Re: [PATCH 6.12 117/337] tracing/mmiotrace: Add NULL check for mmio_trace_array in logging functions

On Sat, 8 Aug 2026 22:53:31 +0530
Harshit Mogalapalli <[email protected]> wrote:

> I ran an AI-assisted backport review and then checked the 6.12.y code. I
> think this backport is missing a small prerequisite.
> 
> Upstream's wrappers pass tr directly to __trace_mmiotrace_rw() and
> __trace_mmiotrace_map().Each function checks tr before dereferencing it:
> 
> if (!tr)
> 	return;
> buffer = tr->array_buffer.buffer;
> 
> 
> The upstream fix follows commit: 6936298393d8 ("tracing/mmiotrace: 
> Remove reference to unused per CPU data pointer"), which removes the 
> unused data argument.
> 
> 6.12.y does not have that commit, so both functions still have that 
> argument and their wrappers compute it first:
> 
> struct trace_array *tr = mmio_trace_array;
> struct trace_array_cpu *data =
> 	per_cpu_ptr(tr->array_buffer.data, smp_processor_id());
> __trace_mmiotrace_rw(tr, data, rw);
> 
> 
> Consequently, if mmio_trace_array is NULL, the dereference in
> per_cpu_ptr(tr->array_buffer.data, ...) occurs before either function's 
> new NULL check.
> 
> I think 6.12.y needs commit: 6936298393d8 ("tracing/mmiotrace: Remove
> reference to unused per CPU data pointer") before this fix.
> 
> Thoughts?
> 

Yeah, it looks like the kernel could crash for the same reasons without
commit 6936298393d8.

Greg, care to backport that one too?

Thanks,

-- Steve