On Thu, 18 Jun 2020 at 18:04, Richard Henderson <richard.hender...@linaro.org> wrote: > First, this could definitely be delayed to the follow-on linux-user patch set. > > Second, in the linux-user patch set, I decode the syndrome data to determine > what kind of segv to deliver for MTE synchronous faults. It would be easy to > extend that just a little to notice the usual syndrome for unaligned accesses. > Which may be less confusing than abusing the v7m exception code?
Yeah, if we're going to look at syndrome data anyway that might be clearer. The other thing that really it would be nice if we were able to feed through (via syndrome info or otherwise) is the difference between SIGSEGV with si_code == SEGV_ACCERR vs SEGV_MAPERR. At the moment handle_cpu_signal() knows the difference, but it doesn't have a way to pass this through to tlb_fill, and then cpu_loop() has to make up a si_code when it gets the EXCP_DATA_ABORT. I mention this mostly in case it affects how you want to design how you treat alignment and MTE faults -- it might be that the si_code stuff is better dealt with entirely differently. thanks -- PMM
