mm_takedown+0x2e/0x30 [drm]() [ 170.518064] Memory manager not clean during takedown. [ 170.518941] CPU: 1 PID: 105 Comm: kworker/1:1 Not tainted 3.13.0-rc4-BEN+ #28 [ 170.519787] Hardware name: Hewlett-Packard HP EliteBook 8470p/179B, BIOS 68ICF Ver. F.02 04/27/2012 [ 170.520662] Call Trace: [ 170.521517] [<ffffffff814f0589>] dump_stack+0x4e/0x7a [ 170.522373] [<ffffffff81049e6d>] warn_slowpath_common+0x7d/0xa0 [ 170.523227] [<ffffffff81049edc>] warn_slowpath_fmt+0x4c/0x50 [ 170.524079] [<ffffffffa06c414e>] drm_mm_takedown+0x2e/0x30 [drm] [ 170.524934] [<ffffffffa07213f3>] gen6_ppgtt_cleanup+0x23/0x110 [i915] [ 170.525777] [<ffffffffa07837ed>] ppgtt_release.part.5+0x24/0x29 [i915] [ 170.526603] [<ffffffffa071aaa5>] i915_gem_context_free+0x195/0x1a0 [i915] [ 170.527423] [<ffffffffa071189d>] i915_gem_free_request+0x9d/0xb0 [i915] [ 170.528247] [<ffffffffa0718af9>] i915_gem_reset+0x1f9/0x3f0 [i915] [ 170.529065] [<ffffffffa0700cce>] i915_reset+0x4e/0x180 [i915] [ 170.529870] [<ffffffffa070829d>] i915_error_work_func+0xcd/0x120 [i915] [ 170.530666] [<ffffffff8106c13a>] process_one_work+0x1fa/0x6d0 [ 170.531453] [<ffffffff8106c0d8>] ? process_one_work+0x198/0x6d0 [ 170.532230] [<ffffffff8106c72b>] worker_thread+0x11b/0x3a0 [ 170.532996] [<ffffffff8106c610>] ? process_one_work+0x6d0/0x6d0 [ 170.533771] [<ffffffff810743ef>] kthread+0xff/0x120 [ 170.534548] [<ffffffff810742f0>] ? insert_kthread_work+0x80/0x80 [ 170.535322] [<ffffffff814f97ac>] ret_from_fork+0x7c/0xb0 [ 170.536089] [<ffffffff810742f0>] ? insert_kthread_work+0x80/0x80 [ 170.536847] ---[ end trace 3d4c12892e42d58f ]--- v2: Whitespace fix. (Chris) Note: This is a bug that only hits the ppgtt topic branch but I've figured that doing the request cleanup in this order is generally the right thing to do. Signed-off-by:
-
