[PATCH v8 00/15] perf annotate arm64: Support instruction tracking for data type profiling
From: Tengda Wu
Date: Tue Oct 20 2026 - 09:54:39 EST
This series adds arm64 support to perf's data type profiling by
implementing instruction tracking for the arm64 architecture.
The main changes since v7 include:
v7: https://lore.kernel.org/all/cover.1791629326.git.wutengda@xxxxxxxxxxxxxxx/
* Patch 02: Minor optimizations to update_insn_state_arm64 to facilitate
future extension, including:
- Encapsulating call instruction handling into update_call_insn_state()
- Encapsulating invalidate dst regs handling into invalidate_unhandled_insn_regs()
- Renaming is_readonly_branch_or_cmp() to is_readonly_insn()
No functional changes.
* Fix issues raised by Sashiko:
- Patch 06: Complete the status register invalidation logic for
exclusive stores.
- Patch 13 & 15: Complete the lifetime invalidation logic for adrp
and mrs instructions.
Regarding the status register invalidation issue for instructions
such as cas/swp as well, we plan to support this incrementally
after the relevant instructions are refined; this patch temporarily
ignores and does not handle it (a TODO marker has already been
reserved in patch 02)
Patch organization
==================
Patches 1-3: Shared infrastructure. Extract invalidate_reg_state() as
a common helper, add the arch_get_reg_offset() helper for
computing operand base offsets, and wire up the arm64
entry points (arch_supports_insn_tracking(),
init_type_state(), and update_insn_state_arm64()).
Patches 4-6: Core instruction tracking for arm64: function return
types after calls, load instructions (ldr/ldp and their
addressing modes), and store instructions (str/stp).
Patches 7-9: Shared data structure improvements and an x86 fix.
Expand type_state_reg.imm_value to u64 so 64-bit
addresses can be tracked, preserve imm_value for stack
variables, and delete stale stack state on store of an
untracked register on x86.
Patches 10-15: Remaining arm64 patterns: stack variable tracking,
'mov' (register and immediate forms), 'add' member
address calculation, 'adrp' global variable tracking,
per-cpu variable access, and 'mrs sp_el0' to resolve the
'current' pointer.
Each patch's type profiling results are as follows:
Patch | Feature | no_sym | no_var | no_typeinfo | bad_offset | insn_track | ok(%)
------+----------------------------+--------+--------+-------------+------------+------------+------
base | baseline | 6 | 493 | - | - | - | 56.2%
0002 | enable insn tracking | 6 | 42 | 437 | 2 | 12 | 57.2%
0005 | support 'load' insn | 6 | 42 | 398 | 1 | 52 | 60.7%
0006 | support 'store' insn | 6 | 42 | 398 | 1 | 52 | 60.7%
0010 | support stack variable | 6 | 42 | 391 | 1 | 59 | 61.3%
0011 | support 'mov' insn | 6 | 42 | 372 | 3 | 76 | 62.8%
0012 | support 'add' insn | 6 | 42 | 318 | 7 | 126 | 67.2%
0013 | support 'adrp' insn | 6 | 42 | 238 | 8 | 205 | 74.2%
0014 | support per-cpu variable | 6 | 42 | 233 | 8 | 210 | 74.6%
0015 | support 'mrs' insn | 6 | 42 | 230 | 8 | 213 | 74.9%
Testing
=======
Tested on arm64 (all passed):
# perf test -v "perf data type profiling tests"
81: perf data type profiling tests : Ok
=== Test Summary ===
Passed main tests : 1
Passed subtests : 0
Skipped tests : 0
Failed tests : 0
Tested on x86. The profiling results show no change before/after applying
this patch series:
before : total 880, ok 711 (80.8%), bad 169 (19.2%)
after : total 880, ok 711 (80.8%), bad 169 (19.2%)
Changelog
=========
v7 -> v8:
- v7: https://lore.kernel.org/all/cover.1791629326.git.wutengda@xxxxxxxxxxxxxxx/
- Encapsulate invalidate dst regs handling into invalidate_unhandled_insn_regs()
- Encapsulate invalidate dst regs handling into invalidate_unhandled_insn_regs()
- Rename is_readonly_branch_or_cmp() to is_readonly_insn()
- Complete the status register invalidation logic for exclusive stores.
- Complete the lifetime invalidation logic for adrp and mrs instructions.
v6 -> v7:
- v6: https://lore.kernel.org/all/cover.1789521520.git.wutengda@xxxxxxxxxxxxxxx/
- Invalidate lifetime in propagate_load_reg_state().
- Remove the has_reg_type() guard in update_store_insn_state().
- Fix the issue where TSR_KIND_PERCPU_BASE fails to retry as expected.
Please let me know if you have any feedback.
Thanks,
Tengda
Tengda Wu (15):
perf annotate-data: Extract invalidate_reg_state() as a common helper
perf annotate-arm64: Enable instruction tracking support
perf annotate-data: Add arch_get_reg_offset helper
perf annotate-arm64: Track return type after call instructions
perf annotate-arm64: Support load instruction tracking
perf annotate-arm64: Support store instruction tracking
perf annotate-data: Expand type_state_reg imm_value to u64
perf annotate-data: Track imm_value for stack variables
perf annotate-x86: Delete stale stack state on store of untracked
register
perf annotate-arm64: Support stack variable tracking
perf annotate-arm64: Support 'mov' instruction tracking
perf annotate-arm64: Support 'add' instruction tracking
perf annotate-arm64: Support 'adrp' instruction to track global
variables
perf annotate-arm64: Support per-cpu variable access tracking
perf annotate-arm64: Support 'mrs' instruction to track 'current'
pointer
.../perf/util/annotate-arch/annotate-arm64.c | 774 ++++++++++++++++++
tools/perf/util/annotate-arch/annotate-x86.c | 38 +-
tools/perf/util/annotate-data.c | 209 ++++-
tools/perf/util/annotate-data.h | 14 +-
4 files changed, 987 insertions(+), 48 deletions(-)
base-commit: 2efb7a11f1aa4d2dc285fe2680cca22cb9f72f64
--
2.34.1