We need to wait for the GPUVM flush to complete.  There
was some confusion as to how this mechanism was supposed
to work.  The operation is not atomic.  For GPU initiated
invalidations you need to read back a VM register to
introduce enough latency for the update to complete.

v2: drop gart changes
v3: just read back rather than polling
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Use multiple VMIDs for each VM, one for each ring. That allows
us to execute flushes separately on each ring, still not ideal
cause in a lot of cases rings can share IDs.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Use ring structure instead of index and provide vm_id and pd_addr separately.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This fixes a problem with GPU resets and TLB flushes on SI/CIK.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

It isn't necessary for command streams generated by the kernel (at least
not while we aren't storing ring or indirect buffers in VRAM).
Signed-off-by: NMichel Dänzer <michel.daenzer@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Doesn't seem necessary, the GART table memory should be persistent.
Signed-off-by: NMichel Dänzer <michel.daenzer@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Query to find out how many compute units on a GPU.
Useful for OpenCL usermode drivers.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Replace occurrences of "v & 0xffffffff" with lower_32_bits(v)
when it's next to an upper_32_bits(v). Also remove unnecessary
"upper_32_bits(v) & 0xffffffff" code snippets.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This patch makes it possible to decide how many address
bits are spend on the page directory vs the page tables.

v2: remove unintended change
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This patch implements support for VRAM page table entry compression.
PTE construction is enhanced to identify physically contiguous page
ranges and mark them in the PTE fragment field. L1/L2 TLB support is
enabled for 64KB (SI/CIK) and 256KB (NI) PTE fragments, significantly
improving TLB utilization for VRAM allocations.

Linear store bandwidth is improved from 60GB/s to 125GB/s on Pitcairn.
Unigine Heaven 3.0 sees an average improvement from 24.7 to 27.7 FPS
on default settings at 1920x1200 resolution with vsync disabled.

See main comment in radeon_vm.c for a technical description.

v2 (chk): rebased and simplified.
v3 (chk): add missing hw setup
v4 (chk): rebased on current drm-fixes-3.15
v5 (chk): fix comments and commit text
Signed-off-by: NJay Cornwall <jay@jcornwall.me>
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Moving the pm resume up in the init order to fix
dpm seems to have regressed somes cases with the old
pm code.  Move it back to late resume.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

The reason for the false positives was fixed quite some time ago and since
most engines can still execute NOPs while being locked up it leads to false
negatives.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>

In all cases where it really matters we are using the read functions anyway.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>

If we are not able to properly initialize one of the gpu
engines for buffer paging, we limit vram to the size of
the cpu visible aperture.  We generally either use the gfx
or dma engine to do this.  Clean up the size limiting code
to only adjust the size based on what ring is selected
for buffer paging rather than making assumptions about which
engine is selected for paging.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>

We need to set the engine bit to select the ME and
also set the full cache bit.  Should help stability
on TN and cayman.

V2: fix up surface sync in ib execute as well
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

pci config reset is a low level reset that resets
the entire chip from the bus interface.  It can
be more reliable if soft reset fails.

v2: put behind module parameter
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Fill in asic family specific versions rather than
using the generic version.  This lets us handle asic
specific differences more easily.  In this case, we
disable sw swapping of the rtpr writeback value on
r6xx+ since the hw does it for us.  Fixes bogus
rptr readback on BE systems.

v2: remove missed cpu_to_le32(), add comments
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We need more control over the ordering of dpm init with
respect to the rest of the asic.  Specifically, the SMC
has to be initialized before the rlc and cg/pg.  The pm
code currently initializes late in the driver, but we need
it to happen much earlier so move pm handling into the asic
specific callbacks.

This makes dpm more reliable and makes clockgating work
properly on CIK parts and should help on SI parts as well.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We need to reorder the driver init sequence to better accomodate
dpm which needs to be loaded earlier in the init sequence.  Move
fw init up so that it's available for dpm init.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Fixes gfx corruption on certain TN/RL parts.

bug:
https://bugs.freedesktop.org/show_bug.cgi?id=60389Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

The DMA ring seems to be stable now.

v2: remove pt_ring_index as well
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Prevent driver load problems if the smc is missing.

bug:
https://bugzilla.kernel.org/show_bug.cgi?id=63011Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Tested-by: NMikko Rapeli <mikko.rapeli@iki.fi>
Cc: stable@vger.kernel.org

The vram scratch buffer needs to be initialized
before the mc is programmed otherwise we program
0 as the GPU address of the default GPU fault
page.  In most cases we put vram at zero anyway and
reserve a page for the legacy vga buffer so in practice
this shouldn't cause any problems, but better to make
it correct.

Was changed in:
6fab3febReported-by: NFrankR Huang <FrankR.Huang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Similar to DCE4/5, but supports multiple audio pins
which can be assigned per afmt block.

v2: rework the driver to handle more than one audio
pin.
v3: try different dto reg
v4: properly program dto
v5 (ck): change dto programming order
v6: program speaker allocation block
v7: rebase
v8: rebase on Rafał's changes
v9: integrated Rafał's comments, update to latest
    drm_edid_to_speaker_allocation API
v10: add missing line break in error message
v11: add back audio enabled messages
v12: fix copy paste typo in r600_audio_enable
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NRafał Miłecki <zajec5@gmail.com>

Similar to separating the UVD code, just put the DMA
functions into separate files.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Our different hardware blocks are actually completely
separated, so it doesn't make much sense any more to
structure the code by pure chipset generations.

Start restructuring the code by separating our the UVD block.

v2: updated commit message
v3: rebased and restructurized start/stop functions for kv dpm.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Now that we have callbacks for [rw]ptr handling we can
remove the special handling for the DMA rings and use
the callbacks instead.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

The hardware just doesn't support this correctly.
Disable it before we accidentally write anywhere we shouldn't.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

When we PG (powergate) UVD, we need to re-initialize it
before we can use it again.

v2: rebase on UVD stop fixes
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

and remove duplicate si_rlc functions.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

No longer used now that we use the async dma engines or
CP DMA for bo copies.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

The smc ucode is required for dpm (dynamic power
management), but if it's missing just skip dpm setup
and don't disable acceleration.

Should fix:
https://bugs.freedesktop.org/show_bug.cgi?id=67876Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

For r6xx+ asics.  This mirrors the behavior of pre-r6xx
asics.  We need to program the MC even if something
else in startup() fails.  Failure to do so results in
an unusable GPU.

Based on a fix from: Mark Kettenis <kettenis@openbsd.org>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Removing the clock/power or resetting the VCPU can cause
hangs if that happens in the middle of a register write.

Stall the memory and register bus before putting the VCPU
into reset. Keep it in reset when unloading the module or
suspending.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Last driver and pretty obviously a major user of this little function.
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NDave Airlie <airlied@gmail.com>

Helpful for debugging GPUVM errors as we can see what
hw block and page generated the fault in the log.

v2: simplify fault decoding
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>

Avoid creating temporary platform device that will lead to issue
when several radeon gpu are in same computer. Instead directly use
the radeon device for requesting firmware.
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NJerome Glisse <jglisse@redhat.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

TN has some additional powergating features beyond what is
supported on ON/LN.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Enables PCIE ASPM (Active State Power Management) on
evergreen-cayman asics.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>