Re: [PATCH v5 6/6] powerpc64/bpf: Additional NVR handling for bpf_throw

[Hari Bathini]

On 24/01/26 1:53 pm, bot+bpf-ci@xxxxxxxxxx wrote:
> > diff --git a/arch/powerpc/net/bpf_jit_comp64.c b/arch/powerpc/net/bpf_jit_comp64.c
> > index db121b1404fe..17de8b53a962 100644
> > --- a/arch/powerpc/net/bpf_jit_comp64.c
> > +++ b/arch/powerpc/net/bpf_jit_comp64.c
>
> [ ... ]
>
> > +/*
> > + * For exception boundary & exception_cb progs:
> > + *     return increased size to accommodate additional NVRs.
> > + */
> > +static int bpf_jit_stack_size(struct codegen_context *ctx)
> > +{
> > +	return ctx->exception_boundary || ctx->exception_cb ?
> > +					BPF_PPC_EXC_STACKFRAME :
> > +					BPF_PPC_STACKFRAME;
> > +}
>
> [ ... ]
>
> > @@ -198,7 +249,19 @@ void bpf_jit_build_prologue(u32 *image, struct codegen_context *ctx)
> >   			EMIT(PPC_RAW_STD(_R0, _R1, PPC_LR_STKOFF));
> >   		}
> >
> > -		EMIT(PPC_RAW_STDU(_R1, _R1, -(BPF_PPC_STACKFRAME + ctx->stack_size)));
> > +		EMIT(PPC_RAW_STDU(_R1, _R1,
> > +				-(bpf_jit_stack_size(ctx) + ctx->stack_size)));
> > +	}
>
> [ ... ]
>
> > @@ -248,9 +311,19 @@ static void bpf_jit_emit_common_epilogue(u32 *image, struct codegen_context *ctx)
> >   		EMIT(PPC_RAW_LD(bpf_to_ppc(ARENA_VM_START), _R1,
> >   				bpf_jit_stack_offsetof(ctx, bpf_to_ppc(ARENA_VM_START))));
> >
> > +	if (ctx->exception_cb) {
> > +		/*
> > +		 * Recover additionally saved non volatile registers from stack
> > +		 * frame of exception boundary program.
> > +		 */
> > +		for (i = _R14; i <= _R25; i++)
> > +			EMIT(PPC_RAW_LD(i, _R1, bpf_jit_stack_offsetof(ctx, i)));
> > +	}
> > +
> >   	/* Tear down our stack frame */
> >   	if (bpf_has_stack_frame(ctx)) {
> > -		EMIT(PPC_RAW_ADDI(_R1, _R1, BPF_PPC_STACKFRAME + ctx->stack_size));
> > +		EMIT(PPC_RAW_ADDI(_R1, _R1, bpf_jit_stack_size(ctx) + ctx->stack_size));
>
> For the exception callback case, doesn't this use the wrong stack_size?
>
> In bpf_jit_build_prologue(), the exception callback does not create its own
> frame - it sets R1 to the boundary's frame pointer via EMIT(PPC_RAW_MR(_R1,
> _R5)). However here in the epilogue, the frame teardown uses ctx->stack_size
> which is the callback's stack depth, not the boundary's.
>
> The boundary program creates its frame with:
>      STDU R1, R1, -(bpf_jit_stack_size(boundary_ctx) + boundary_stack_size)
>
> But the callback tears it down with:
>      ADDI R1, R1, bpf_jit_stack_size(callback_ctx) + callback_stack_size
>
> Since each BPF subprogram can have a different stack_depth (set per-subprog
> in kernel/bpf/verifier.c via func[i]->aux->stack_depth), if the exception
> callback has a different stack depth than the boundary, R1 will end up at
> the wrong location after this ADDI.
>
> After the wrong R1 adjustment, the LR restore below loads from R1+16 which
> is now pointing to the wrong memory location, potentially causing a return
> to an invalid address.
>
> Would it be correct to store the boundary's stack_size somewhere accessible
> to the callback, or ensure the callback uses the same stack layout as the
> boundary?
>
> > +
> >   		if (ctx->seen & SEEN_FUNC || ctx->exception_cb) {
> >   			EMIT(PPC_RAW_LD(_R0, _R1, PPC_LR_STKOFF));
> >   			EMIT(PPC_RAW_MTLR(_R0));
>
>
> ---
> AI reviewed your patch. Please fix the bug or email reply why it's not a bug.

I will post the fix for this issue as a follow-up.

Thanks
Hari