[ Upstream commit 84a833d90635e4b846333e2df0ae72f9cbecac39 ]
When slicing a BO, we need to iterate through the BO's sgt to find the
right pieces to construct the slice. Some of the data types chosen for
this process are incorrectly too small, and can overflow. This can
result in the incorrect slice construction, which can lead to data
corruption in workload execution.
The device can only handle 32-bit sized transfers, and the scatterlist
struct only supports 32-bit buffer sizes, so our upper limit for an
individual transfer is an unsigned int. Using an int is incorrect due to
the reservation of the sign bit. Upgrade the length of a scatterlist
entry and the offsets into a scatterlist entry to unsigned int for a
correct representation.
While each transfer may be limited to 32-bits, the overall BO may exceed
that size. For counting the total length of the BO, we need a type that
can represent the largest allocation possible on the system. That is the
definition of size_t, so use it.
Fixes: ff13be8303 ("accel/qaic: Add datapath")
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Jeff Hugo <jeff.hugo@oss.qualcomm.com>
Reviewed-by: Lizhi Hou <lizhi.hou@amd.com>
Reviewed-by: Troy Hanson <quic_thanson@quicinc.com>
Reviewed-by: Youssef Samir <quic_yabdulra@quicinc.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20250306171959.853466-1-jeff.hugo@oss.qualcomm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 0165994c215f321e2d055368f89b424756e340eb ]
There is a potential race between user thread seeking to re-use
a timestamp record with new interrupt id, while this record is still
in the middle of interrupt handling and it is about to be freed.
Imagine the driver set the record in_use to 0 and only then fill the
free_node information. This might lead to unpleasant scenario where
the new registration thread detects the record as free to use, and
change the cq buff address. That will cause the free_node to get
the wrong buffer address to put refcount to.
Signed-off-by: farah kassabri <fkassabri@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 0b75cb5b240fddf181c284d415ee77ef61b418d6 ]
It is currently allowed for a user to export dma-buf with size and
offset that are not multiples of PAGE_SIZE.
The exported memory is mapped for the importer device, and there it will
be rounded to PAGE_SIZE, leading to actually exporting more than the
user intended to.
To make the user be aware of it, accept only size and offset which are
multiple of PAGE_SIZE.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 6309727ef27162deabd5c095c11af24970fba5a2 ]
Add a kthread_stop_put() helper that stops a thread and puts its task
struct. Use it to replace the various instances of kthread_stop()
followed by put_task_struct().
Remove the kthread_stop_put() macro in usbip that is similar but doesn't
return the result of kthread_stop().
[agruenba@redhat.com: fix kerneldoc comment]
Link: https://lkml.kernel.org/r/20230911111730.2565537-1-agruenba@redhat.com
[akpm@linux-foundation.org: document kthread_stop_put()'s argument]
Link: https://lkml.kernel.org/r/20230907234048.2499820-1-agruenba@redhat.com
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Stable-dep-of: bb9025f4432f ("dma-mapping: benchmark: fix up kthread-related error handling")
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 42422993cf28d456778ee9168d73758ec037cd51 ]
Gaudi2 with PCI revision ID with the value of '3' represents Gaudi2C
device and should be detected and initialized as Gaudi2.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit a9f07790a4b2250f0140e9a61c7f842fd9b618c7 ]
This function may copy the pad0 field of struct hl_info_sec_attest to user
mode which has not been initialized, resulting in leakage of kernel heap
data to user mode. To prevent this, use kzalloc() to allocate and zero out
the buffer, which can also eliminate other uninitialized holes, if any.
Fixes: 0c88760f8f ("habanalabs/gaudi2: add secured attestation info uapi")
Signed-off-by: Xingyuan Mo <hdthky0@gmail.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 4c8874c2a6512b9fe7285cab1a6910d9211a6cfb ]
The SOC_HW_VERSION register in the BHI space is not correctly initialized
by the device and in many cases contains uninitialized data. The register
could contain 0xFFFFFFFF which is a special value to indicate a link
error in PCIe, therefore if observed, we could incorrectly think the
device is down.
Intercept reads for this register, and provide the correct value - every
production instance would read 0x60110200 if the device was operating as
intended.
Fixes: a36bf7af86 ("accel/qaic: Add MHI controller")
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231208163101.1295769-3-quic_jhugo@quicinc.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 35c49cfc8b702eda7a0d3f05497b16f81b69e289 ]
Using PCI Device ID/Revision to initialize the interrupt_clear_with_0
workaround is problematic - there are many pre-production
steppings with different behavior, even with the same PCI ID/Revision
Instead of checking for PCI Device ID/Revision, check the VPU
buttress interrupt status register behavior - if this register
is not zero after writing 1s it means there register is RW
instead of RW1C and we need to enable the interrupt_clear_with_0
workaround.
Fixes: 7f34e01f77 ("accel/ivpu: Clear specific interrupt status bits on C0")
Signed-off-by: Andrzej Kacprowski <Andrzej.Kacprowski@intel.com>
Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Link: https://lore.kernel.org/all/20231204122331.40560-1-jacek.lawrynowicz@linux.intel.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 3f7c0634926daf48cd2f6db6c1197a1047074088 ]
There is no need to call MMIO reset using VPU_37XX_BUTTRESS_VPU_IP_RESET
register. IP will be reset by FLR or by entering d0i3. Also IP reset
during power_up is not needed as the VPU is already in reset.
Removing MMIO reset improves stability as it a partial device reset
that is not safe in some corner cases.
This change also brings back ivpu_boot_pwr_domain_disable() that
helps to properly power down VPU when it is hung by a buggy workload.
Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Fixes: 828d63042a ("accel/ivpu: Don't enter d0i3 during FLR")
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231115111004.1304092-1-jacek.lawrynowicz@linux.intel.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 90f3de616259cbb5666f00f8daf5420cd7d71d18 ]
snprintf() returns the "number of characters which *would* be generated for
the given input", not the size *really* generated.
In order to avoid too large values for 'str_size' (and potential negative
values for "PSOC_RAZWI_ENG_STR_SIZE - str_size") use scnprintf()
instead of snprintf().
Fixes: c0e6df9160 ("accel/habanalabs: fix address decode RAZWI handling")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Create buffers with cache coherency on the CPU side (write-back) while
disabling snooping on the VPU side. These buffers require an explicit
cache flush after each CPU-side modification.
Configuring pages as write-combined may introduce significant delays,
potentially taking hundreds of milliseconds for 64 MB buffers.
Added internal DRM_IVPU_BO_NOSNOOP mask which disables snooping on the
VPU side. Allocate FW runtime memory buffer (64 MB) as cached with
snooping-disabled.
This fixes random long FW loading times and boot params memory
corruption on warmboot (due to missed wmb).
Fixes: 02d5b0aacd ("accel/ivpu: Implement firmware parsing and booting")
Signed-off-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230926120943.GD846747@linux.intel.com
We need to wait for the CLOCK_RESOURCE_OWN_ACK bit to be set
after configuring the workpoint. This step ensures that the VPU
microcontroller clock is actively toggling and ready for operation.
Previously, we relied solely on the READY bit in the VPU_STATUS
register, which indicated the completion of the workpoint download.
However, this approach was insufficient, as the READY bit could be set
while the device was still running on a sideband clock until the PLL
locked. To guarantee that the PLL is locked and the device is running on
the main clock source, we now wait for the CLOCK_RESOURCE_OWN_ACK before
proceeding with the remainder of the power-up sequence.
Fixes: 79cdc56c4a ("accel/ivpu: Add initial support for VPU 4")
Signed-off-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230925121137.872158-4-stanislaw.gruszka@linux.intel.com
If we receive signal when waiting for IPC message response in
ivpu_ipc_receive() we return error and continue to operate.
Then the driver can send another IPC messages and re-use occupied
slot of the message still processed by the firmware. This can result
in corrupting firmware memory and following FW crash with messages:
[ 3698.569719] intel_vpu 0000:00:0b.0: [drm] ivpu_ipc_send_receive_internal(): IPC receive failed: type 0x1103, ret -512
[ 3698.569747] intel_vpu 0000:00:0b.0: [drm] ivpu_jsm_unregister_db(): Failed to unregister doorbell 3: -512
[ 3698.569756] intel_vpu 0000:00:0b.0: [drm] ivpu_ipc_tx_prepare(): IPC message vpu:0x88980000 not released by firmware
[ 3698.569763] intel_vpu 0000:00:0b.0: [drm] ivpu_ipc_tx_prepare(): JSM message vpu:0x88980040 not released by firmware
[ 3698.570234] intel_vpu 0000:00:0b.0: [drm] ivpu_ipc_send_receive_internal(): IPC receive failed: type 0x110e, ret -512
[ 3698.570318] intel_vpu 0000:00:0b.0: [drm] *ERROR* ivpu_mmu_dump_event(): MMU EVTQ: 0x10 (Translation fault) SSID: 0 SID: 3, e[2] 00000000, e[3] 00000208, in addr: 0x88988000, fetch addr: 0x0
To fix the issue don't use interruptible variant of wait event to
allow firmware to finish IPC processing.
Fixes: 5d7422cfb4 ("accel/ivpu: Add IPC driver and JSM messages")
Reviewed-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230925121137.872158-2-stanislaw.gruszka@linux.intel.com
The encode_dma() function has some validation on in_trans->size but it
would be more clear to move those checks to find_and_map_user_pages().
The encode_dma() had two checks:
if (in_trans->addr + in_trans->size < in_trans->addr || !in_trans->size)
return -EINVAL;
The in_trans->addr variable is the starting address. The in_trans->size
variable is the total size of the transfer. The transfer can occur in
parts and the resources->xferred_dma_size tracks how many bytes we have
already transferred.
This patch introduces a new variable "remaining" which represents the
amount we want to transfer (in_trans->size) minus the amount we have
already transferred (resources->xferred_dma_size).
I have modified the check for if in_trans->size is zero to instead check
if in_trans->size is less than resources->xferred_dma_size. If we have
already transferred more bytes than in_trans->size then there are negative
bytes remaining which doesn't make sense. If there are zero bytes
remaining to be copied, just return success.
The check in encode_dma() checked that "addr + size" could not overflow
and barring a driver bug that should work, but it's easier to check if
we do this in parts. First check that "in_trans->addr +
resources->xferred_dma_size" is safe. Then check that "xfer_start_addr +
remaining" is safe.
My final concern was that we are dealing with u64 values but on 32bit
systems the kmalloc() function will truncate the sizes to 32 bits. So
I calculated "total = in_trans->size + offset_in_page(xfer_start_addr);"
and returned -EINVAL if it were >= SIZE_MAX. This will not affect 64bit
systems.
Fixes: 129776ac2e ("accel/qaic: Add control path")
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Carl Vanderlip <quic_carlv@quicinc.com>
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Link: https://patchwork.freedesktop.org/patch/msgid/24d3348b-25ac-4c1b-b171-9dae7c43e4e0@moroto.mountain
