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, 235 insertions, 0 deletions

diff --git a/Documentation/i2c/slave-testunit-backend.rst b/Documentation/i2c/slave-testunit-backend.rst
new file mode 100644
index 000000000..d752f433b
--- /dev/null
+++ b/Documentation/i2c/slave-testunit-backend.rst
@@ -0,0 +1,235 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+================================
+Linux I2C slave testunit backend
+================================
+
+by Wolfram Sang <wsa@sang-engineering.com> in 2020
+
+This backend can be used to trigger test cases for I2C bus masters which
+require a remote device with certain capabilities (and which are usually not so
+easy to obtain). Examples include multi-master testing, and SMBus Host Notify
+testing. For some tests, the I2C slave controller must be able to switch
+between master and slave mode because it needs to send data, too.
+
+Note that this is a device for testing and debugging. It should not be enabled
+in a production build. And while there is some versioning and we try hard to
+keep backward compatibility, there is no stable ABI guaranteed!
+
+Instantiating the device is regular. Example for bus 0, address 0x30::
+
+  # echo "slave-testunit 0x1030" > /sys/bus/i2c/devices/i2c-0/new_device
+
+Or using firmware nodes. Here is a devicetree example (note this is only a
+debug device, so there are no official DT bindings)::
+
+  &i2c0	{
+        ...
+
+	testunit@30 {
+		compatible = "slave-testunit";
+		reg = <(0x30 | I2C_OWN_SLAVE_ADDRESS)>;
+	};
+  };
+
+After that, you will have the device listening. Reading will return a single
+byte. Its value is 0 if the testunit is idle, otherwise the command number of
+the currently running command.
+
+When writing, the device consists of 4 8-bit registers and, except for some
+"partial" commands, all registers must be written to start a testcase, i.e. you
+usually write 4 bytes to the device. The registers are:
+
+.. csv-table::
+  :header: "Offset", "Name", "Description"
+
+  0x00, CMD, which test to trigger
+  0x01, DATAL, configuration byte 1 for the test
+  0x02, DATAH, configuration byte 2 for the test
+  0x03, DELAY, delay in n * 10ms until test is started
+
+Using 'i2cset' from the i2c-tools package, the generic command looks like::
+
+  # i2cset -y <bus_num> <testunit_address> <CMD> <DATAL> <DATAH> <DELAY> i
+
+DELAY is a generic parameter which will delay the execution of the test in CMD.
+While a command is running (including the delay), new commands will not be
+acknowledged. You need to wait until the old one is completed.
+
+The commands are described in the following section. An invalid command will
+result in the transfer not being acknowledged.
+
+Commands
+--------
+
+0x00 NOOP
+~~~~~~~~~
+
+Reserved for future use.
+
+0x01 READ_BYTES
+~~~~~~~~~~~~~~~
+
+.. list-table::
+  :header-rows: 1
+
+  * - CMD
+    - DATAL
+    - DATAH
+    - DELAY
+
+  * - 0x01
+    - address to read data from (lower 7 bits, highest bit currently unused)
+    - number of bytes to read
+    - n * 10ms
+
+Also needs master mode. This is useful to test if your bus master driver is
+handling multi-master correctly. You can trigger the testunit to read bytes
+from another device on the bus. If the bus master under test also wants to
+access the bus at the same time, the bus will be busy. Example to read 128
+bytes from device 0x50 after 50ms of delay::
+
+  # i2cset -y 0 0x30 1 0x50 0x80 5 i
+
+0x02 SMBUS_HOST_NOTIFY
+~~~~~~~~~~~~~~~~~~~~~~
+
+.. list-table::
+  :header-rows: 1
+
+  * - CMD
+    - DATAL
+    - DATAH
+    - DELAY
+
+  * - 0x02
+    - low byte of the status word to send
+    - high byte of the status word to send
+    - n * 10ms
+
+Also needs master mode. This test will send an SMBUS_HOST_NOTIFY message to the
+host. Note that the status word is currently ignored in the Linux Kernel.
+Example to send a notification with status word 0x6442 after 10ms::
+
+  # i2cset -y 0 0x30 2 0x42 0x64 1 i
+
+If the host controller supports HostNotify, this message with debug level
+should appear (Linux 6.11 and later)::
+
+  Detected HostNotify from address 0x30
+
+0x03 SMBUS_BLOCK_PROC_CALL
+~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. list-table::
+  :header-rows: 1
+
+  * - CMD
+    - DATAL
+    - DATAH
+    - DELAY
+
+  * - 0x03
+    - 0x01 (i.e. one further byte will be written)
+    - number of bytes to be sent back
+    - leave out, partial command!
+
+Partial command. This test will respond to a block process call as defined by
+the SMBus specification. The one data byte written specifies how many bytes
+will be sent back in the following read transfer. Note that in this read
+transfer, the testunit will prefix the length of the bytes to follow. So, if
+your host bus driver emulates SMBus calls like the majority does, it needs to
+support the I2C_M_RECV_LEN flag of an i2c_msg. This is a good testcase for it.
+The returned data consists of the length first, and then of an array of bytes
+from length-1 to 0. Here is an example which emulates
+i2c_smbus_block_process_call() using i2ctransfer (you need i2c-tools v4.2 or
+later)::
+
+  # i2ctransfer -y 0 w3@0x30 3 1 0x10 r?
+  0x10 0x0f 0x0e 0x0d 0x0c 0x0b 0x0a 0x09 0x08 0x07 0x06 0x05 0x04 0x03 0x02 0x01 0x00
+
+0x04 GET_VERSION_WITH_REP_START
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. list-table::
+  :header-rows: 1
+
+  * - CMD
+    - DATAL
+    - DATAH
+    - DELAY
+
+  * - 0x04
+    - currently unused
+    - currently unused
+    - leave out, partial command!
+
+Partial command. After sending this command, the testunit will reply to a read
+message with a NUL terminated version string based on UTS_RELEASE. The first
+character is always a 'v' and the length of the version string is at maximum
+128 bytes. However, it will only respond if the read message is connected to
+the write message via repeated start. If your controller driver handles
+repeated start correctly, this will work::
+
+  # i2ctransfer -y 0 w3@0x30 4 0 0 r128
+  0x76 0x36 0x2e 0x31 0x31 0x2e 0x30 0x2d 0x72 0x63 0x31 0x2d 0x30 0x30 0x30 0x30 ...
+
+If you have i2c-tools 4.4 or later, you can print out the data right away::
+
+  # i2ctransfer -y -b 0 w3@0x30 4 0 0 r128
+  v6.11.0-rc1-00009-gd37a1b4d3fd0
+
+STOP/START combinations between the two messages will *not* work because they
+are not equivalent to a REPEATED START. As an example, this returns just the
+default response::
+
+  # i2cset -y 0 0x30 4 0 0 i; i2cget -y 0 0x30
+  0x00
+
+0x05 SMBUS_ALERT_REQUEST
+~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. list-table::
+  :header-rows: 1
+
+  * - CMD
+    - DATAL
+    - DATAH
+    - DELAY
+
+  * - 0x05
+    - response value (7 MSBs interpreted as I2C address)
+    - currently unused
+    - n * 10ms
+
+This test raises an interrupt via the SMBAlert pin which the host controller
+must handle. The pin must be connected to the testunit as a GPIO. GPIO access
+is not allowed to sleep. Currently, this can only be described using firmware
+nodes. So, for devicetree, you would add something like this to the testunit
+node::
+
+  gpios = <&gpio1 24 GPIO_ACTIVE_LOW>;
+
+The following command will trigger the alert with a response of 0xc9 after 1
+second of delay::
+
+  # i2cset -y 0 0x30 5 0xc9 0x00 100 i
+
+If the host controller supports SMBusAlert, this message with debug level
+should appear::
+
+  smbus_alert 0-000c: SMBALERT# from dev 0x64, flag 1
+
+This message may appear more than once because the testunit is software not
+hardware and, thus, may not be able to react to the response of the host fast
+enough. The interrupt count should increase only by one, though::
+
+  # cat /proc/interrupts | grep smbus_alert
+   93:          1  gpio-rcar  26 Edge      smbus_alert
+
+If the host does not respond to the alert within 1 second, the test will be
+aborted and the testunit will report an error.
+
+For this test, the testunit will shortly drop its assigned address and listen
+on the SMBus Alert Response Address (0x0c). It will reassign its original
+address afterwards.