For tasks that execute very quickly, but not quickly enough to be
already complete by the time execution reaches host1x_syncpt_wait,
the proportion of time spent in allocating a fence and invoking
dma_fence_wait_timeout becomes dominating in comparison to the
time it actually takes to execute the task.
To improve wait latency in these cases, replace the current
"is threshold already reached" check with a short spin loop
to catch these situations before going to the heavy machinery.
For longer waits, since this function blocks anyway, the only
negative effect is slightly increased CPU consumption due to
the loop.
Bug 4001325
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
Change-Id: I44e99cda88b4bcb33f190884d1a2e5f7588cb775
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2881716
Reviewed-by: Santosh BS <santoshb@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2916412
The current submission opcode sequence first takes the engine MLOCK,
and then switches to HOST1X class to wait prefences. This is fine
while we only use a single channel per engine and there is no
virtualization, since jobs are serialized on that one channel anyway.
However, when that assumption doesn't hold, we are keeping the
engine locked while not running anything on it while waiting for
prefences to complete.
To resolve this, execute wait commands in the beginning of the job
outside the engine MLOCK. We still take the HOST1X MLOCK because
recent hardware requires register opcodes to be executed within some
MLOCK, but the hardware also allows unlimited channels to take the
HOST1X MLOCK at the same time.
Jira HOSTX-4687
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
Change-Id: I783cbc7f1bbd7415fbf0e61163935186b2ba0a44
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2911124
Reviewed-by: Santosh BS <santoshb@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
The integer argument returned by the NVHOST interrupt notifier has never
been supported by the host1x-nvhost driver. The DLA driver's
nvdla_queue_update() function that is called by this notifier has been
update so that it no longer uses this argument. Given that there are no
other users of this notifier for the host1x-nvhost driver, let's remove
this unused variable.
Bug 4059530
Change-Id: I04dda2ea7338f0c7ee55e6e968b46f063eecae90
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2892312
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Amit Sharma (SW-TEGRA) <amisharma@nvidia.com>
Reviewed-by: Mitch Harwell <mharwell@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
[1] The DLA firmware configures the DEST and MASK registers during
the initialization.
[2] In OOT, the DLA KMD is accidentally setting the interrupts. This
will result in enabling additional interrupts that the firmware
does not handle (like NVDLA_GENERIC_INTR_FAULT_CRITICAL).
[3] This commit fixes the issue by avoiding the interrupt settings
at the DLA KMD.
Bug 3960841
Change-Id: Ied77ac7564ae3b3202b9cceaf08ad3c7da9f947c
Signed-off-by: Arvind M <am@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2861253
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
Reviewed-by: Amit Sharma (SW-TEGRA) <amisharma@nvidia.com>
Reviewed-by: Jonathan Hunter <jonathanh@nvidia.com>
Reviewed-by: svcacv <svcacv@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
The syncpt interface APIs in the host1x-nvhost driver slightly differ
from those in the legacy nvhost driver because instead of passing the
platform device structure for the host1x device, the platform device
structure for the host1x client device is passed.
By aligning the APIs so that we pass the platform device structure for
the host1x in all implementations, we can simplify the PVA driver and
directly use the APIs in the NVIDIA display driver.
The NVIDIA display driver requires some additional syncpt interface APIs
and so implement these as well.
Bug 3713048
Change-Id: I507e6fd066e6e22c0c47c20ba0dd3be5fa033c59
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2813827
Reviewed-by: Bibek Basu <bbasu@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: svc_kernel_abi <svc_kernel_abi@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
With Linux v6.1, if CONFIG_PCI_P2PDMA is enabled, then mapping the
syncpt shim causes a kernel paging request fault. The fault occurs when
dma_map_sg_attrs() is called in by nvhost_syncpt_unit_interface_init(),
which is attempting to map the syncpt shim MMIO space.
The function dma_map_sg_attrs() calls iomm_dma_map_sg() and since
upstream Linux commit 30280eee2db1 ("iommu/dma: support PCI P2PDMA pages
in dma-iommu map_sg") was added this now calls is_pci_p2pdma_page(). The
function is_pci_p2pdma_page() requires that there is a 'struct page'
associated with the sg, but this is not the case and so deferencing the
'struct page' pointer returned by sg_page() results in a paging fault.
Using dma_map_sg_attrs() to map MMIO that is not associated with valid
page structures is not guaranteed to work. Therefore, fix this by using
dma_map_resource() to map the syncpt shim instead.
Bug 3871422
Change-Id: I2b08a3b78be0850bff44e62bfc93c8d095eb82a2
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2818417
(cherry picked from commit 97f1282f2959c208effebf93cc5061d6e7fa6c55)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2819086
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Drivers, such as the PVA and DLA driver, that call
nvhost_syncpt_unit_interface_init() during probe are missing a call to
clean-up the DMA mappings this function may create if the driver probe
fails or if the driver is removed. The function
nvhost_syncpt_unit_interface_init() may make a call to
dma_map_sg_attrs() and we need to ensure that dma_unmap_sg_attrs() is
called if the probe of the driver fails or if the driver is unloaded.
Add a new function, nvhost_syncpt_unit_interface_deinit(), that calls
dma_unmap_sg_attrs() if needed for drivers to call.
Bug 3800349
Change-Id: I62a4e19cd42878dac54fa623509440596ffdf17f
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2801280
(cherry picked from commit cee12fb989186611aa9deb5ada45831ab5f783aa)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2801934
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
When building the host1x-nvhost driver as an external module, the driver
source is now copied into a common location with all the other external
modules. Therefore, we can now use the 'srctree.nvidia' path for finding
the necessary header files for Host1x. Update the include search paths
to use 'srctree.nvidia' when building the driver.
Bug 3817518
Change-Id: I443dd4cb6d01c4e9be8f3c02550b1050d0bcfd28
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2785956
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
1. Replace nvhost print functions with PVA driver's private
implementation with similar functionalities.
2. Change syncpoint from host managed to client managed so that
syncpoint max value is tracked inside PVA driver. This change is needed
because upstream host1x driver doesn't allow decrement of max value.
3. Various minor changes for adapting to upstream kernel APIs.
Bug 3662478
Change-Id: I16eff6c06c51afe0f274598e1273404924d18684
Signed-off-by: Nan Wang <nanwa@nvidia.com>
Signed-off-by: Abhinaya Agrawal <abhinayaa@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2710684
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Both host1x and nvhost expose APIs that deinitialize the debugfs
and remove debugfs files/directories that were created. At present,
debugfs_remove() is inovked to do this. However,
debugfs_remove_recursive() should instead be used to ensure that
all files within a debugfs are recursively removed before the
directory is removed.
Bug 3662478
Change-Id: Idb7d1458c02f57f6d656a7bfe1df3adf5e0497aa
Signed-off-by: Abhinaya Agrawal <abhinayaa@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2714042
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: Omar Nemri <onemri@nvidia.com>
GVS: Gerrit_Virtual_Submit
Some host1x clients, such as PVA, do not have a falcon microprocessor
and currently the host1x-nvhost shim driver assumes that all clients do.
Fix this by moving the falcon initialisation code from the
nvhost_module_init() function and into the
nvhost_flcn_finalize_poweron() function. This aligns with the
implementation in the downstream nvhost driver.
JIRA LS-493
Change-Id: Ic94c3489cefa6fd256f111d8e6c7afb539b5bc8a
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2710548
Reviewed-by: Nan Wang <nanwa@nvidia.com>
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
GVS: Gerrit_Virtual_Submit
After unloading the NVDLA driver and then reloading the NVDLA, the
kernel is crashing. The host1x-nvhost driver is missing a call to
class_destroy() in nvhost_client_device_release() which gets called when
removing the DLA driver and this is preventing the DLA driver from
creating the class again when reloading. The crash then occurs because
when the driver is reloaded, creating the class for the DLA driver
fails and then the host1x-nvhost driver incorrectly calls
device_destroy() even though the device has not been created yet. Fix
this by ensuring the class_destroy() is called by
nvhost_client_device_release() and in the necessary error paths and then
remove the call to device_destroy().
Finally, replace request_irq() with devm_request_irq() to ensure that
the interrupts are also released as necessary on removal.
Bug 3641820
Change-Id: Ia328bf63d528e8c31bff1d7b3ac0d5dddc22f1f5
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2710546
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
GVS: Gerrit_Virtual_Submit
Sparse reports the following warnings for the host1x-nvhost driver ...
drivers/gpu/host1x-nvhost/falcon.c:65:27: warning: cast to restricted __le32
drivers/gpu/host1x-nvhost/nvhost.c:205:22: warning: symbol
'nvhost_syncpt_get_by_id' was not declared. Should it be static?
drivers/gpu/host1x-nvhost/nvhost.c:768:5: warning: symbol
'nvhost_module_runtime_resume' was not declared. Should it be static?
drivers/gpu/host1x-nvhost/nvhost.c:796:5: warning: symbol
'nvhost_module_runtime_suspend' was not declared. Should it be static?
Fix the above by correcting the cast in falcon_copy_firmware_image(), remove
the function nvhost_syncpt_get_by_id() because it is not used, and make the
nvhost_module_runtime_resume/suspend functions static.
JIRA LS-410
Change-Id: I2c923fccd0b6e53d17198617b90905c6cd06eb5c
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2670180
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
GVS: Gerrit_Virtual_Submit
When compiling the host1x-nvhost driver for Linux v5.10 it is currently
failing and the following errors are seen ...
host1x-nvhost/nvhost.c: In function ‘nvhost_intr_do_work’:
nvhost.c:554:2: error: implicit declaration of function ‘kfree’
[-Werror=implicit-function-declaration]
kfree(host1x_cb);
^~~~~
host1x-nvhost/nvhost.c: In function ‘nvhost_intr_register_notifier’:
nvhost.c:579:7: error: implicit declaration of function ‘kzalloc’
[-Werror=implicit-function-declaration]
cb = kzalloc(sizeof(*cb), GFP_KERNEL);
^~~~~~~
Fix this by including slab.h.
JIRA LS-410
Change-Id: I8a56d33933d3e57c04b011e2c6da03d5f2bbc707
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2667882
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
GVS: Gerrit_Virtual_Submit
The host1x-nvhost driver by default configures the clock rates for
the host1x client devices to the maximum supported rate because
there is runtime clock scaling support yet. However, currently the
host1x-nvhost driver is only setting the rate for the client's first
clock and not the others. Fix this by correcting the index used when
iterating through the list of clocks.
JIRA LS-410
Change-Id: If410ee25548c0dddecc5b0de9d8b6f4f687fd73c
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2657897
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
