- 26 Aug, 2015 34 commits
-
-
Paulo Zanoni authored
Dear git bisect user, Even though this is the patch that introduced the WARN() you're bisecting, please notice that it's very likely that the problem you're facing was already present before this commit. In other words: this commit adds code to detect errors and give WARN()s about them, but the errors were already there. In order to continue your debug, please use the i915.mmio_debug option, check the backtraces and try to discover which read or write operation is causing the error message. Then check if this is happening because the register does not exist or because its power well is down when the operation is being done. On my SKL machine, if I use i915.mmio_debug=999, this patch triggers 42 WARNs just by booting. I didn't investigate them yet. Normal users are only going to get a single WARN due to the default i915.mmio_debug setting. Thank you for your comprehension, Paulo Signed-off-by:
Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by:
Daniel Vetter <daniel.vetter@ffwll.ch>
-
Ville Syrjälä authored
We can choose to leave the display PHY CL2 powerdown up to some hardware signals, or we can force it. The BXT code forces the nonexistent CL2 in the x1 PHY to power down. Follow suit on CHV. Maybe it can still save some extra power by disabling some extra logic in CL1, or something. Signed-off-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by:
Deepak S <deepak.s@linux.intel.com> Signed-off-by:
-
Ville Syrjälä authored
CHV has supports some form of automagic clock gating for the DPIO SUS clock. We can simply enable the magic bits and the hardware should take care of the rest. Signed-off-by:
-
Ville Syrjälä authored
With DPIO powergating active the DPLL can't be accessed unless something else is keeping the common lane in the channel on. That means the PPS kick procedure could fail to enable the PLL. Power up some data lanes to force the common lane to power up so that the PLL can be enabled temporarily. v2: Avoid gcc uninitilized variable warning Signed-off-by:
-
Ville Syrjälä authored
Normmally the common lane in a PHY channel gets powered up when some of the data lanes get powered up. But when we're driving port B with pipe B we don't want to enabled any of the data lanes, and just want the DPLL in the common lane to be active. To make that happens we have to temporarily enable some data lanes after which we can access the DPLL registers in the common lane. Once the pipe is up and running we can drop the power override on the data lanes allowing them to shut down. From this point forward the common lane will in fact stay powered on until the data lanes in the other channel get powered down. Ville's extended explanation from the review thread: On Wed, Aug 19, 2015 at 07:47:41AM +0530, Deepak wrote: > One Q, why only for port B? Port C is also in same common lane right? Port B is in the first PHY channel which also houses CL1. CL1 always powers up whenever any lanes in either PHY channel are powered up. CL2 only powers up if lanes in the second channel (ie. the one with port C) powers up. So in this scenario (pipe B->port B) we want the DPLL from CL2, but ideally we only want to power up the lanes for port B. Powering up port B lanes will only power up CL1, but as we need CL2 instead we need to, temporarily, power up some lanes in port C as well. Crossing the streams the other way (pipe A->port C) is not a problem since CL1 powers up whenever anything else powers up. So powering up some port C lanes is enough on its own to make the CL1 DPLL operational, even though CL1 and the lanes live in separate channels. Signed-off-by:
-
Ville Syrjälä authored
Powergate the PHY lanes when they're not needed. For HDMI all four lanes are needed always, but for DP we can enable only the needed lanes. To power down the unused lanes we use some power down override bits in the DISPLAY_PHY_CONTROL register. Without the overrides it appears that the hardware always powers on all the lanes. When the port is disabled the power down override is not needed and the lanes will shut off on their own. That also means the override is critical to actually be able to access the DPIO registers before the port is actually enabled. Additionally the common lanes will power down when not needed. CL1 remains on as long as anything else is on, CL2 will shut down when all the lanes in the same channel will shut down. There is one exception for CL2 that will be dealt in a separate patch for clarity. With potentially some lanes powered down, the DP code now has to check the number of active lanes before accessing PCS/TX registers. All registers in powered down blocks will reads as 0xffffffff, and soe we would drown in warnings from vlv_dpio_read() if we allowed the code to access all those registers. Another important detail in the DP code is the "TX latency optimal" setting. Normally the second TX lane acts as some kind of reset master, with the other lanes as slaves. But when only a single lane is enabled, that single lane obviously has to be the master. A bit of extra care is needed to reconstruct the initial state of the DISPLAY_PHY_CONTROL register since it can't be read safely. So instead read the actual lane status from the DPLL/PHY_STATUS registers and use that to determine which lanes ought to be powergated initially. We also need to switch the PHY power modes to "deep PSR" to avoid a hard system hang when powering down the single channel PHY. Also sprinkle a few debug prints around so that we can monitor the DISPLAY_PHY_STATUS changes without having to read it and risk corrupting it. v2: Add locking to chv_powergate_phy_lanes() v3: Actually enable dynamic powerdown in the PHY and deal with the fallout Signed-off-by: -
Ville Syrjälä authored
Bunch of stuff needs the DPLL ref/cri clocks on both VLV and CHV, and having VGA mode enabled causes some problems for CHV. So let's just pull the code to configure those bits into the disp2d well enable hook. With the DPLL disable code also fixed to leave those bits alone we should now have a consistent DPLL state all the time even if the DPLL is disabled. This also neatly removes some duplicated code between the VLV and CHV codepaths. Signed-off-by:
Sivakumar Thulasimani <sivakumar.thulasimani@intel.com> Signed-off-by:
-
Ville Syrjälä authored
Most of our char* arrays are markes as const already, but a few slipped through the cracks. Fix it. Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Ville Syrjälä authored
A couple of hand rolled ARRAY_SIZE()s caught my eye. Get rid of them. Signed-off-by:
-
Ville Syrjälä authored
Simple one: drivers/gpu/drm/i915/i915_debugfs.c:2449:57: warning: Using plain integer as NULL pointer And something a bit more peculiar: drivers/gpu/drm/i915/i915_debugfs.c:4953:18: warning: Variable length array is used. drivers/gpu/drm/i915/i915_debugfs.c:4953:32: warning: Variable length array is used. We pass a 'const int' as the array size which results in the warning, dropping the const gets rid of the warning. Weird, but I think getting rid of the warnings is better than holding on to the const. Signed-off-by:
-
Jani Nikula authored
BXT platform uses live status bits from 0x44440 register to obtain DP status on hotplug. The existing g4x_digital_port_connected() uses a different register and hence misses DP hotplug events on BXT platform. This patch fixes it by using the appropriate register(0x44440) and live status bits(3:5). Based on a patch by Durgadoss R <durgadoss.r@intel.com>, from whom the commit message is shamelessly copy pasted. Reported-by:
Durgadoss R <durgadoss.r@intel.com> Signed-off-by:
Jani Nikula <jani.nikula@intel.com> Reviewed-by:
-
Jani Nikula authored
Choose the right function at the intel_digital_port_connected level. Signed-off-by:
-
Jani Nikula authored
Choose the right function at the intel_digital_port_connected level. Signed-off-by:
-
Jani Nikula authored
Add a common intel_digital_port_connected() that splits out to functions for different platforms. No functional changes. v2: make the function return a boolean Signed-off-by:
-
Jani Nikula authored
With the case added for eDP on port A (always connected from this function's point of view), we should not be hitting any of the default cases in ibx_digital_port_connected, so add MISSING_CASE annotation. Signed-off-by:
-
Jani Nikula authored
We should not be hitting any of the default cases in g4x_digital_port_connected, so add MISSING_CASE annotation and return boolean status. The current behaviour is just cargo culting from the days of yonder when the display port support was added to i915. Signed-off-by:
-
Jani Nikula authored
The function can be made static there. No functional changes. Reviewed-by:
-
Mika Kahola authored
It is possible the we request to have a mode that has higher pixel clock than our HW can support. This patch checks if requested pixel clock is lower than the one supported by the HW. The requested mode is discarded if we cannot support the requested pixel clock. This patch applies to DVO. V2: - removed computation for max pixel clock V3: - cleanup by removing unnecessary lines V4: - clock check against max dotclock moved inside 'if (fixed_mode)' V5: - dot clock check against fixed_mode clock when available Signed-off-by:
Mika Kahola <mika.kahola@intel.com> Reviewed-by:
-
Mika Kahola authored
It is possible the we request to have a mode that has higher pixel clock than our HW can support. This patch checks if requested pixel clock is lower than the one supported by the HW. The requested mode is discarded if we cannot support the requested pixel clock. This patch applies to DSI. V2: - removed computation for max pixel clock V3: - cleanup by removing unnecessary lines V4: - max_pixclk variable renamed as max_dotclk - moved dot clock checking inside 'if (fixed_mode)' V5: - dot clock checked against fixed_mode clock Signed-off-by:
-
Mika Kahola authored
It is possible the we request to have a mode that has higher pixel clock than our HW can support. This patch checks if requested pixel clock is lower than the one supported by the HW. The requested mode is discarded if we cannot support the requested pixel clock. This patch applies to LVDS. V2: - removed computation for max pixel clock V3: - cleanup by removing unnecessary lines V4: - moved supported dotclock check from mode_valid() to intel_lvds_init() V5: - dotclock check moved back to mode_valid() function - dotclock check for fixed mode Signed-off-by:
-
Mika Kahola authored
Store max dotclock into dev_priv structure so we are able to filter out the modes that are not supported by our platforms. V2: - limit the max dot clock frequency to max CD clock frequency for the gen9 and above - limit the max dot clock frequency to 90% of the max CD clock frequency for the older gens - for Cherryview the max dot clock frequency is limited to 95% of the max CD clock frequency - for gen2 and gen3 the max dot clock limit is set to 90% of the 2X max CD clock frequency V3: - max_dotclk variable renamed as max_dotclk_freq in i915_drv.h - in intel_compute_max_dotclk() the rounding method changed from round up to round down when computing max dotclock V4: - Haswell and Broadwell supports now dot clocks up to max CD clock frequency Signed-off-by:
-
Ville Syrjälä authored
Add vlv_dport_to_phy() and fix up the return values of vlv_dport_to_channel() and vlv_pipe_to_channel() to use the appropriate enums. Signed-off-by:
-
Ville Syrjälä authored
With DPIO powergating active on CHV, we can't even access the DPIO PLL registers until the lane power state overrides have been enabled. That will happen from the encoder .pre_pll_enable() hook, so move chv_prepare_pll() to happen after that point, which puts it just before chv_enable_pll() actually. Do the same for VLV to avoid accumulating weird differences between the platforms. Both platforms seem happy with the new arrangement. Signed-off-by:
-
Ville Syrjälä authored
dev_priv->chv_phy_control is protected by the power_domains->lock elsewhere, so also grab it when initializing chv_phy_control. Signed-off-by:
-
Ville Syrjälä authored
To implement DPIO lane power gating on CHV we're going to need to access DPIO registers from the cmn power well enable hook. That gets called rather early, so we need to move the DPIO port IOSF sideband port assignment earlier as well. Signed-off-by:
-
Ville Syrjälä authored
Move the CHV clock buffer disable from chv_disable_pll() to the new encoder .post_pll_disable() hook. This is more symmetric since the clock buffer enable happens from the .pre_pll_enable() hook. We'll have more use for the new hook soon. Signed-off-by:
-
Ville Syrjälä authored
The docs give you the impression that the unique transition scale value shouldn't matter when unique transition scale is enabled. But as Imre found on BXT (and I verfied also on BSW) the value does matter. So from now on just program the same value 0x9a always. Cc: Imre Deak <imre.deak@intel.com> Signed-off-by:
-
Ville Syrjälä authored
When fractional m2 divider isn't used on CHV the fractional part is ignore by the hardware. Despite that, program the fractional value (0 in this case) to the hardware register just to keep things a bit more consistent. Might at least make register dumps a bit less confusing when there isn't some stale fractional part hanging around. Signed-off-by:
-
Dave Gordon authored
The current versions of these two macros don't work correctly if the argument expression happens to contain a modulo operator (%) -- when stringified, it gets interpreted as a printf formatting character! With a specifically crafted parameter, this could probably cause a kernel OOPS; consider WARN_ON(p%s) or WARN_ON(f %*pEp). Instead, we should use an explicit "%s" format, with the stringified expression as the coresponding literal-string argument. Signed-off-by:
Dave Gordon <david.s.gordon@intel.com> Reviewed-by:
-
Ville Syrjälä authored
With MST there won't be a crtc assigned to the main link encoder, so trying to dig up the pipe_config from there is a recipe for an oops. Instead store the parameters (lane_count and link_rate) in the encoder, and use those values during link training etc. Since those parameters are now assigned only when the link is actually enabled, .compute_config() won't clobber them as it did before. Hardware state readout is still bonkers though as we don't transfer the link parameters from pipe_config intel_dp. We should do that during encoder sanitation. But since we don't even do a proper job of reading out the main link encoder state for MST there's littel point in worrying about this now. Fixes a regression with MST caused by: commit 90a6b7b0 Author: Ville Syrjälä <ville.syrjala@linux.intel.com> Date: Mon Jul 6 16:39:15 2015 +0300 drm/i915: Move intel_dp->lane_count into pipe_config v2: Different apporoach that should keep intel_dp_check_mst_status() somewhat less oopsy Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Reported-by:
Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Signed-off-by:
-
Imre Deak authored
Due to a coherency issue on BXT A steppings we can't guarantee a coherent view of cached (CPU snooped) GPU mappings, so fail such requests. User space is supposed to fall back to uncached mappings in this case. v2: - limit the WA to A steppings, on later stepping this HW issue is fixed v3: - return error instead of trying to work around the issue in kernel, since that could confuse user space (Chris) Testcast: igt/gem_store_dword_batches_loop/cached-mapping Signed-off-by:
Imre Deak <imre.deak@intel.com> Reviewed-by:
-
Imre Deak authored
By running igt/store_dword_loop_render on BXT we can hit a coherency problem where the seqno written at GPU command completion time is not seen by the CPU. This results in __i915_wait_request seeing the stale seqno and not completing the request (not considering the lost interrupt/GPU reset mechanism). I also verified that this isn't a case of a lost interrupt, or that the command didn't complete somehow: when the coherency issue occured I read the seqno via an uncached GTT mapping too. While the cached version of the seqno still showed the stale value the one read via the uncached mapping was the correct one. Work around this issue by clflushing the corresponding CPU cacheline following any store of the seqno and preceding any reading of it. When reading it do this only when the caller expects a coherent view. v2: - fix using the proper logical && instead of a bitwise & (Jani, Mika) - limit the workaround to A stepping, on later steppings this HW issue is fixed v3: - use a separate get_seqno/set_seqno vfunc (Chris) Testcase: igt/store_dword_loop_render Signed-off-by:
-
Rodrigo Vivi authored
We also need to call the frontbuffer flip to trigger proper invalidations when disabling planes. Otherwise we will miss screen updates when disabling sprites or cursor. On core platforms where HW tracking also works, this issue is totally masked because HW tracking triggers PSR exit however on VLV/CHV that has only SW tracking we miss screen updates when disabling planes. It was caught with kms_psr_sink_crc sprite_plane_onoff and cursor_plane_onoff subtests running on VLV/CHV. This is probably a regression since I can also get this with the manual test case, but with so many changes on atomic modeset I couldn't track exactly when this was introduced. Reviewed-by:
Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by:
-
Arun Siluvery authored
MI_STORE_REGISTER_MEM, MI_LOAD_REGISTER_MEM instructions are not really variable length instructions unlike MI_LOAD_REGISTER_IMM where it expects (reg, addr) pairs so use fixed length for these instructions. v2: rebase Cc: Dave Gordon <david.s.gordon@intel.com> Signed-off-by:
Arun Siluvery <arun.siluvery@linux.intel.com> Reviewed-by:
Mika Kuoppala <mika.kuoppala@intel.com> [danvet: Appease checkpatch as Mika spotted in i915_reg.h - it seems terminally unhappy about i915_cmd_parser.c so that would be a separate patch.] Signed-off-by:
-
- 25 Aug, 2015 1 commit
-
-
Graham Whaley authored
In commit d1675198: drm/i915: Integrate GuC-based command submission the drm.tmpl include lines reference the intel_guc_submission.c but the patch adds the file i915_guc_submission.c. drm.tmpl fails to build with: docproc: .//drivers/gpu/drm/i915/intel_guc_submission.c: No such file or directory Change the file reference to the actual file. Signed-off-by:
Graham Whaley <graham.whaley@linux.intel.com> Signed-off-by:
-
- 14 Aug, 2015 5 commits
-
-
Jani Nikula authored
There's so much scaler debugging messages that it makes other debugging hard. Remove them. Signed-off-by:
-
Dave Gordon authored
This provides a means of reading status and counts relating to GuC actions and submissions. v2: Remove surplus blank line in output [Chris Wilson] v5: Added GuC per-engine submission & seqno statistics v6: Add per-ring statistics to client, refactor client-dumper. Signed-off-by:Alex Dai <yu.dai@intel.com> Reviewed-by:
Tom O'Rourke <Tom.O'Rourke@intel.com> Signed-off-by:
-
Alex Dai authored
GuC-based submission is mostly the same as execlist mode, up to intel_logical_ring_advance_and_submit(), where the context being dispatched would be added to the execlist queue; at this point we submit the context to the GuC backend instead. There are, however, a few other changes also required, notably: 1. Contexts must be pinned at GGTT addresses accessible by the GuC i.e. NOT in the range [0..WOPCM_SIZE), so we have to add the PIN_OFFSET_BIAS flag to the relevant GGTT-pinning calls. 2. The GuC's TLB must be invalidated after a context is pinned at a new GGTT address. 3. GuC firmware uses the one page before Ring Context as shared data. Therefore, whenever driver wants to get base address of LRC, we will offset one page for it. LRC_PPHWSP_PN is defined as the page number of LRCA. 4. In the work queue used to pass requests to the GuC, the GuC firmware requires the ring-tail-offset to be represented as an 11-bit value, expressed in QWords. Therefore, the ringbuffer size must be reduced to the representable range (4 pages). v2: Defer adding #defines until needed [Chris Wilson] Rationalise type declarations [Chris Wilson] v4: Squashed kerneldoc patch into here [Daniel Vetter] v5: Update request->tail in code common to both GuC and execlist modes. Add a private version of lr_context_update(), as sharing the execlist version leads to race conditions when the CPU and the GuC both update TAIL in the context image. Conversion of error-captured HWS page to string must account for offset from start of object to actual HWS (LRC_PPHWSP_PN). Issue: VIZ-4884 Signed-off-by: -
Dave Gordon authored
Turn on interrupt steering to route necessary interrupts to GuC. v6: Rebased Issue: VIZ-4884 Signed-off-by: -
Dave Gordon authored
A GuC client has its own doorbell and workqueue. It maintains the doorbell cache line, process description object and work queue item. A default guc_client is created for the i915 driver to use for normal-priority in-order submission. Note that the created client is not yet ready for use; doorbell allocation will fail as we haven't yet linked the GuC's context descriptor to the default contexts for each ring (see later patch). v2: Defer adding structure members until needed [Chris Wilson] Rationalise type declarations [Chris Wilson] v5: Add GuC per-engine submission & seqno statistics. Move wq locking to encompass both get_space() and add_item(). Take forcewake lock in host2guc_action() [Tom O'Rourke] v6: Fix GuC doorbell cacheline selection code (the cacheline-within-page calculation was wrong). Rename GuC priorities to make them closer to the names used in the GuC firmware source, matching what the autogenerated versions will (probably) be. Add per-ring statistics to client. Issue: VIZ-4884 Signed-off-by:
-
