anolis-cloud-kernel/kernel/kcsan
Marco Elver d70951b88c kcsan: Turn report_filterlist_lock into a raw_spinlock
ANBZ: #13070

commit 59458fa4dd upstream.

Ran Xiaokai reports that with a KCSAN-enabled PREEMPT_RT kernel, we can see
splats like:

| BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:48
| in_atomic(): 1, irqs_disabled(): 1, non_block: 0, pid: 0, name: swapper/1
| preempt_count: 10002, expected: 0
| RCU nest depth: 0, expected: 0
| no locks held by swapper/1/0.
| irq event stamp: 156674
| hardirqs last  enabled at (156673): [<ffffffff81130bd9>] do_idle+0x1f9/0x240
| hardirqs last disabled at (156674): [<ffffffff82254f84>] sysvec_apic_timer_interrupt+0x14/0xc0
| softirqs last  enabled at (0): [<ffffffff81099f47>] copy_process+0xfc7/0x4b60
| softirqs last disabled at (0): [<0000000000000000>] 0x0
| Preemption disabled at:
| [<ffffffff814a3e2a>] paint_ptr+0x2a/0x90
| CPU: 1 UID: 0 PID: 0 Comm: swapper/1 Not tainted 6.11.0+ #3
| Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-0-ga698c8995f-prebuilt.qemu.org 04/01/2014
| Call Trace:
|  <IRQ>
|  dump_stack_lvl+0x7e/0xc0
|  dump_stack+0x1d/0x30
|  __might_resched+0x1a2/0x270
|  rt_spin_lock+0x68/0x170
|  kcsan_skip_report_debugfs+0x43/0xe0
|  print_report+0xb5/0x590
|  kcsan_report_known_origin+0x1b1/0x1d0
|  kcsan_setup_watchpoint+0x348/0x650
|  __tsan_unaligned_write1+0x16d/0x1d0
|  hrtimer_interrupt+0x3d6/0x430
|  __sysvec_apic_timer_interrupt+0xe8/0x3a0
|  sysvec_apic_timer_interrupt+0x97/0xc0
|  </IRQ>

On a detected data race, KCSAN's reporting logic checks if it should
filter the report. That list is protected by the report_filterlist_lock
*non-raw* spinlock which may sleep on RT kernels.

Since KCSAN may report data races in any context, convert it to a
raw_spinlock.

This requires being careful about when to allocate memory for the filter
list itself which can be done via KCSAN's debugfs interface. Concurrent
modification of the filter list via debugfs should be rare: the chosen
strategy is to optimistically pre-allocate memory before the critical
section and discard if unused.

Link: https://lore.kernel.org/all/20240925143154.2322926-1-ranxiaokai627@163.com/
Reported-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
Tested-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
Signed-off-by: Marco Elver <elver@google.com>

Fixes: CVE-2024-56610
Signed-off-by: Xiao Long <xiaolong@openanolis.org>
Signed-off-by: Qinyun Tan <qinyuntan@linux.alibaba.com>
Reviewed-by: Xunlei Pang <xlpang@linux.alibaba.com>
Link: https://gitee.com/anolis/cloud-kernel/pulls/4898
2025-03-27 04:17:14 +00:00
..
Makefile kcsan: Disable branch tracing in core runtime 2020-06-29 12:04:48 -07:00
atomic.h kcsan: Remove existing special atomic rules 2020-06-29 12:04:48 -07:00
core.c kcsan, debugfs: Move debugfs file creation out of early init 2021-05-11 14:47:33 +02:00
debugfs.c kcsan: Turn report_filterlist_lock into a raw_spinlock 2025-03-27 04:17:14 +00:00
encoding.h kernel/: fix repeated words in comments 2020-10-16 11:11:19 -07:00
kcsan-test.c kcsan: Test support for compound instrumentation 2020-08-24 15:09:58 -07:00
kcsan.h kcsan, debugfs: Move debugfs file creation out of early init 2021-05-11 14:47:33 +02:00
report.c kcsan: Optimize debugfs stats counters 2020-08-24 15:10:23 -07:00
selftest.c kcsan: Use pr_fmt for consistency 2020-08-24 15:10:23 -07:00