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

The primary interfaces to NVKM are through NVIF, but there are a small
number of functions which are called directly.

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

The move and renames were kept separate for cleaner diffs.

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

Previously, the DRM driver was the entry-point for the pci/platform
bus probe() functions, and calls into NVKM to create an nvkm_device
from pci/platform devices, then continues on with DRM init.

Most of the code handling PCI/Tegra-specific functions has now been
moved to NVKM (though prior to this commit, still *called* from DRM)
and NVKM registers devices named "nvkm.device.<n>" on the auxiliary
bus for each PCI or Tegra GPU in the system.

The final step is to move pci/platform driver registration to NVKM,
and have the DRM driver register as an auxiliary driver, from where
it can gain access to the nvkm_device and proceed as it did before.

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

This wraps direct calls to pm_runtime functions to cleanup some common
usage patterns, and keep the logic in one place.

From the next commit, the pm_runtime calls will need to operate against
the DRM driver's auxiliary device instead of directly on the underlying
PCI device.  This commit will help make that switch cleaner.

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

This commit registers a device on the aux bus after core NVKM init
has finished, which will be used by a later commit to have the DRM
driver probe() against an aux device instead of a PCI or platform
device (in the case of Tegra).

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

Moves the PCI-specific portions of nouveau_pmops to NVKM, leaving the
DRM pieces where they are.  The NVKM functions are called through the
the nvkm_device_pci_driver struct from DRM for now, but will be fully
separated once the DRM driver is implemented on an auxiliary device.

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

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

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

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

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

Rename, and modify the signatures of nvkm_device_{pci,tegra}_new() to be
consistent with pci/platform driver probe() functions.

The DRM driver is still calling the functions as it did before, just via
the new pci/platform driver structs in NVKM that will eventually be used
for driver registration.

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

The DRM driver needs to do things in response to vga_switcheroo_client
callbacks, so this move isn't a direct copy+paste.

3 new callbacks are added to nvif_driver_func, which are basically the
same as the switcheroo callbacks, except NVKM handles all the PCI/ACPI
parts, and calls into the DRM driver for the rest.

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

The DRM driver uses "nouveau_is_optimus() || nouveau_is_v1_dsm()", which
have been moved to NVKM, to determine support for runtime pm.

Replace with a feature flag returned when allocating an nvif_device.

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

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

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

These previously depended on the caller calling nvkm_device_del() to
cleanup if nvkm_device_*_new() fails.

That's a little odd for starters, but only the Tegra path cleaned up,
and the PCI path would have leaked the memory, FWs etc that had been
allocated as NVKM ran each subdev's ctor().

A later patch turns these into pci/platform driver probe() functions,
so they need to learn how to clean up after themselves regardless.

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

Add stub init()/exit() functions that subsequent patches will use as
they move the pci/platform-specific code out of the DRM driver.

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

A later patch in this series will move the code that binds to a PCI/
Tegra device into NVKM, and requires a couple of callbacks to handle
switcheroo, etc.

nvif_event is generally used for NVKM->DRM callbacks, but these will
need to be registered before nvif_event is ready.

This patch replaces the event() function pointer that was passed to
nvkm_driver_ctor() during init with a nvif_driver_func struct, that
contains the existing event() callback.

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

Return an enum rather than an int from nvif_event callbacks.

Also moves the NVIF_EVENT_* definitions to driverif.h, with the
rest of the DRM<->NVKM interfaces.

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

A later patch in this series moves the nvif_event callback function,
but all the callers directly dereference object->client->event().

Add a helper function so this only has to be changed in one place.

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

A later patch in the series converts nvkm_device_{pci_tegra}_new into
PCI/platform device probe functions, as a step towards moving all the
PCI/Tegra-specific handling into NVKM.

nouveau.ko has two module options (nouveau.config/nouveau.debug) that
affect the behaviour of NVKM, however, and the probe() functions will
not have access to these, which would break a user's configuration if
they depend on any of the options to workaround a problem, etc.

To avoid this, we delay calling constructors for each subdev (which
could depend on module parameters) until allocation of the first
nvif_device, which will allow the DRM driver a chance to override
device.{cfg,dbg}opt before they're needed.

v2:
- move pci dma_set_mask() to oneinit, so mmu.dma_bits is initialised

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

The NVIF/NVKM comments were backwards.

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

- complete the switch to nouveau_drm.cli
- renames drm._client to drm.client

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

- accesses to PRI are now via nvif_device.map

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

- complete the switch to nvif_device.impl

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

- this protected the drm.master object rb, which no longer exists

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

- nvkm_object_search() no longer exists to be called from irq context

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

- no more users

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

- transition from "ioctl" interface

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

- transition from "ioctl" interface

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

- no more users

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

- transition from "ioctl" interfaces

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

- transition from "ioctl" interface

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

- transition from "ioctl" interfaces

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

- transition from "ioctl" interfaces

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

- transition from "ioctl" interface

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

- no more users

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

- no more users

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

- no more users

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