There are cases when we want to have separate DRM devices for GPU and
display pipelines.
One example is development, when it is beneficial to be able to bind the
GPU driver separately, without the display pipeline (and without the
hacks adding "amd,imageon" to the compatible string).
Another example is some of Qualcomm platforms, which have two MDSS
units, but only one GPU. With current approach it is next to impossible
to support this usecase properly, while separate binding allows users to
have three DRM devices: two for MDSS units and a single headless GPU.
Add kernel param msm.separate_gpu_kms, which if set to true forces
creation of separate display and GPU DRM devices. Mesa supports this
setup by using the kmsro wrapper.
The param is disabled by default, in order to be able to test userspace
for the compatibility issues. Simple clients are able to handle this
setup automatically.
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662590/
[Rob: renamed the modparam to separate_gpu_kms, and add missing
DRIVER_GEM_GPUVA]
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Currently the msm driver creates an extra interim platform device for
Imageon GPUs. This is not ideal, as the device doesn't have
corresponding OF node. If the headless mode is used for newer GPUs, then
the msm_use_mmu() function can not detect corresponding IOMMU devices.
Also the DRM device (although it's headless) is created with modesetting
flags being set.
To solve all these issues, rework the way the Imageon devices are bound.
Remove the interim device, don't register a component and instead use a
cut-down version of the normal functions to probe or remove the driver.
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662584/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
As discussed a lot in the past, the UBWC config must be coherent across
a number of IP blocks (currently display and GPU, but it also may/will
concern camera/video as the drivers evolve).
So far, we've been trying to keep the values reasonable in each of the
two drivers separately, but it really make sense to do so centrally,
especially given certain fields (e.g. HBB) may need to be gathered
dynamically.
To reduce room for error, move to fetching the config from a central
source, so that the data programmed into the hardware is consistent
across all multimedia blocks that request it.
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660963/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
A large number of (unsorted or separate) small (<2MB) mappings can cause
a lot of, probably unnecessary, prealloc pages. Ie. a single 4k page
size mapping will pre-allocate 3 pages (for levels 2-4) for the
pagetable. Which can chew up a large amount of unneeded memory. So add
a mechanism to put an upper bound on the # of pre-alloc pages.
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661529/
With the conversion to drm_gpuvm, we lost the lazy VMA cleanup, which
means that fb cleanup/unpin when pageflipping to new scanout buffers
immediately unmaps the scanout buffer. This is costly (with tlbinv,
it can be 4-6ms for a 1080p scanout buffer, and more for higher
resolutions)!
To avoid this, introduce a vma_ref, which is incremented whenever
userspace has a GEM handle or dma-buf fd. When unpinning if the
vm is the kms->vm we defer tearing down the VMA until the vma_ref
drops to zero. If the buffer is still part of a flip-chain then
userspace will be holding some sort of reference to the BO, either
via a GEM handle and/or dma-buf fd. So this avoids unmapping the VMA
when there is a strong possibility that it will be needed again.
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661538/
This resolves a potential deadlock vs msm_gem_vm_close(). Otherwise for
_NO_SHARE buffers msm_gem_describe() could be trying to acquire the
shared vm resv, while already holding priv->obj_lock. But _vm_close()
might drop the last reference to a GEM obj while already holding the vm
resv, and msm_gem_free_object() needs to grab priv->obj_lock, a locking
inversion.
OTOH this is only for debugfs and it isn't critical if we undercount by
skipping a locked obj. So just use trylock() and move along if we can't
get the lock.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661525/
When userspace opts in to VM_BIND, the submit no longer holds references
keeping the VMA alive. This makes it difficult to distinguish between
UMD/KMD/app bugs. So add a debug option for logging the most recent VM
updates and capturing these in GPU devcoredumps.
The submitqueue id is also captured, a value of zero means the operation
did not go via a submitqueue (ie. comes from msm_gem_vm_close() tearing
down the remaining mappings when the device file is closed.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661518/
With async VM_BIND, the actual pgtable updates are deferred.
Synchronously, a list of map/unmap ops will be generated, but the
actual pgtable changes are deferred. To support that, split out
op handlers and change the existing non-VM_BIND paths to use them.
Note in particular, the vma itself may already be destroyed/freed
by the time an UNMAP op runs (or even a MAP op if there is a later
queued UNMAP). For this reason, the op handlers cannot reference
the vma pointer.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661516/
With user managed VMs and multiple queues, it is in theory possible to
trigger map/unmap errors. These will (in a later patch) mark the VM as
unusable. But we want to tell the io-pgtable helpers not to spam the
log. In addition, in the unmap path, we don't want to bail early from
the unmap, to ensure we don't leave some dangling pages mapped.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661520/
Rob Clark
Dmitry Baryshkov
Konrad Dybcio
Rob Clark