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

5 files changed, 288 insertions, 0 deletions

diff --git a/Documentation/arch/xtensa/atomctl.rst b/Documentation/arch/xtensa/atomctl.rst
new file mode 100644
index 000000000..75d174169
--- /dev/null
+++ b/Documentation/arch/xtensa/atomctl.rst
@@ -0,0 +1,51 @@
+===========================================
+Atomic Operation Control (ATOMCTL) Register
+===========================================
+
+We Have Atomic Operation Control (ATOMCTL) Register.
+This register determines the effect of using a S32C1I instruction
+with various combinations of:
+
+     1. With and without an Coherent Cache Controller which
+        can do Atomic Transactions to the memory internally.
+
+     2. With and without An Intelligent Memory Controller which
+        can do Atomic Transactions itself.
+
+The Core comes up with a default value of for the three types of cache ops::
+
+      0x28: (WB: Internal, WT: Internal, BY:Exception)
+
+On the FPGA Cards we typically simulate an Intelligent Memory controller
+which can implement  RCW transactions. For FPGA cards with an External
+Memory controller we let it to the atomic operations internally while
+doing a Cached (WB) transaction and use the Memory RCW for un-cached
+operations.
+
+For systems without an coherent cache controller, non-MX, we always
+use the memory controllers RCW, though non-MX controllers likely
+support the Internal Operation.
+
+CUSTOMER-WARNING:
+   Virtually all customers buy their memory controllers from vendors that
+   don't support atomic RCW memory transactions and will likely want to
+   configure this register to not use RCW.
+
+Developers might find using RCW in Bypass mode convenient when testing
+with the cache being bypassed; for example studying cache alias problems.
+
+See Section 4.3.12.4 of ISA; Bits::
+
+                             WB     WT      BY
+                           5   4 | 3   2 | 1   0
+
+=========    ==================      ==================      ===============
+  2 Bit
+  Field
+  Values     WB - Write Back         WT - Write Thru         BY - Bypass
+=========    ==================      ==================      ===============
+    0        Exception               Exception               Exception
+    1        RCW Transaction         RCW Transaction         RCW Transaction
+    2        Internal Operation      Internal Operation      Reserved
+    3        Reserved                Reserved                Reserved
+=========    ==================      ==================      ===============
diff --git a/Documentation/arch/xtensa/booting.rst b/Documentation/arch/xtensa/booting.rst
new file mode 100644
index 000000000..e1b83707e
--- /dev/null
+++ b/Documentation/arch/xtensa/booting.rst
@@ -0,0 +1,22 @@
+=====================================
+Passing boot parameters to the kernel
+=====================================
+
+Boot parameters are represented as a TLV list in the memory. Please see
+arch/xtensa/include/asm/bootparam.h for definition of the bp_tag structure and
+tag value constants. First entry in the list must have type BP_TAG_FIRST, last
+entry must have type BP_TAG_LAST. The address of the first list entry is
+passed to the kernel in the register a2. The address type depends on MMU type:
+
+- For configurations without MMU, with region protection or with MPU the
+  address must be the physical address.
+- For configurations with region translarion MMU or with MMUv3 and CONFIG_MMU=n
+  the address must be a valid address in the current mapping. The kernel will
+  not change the mapping on its own.
+- For configurations with MMUv2 the address must be a virtual address in the
+  default virtual mapping (0xd0000000..0xffffffff).
+- For configurations with MMUv3 and CONFIG_MMU=y the address may be either a
+  virtual or physical address. In either case it must be within the default
+  virtual mapping. It is considered physical if it is within the range of
+  physical addresses covered by the default KSEG mapping (XCHAL_KSEG_PADDR..
+  XCHAL_KSEG_PADDR + XCHAL_KSEG_SIZE), otherwise it is considered virtual.
diff --git a/Documentation/arch/xtensa/features.rst b/Documentation/arch/xtensa/features.rst
new file mode 100644
index 000000000..28dcce175
--- /dev/null
+++ b/Documentation/arch/xtensa/features.rst
@@ -0,0 +1,3 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+.. kernel-feat:: features xtensa
diff --git a/Documentation/arch/xtensa/index.rst b/Documentation/arch/xtensa/index.rst
new file mode 100644
index 000000000..69952446a
--- /dev/null
+++ b/Documentation/arch/xtensa/index.rst
@@ -0,0 +1,14 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+===================
+Xtensa Architecture
+===================
+
+.. toctree::
+   :maxdepth: 1
+
+   atomctl
+   booting
+   mmu
+
+   features
diff --git a/Documentation/arch/xtensa/mmu.rst b/Documentation/arch/xtensa/mmu.rst
new file mode 100644
index 000000000..450573afa
--- /dev/null
+++ b/Documentation/arch/xtensa/mmu.rst
@@ -0,0 +1,198 @@
+=============================
+MMUv3 initialization sequence
+=============================
+
+The code in the initialize_mmu macro sets up MMUv3 memory mapping
+identically to MMUv2 fixed memory mapping. Depending on
+CONFIG_INITIALIZE_XTENSA_MMU_INSIDE_VMLINUX symbol this code is
+located in addresses it was linked for (symbol undefined), or not
+(symbol defined), so it needs to be position-independent.
+
+The code has the following assumptions:
+
+  - This code fragment is run only on an MMU v3.
+  - TLBs are in their reset state.
+  - ITLBCFG and DTLBCFG are zero (reset state).
+  - RASID is 0x04030201 (reset state).
+  - PS.RING is zero (reset state).
+  - LITBASE is zero (reset state, PC-relative literals); required to be PIC.
+
+TLB setup proceeds along the following steps.
+
+  Legend:
+
+    - VA = virtual address (two upper nibbles of it);
+    - PA = physical address (two upper nibbles of it);
+    - pc = physical range that contains this code;
+
+After step 2, we jump to virtual address in the range 0x40000000..0x5fffffff
+or 0x00000000..0x1fffffff, depending on whether the kernel was loaded below
+0x40000000 or above. That address corresponds to next instruction to execute
+in this code. After step 4, we jump to intended (linked) address of this code.
+The scheme below assumes that the kernel is loaded below 0x40000000.
+
+ ====== =====  =====  =====  =====   ====== =====  =====
+ -      Step0  Step1  Step2  Step3          Step4  Step5
+
+   VA      PA     PA     PA     PA     VA      PA     PA
+ ====== =====  =====  =====  =====   ====== =====  =====
+ E0..FF -> E0  -> E0  -> E0          F0..FF -> F0  -> F0
+ C0..DF -> C0  -> C0  -> C0          E0..EF -> F0  -> F0
+ A0..BF -> A0  -> A0  -> A0          D8..DF -> 00  -> 00
+ 80..9F -> 80  -> 80  -> 80          D0..D7 -> 00  -> 00
+ 60..7F -> 60  -> 60  -> 60
+ 40..5F -> 40         -> pc  -> pc   40..5F -> pc
+ 20..3F -> 20  -> 20  -> 20
+ 00..1F -> 00  -> 00  -> 00
+ ====== =====  =====  =====  =====   ====== =====  =====
+
+The default location of IO peripherals is above 0xf0000000. This may be changed
+using a "ranges" property in a device tree simple-bus node. See the Devicetree
+Specification, section 4.5 for details on the syntax and semantics of
+simple-bus nodes. The following limitations apply:
+
+1. Only top level simple-bus nodes are considered
+
+2. Only one (first) simple-bus node is considered
+
+3. Empty "ranges" properties are not supported
+
+4. Only the first triplet in the "ranges" property is considered
+
+5. The parent-bus-address value is rounded down to the nearest 256MB boundary
+
+6. The IO area covers the entire 256MB segment of parent-bus-address; the
+   "ranges" triplet length field is ignored
+
+
+MMUv3 address space layouts.
+============================
+
+Default MMUv2-compatible layout::
+
+                        Symbol                   VADDR       Size
+  +------------------+
+  | Userspace        |                           0x00000000  TASK_SIZE
+  +------------------+                           0x40000000
+  +------------------+
+  | Page table       |  XCHAL_PAGE_TABLE_VADDR   0x80000000  XCHAL_PAGE_TABLE_SIZE
+  +------------------+
+  | KASAN shadow map |  KASAN_SHADOW_START       0x80400000  KASAN_SHADOW_SIZE
+  +------------------+                           0x8e400000
+  +------------------+
+  | VMALLOC area     |  VMALLOC_START            0xc0000000  128MB - 64KB
+  +------------------+  VMALLOC_END
+  +------------------+
+  | Cache aliasing   |  TLBTEMP_BASE_1           0xc8000000  DCACHE_WAY_SIZE
+  | remap area 1     |
+  +------------------+
+  | Cache aliasing   |  TLBTEMP_BASE_2                       DCACHE_WAY_SIZE
+  | remap area 2     |
+  +------------------+
+  +------------------+
+  | KMAP area        |  PKMAP_BASE                           PTRS_PER_PTE *
+  |                  |                                       DCACHE_N_COLORS *
+  |                  |                                       PAGE_SIZE
+  |                  |                                       (4MB * DCACHE_N_COLORS)
+  +------------------+
+  | Atomic KMAP area |  FIXADDR_START                        KM_TYPE_NR *
+  |                  |                                       NR_CPUS *
+  |                  |                                       DCACHE_N_COLORS *
+  |                  |                                       PAGE_SIZE
+  +------------------+  FIXADDR_TOP              0xcffff000
+  +------------------+
+  | Cached KSEG      |  XCHAL_KSEG_CACHED_VADDR  0xd0000000  128MB
+  +------------------+
+  | Uncached KSEG    |  XCHAL_KSEG_BYPASS_VADDR  0xd8000000  128MB
+  +------------------+
+  | Cached KIO       |  XCHAL_KIO_CACHED_VADDR   0xe0000000  256MB
+  +------------------+
+  | Uncached KIO     |  XCHAL_KIO_BYPASS_VADDR   0xf0000000  256MB
+  +------------------+
+
+
+256MB cached + 256MB uncached layout::
+
+                        Symbol                   VADDR       Size
+  +------------------+
+  | Userspace        |                           0x00000000  TASK_SIZE
+  +------------------+                           0x40000000
+  +------------------+
+  | Page table       |  XCHAL_PAGE_TABLE_VADDR   0x80000000  XCHAL_PAGE_TABLE_SIZE
+  +------------------+
+  | KASAN shadow map |  KASAN_SHADOW_START       0x80400000  KASAN_SHADOW_SIZE
+  +------------------+                           0x8e400000
+  +------------------+
+  | VMALLOC area     |  VMALLOC_START            0xa0000000  128MB - 64KB
+  +------------------+  VMALLOC_END
+  +------------------+
+  | Cache aliasing   |  TLBTEMP_BASE_1           0xa8000000  DCACHE_WAY_SIZE
+  | remap area 1     |
+  +------------------+
+  | Cache aliasing   |  TLBTEMP_BASE_2                       DCACHE_WAY_SIZE
+  | remap area 2     |
+  +------------------+
+  +------------------+
+  | KMAP area        |  PKMAP_BASE                           PTRS_PER_PTE *
+  |                  |                                       DCACHE_N_COLORS *
+  |                  |                                       PAGE_SIZE
+  |                  |                                       (4MB * DCACHE_N_COLORS)
+  +------------------+
+  | Atomic KMAP area |  FIXADDR_START                        KM_TYPE_NR *
+  |                  |                                       NR_CPUS *
+  |                  |                                       DCACHE_N_COLORS *
+  |                  |                                       PAGE_SIZE
+  +------------------+  FIXADDR_TOP              0xaffff000
+  +------------------+
+  | Cached KSEG      |  XCHAL_KSEG_CACHED_VADDR  0xb0000000  256MB
+  +------------------+
+  | Uncached KSEG    |  XCHAL_KSEG_BYPASS_VADDR  0xc0000000  256MB
+  +------------------+
+  +------------------+
+  | Cached KIO       |  XCHAL_KIO_CACHED_VADDR   0xe0000000  256MB
+  +------------------+
+  | Uncached KIO     |  XCHAL_KIO_BYPASS_VADDR   0xf0000000  256MB
+  +------------------+
+
+
+512MB cached + 512MB uncached layout::
+
+                        Symbol                   VADDR       Size
+  +------------------+
+  | Userspace        |                           0x00000000  TASK_SIZE
+  +------------------+                           0x40000000
+  +------------------+
+  | Page table       |  XCHAL_PAGE_TABLE_VADDR   0x80000000  XCHAL_PAGE_TABLE_SIZE
+  +------------------+
+  | KASAN shadow map |  KASAN_SHADOW_START       0x80400000  KASAN_SHADOW_SIZE
+  +------------------+                           0x8e400000
+  +------------------+
+  | VMALLOC area     |  VMALLOC_START            0x90000000  128MB - 64KB
+  +------------------+  VMALLOC_END
+  +------------------+
+  | Cache aliasing   |  TLBTEMP_BASE_1           0x98000000  DCACHE_WAY_SIZE
+  | remap area 1     |
+  +------------------+
+  | Cache aliasing   |  TLBTEMP_BASE_2                       DCACHE_WAY_SIZE
+  | remap area 2     |
+  +------------------+
+  +------------------+
+  | KMAP area        |  PKMAP_BASE                           PTRS_PER_PTE *
+  |                  |                                       DCACHE_N_COLORS *
+  |                  |                                       PAGE_SIZE
+  |                  |                                       (4MB * DCACHE_N_COLORS)
+  +------------------+
+  | Atomic KMAP area |  FIXADDR_START                        KM_TYPE_NR *
+  |                  |                                       NR_CPUS *
+  |                  |                                       DCACHE_N_COLORS *
+  |                  |                                       PAGE_SIZE
+  +------------------+  FIXADDR_TOP              0x9ffff000
+  +------------------+
+  | Cached KSEG      |  XCHAL_KSEG_CACHED_VADDR  0xa0000000  512MB
+  +------------------+
+  | Uncached KSEG    |  XCHAL_KSEG_BYPASS_VADDR  0xc0000000  512MB
+  +------------------+
+  | Cached KIO       |  XCHAL_KIO_CACHED_VADDR   0xe0000000  256MB
+  +------------------+
+  | Uncached KIO     |  XCHAL_KIO_BYPASS_VADDR   0xf0000000  256MB
+  +------------------+