The core/mem/shader clocks don't support the fractional feedback divider,
causing our calculated clocks to be off by quite a lot in some cases.  To
solve this we will switch to a search-based algorithm when fN is NULL.

For my NVA8 at PL3, this actually generates identical cooefficients to
the binary driver.  Hopefully that's a good sign, and that does not
break VPLL calculation for someone..
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Hopefully this is how we're supposed to correctly handle when the RAMCFG
strap is above the number of entries in timing-related tables.

It's rather difficult to confirm without finding a configuration where
the ram restrict table doesn't map 8-15 back onto 0-7 anyway.  There's
not a single vbios in the repo which is configured differently..

In any case, this is probably still better than potentially reading
outside of the bounds of various tables..
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

We need to parse some of these other entries still, but I've yet to
determine exactly which PLLs the rest map to.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Really not necessary here, we want to be able to see if/how we managed to
match a timingset to a performance level, even if we can't currently
program it.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

v2 (Ben Skeggs): fix ramcfg strap, and remove bogus handling of perf 0x40
Signed-off-by: Martin Peres <martin.peres@ensi-bourges.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

I still discourage anyone from actually doing this yet.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Again, doesn't quite match NVIDIA's, but not sure it really matters.  This
will however, match the same rules we use to calculate the other related
grctx magics.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

After looking at a number of different logs, it appears 0x41 likely
indicates the presense of an LVDS panel following the SPWG spec
(http://www.spwg.org/)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Looks like a false positive to me, but, anyways!
Reported-by: Jimmy Rentz <jb17bsome@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

NV40 and older cards (pre NV50) reserve a vram bo for the vga memory at
card init. This bo is then freed at card shutdown.  The problem is that
the ttm bo vram manager was already freed. So a crash occurs when the
vga bo is freed. The fix is to free the vga bo prior to freeing the ttm
bo vram manager. There might be other solutions but this seemed the
simplest to me.
Signed-off-by: Jimmy Rentz <jb17bsome@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Add a missing nv04 instmem ramin heap shutdown call.
Signed-off-by: Jimmy Rentz <jb17bsome@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Definitely not 100% correct, but, for the configurations I've seen used
it'll read back the correct clocks now.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Many of the nv50 cards have their shader and/or memory pll
disabled at some stage.
This patch addresses those cases, so that the function
returns the correct frequency.

When the shader pll is disabled, the blob reports 2*core clock
Whereas for memory, the data stored in the vbios. This action
is incorrect as some vbioses store a clock value that is less
than the refference clock of the pll.

Thus we are reporting the reff_clk as it is the frequency the
pll actually operates

v2 - Convert NV_INFO() messages to NV_DEBUG()
Provide more information in the actuall message
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Martin Peres <martin.peres@ensi-bourges.fr>
Reported-by: Stratos Psomadakis <psomas@ece.ntua.gr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Contents will now be preserved across a suspend, unlike a pinned bo
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

PFIFO kickoff should have handled this for us.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Allows per-chipset firmware to be installed, and keeps a copy in memory
for suspend/resume purposes.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

In preparation for adding 0x50 support.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

The nouveau_wait_for_idle() call should hopefully not have been actually
necessary, we *do* wait for the channel to go idle already.  If it's
an issue somehow, the chipset-specific hooks can wait for idle themselves
before taking the lock.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

In the very least VPE (PMPEG and friends) also has this style of tile
region regs, lets make them just work if/when they get added.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Like nv10-nv50, needs cleanup.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Like nv20-nv50, needs cleanup.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

A bit of cleanup done along the way, but, like nv40/nv50, needs more.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Like nv50, this needs a good cleanup.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Much nicer to do that nv50, the code was pretty much written to expect
such a change in the future.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>