This patch adds a new parameter to the amdkfd driver. This parameter enables
the user to select the scheduling policy of the CP. The choices are:

* CP Scheduling with support for over-subscription
* CP Scheduling without support for over-subscription
* Without CP Scheduling

Note that the third option (Without CP scheduling) is only for debug purposes
and bringup of new H/W. As such, it is _not_ guaranteed to work at all times on
all H/W versions.

v3: Fixed description of parameter, changed the permissions to read_only, added
a verification of the value and added documentation

v5: Set default sched_policy to HWS as it is now supported by firmware
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

The kernel queue module enables the amdkfd to establish kernel queues, not
exposed to user space.

The kernel queues are used for HIQ (HSA Interface Queue) and DIQ (Debug
Interface Queue) operations

v3: Removed use of internal typedefs and added use of the new gart allocation
functions

v4: Fixed a miscalculation in kernel queue wrapping

v5:

Move amdkfd from drm/radeon/ to drm/amd/
Change format of mqd structure to match latest KV firmware
Add support for AQL queues creation to enable working with open-source HSA
runtime
Add define for kernel queue size
Various fixes
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

The mqd_manager module handles MQD data structures.
MQD stands for Memory Queue Descriptor, which is used by the H/W to
keep the usermode queue state in memory.

v3:

Removed new typedefs
Removed pragma pack 4
Remove cik_mqds.h file
Changed lower_32/upper_32 calls to use linux macros
Used new gart allocation functions
Added documentation

v4:

Added missing initialization of the addr field in init_mqd()

Setting the hqd persistent.preload_req bit ON so that when queues switches
on/off, their context will kept and read from the mqd when the cp reassign
them, and thus the dispatched workload context kept consistent without any
interrupts.

v5:

Move amdkfd from drm/radeon/ to drm/amd/
Change format of mqd structure to match latest KV firmware
Add support for AQL queues creation to enable working with open-source HSA
runtime.
Various fixes
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

The queue module enables allocating and initializing queues uniformly.

v3: Removed typedef and redundant memset call. Broke long pr_debug print to one
liners and Added documentation.

v5: Move amdkfd from drm/radeon/ to drm/amd/
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

This patch adds the functions to bind and unbind pasid
from a device through the amd_iommu driver.

The unbind function is called when the mm_struct of the
process is released.

The bind function is not called here because it is called
only in the IOCTLs which are not yet implemented at this
stage of the patchset.
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

This patch adds the process module and three helper modules:

- kfd_process, which handles process which open /dev/kfd

- kfd_doorbell, which provides helper functions for doorbell allocation,
  release and mapping to userspace

- kfd_pasid, which provides helper functions for pasid allocation and release

- kfd_aperture, which provides helper functions for managing the LDS, Local GPU
  memory and Scratch memory apertures of the process

This patch only contains the basic kfd_process module, which doesn't contain
the reference to the queue scheduler. This was done to allow easier code review.

Also, this patch doesn't contain the calls to the IOMMU driver for binding the
pasid to the device. Again, this was done to allow easier code review

The kfd_process object is created when a process opens /dev/kfd and is closed
when the mm_struct of that process is teared-down.

v3:

Removed kfd_vidmem.c file
Replaced direct mmput call to mmu_notifier release
Removed typedefs
Moved bool field to end of the structure
Added new kernel params for gart usage limitation
Added initialization of sa manager
Fixed debug messages
Remove support for LDS in 32 bit
Changed code to support mmap of doorbell pages from userspace
Added documentation for apertures

v4: Replaced RCU by SRCU for kfd_process list management

v5:

Move amdkfd from drm/radeon/ to drm/amd/
Rename kfd_aperture.c to kfd_flat_memory.c
Protect against multiple init calls
MQD size is H/W dependent so moved it to device info structure
Rename kfd_mem_obj structure's members
Use delayed function for process tear-down
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

This patch adds the topology module to the driver. The topology is exposed to
userspace through the sysfs.

The calls to add and remove a device to/from topology are done by the radeon
driver.

v3:

The CPU information, that is provided in the topology section of the amdkfd
driver, is extracted from the CRAT table. Unlike the CPU information located
in /sys/devices/system/cpu/cpu*, which is extracted from the SRAT table.

While the CPU information provided by the CRAT and the SRAT tables might be
identical, the node topology might be different. The SRAT table contains the
topology of CPU nodes only. The CRAT table contains the topology of CPU and GPU
nodes together (and can be interleaved). For example CPU node 1 in SRAT can be
CPU node 3 in CRAT. Furthermore it's worth to mention that the CRAT table
contains only HSA compatible nodes (nodes which are compliant with the HSA
spec).

To recap, amdkfd exposes a different kind of topology than the one exposed by
/sys/devices/system/cpu/cpu even though it may contain similar information.

v4:

The topology module doesn't support uevent handling and doesn't notify the
userspace about runtime modifications. It is up to the userspace to acquire
snapshots of the topology information created by the amdkfd and exposed
in sysfs.

The following is an example of how the topology looks on a Kaveri A10-7850K
system with amdkfd installed:

/sys/devices/virtual/kfd/kfd/
|
--- topology/
      |
      |--- generation_id
      |--- system_properties
      |--- nodes/
            |
            |--- 0/
                 |
                 |--- gpu_id
                 |--- name
                 |--- properties
                 |--- caches/
                      |
                      |--- 0/
                           |
                           |--- properties
                      |--- 1/
                           |
                           |--- properties
                      |--- 2/
                           |
                           |--- properties
                 |--- io_links/
                      |
                 |--- mem_banks/
                      |
                      |--- 0/
                           |
                           |--- properties
                      |--- 1/
                           |
                           |--- properties
                      |--- 2/
                           |
                           |--- properties
                      |--- 3/
                           |
                           |--- properties

v5:

Move amdkfd from drm/radeon/ to drm/amd/

Add a check if dev->gpu pointer is null before accessing it in the
node_show function in kfd_topology.c
This situation may occur when amdkfd is loaded and there is a GPU with a CRAT
table, but that GPU isn't supported by amdkfd
Signed-off-by: Evgeny Pinchuk <evgeny.pinchuk@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

This patch adds the amdkfd skeleton driver. The driver does nothing except
define a /dev/kfd device.

It returns -ENODEV on all amdkfd IOCTLs.

v3: Move bool field to the end of structure, removed the pmc ioctls and added
a meaningful error message for ioctl error.

v5:

Create a new folder drm/amd and move amdkfd from drm/radeon/ to drm/amd/
Remove scheduler_class from kfd_priv.h as it was never used
Add skeleton implementation of the Get Version IOCTL

v6:
Update module version to the correct number and remove the "default m" from the
Kconfig file
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

- KFD_IOC_GET_VERSION:
	Retrieves the interface version of amdkfd

- KFD_IOC_CREATE_QUEUE:
	Creates a usermode queue that runs on a specific GPU device

- KFD_IOC_DESTROY_QUEUE:
	Destroys an existing usermode queue

- KFD_IOC_SET_MEMORY_POLICY:
	Sets the memory policy of the default and alternate aperture of the
        calling process

- KFD_IOC_GET_CLOCK_COUNTERS:
	Retrieves counters (timestamps) of CPU and GPU

- KFD_IOC_GET_PROCESS_APERTURES:
	Retrieves information about process apertures that were initialized
        during the open() call of the amdkfd device

- KFD_IOC_UPDATE_QUEUE:
	Updates configuration of an existing usermode queue

v3: Remove pragma pack and pmc ioctls. Added parameter for doorbell offset and
a comment on counters

v5:

Add define for AQL queues.
Fix arguments of Get Version IOCTL
Make IOCTL's structures to be the same size on 32/64 bit

v6: Change the version of the amdkfd-thunk interface
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

v6: Update entries to reflect new name & location of driver
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

This patch adds the interface between the radeon driver and the amdkfd driver.
The interface implementation is contained in radeon_kfd.c and radeon_kfd.h.

The interface itself is represented by a pointer to struct
kfd_dev. The pointer is located inside radeon_device structure.

All the register accesses that amdkfd need are done using this interface. This
allows us to avoid direct register accesses in amdkfd proper,  while also
avoiding locking between amdkfd and radeon.

The single exception is the doorbells that are used in both of the drivers.
However, because they are located in separate pci bar pages, the danger of
sharing registers between the drivers is minimal.

Having said that, we are planning to move the doorbells as well to radeon.

v3:

Add interface for sa manager init and fini. The init function will allocate a
buffer on system memory and pin it to the GART address space via the radeon sa
manager.

All mappings of buffers to GART address space are done via the radeon sa
manager. The interface of allocate memory will use the radeon sa manager to sub
allocate from the single buffer that was allocated during the init function.

Change lower_32/upper_32 calls to use linux macros

Add documentation for the interface

v4:

Change ptr field type in kgd_mem from uint32_t* to void* to match to type that
is returned by radeon_sa_bo_cpu_addr

v5:

Change format of mqd structure to work with latest KV firmware
Add support for AQL queues creation to enable working with open-source HSA
runtime.
Move generic kfd-->kgd interface and other generic kgd definitions to a generic
header file that will be used by AMD's radeon and amdgpu drivers
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

Implementing a lock for selecting and accessing shader engines and arrays.
This lock will make sure that radeon and amdkfd are not colliding when
accessing shader engines and arrays with GRBM_GFX_INDEX register.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

radeon and amdkfd share the doorbell aperture.
radeon sets it up, takes the doorbells required for its own rings
and reports the setup to amdkfd.
radeon reserved doorbells are at the start of the doorbell aperture.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

amdkfd should set interrupts for pipes 1-7.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

To support HSA on KV, we need to limit the number of vmids and pipes
that are available for radeon's use with KV.

This patch reserves VMIDs 8-15 for amdkfd (so radeon can only use VMIDs
0-7) and also makes radeon thinks that KV has only a single MEC with a single
pipe in it

v3: Use define for static vmid allocation in radeon
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

This patch fixes a bug in the accounting of the device_state.
In the current code, the device_state was put (decremented) too many times,
which sometimes lead to the driver getting stuck permanently in
put_device_state_wait(). That happen because the device_state->count would go
below zero, which is never supposed to happen.

The root cause is that the device_state was decremented in put_pasid_state()
and put_pasid_state_wait() but also in all the functions that call those
functions. Therefore, the device_state was decremented twice in each of these
code paths.

The fix is to decouple the device_state accounting from the pasid_state
accounting - remove the call to put_device_state() from the
put_pasid_state() and the put_pasid_state_wait())
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

Make use of the new invalidate_range mmu_notifier call-back and remove the
old logic of assigning an empty page-table between invalidate_range_start
and invalidate_range_end.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Rik van Riel <riel@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Jay Cornwall <Jay.Cornwall@amd.com>
Cc: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

Now that the mmu_notifier_invalidate_range() calls are in place, add the
callback to allow subsystems to register against it.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Rik van Riel <riel@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Jay Cornwall <Jay.Cornwall@amd.com>
Cc: Oded Gabbay <Oded.Gabbay@amd.com>
Cc: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

Add calls to the new mmu_notifier_invalidate_range() function to all
places in the VMM that need it.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Rik van Riel <riel@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Jay Cornwall <Jay.Cornwall@amd.com>
Cc: Oded Gabbay <Oded.Gabbay@amd.com>
Cc: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

This notifier closes an important gap in the current mmu_notifier
implementation, the existing callbacks are called too early or too late to
reliably manage a non-CPU TLB.  Specifically, invalidate_range_start() is
called when all pages are still mapped and invalidate_range_end() when all
pages are unmapped and potentially freed.

This is fine when the users of the mmu_notifiers manage their own SoftTLB,
like KVM does.  When the TLB is managed in software it is easy to wipe out
entries for a given range and prevent new entries to be established until
invalidate_range_end is called.

But when the user of mmu_notifiers has to manage a hardware TLB it can
still wipe out TLB entries in invalidate_range_start, but it can't make
sure that no new TLB entries in the given range are established between
invalidate_range_start and invalidate_range_end.

To avoid silent data corruption the entries in the non-CPU TLB need to be
flushed when the pages are unmapped (at this point in time no _new_ TLB
entries can be established in the non-CPU TLB) but not yet freed (as the
non-CPU TLB may still have _existing_ entries pointing to the pages about
to be freed).

To fix this problem we need to catch the moment when the Linux VMM flushes
remote TLBs (as a non-CPU TLB is not very CPU TLB), as this is the point
in time when the pages are unmapped but _not_ yet freed.

The mmu_notifier_invalidate_range() function aims to catch that moment.

IOMMU code will be one user of the notifier-callback.  Currently this is
only the AMD IOMMUv2 driver, but its code is about to be more generalized
and converted to a generic IOMMU-API extension to fit the needs of similar
functionality in other IOMMUs as well.

The current attempt in the AMD IOMMUv2 driver to work around the
invalidate_range_start/end() shortcoming is to assign an empty page table
to the non-CPU TLB between any invalidata_range_start/end calls.  With the
empty page-table assigned, every page-table walk to re-fill the non-CPU
TLB will cause a page-fault reported to the IOMMU driver via an interrupt,
possibly causing interrupt storms.

The page-fault handler in the AMD IOMMUv2 driver doesn't handle the fault
if an invalidate_range_start/end pair is active, it just reports back
SUCCESS to the device and let it refault the page.  But existing hardware
(newer Radeon GPUs) that makes use of this feature don't re-fault
indefinitly, after a certain number of faults for the same address the
device enters a failure state and needs to be resetted.

To avoid the GPUs entering a failure state we need to get rid of the
empty-page-table workaround and use the mmu_notifier_invalidate_range()
function introduced with this patch.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Rik van Riel <riel@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Jay Cornwall <Jay.Cornwall@amd.com>
Cc: Oded Gabbay <Oded.Gabbay@amd.com>
Cc: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>

Radeon patches for 3.19.  Christian has a number of GPUVM improvements
slated as well, but I'd like to wait until he gets back to work next week
to pull those in. Highlights of this pull:
- ttm performance improvements
- CI dpm fixes

* 'drm-next-3.19' of git://people.freedesktop.org/~agd5f/linux: (26 commits)
  drm/radeon/si/ci: make u8 static arrays constant
  drm/radeon: set power control in ci dpm enable
  drm/radeon: powertune fixes for hawaii
  drm/radeon: fix dpm mc init for certain hawaii boards
  drm/radeon: set bootup pcie level to max for ci dpm
  drm/radeon: fix default dpm state setup
  drm/radeon: workaround a hw bug in bonaire pcie dpm
  drm/radeon: fix mclk vddc configuration for cards for hawaii
  drm/radeon: fix sclk DS enablement
  drm/radeon: fix activity settings for sclk and mclk for CI
  drm/radeon: improve mclk param calcuations for ci dpm
  drm/radeon: fix dram timing for certain hawaii boards
  drm/radeon: switch force state commands for CI
  drm/radeon: fix for memory training on bonaire 0x6649
  drm/radeon/ci: handle gpio controlled dpm features properly
  drm/radeon: store the gpio shift as well
  drm/radeon: export radeon_atombios_lookup_gpio
  drm/radeon: fix typo in CI dpm disable
  drm/radeon: rework CI dpm thermal setup
  drm/radeon: rework SI dpm thermal setup
  ...

These two arrays don't change, just make them constant,
reduces data segment by a few bytes.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Necessary for poper operation.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

- bapm is not available on hawaii
- update pt defaults
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Needs special overrides for certain vram configurations.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Avoids problems when re-loading the driver.  Does not
affect power saving when dpm is enabled.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Only enable the first levels for mclk and sclk.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Some boards get stuck in pcie x1 otherwise.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Need to use vddc0 for vdcc1 for certain hawaii configurations.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Only enable it for levels 0 and 1.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Only need to be enabled on the first level.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Properly take into account the post divider.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Certain memory configurations need a fix.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Use the preferred SMC commands for forcing state on CI.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Workaround for memory link training on certain variants
of 0x6649.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Certain feature enablement depends on entries in the atom
gpio pin table.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

We need this in the dpm code.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

We need it for dpm.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Need to disable DS, not enable it when disabling dpm.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

In preparation for fan control.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>