Perf can record user stack data in response to a synchronous request, such
as a tracepoint firing. If this happens under set_fs(KERNEL_DS), then we
end up reading user stack data using __copy_from_user_inatomic() under
set_fs(KERNEL_DS). I think this conflicts with the intention of using
set_fs(KERNEL_DS). And it is explicitly forbidden by hardware on ARM64
when both CONFIG_ARM64_UAO and CONFIG_ARM64_PAN are used.

So fix this by forcing USER_DS when recording user stack data.

Signed-off-by: Yabin Cui <yab...@google.com>
---
 kernel/events/core.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 2a62b96600ad..9bc047421e75 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -5948,6 +5948,7 @@ perf_output_sample_ustack(struct perf_output_handle 
*handle, u64 dump_size,
                unsigned long sp;
                unsigned int rem;
                u64 dyn_size;
+               mm_segment_t fs;
 
                /*
                 * We dump:
@@ -5965,7 +5966,10 @@ perf_output_sample_ustack(struct perf_output_handle 
*handle, u64 dump_size,
 
                /* Data. */
                sp = perf_user_stack_pointer(regs);
+               fs = get_fs();
+               set_fs(USER_DS);
                rem = __output_copy_user(handle, (void *) sp, dump_size);
+               set_fs(fs);
                dyn_size = dump_size - rem;
 
                perf_output_skip(handle, rem);
-- 
2.19.0.rc0.228.g281dcd1b4d0-goog

Reply via email to