There is a regression from 2.6.34 related to the recent radeon power
management changes, caused by attempting to cancel a delayed work
item that's never been scheduled. However, the code as is has some
other issues potentially leading to visible problems.
First, the mutex around cancel_delayed_work() in radeon_pm_suspend()
doesn't really serve any purpose, because cancel_delayed_work() only
tries to delete the work's timer. Moreover, it doesn't prevent the
work handler from running, so the handler can do some wrong things if
it wins the race and in that case it will rearm itself to do some
more wrong things going forward. So, I think it's better to wait for
the handler to return in case it's already been queued up for
execution. Also, it should be prevented from rearming itself in that
case.
Second, in radeon_set_pm_method() the cancel_delayed_work() is not
sufficient to prevent the work handler from running and queing up
itself for the next run (the failure scenario is that
cancel_delayed_work() returns 0, so the handler is run, it waits on
the mutex and then rearms itself after the mutex has been released),
so again the work handler should be prevented from rearming itself in
that case..
Finally, there's a potential deadlock in radeon_pm_fini(), because
cancel_delayed_work_sync() is called under rdev->pm.mutex, but the
work handler tries to acquire the same mutex (if it wins the race).
Fix the issues described above.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Some IGP systems specify the system memory clock in the Firmware
table rather than the IGP info table. Check both and make sure
we have a value system memory clock value.
v2: make sure rs690_pm_info is called on rs780/rs880 as well.
fixes a regression since 07d4190327b02ab3aaad25a2d168f79d92e8f8c2.
Reported-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Tested-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
track the current voltage level and avoid setting it
if the requested voltage is already set.
v2: check voltage type before checking current voltage
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This adds an additional profile, mid, to the pm profile
code which takes the place of the old low profile. The default
behavior remains the same, e.g., auto profile now selects between
mid and high profiles based on power source, however, you can now
manually force the low profile which was previously only available
as a dpms off state. Enabling the low profile when the displays
are on has been known to cause display corruption in some cases.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
- This enables voltage adjustment on r6xx+ and certain
r5xx asics.
- Voltage drop support is already available for most
r1xx-r5xx asics.
V2: endian fix for voltage table.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
I think it's good to release the AGP bridge at suspend
and reacquire it at resume. Also fix :
https://bugzilla.kernel.org/show_bug.cgi?id=15969
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Cc: stable <stable@kernel.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
list reservation was too optimistic about ttm object reservation
and could think that an object reserved by some other process
as reserved by the list reservation which was false. Thus when
unreserving the list it might unreserve object that it didn't
reserved in the list. Sorry if it's hard to follow but this
kind of things are just causing headheck.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
- Separate dynpm and profile based power management methods. You can select the pm method
by echoing the selected method ("dynpm" or "profile") to power_method in sysfs.
- Expose basic 4 profile in profile method
"default" - default clocks
"auto" - select between low and high based on ac/dc state
"low" - DC, low power mode
"high" - AC, performance mode
The current base profile is "default", but it should switched to "auto" once we've tested
on more systems. Switching the state is a matter of echoing the requested profile to
power_profile in sysfs. The lowest power states are selected automatically when dpms turns
the monitors off in all states but default.
- Remove dynamic fence-based reclocking for the moment. We can revisit this later once we
have basic pm in.
- Move pm init/fini to modesetting path. pm is tightly coupled with display state. Make sure
display side is initialized before pm.
- Add pm suspend/resume functions to make sure pm state is properly reinitialized on resume.
- Remove dynpm module option. It's now selectable via sysfs.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The lowest power states often cause display problems, so only enable
them when all displays are off.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We need to handle the ring while we've already locked it, so split out
the allocation and commit functions in order to allow them to be used.
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The vblank interrupt on r600 doesn't seem to be especially reliable, so
perform some sanity checks before the actual reclock.
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Touching vram while the card is reclocking can lead to lockups. Unmap
any pages that could be touched by the CPU and block any accesses to
vram until the reclocking is complete.
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Add two new sysfs attributes:
- dynpm
- power_state
Echoing 0/1 to dynpm disables/enables dynamic power management.
The driver scales the sclk dynamically based on the number of
queued fences. dynpm only scales sclk dynamically in single head
mode.
Echoing x.y to power_state selects a static power state (x) and clock
mode (y). This allows you to statically select a power state and clock
mode. Selecting a static clock mode will disable dynpm.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
- pm_misc() - handles voltage, pcie lanes, and other non
clock related power mode settings. Currently disabled.
Needs further debugging
- pm_prepare() - disables crtc mem requests right now.
All memory clients need to be disabled when changing
memory clocks. This function can be expanded to include
disabling fb access as well.
- pm_finish() - enable active memory clients.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
- remove non_clock_info struct
- track power state misc flags
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This also simplifies the code and enables reclocking with multiple heads
active by tracking whether the power states are single or multi-head
capable.
Eventually, we will want to select a power state based on external
factors (AC/DC state, user selection, etc.).
(v2) Update for evergreen
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Useful for certain power management operations. You
need to wait for the GUI engine (2D, 3D, CP, etc.) to be
idle before changing clocks or adjusting engine parameters.
(v2) Fix gui idle enable on pre-r6xx asics
(v3) The gui idle interrrupt status bit is permanently asserted
on pre-r6xx chips, but the interrrupt is still generated.
workaround it in the driver.
(v4) Add support for evergreen
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Check to see if the GUI engine and related blocks
(2D, 3D, CP, etc) are idle or not. There are a number
of cases when we need to know if the drawing engine
is busy.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Implements irq support for HDMI audio output. Now the polling timer
is only enabled if irq support isn't available.
Signed-off-by: Christian König <deathsimple@vodafone.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Rework HDMI audio polling timer, only enable it when
at least one HDMI encoder needs it. Preparation for
replacing it with irq support.
Signed-off-by: Christian König <deathsimple@vodafone.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
* drm-radeon-evergreen-accel:
drm/radeon: fix cypress firmware typo.
drm/radeon/kms/evergreen: add hpd support
drm/radeon/kms/evergreen: implement irq support
drm/radeon/kms/evergreen: setup and enable the CP
drm/radeon/kms/evergreen: implement gfx init
drm/radeon/kms/evergreen: add soft reset function
drm/radeon/kms/evergreen: add gart support
drm/radeon/kms: add support for evergreen power tables
drm/radeon/kms: update atombios.h power tables for evergreen
* drm-fbdev-cleanup:
drm/fb: remove drm_fb_helper_setcolreg
drm/kms/fb: use slow work mechanism for normal hotplug also.
drm/kms/fb: add polling support for when nothing is connected.
drm/kms/fb: provide a 1024x768 fbcon if no outputs found.
drm/kms/fb: separate fbdev connector list from core drm connectors
drm/kms/fb: move to using fb helper crtc grouping instead of core crtc list
drm/fb: fix fbdev object model + cleanup properly.
Conflicts:
drivers/gpu/drm/i915/i915_drv.h
drivers/gpu/drm/nouveau/nouveau_drv.h
The command processor (CP) fetches command buffers and
feeds the GPU. This patch requires the evergreen
family me and pfp ucode files.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This initializes the gfx engine so accel can
eventually be used.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The fbdev layer in the kms code should act like a consumer of the kms services and avoid having relying on information being store in the kms core structures in order for it to work.
This patch
a) removes the info pointer/psuedo palette from the core drm_framebuffer structure and moves it to the fbdev helper layer, it also removes the core drm keeping a list of kernel kms fbdevs.
b) migrated all the fb helper functions out of the crtc helper file into the fb helper file.
c) pushed the fb probing/hotplug control into the driver
d) makes the surface sizes into a structure for ease of passing
This changes the intel/radeon/nouveau drivers to use the new helper.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This simplify and improve GPU reset for R1XX-R6XX hw, it's
not 100% reliable here are result:
- R1XX/R2XX works bunch of time in a row, sometimes it
seems it can work indifinitly
- R3XX/R3XX the most unreliable one, sometimes you will be
able to reset few times, sometimes not even once
- R5XX more reliable than previous hw, seems to work most
of the times but once in a while it fails for no obvious
reasons (same status than previous reset just no same
happy ending)
- R6XX/R7XX are lot more reliable with this patch, still
it seems that it can fail after a bunch (reset every
2sec for 3hour bring down the GPU & computer)
This have been tested on various hw, for some odd reasons
i wasn't able to lockup RS480/RS690 (while they use to
love locking up).
Note that on R1XX-R5XX the cursor will disapear after
lockup haven't checked why, switch to console and back
to X will restore cursor.
Next step is to record the bogus command that leaded to
the lockup.
V2 Fix r6xx resume path to avoid reinitializing blit
module, use the gpu_lockup boolean to avoid entering
inifinite waiting loop on fence while reiniting the GPU
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Patch rename gpu_reset to asic_reset in prevision of having
gpu_reset doing more stuff than just basic asic reset.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This patch cleanup the fence code, it drops the timeout field of
fence as the time to complete each IB is unpredictable and shouldn't
be bound.
The fence cleanup lead to GPU lockup detection improvement, this
patch introduce a callback, allowing to do asic specific test for
lockup detection. In this patch the CP is use as a first indicator
of GPU lockup. If CP doesn't make progress during 1second we assume
we are facing a GPU lockup.
To avoid overhead of testing GPU lockup frequently due to fence
taking time to be signaled we query the lockup callback every
500msec. There is plenty code comment explaining the design & choise
inside the code.
This have been tested mostly on R3XX/R5XX hw, in normal running
destkop (compiz firefox, quake3 running) the lockup callback wasn't
call once (1 hour session). Also tested with forcing GPU lockup and
lockup was reported after the 1s CP activity timeout.
V2 switch to 500ms timeout so GPU lockup get call at least 2 times
in less than 2sec.
V3 store last jiffies in fence struct so on ERESTART, EBUSY we keep
track of how long we already wait for a given fence
V4 make sure we got up to date cp read pointer so we don't have
false positive
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Turn off hw i2c by default except for mm i2c which
is hw only until we sort out the remaining prescale
issues on older chips. hw i2c can be enabled with
hw_i2c=1.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
- Add module option to force the display priority
0 = auto, 1 = normal, 2 = high
- Default to high on r3xx/r4xx/rv515 chips
Fixes flickering problems during heavy acceleration
due to underflow to the display controllers
- Fill in minimal support for RS600
v2 - update display priority when bandwidth is updated
so the user can change the parameter at runtime and it
will take affect on the next modeset.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
- rs780/880 were using the wrong bandwidth functions
- convert r1xx-r4xx to use the same pm sclk/mclk structs as
r5xx+
- move bandwidth setup to a common function
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Look up i2c bus in the power table and expose it.
You'll need to load a hwmon driver for any chips
on the bus, this patch just exposes the bus.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
This just an example to show what radeon_asic.h might be good for.
Before Jerome kills it ;)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
And move asic init plus a few related functions from radeon_device.c
to it. This file will hold all the asic structures in the future,
but atm they're still stuck in radeon_asic.h.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We tried to implement interruptible waiting with timeout (it was broken
anyway) which was not a good idea as explained by Andrew. It's possible
to avoid using additional variable but actually it inroduces using more
complex in-kernel tools. So simply add one variable for condition.
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We almost always used first HDMI block for first encoder and second for sencod.
Exception was KLDSCP_LVTMA. Analyzing code picking DIG encoder shows the same
behaviour. It shows HDMI block are related to DIGs, which relation we now use.
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
radeon was always including the atpx code unnecessarily, also core
switcheroo was including acpi headers.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Many new laptops now come with 2 gpus, one to be used for low power
modes and one for gaming/on-ac applications. These GPUs are typically
wired to the laptop panel and VGA ports via a multiplexer unit which
is controlled via ACPI methods.
4 combinations of systems typically exist - with 2 ACPI methods.
Intel/ATI - Lenovo W500/T500 - use ATPX ACPI method
ATI/ATI - some ASUS - use ATPX ACPI Method
Intel/Nvidia - - use _DSM ACPI method
Nvidia/Nvidia - - use _DSM ACPI method.
TODO:
This patch adds support for the ATPX method and initial bits
for the _DSM methods that need to written by someone with
access to the hardware.
Add a proper non-debugfs interface - need to get some proper
testing first.
v2: add power up/down support for both devices
on W500 puts i915/radeon into D3 and cuts power to radeon.
v3: redo probing methods, no DMI list, drm devices call to
register with switcheroo, it tries to find an ATPX method on
any device and once there is two devices + ATPX it inits the
switcher.
v4: ATPX msg handling using buffers - should work on more machines
v5: rearchitect after more mjg59 discussion - move ATPX handling to
radeon driver.
v6: add file headers + initial nouveau bits (to be filled out).
v7: merge delayed switcher code.
v8: avoid suspend/resume of gpu that is off
v9: rearchitect - mjg59 is always right. - move all ATPX code to
radeon, should allow simpler DSM also proper ATRM handling
v10: add ATRM support for radeon BIOS, add mutex to lock vgasr_priv
v11: fix bug in resuming Intel for 2nd time.
v12: start fixing up nvidia code blindly.
v13: blindly guess at finishing nvidia code
v14: remove radeon audio hacks - fix up intel resume more like upstream
v15: clean up printks + remove unnecessary igd/dis pointers
mount debugfs
/sys/kernel/debug/vgaswitcheroo/switch - should exist if ATPX detected
+ 2 cards.
DIS - immediate change to discrete
IGD - immediate change to IGD
DDIS - delayed change to discrete
DIGD - delayed change to IGD
ON - turn on not in use
OFF - turn off not in use
Tested on W500 (Intel/ATI) and T500 (Intel/ATI)
Signed-off-by: Dave Airlie <airlied@redhat.com>
We kept pointers to requested and current clock modes in every power state.
That was useless, more /global/ pointers in power struct are enough.
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
There is 3 different distinct states for an indirect buffer (IB) :
1- free with no fence
2- free with a fence
3- non free (fence doesn't matter)
Previous code mixed case 2 & 3 in a single one leading to possible
catastrophique failure. This patch rework the handling and properly
separate each case. So when you get ib we set the ib as non free and
fence status doesn't matter. Fence become active (ie has a meaning
for the ib code) once the ib is scheduled or free. This patch also
get rid of the alloc bitmap as it was overkill, we know go through
IB pool list like in a ring buffer as the oldest IB is the first
one the will be free.
Fix :
https://bugs.freedesktop.org/show_bug.cgi?id=26438
and likely other bugs.
V2 remove the scheduled list, it's useless now, fix free ib scanning
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Get rid of _location and use _start/_end also simplify the
computation of vram_start|end & gtt_start|end. For R1XX-R2XX
we place VRAM at the same address of PCI aperture, those GPU
shouldn't have much memory and seems to behave better when
setup that way. For R3XX and newer we place VRAM at 0. For
R6XX-R7XX AGP we place VRAM before or after AGP aperture this
might limit to limit the VRAM size but it's very unlikely.
For IGP we don't change the VRAM placement.
Tested on (compiz,quake3,suspend/resume):
PCI/PCIE:RV280,R420,RV515,RV570,RV610,RV710
AGP:RV100,RV280,R420,RV350,RV620(RPB*),RV730
IGP:RS480(RPB*),RS690,RS780(RPB*),RS880
RPB: resume previously broken
V2 correct commit message to reflect more accurately the bug
and move VRAM placement to 0 for most of the GPU to avoid
limiting VRAM.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
There is 3 different distinct states for an indirect buffer (IB) :
1- free with no fence
2- free with a fence
3- non free (fence doesn't matter)
Previous code mixed case 2 & 3 in a single one leading to possible
catastrophique failure. This patch rework the handling and properly
separate each case. So when you get ib we set the ib as non free and
fence status doesn't matter. Fence become active (ie has a meaning
for the ib code) once the ib is scheduled or free. This patch also
get rid of the alloc bitmap as it was overkill, we know go through
IB pool list like in a ring buffer as the oldest IB is the first
one the will be free.
Fix :
https://bugs.freedesktop.org/show_bug.cgi?id=26438
and likely other bugs.
V2 remove the scheduled list, it's useless now, fix free ib scanning
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
