platform/kernel/linux-starfive.git
4 years agodrm/nouveau: signal pending fences when channel has been killed
Ben Skeggs [Thu, 23 Jan 2020 05:39:27 +0000 (15:39 +1000)]
drm/nouveau: signal pending fences when channel has been killed

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau: reject attempts to submit to dead channels
Ben Skeggs [Thu, 23 Jan 2020 03:47:12 +0000 (13:47 +1000)]
drm/nouveau: reject attempts to submit to dead channels

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau: zero vma pointer even if we only unreference it rather than free
Ben Skeggs [Thu, 23 Jan 2020 03:55:19 +0000 (13:55 +1000)]
drm/nouveau: zero vma pointer even if we only unreference it rather than free

I'm not sure this affects anything, but best be safe.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau: Add HD-audio component notifier support
Takashi Iwai [Mon, 13 Jan 2020 14:17:21 +0000 (15:17 +0100)]
drm/nouveau: Add HD-audio component notifier support

This patch adds the support for the notification of HD-audio hotplug
via the already existing drm_audio_component framework.  This allows
us more reliable hotplug notification and ELD transfer without
accessing HD-audio bus; it's more efficient, and more importantly, it
works without waking up the runtime PM.

The implementation is rather simplistic: nouveau driver provides the
get_eld ops for HD-audio, and it notifies the audio hotplug via
pin_eld_notify callback upon each nv50_audio_enable() and _disable()
call.  As the HD-audio pin assignment seems corresponding to the CRTC,
the crtc->index number is passed directly as the zero-based port
number.

The bind and unbind callbacks handle the device-link so that it
assures the PM call order.

Link: https://lore.kernel.org/r/20190722143815.7339-3-tiwai@suse.de
Reviewed-by: Lyude Paul <lyude@redhat.com>
Cc: Jaroslav Kysela <perex@perex.cz>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau: fix build error without CONFIG_IOMMU_API
Chen Zhou [Thu, 16 Jan 2020 12:50:10 +0000 (20:50 +0800)]
drm/nouveau: fix build error without CONFIG_IOMMU_API

If CONFIG_IOMMU_API is n, build fails:

vers/gpu/drm/nouveau/nvkm/subdev/ltc/gp10b.c:37:9: error: implicit declaration of function dev_iommu_fwspec_get; did you mean iommu_fwspec_free? [-Werror=implicit-function-declaration]
  spec = dev_iommu_fwspec_get(device->dev);
         ^~~~~~~~~~~~~~~~~~~~
         iommu_fwspec_free
drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gp10b.c:37:7: warning: assignment makes pointer from integer without a cast [-Wint-conversion]
  spec = dev_iommu_fwspec_get(device->dev);
       ^
drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gp10b.c:39:17: error: struct iommu_fwspec has no member named ids
   u32 sid = spec->ids[0] & 0xffff;

Seletc IOMMU_API under config DRM_NOUVEAU to fix this.

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Chen Zhou <chenzhou10@huawei.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/kms/nv04: remove set but not used variable 'width'
YueHaibing [Fri, 17 Jan 2020 02:01:02 +0000 (10:01 +0800)]
drm/nouveau/kms/nv04: remove set but not used variable 'width'

drivers/gpu/drm/nouveau/dispnv04/arb.c: In function nv04_calc_arb:
drivers/gpu/drm/nouveau/dispnv04/arb.c:56:21: warning:
 variable width set but not used [-Wunused-but-set-variable]

'width' is never used, so remove it.

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/kms/nv50: remove set but not unused variable 'nv_connector'
YueHaibing [Fri, 17 Jan 2020 03:36:42 +0000 (11:36 +0800)]
drm/nouveau/kms/nv50: remove set but not unused variable 'nv_connector'

drivers/gpu/drm/nouveau/dispnv50/disp.c: In function nv50_pior_enable:
drivers/gpu/drm/nouveau/dispnv50/disp.c:1672:28: warning:
 variable nv_connector set but not used [-Wunused-but-set-variable]

commit ac2d9275f371 ("drm/nouveau/kms/nv50-: Store the
bpc we're using in nv50_head_atom") left behind this.

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/mmu: fix comptag memory leak
Ben Skeggs [Wed, 22 Jan 2020 22:23:06 +0000 (08:23 +1000)]
drm/nouveau/mmu: fix comptag memory leak

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gp10b: Use gp100_grctx and gp100_gr_zbc
Thierry Reding [Wed, 15 Jan 2020 14:07:56 +0000 (15:07 +0100)]
drm/nouveau/gr/gp10b: Use gp100_grctx and gp100_gr_zbc

gp10b doesn't have all the registers that gp102_gr_zbc wants to access,
which causes IBUS MMIO faults to occur. Avoid this by using the gp100
variants of grctx and gr_zbc.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/pmu/gm20b,gp10b: Fix Falcon bootstrapping
Thierry Reding [Wed, 15 Jan 2020 14:06:13 +0000 (15:06 +0100)]
drm/nouveau/pmu/gm20b,gp10b: Fix Falcon bootstrapping

The low-level Falcon bootstrapping callbacks are expected to return 0 on
success or a negative error code on failure. However, the implementation
on Tegra returns the ID or mask of the Falcons that were bootstrapped on
success, thus breaking the calling code, which treats this as failure.

Fix this by making sure we only return 0 or a negative error code, just
like the code for discrete GPUs does.

Fixes: 86ce2a71539c ("drm/nouveau/flcn/cmdq: move command generation to subdevs")
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/tu10x: initial support
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/gr/tu10x: initial support

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/acr/tu10x: initial support
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/acr/tu10x: initial support

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/core: remove previous versioned fw loader
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/core: remove previous versioned fw loader

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/secboot: remove
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/secboot: remove

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/acr: implement new subdev to replace "secure boot"
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/acr: implement new subdev to replace "secure boot"

ACR is responsible for managing the firmware for LS (Low Secure) falcons,
this was previously handled in the driver by SECBOOT.

This rewrite started from some test code that attempted to replicate the
procedure RM uses in order to debug early Turing ACR firmwares that were
provided by NVIDIA for development.

Compared with SECBOOT, the code is structured into more individual steps,
with the aim of making the process easier to follow/debug, whilst making
it possible to support newer firmware versions that may have a different
binary format or API interface.

The HS (High Secure) binary(s) are now booted earlier in device init, to
match the behaviour of RM, whereas SECBOOT would delay this until we try
to boot the first LS falcon.

There's also additional debugging features available, with the intention
of making it easier to solve issues during FW/HW bring-up in the future.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/fb/gp102-: unlock VPR as part of FB init
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/fb/gp102-: unlock VPR as part of FB init

We perform memory allocations long before we hit the code in SECBOOT that
would unlock the VPR, which could potentially result in memory allocation
within the locked region.

Run the scrubber binary right after VRAM init to ensure we don't.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/core/memory: add macros to read/write blocks from objects
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/core/memory: add macros to read/write blocks from objects

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/secboot: move code to boot LS falcons to subdevs
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/secboot: move code to boot LS falcons to subdevs

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/msgq: rename msgq-related nvkm_msgqueue_queue to nvkm_falcon_msgq
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/msgq: rename msgq-related nvkm_msgqueue_queue to nvkm_falcon_msgq

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/msgq: pass explicit message queue pointer to recv()
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/msgq: pass explicit message queue pointer to recv()

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/msgq: move handling of init message to subdevs
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/msgq: move handling of init message to subdevs

When the PMU/SEC2 LS FWs have booted, they'll send a message to the host
with various information, including the configuration of message/command
queues that are available.

Move the handling for this to the relevant subdevs.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/msgq: drop nvkm_msgqueue argument to functions
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/msgq: drop nvkm_msgqueue argument to functions

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/msgq: switch to falcon queue printk macros
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/msgq: switch to falcon queue printk macros

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/msgq: simplify msg_queue_pop() error handling
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/msgq: simplify msg_queue_pop() error handling

We always want at least requested size, make anything less a more direct
error condition.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/msgq: remove error handling for msg_queue_open(), it can't fail
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/msgq: remove error handling for msg_queue_open(), it can't fail

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/cmdq: move command generation to subdevs
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/cmdq: move command generation to subdevs

This moves the code to generate commands for the ACR unit of the PMU/SEC2 LS
firmwares to those subdevs.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/cmdq: rename cmdq-related nvkm_msqqueue_queue to nvkm_falcon_cmdq
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/cmdq: rename cmdq-related nvkm_msqqueue_queue to nvkm_falcon_cmdq

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/cmdq: implement a more explicit send() interface
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/cmdq: implement a more explicit send() interface

Takes the command queue pointer directly instead of requiring a function to
lookup based on an queue type, as well as an explicit timeout value.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/cmdq: drop nvkm_msgqueue argument to functions
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/cmdq: drop nvkm_msgqueue argument to functions

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/cmdq: switch to falcon queue printk macros
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/cmdq: switch to falcon queue printk macros

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/cmdq: cmd_queue_close always commits, simplify it
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/cmdq: cmd_queue_close always commits, simplify it

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/cmdq: cmd_queue_push can't fail, remove error handling for it
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/cmdq: cmd_queue_push can't fail, remove error handling for it

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/cmdq: split the condition for queue readiness vs pmu acr readiness
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/cmdq: split the condition for queue readiness vs pmu acr readiness

This is to allow for proper separation of the LS interface code from the
queue handling code.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/qmgr: rename remaining nvkm_msgqueue bits to nvkm_falcon_qmgr
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/qmgr: rename remaining nvkm_msgqueue bits to nvkm_falcon_qmgr

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/qmgr: support syncronous command submission from common code
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/qmgr: support syncronous command submission from common code

Functions implementing FW commands had to implement this themselves, let's
move that to common code and plumb the return code from callbacks through.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/qmgr: allow arbtrary priv + return code for callbacks
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/qmgr: allow arbtrary priv + return code for callbacks

Code to interface with LS firmwares is being moved to the subdevs where it
belongs, rather than living in the common falcon code.

Arbitrary private data passed to callbacks is to allow for something other
than struct nvkm_msgqueue to be passed into the callback (like the pointer
to the subdev itself, for example), and the return code will be used where
we'd like to detect failure from synchronous messages.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/qmgr: move sequence tracking from nvkm_msgqueue to nvkm_falcon_qmgr
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/qmgr: move sequence tracking from nvkm_msgqueue to nvkm_falcon_qmgr

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/msgq: explicitly create message queue from subdevs
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/msgq: explicitly create message queue from subdevs

Code to interface with LS firmwares is being moved to the subdevs where it
belongs, rather than living in the common falcon code.

This is an incremental step towards that goal.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/cmdq: explicitly create command queue(s) from subdevs
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/cmdq: explicitly create command queue(s) from subdevs

Code to interface with LS firmwares is being moved to the subdevs where it
belongs, rather than living in the common falcon code.

This is an incremental step towards that goal.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn/qmgr: explicitly create queue manager from subdevs
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn/qmgr: explicitly create queue manager from subdevs

Code to interface with LS firmwares is being moved to the subdevs where it
belongs, rather than living in the common falcon code.

This is an incremental step towards that goal.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: split msgqueue into multiple pieces
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn: split msgqueue into multiple pieces

To make things clearer while modifying the interfaces, split msgqueue into
Queue Manager, Command Queue, and Message Queue.

There should be no code changes here, these will be done incrementally.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: add printk macros
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn: add printk macros

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: reset sec2/gsp falcons harder
Ben Skeggs [Tue, 14 Jan 2020 20:34:22 +0000 (06:34 +1000)]
drm/nouveau/flcn: reset sec2/gsp falcons harder

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: specify queue register offsets from subdev
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/flcn: specify queue register offsets from subdev

Also fixes the values for Turing, even though we don't use it yet.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: specify debug/production register offset from subdev
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/flcn: specify debug/production register offset from subdev

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: specify EMEM address from subdev
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/flcn: specify EMEM address from subdev

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: move bind_context WAR out of common code
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/flcn: move bind_context WAR out of common code

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: specify FBIF offset from subdev
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/flcn: specify FBIF offset from subdev

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/nvenc: add a stub implementation for the GPUs where it should be supported
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/nvenc: add a stub implementation for the GPUs where it should be supported

Mostly so we don't lose info hidden in falcon.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/nvdec/gm107-: add missing engine instances
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/nvdec/gm107-: add missing engine instances

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/nvdec/gm107: rename from gp102 implementation
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/nvdec/gm107: rename from gp102 implementation

NVDEC is available from GM107, and we currently only have a stub
implementation anyway, let's make it explicit.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/nvdec: initialise SW state for falcon from constructor
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/nvdec: initialise SW state for falcon from constructor

This will allow us to register the falcon with ACR, and further customise
its behaviour by providing the nvkm_falcon_func structure directly.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/nvdec: select implementation based on available fw
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/nvdec: select implementation based on available fw

This will allow for further customisation of the subdev depending on what
firmware is available.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/sec2: move interrupt handler to hw-specific module
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/sec2: move interrupt handler to hw-specific module

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/sec2: use falcon funcs
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/sec2: use falcon funcs

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/sec2: initialise SW state for falcon from constructor
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/sec2: initialise SW state for falcon from constructor

This will allow us to register the falcon with ACR, and further customise
its behaviour by providing the nvkm_falcon_func structure directly.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/sec2: select implementation based on available firmware
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/sec2: select implementation based on available firmware

This will allow for further customisation of the subdev depending on what
firmware is available.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/sec2/gp108: split from gp102 implementation
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/sec2/gp108: split from gp102 implementation

ACR LS FW loading is moving out of SECBOOT and into their specific subdevs,
and the available GP108/GV100 FWs differ from the other GP10x boards.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gf100-: initialise SW state for falcon from constructor
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/gr/gf100-: initialise SW state for falcon from constructor

This will allow us to register the falcon with ACR, and further customise
its behaviour by providing the nvkm_falcon_func structure directly.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gf100-: select implementation based on available FW
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/gr/gf100-: select implementation based on available FW

This will allow for further customisation of the subdev depending on what
firmware is available.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gp108: split from gp107
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/gr/gp108: split from gp107

ACR LS FW loading is moving out of SECBOOT and into their specific subdevs,
and the available GP107/GP108 FWs have interface differences.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gf100-: move fecs/gpccs ucode into their substructures
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/gr/gf100-: move fecs/gpccs ucode into their substructures

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gf100-: drop fuc_ prefix on sw init
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/gr/gf100-: drop fuc_ prefix on sw init

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gk20a,gm200-: use nvkm_firmware_load_blob for sw init
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/gr/gk20a,gm200-: use nvkm_firmware_load_blob for sw init

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gf100-: use nvkm_blob structure for fecs/gpccs fw
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/gr/gf100-: use nvkm_blob structure for fecs/gpccs fw

It serves the exact same purpose.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/pmu: initialise SW state for falcon from constructor
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/pmu: initialise SW state for falcon from constructor

This will allow us to register the falcon with ACR, and further customise
its behaviour by providing the nvkm_falcon_func structure directly.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/pmu: select implementation based on available firmware
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/pmu: select implementation based on available firmware

This will allow for further customisation of the subdev depending on what
firmware is available.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/pmu/gp10b: split from gm20b implementation
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/pmu/gp10b: split from gm20b implementation

ACR LS FW loading is moving out of SECBOOT and into their specific subdevs,
and the available GM20B/GP10B FWs have interface differences.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gsp: initialise SW state for falcon from constructor
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/gsp: initialise SW state for falcon from constructor

This will allow us to register the falcon with ACR, and further customise
its behaviour by providing the nvkm_falcon_func structure directly.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gsp: select implementation based on available firmware
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/gsp: select implementation based on available firmware

This will allow for further customisation of the subdev depending on what
firmware is available.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/acr: add loaders for currently available LS firmware images
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/acr: add loaders for currently available LS firmware images

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/acr: add stub implementation for all GPUs currently supported by SECBOOT
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/acr: add stub implementation for all GPUs currently supported by SECBOOT

PMU, SEC2 and GR will be modified to register their falcons with ACR before
the main commit switching everything over.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/core: define ACR subdev
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/core: define ACR subdev

This will replace the current SECBOOT subdev for handling firmware on
secure falcons.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/core: add representation of generic binary objects
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/core: add representation of generic binary objects

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/core: add a macro to better handle multiple firmware versions
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/core: add a macro to better handle multiple firmware versions

Will be used in upcoming commits to allow subdevs to better customise
themselves based on which (if any) firmware is available.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/core: output fw size in debug messages
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/core: output fw size in debug messages

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: export existing funcs
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/flcn: export existing funcs

These will be used in upcoming commits which will provide more customisation.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: fetch PRI address from TOP if not provided by constructor
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/flcn: fetch PRI address from TOP if not provided by constructor

Shortcut to avoid each subdev having to do this itself.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/flcn: move fetching of configuration until first use
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/flcn: move fetching of configuration until first use

We want to be able to register falcons with ACR during the constructor for
the subdev it belongs to, however, we may not have access to the falcon's
registers prior to DEVINIT.

Delay touching registers until the first time the falcon is acquired.

This may temporarily break secboot on non-production boards due to not
being able to determine whether the falcon is in debug or production mode,
the new ACR subdev will not have this issue, and it's not a use-case that's
terribly important for bisectability.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/fault/gv100-: fix memory leak on module unload
Ben Skeggs [Tue, 14 Jan 2020 20:34:21 +0000 (06:34 +1000)]
drm/nouveau/fault/gv100-: fix memory leak on module unload

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/disp/dp: fix typo when determining failsafe link configuration
Ben Skeggs [Tue, 14 Jan 2020 01:31:01 +0000 (11:31 +1000)]
drm/nouveau/disp/dp: fix typo when determining failsafe link configuration

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/drm/ttm: Remove set but not used variable 'mem'
YueHaibing [Fri, 10 Jan 2020 07:28:37 +0000 (15:28 +0800)]
drm/nouveau/drm/ttm: Remove set but not used variable 'mem'

drivers/gpu/drm/nouveau/nouveau_ttm.c: In function nouveau_vram_manager_new:
drivers/gpu/drm/nouveau/nouveau_ttm.c:66:22: warning: variable mem set but not used [-Wunused-but-set-variable]
drivers/gpu/drm/nouveau/nouveau_ttm.c: In function nouveau_gart_manager_new:
drivers/gpu/drm/nouveau/nouveau_ttm.c:106:22: warning: variable mem set but not used [-Wunused-but-set-variable]

They are not used any more, so remove it.

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau: Fix copy-paste error in nouveau_fence_wait_uevent_handler
YueHaibing [Fri, 10 Jan 2020 06:32:01 +0000 (14:32 +0800)]
drm/nouveau: Fix copy-paste error in nouveau_fence_wait_uevent_handler

Like other cases, it should use rcu protected 'chan' rather
than 'fence->channel' in nouveau_fence_wait_uevent_handler.

Fixes: 0ec5f02f0e2c ("drm/nouveau: prevent stale fence->channel pointers, and protect with rcu")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gv100-: modify gr init to match newer version of RM
Ben Skeggs [Thu, 1 Aug 2019 01:59:12 +0000 (11:59 +1000)]
drm/nouveau/gr/gv100-: modify gr init to match newer version of RM

Will be used as a basis for implementing changes needed for Turing.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gk20a,gm200-: add terminators to method lists read from fw
Ben Skeggs [Thu, 9 Jan 2020 01:46:15 +0000 (11:46 +1000)]
drm/nouveau/gr/gk20a,gm200-: add terminators to method lists read from fw

Method init is typically ordered by class in the FW image as ThreeD,
TwoD, Compute.

Due to a bug in parsing the FW into our internal format, we've been
accidentally sending Twod + Compute methods to the ThreeD class, as
well as Compute methods to the TwoD class - oops.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/gr/gf100-: remove dtor
Ben Skeggs [Thu, 31 Oct 2019 05:49:48 +0000 (15:49 +1000)]
drm/nouveau/gr/gf100-: remove dtor

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/fault/tu102: define nvkm_fault_func.pin
Ben Skeggs [Thu, 9 Jan 2020 05:46:29 +0000 (15:46 +1000)]
drm/nouveau/fault/tu102: define nvkm_fault_func.pin

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/core: fix missing newline in fw loader error message
Ben Skeggs [Tue, 7 Jan 2020 06:19:05 +0000 (16:19 +1000)]
drm/nouveau/core: fix missing newline in fw loader error message

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/secboot/gm20b: initialize pointer in gm20b_secboot_new()
Dan Carpenter [Wed, 8 Jan 2020 05:46:01 +0000 (08:46 +0300)]
drm/nouveau/secboot/gm20b: initialize pointer in gm20b_secboot_new()

We accidentally set "psb" which is a no-op instead of "*psb" so it
generates a static checker warning.  We should probably set it before
the first error return so that it's always initialized.

Fixes: 923f1bd27bf1 ("drm/nouveau/secboot/gm20b: add secure boot support")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/mmu: Add correct turing page kinds
James Jones [Tue, 17 Dec 2019 00:58:05 +0000 (16:58 -0800)]
drm/nouveau/mmu: Add correct turing page kinds

Turing introduced a new simplified page kind
scheme, reducing the number of possible page
kinds from 256 to 16.  It also is the first
NVIDIA GPU in which the highest possible page
kind value is not reserved as an "invalid" page
kind.

To address this, the invalid page kind is made
an explicit property of the MMU HAL, and a new
table of page kinds is added to the tu102 MMU
HAL.

One hardware change not addressed here is that
0x00 is technically no longer a supported page
kind, and pitch surfaces are instead intended to
share the block-linear generic page kind 0x06.
However, because that will be a rather invasive
change to nouveau and 0x00 still works fine in
practice on Turing hardware, addressing this new
behavior is deferred.

Signed-off-by: James Jones <jajones@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau: Fix ttm move init with multiple GPUs
James Jones [Tue, 17 Dec 2019 00:56:12 +0000 (16:56 -0800)]
drm/nouveau: Fix ttm move init with multiple GPUs

The pointer used to walk the table of move ops
and pick the right one for the current GPU was
declared static, meaning its state was carried
over between invocations of the function, and also
made the function non-rentrant and thread-unsafe.
Since the table is ordered such that newer GPU
methods are listed first, the result of this was
that initializing newer GPUs after older GPUs
would result in no suitable ttm move acceleration
operations being found, and ttm would fall back
to CPU blits on the older GPUs.

This change declares the walking pointer
separately from the table and makes it non-static
to fix the logic.

Signed-off-by: James Jones <jajones@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau: use NULL for pointer assignment.
Wambui Karuga [Tue, 31 Dec 2019 20:57:34 +0000 (23:57 +0300)]
drm/nouveau: use NULL for pointer assignment.

Replace the use of 0 in the pointer assignment with NULL to address the
following sparse warning:
drivers/gpu/drm/nouveau/nouveau_hwmon.c:744:29: warning: Using plain integer as NULL pointer

Signed-off-by: Wambui Karuga <wambui.karugax@gmail.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/kms/nv04: remove set but unused variable.
Wambui Karuga [Tue, 31 Dec 2019 20:56:07 +0000 (23:56 +0300)]
drm/nouveau/kms/nv04: remove set but unused variable.

The local variable `pclks` is defined and set but not used and can
therefore be removed.
Issue found by coccinelle.

Signed-off-by: Wambui Karuga <wambui.karugax@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/fb/gf100-: declare constants as unsigned long long.
Wambui Karuga [Thu, 2 Jan 2020 12:25:48 +0000 (15:25 +0300)]
drm/nouveau/fb/gf100-: declare constants as unsigned long long.

Explicitly declare constants as unsigned long long to address the
following sparse warnings:
warning: constant is so big it is long

v2: convert to unsigned long long for compatibility with 32-bit
architectures.

Signed-off-by: Wambui Karuga <wambui.karugax@gmail.com>
Suggested by: lia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/kms/nv04-nv4x: Use match_string() helper to simplify the code
YueHaibing [Mon, 30 Dec 2019 02:46:28 +0000 (10:46 +0800)]
drm/nouveau/kms/nv04-nv4x: Use match_string() helper to simplify the code

match_string() returns the array index of a matching string.
Use it instead of the open-coded implementation.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/ce/gp10b: Use correct copy engine
Thierry Reding [Mon, 9 Dec 2019 12:00:05 +0000 (13:00 +0100)]
drm/nouveau/ce/gp10b: Use correct copy engine

gp10b uses the new engine enumeration mechanism introduced in the Pascal
architecture. As a result, the copy engine, which used to be at index 2
for prior Tegra GPU instantiations, has now moved to index 0. Fix up the
index and also use the gp100 variant of the copy engine class because on
gp10b the PASCAL_DMA_COPY_B class is not supported.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/ltc/gp10b: Add custom L2 cache implementation
Thierry Reding [Mon, 9 Dec 2019 12:00:04 +0000 (13:00 +0100)]
drm/nouveau/ltc/gp10b: Add custom L2 cache implementation

There are extra registers that need to be programmed to make the level 2
cache work on GP10B, such as the stream ID register that is used when an
SMMU is used to translate memory addresses.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/secboot/gm20b,gp10b: Read WPR configuration from GPU registers
Thierry Reding [Mon, 9 Dec 2019 12:00:03 +0000 (13:00 +0100)]
drm/nouveau/secboot/gm20b,gp10b: Read WPR configuration from GPU registers

The GPUs found on Tegra SoCs have registers that can be used to read the
WPR configuration. Use these registers instead of reaching into the
memory controller's register space to read the same information.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/tegra: Set clock rate if not set
Thierry Reding [Mon, 9 Dec 2019 12:00:02 +0000 (13:00 +0100)]
drm/nouveau/tegra: Set clock rate if not set

If the GPU clock has not had a rate set, initialize it to the maximum
clock rate to make sure it does run.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
4 years agodrm/nouveau/tegra: Avoid pulsing reset twice
Thierry Reding [Mon, 9 Dec 2019 12:00:01 +0000 (13:00 +0100)]
drm/nouveau/tegra: Avoid pulsing reset twice

When the GPU powergate is controlled by a generic power domain provider,
the reset will automatically be asserted and deasserted as part of the
power-ungating procedure.

On some Jetson TX2 boards, doing an additional assert and deassert of
the GPU outside of the power-ungate procedure can cause the GPU to go
into a bad state where the memory interface can no longer access system
memory.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>