Hi Cédric, Peter, > -----Original Message----- > From: Peter Maydell <peter.mayd...@linaro.org> > Sent: Friday, May 9, 2025 11:28 PM > To: Cédric Le Goater <c...@redhat.com> > Cc: qemu-...@nongnu.org; qemu-devel@nongnu.org; Steven Lee > <steven_...@aspeedtech.com> > Subject: Re: [PULL 22/23] tests/function/aspeed: Add functional test for > ast2700fc > > On Fri, 9 May 2025 at 15:32, Cédric Le Goater <c...@redhat.com> wrote: > > In the ast2700fc machine and the fby35, each SoC has its own memory > > region and, since there is no need for the global system memory > > region, it's empty. But in some places, like > > rom_check_and_register_reset(), it's a problem for the flattened > > global view. > > Generally speaking, board models should set things up so that the "main" SoC > for the board uses the main system memory MR, rather than creating its own > empty MR for that purpose, I think. > > But separately from that, it's weird to get into flatview_lookup with an empty > FlatView here. We called memory_region_find(), which is a wrapper around > memory_region_find_rcu(). That function doesn't make assumptions about > where the MR it's passed is from: it walks up the tree of MRs to find the root > MR, it looks up the AddressSpace for that root MR, and then it gets the > FlatView for that AS and calls flatview_lookup(). If the FlatView we end up > with > doesn't have the MR in it, what's happened ? >
I also realized this was caused by the CA35 memory not being part of the system memory, while fixing the 2700FC Clang sanitizer issue last Wednesday. To fix it, I added the CA35 memory region to the system memory: memory_region_add_subregion(get_system_memory(), 0, &s->ca35_memory); Patch for reference: https://patchwork.kernel.org/project/qemu-devel/patch/20250507101005.1474823-2-steven_...@aspeedtech.com/ This seems to resolve the issue, though I'm not sure whether it's a proper fix or just a workaround. Regards, Steven
