This reverts commit 1addc12648
This commit seems to cause crashes in gk104_fifo_intr_runlist() by
returning 0xbad0da00 when register 0x2a00 is read. Since this commit was
intended for GM20B which is not completely supported yet, let's revert
it for the time being.
Reported-by: Eric Biggers <ebiggers3@gmail.com>
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Tested-by: Afzal Mohammed <afzal.mohd.ma@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Add Maxwell to the switch statement that sets node->memtype, otherwise
all tiling information is ignored for buffers in system memory.
While we are at it, make that switch statement explicitly complain the
next time we meet a non-handled card family.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Userspace has started doing this, which upsets the display class hw
error checking in various unpleasant ways.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Only 58 words get written to the ring, not 59. Also, normalize the accel
init wrt nvc0 and nv04 fbcon impls by firing the ring at accel init time
rather than waiting until "later".
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
The RING_SPACE macro accounts how much space is used up so it's
important to ask it for the right amount. Incorrect accounting of this
can cause page faults down the line as writes are attempted outside of
the ring.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: stable@vger.kernel.org
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
More analysis shows that this is identical to 0x79 except that it loads
the frequency indirectly from elsewhere in the VBIOS.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91025
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Opcode 0x5a is a register write for data looked up from another part of
the VBIOS image. 0x59 is a more complex opcode, but we may as well
recognize it. These occur on a single known instance of Riva TNT2
hardware.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91025
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
The return type of exec_lookup() is struct nvkm_output *, so it should
return NULL rather than 0.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
There is always the possiblity that the ppm->context pointer would get
partially updated and accidentally would equal ctx. This would allow two
contexts to co-exist, which is not acceptable. Moving the test to the
critical section takes care of this problem.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Martin Peres <martin.peres@free.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This fixes a crash when multiple PM engine contexts are created.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Martin Peres <martin.peres@free.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
After submitting a GO_IDLE bundle, one must wait for GR to effectively
be idle before submitting the next bundle. Failure to do so may result
in undefined behavior in some rare cases.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Reported-by: Kary Jin <karyj@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
When emitting the ICMD bundle, wait on the bottom half (bit 3 of the
GR_STATUS register) instead of upper half (bit 2) to make sure methods
are effectively emitted.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Kicking channels is part of their deactivation process. Maxwell chips
are particularly sensitive to this, and can start fetching the previous
pushbuffer of a recycled channel if this is not done.
While we are at it, improve the channel preemption code to only wait for
bit 20 of 0x002634 to turn to 0, as it is the bit indicating a
preempt is pending.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Increase clock timeout for SYS, FPB and GPC in order to avoid operation
failure at high gpcclk rate.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
The lack of IOMMU API support can make nouveau_platform_probe_iommu()
fail to compile because struct iommu_ops is then empty. Fix this by
skipping IOMMU probe in that case - lack of IOMMU on platform devices
is sub-optimal, but is not an error.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
The memory allocated for a nouveau_cli object in nouveau_cli_create() is
never freed. Free the memory in nouveau_cli_destroy() to plug this leak.
kmemleak recorded this after running a couple of nouveau test programs.
Note that kmemleak points at drm_open_helper() because for some reason
it thinks that skipping the first two stack frames is a good idea.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Use kvfree() instead of open-coding it.
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Cc: David Airlie <airlied@linux.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
On the MacBook Pro, power of the gpu is cut by a gmux chip. Sometimes
the gpu gets stuck in powersaving mode and refuses to wake up
("Refused to change power state, currently in D3"). Inserting a
delay between setting the gpu to D3hot and cutting the power seems
to help (most of the time). This issue and its (partial) remediation
by the patch was observed with an Nvidia GT650M (NVE7 / GK107).
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
One more drm-misch pull for 4.1 with mostly simple stuff and boring
refactoring. Even the cursor fix from Matt is just to make a really anal
igt happy.
* tag 'topic/drm-misc-2015-04-15' of git://anongit.freedesktop.org/drm-intel:
drm: fix trivial typo mistake
drm: Make integer overflow checking cover universal cursor updates (v2)
drm: make crtc/encoder/connector/plane helper_private a const pointer
drm/armada: constify struct drm_encoder_helper_funcs pointer
drm/radeon: constify more struct drm_*_helper funcs pointers
drm/edid: add #defines for ELD versions
drm/atomic: Add for_each_{connector,crtc,plane}_in_state helper macros
drm: Use kref_put_mutex in drm_gem_object_unreference_unlocked
drm/drm: constify all struct drm_*_helper funcs pointers
drm/qxl: constify all struct drm_*_helper funcs pointers
drm/nouveau: constify all struct drm_*_helper funcs pointers
drm/radeon: constify all struct drm_*_helper funcs pointers
drm/gma500: constify all struct drm_*_helper funcs pointers
drm/mgag200: constify all struct drm_*_helper funcs pointers
drm/exynos: constify all struct drm_*_helper funcs pointers
drm: Fix some typos
nvbios_extend() returns 1 to indicate "extended the array" and 0 to
indicate the array is already big enough. This is used by the core
shadowing code to prevent re-fetching chunks of the image that have
already been shadowed.
The ACPI fetching code may possibly need to extend this further due
to requiring fetches to happen in 4KiB chunks.
Under certain circumstances (that happen if the total image size is
a multiple of 4KiB), the memory allocated to store the shadow will
already be big enough, causing the ACPI code's nvbios_extend() call
to return 0, which is misinterpreted as a failure.
The fix is simple, accept >= 0 as a successful condition here. The
core will have already made sure that we're not re-fetching data we
already have.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89047
v2 (Ben Skeggs):
- dropped hunk which would cause unnecessary re-fetching
- more descriptive explanation
Signed-off-by: Jan Vesely <jano.vesely@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Uncertain whether the GPC pack change is due to a newer driver version,
or a legitimate difference from GM204. My GM204 has broken vram, so
can't currently try a newer binary driver on it to confirm.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Under certain circumstances the trapped address will contain subc 7,
which GK104 GR doesn't have anymore.
Notice this case to avoid causing additional priv ring faults.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
No idea if "3" is a constant or derived from something else, but the
value is unchanged in the limited traces of gm107/gm204 I have here.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Make static a few functions and structures that should be.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
A "return 0" found its way in the middle of the error path of
nouveau_platform_probe(), remove it as it will make the kernel crash if
we try to unload the module afterwards.
While we are at it, also remove the IOMMU domain if it has been created,
as we should.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
nvkm_mm_fini() was not called when exiting the driver, resulting in a
memory leak. Fix this.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
On some of these chipsets, reading NV_PGRAPH_GPC_GPM_PD_PES_TPC_ID_MASK
can trigger a PRI fault and return an error code instead of a TPC mask,
unless PGOB has been disabled first.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Before we moved gk110's implementation of this to pmu, the functions were
identical. This commit just switches GK208 to use the new (more complete)
implementation of the power-up sequence.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Turns out the PTHERM part of this dance is bracketed by the same PMU
fiddling that occurs on GK104/6, let's assume it's also PGOB.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
