x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 109 insertions, 0 deletions

diff --git a/Documentation/admin-guide/media/raspberrypi-pisp-be.rst b/Documentation/admin-guide/media/raspberrypi-pisp-be.rst
new file mode 100644
index 000000000..0fcf46f26
--- /dev/null
+++ b/Documentation/admin-guide/media/raspberrypi-pisp-be.rst
@@ -0,0 +1,109 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+=========================================================
+Raspberry Pi PiSP Back End Memory-to-Memory ISP (pisp-be)
+=========================================================
+
+The PiSP Back End
+=================
+
+The PiSP Back End is a memory-to-memory Image Signal Processor (ISP) which reads
+image data from DRAM memory and performs image processing as specified by the
+application through the parameters in a configuration buffer, before writing
+pixel data back to memory through two distinct output channels.
+
+The ISP registers and programming model are documented in the `Raspberry Pi
+Image Signal Processor (PiSP) Specification document`_
+
+The PiSP Back End ISP processes images in tiles. The handling of image
+tessellation and the computation of low-level configuration parameters is
+realized by a free software library called `libpisp
+<https://github.com/raspberrypi/libpisp>`_.
+
+The full image processing pipeline, which involves capturing RAW Bayer data from
+an image sensor through a MIPI CSI-2 compatible capture interface, storing them
+in DRAM memory and processing them in the PiSP Back End to obtain images usable
+by an application is implemented in `libcamera <https://libcamera.org>`_ as
+part of the Raspberry Pi platform support.
+
+The pisp-be driver
+==================
+
+The Raspberry Pi PiSP Back End (pisp-be) driver is located under
+drivers/media/platform/raspberrypi/pisp-be. It uses the `V4L2 API` to register
+a number of video capture and output devices, the `V4L2 subdev API` to register
+a subdevice for the ISP that connects the video devices in a single media graph
+realized using the `Media Controller (MC) API`.
+
+The media topology registered by the `pisp-be` driver is represented below:
+
+.. _pips-be-topology:
+
+.. kernel-figure:: raspberrypi-pisp-be.dot
+    :alt:   Diagram of the default media pipeline topology
+    :align: center
+
+
+The media graph registers the following video device nodes:
+
+- pispbe-input: output device for images to be submitted to the ISP for
+  processing.
+- pispbe-tdn_input: output device for temporal denoise.
+- pispbe-stitch_input: output device for image stitching (HDR).
+- pispbe-output0: first capture device for processed images.
+- pispbe-output1: second capture device for processed images.
+- pispbe-tdn_output: capture device for temporal denoise.
+- pispbe-stitch_output: capture device for image stitching (HDR).
+- pispbe-config: output device for ISP configuration parameters.
+
+pispbe-input
+------------
+
+Images to be processed by the ISP are queued to the `pispbe-input` output device
+node. For a list of image formats supported as input to the ISP refer to the
+`Raspberry Pi Image Signal Processor (PiSP) Specification document`_.
+
+pispbe-tdn_input, pispbe-tdn_output
+-----------------------------------
+
+The `pispbe-tdn_input` output video device receives images to be processed by
+the temporal denoise block which are captured from the `pispbe-tdn_output`
+capture video device. Userspace is responsible for maintaining queues on both
+devices, and ensuring that buffers completed on the output are queued to the
+input.
+
+pispbe-stitch_input, pispbe-stitch_output
+-----------------------------------------
+
+To realize HDR (high dynamic range) image processing the image stitching and
+tonemapping blocks are used. The `pispbe-stitch_output` writes images to memory
+and the `pispbe-stitch_input` receives the previously written frame to process
+it along with the current input image. Userspace is responsible for maintaining
+queues on both devices, and ensuring that buffers completed on the output are
+queued to the input.
+
+pispbe-output0, pispbe-output1
+------------------------------
+
+The two capture devices write to memory the pixel data as processed by the ISP.
+
+pispbe-config
+-------------
+
+The `pispbe-config` output video devices receives a buffer of configuration
+parameters that define the desired image processing to be performed by the ISP.
+
+The format of the ISP configuration parameter is defined by
+:c:type:`pisp_be_tiles_config` C structure and the meaning of each parameter is
+described in the `Raspberry Pi Image Signal Processor (PiSP) Specification
+document`_.
+
+ISP configuration
+=================
+
+The ISP configuration is described solely by the content of the parameters
+buffer. The only parameter that userspace needs to configure using the V4L2 API
+is the image format on the output and capture video devices for validation of
+the content of the parameters buffer.
+
+.. _Raspberry Pi Image Signal Processor (PiSP) Specification document: https://datasheets.raspberrypi.com/camera/raspberry-pi-image-signal-processor-specification.pdf