On Mon, Nov 02, 2020 at 11:44:45AM +0000, Catalin Marinas wrote: > On Fri, Oct 30, 2020 at 04:13:53PM +0000, Will Deacon wrote: > > On Fri, Oct 30, 2020 at 11:18:47AM +0000, Catalin Marinas wrote: > > > On Thu, Oct 29, 2020 at 10:20:48PM +0000, Will Deacon wrote: > > > > This means that if the first 32-bit-capable core is onlined late, then > > > > it will only get the base capabilities, but I think that's fine and > > > > consistent with our overall handling of hwcaps (which cannot appear > > > > dynamically to userspace). > > > > > > Yes but such bare 32-bit mode is entirely useless and I don't think we > > > should even pretend we have 32-bit. The compat hwcaps here would be > > > "half thumb fastmult edsp tls idiva idivt lpae evtstrm", statically > > > filled in. It's missing major bits like "vfp" and "neon" which are > > > necessary for the general purpose 32-bit EABI. > > > > So? If we found such a CPU during boot, would we refuse to online it because > > we consider it "entirely useless"? No! > > We _do_ online it but as a 64-bit only CPU if there were no early 32-bit > CPUs since we are not updating the compat hwcaps anyway (and that's > handled automatically by WEAK_LOCAL_CPU_FEATURE; we do this in a few > places already). > > > That said, given that it's _very_ > > likely for the late CPUs to support vfp and neon, we could set those caps > > speculatively if the 64-bit cores have fpsimd (late onlining would be > > prevented for cores lacking those). Does the architecture allow you to > > implement both AArch64 and AArch32 at EL0, but only have fpsimd for AArch64? > > Probably not but I don't want to butcher the cpufeature support further > and have compat hwcaps derived from ID_AA64* regs. I find this hack even > worse and I'd rather live with the partial hwcap information (and hope > user space doesn't read hwcaps anyway ;)). > > I don't see why we should change this code further when the requirement > to the mobile vendors is to simply allow a 32-bit CPU to come up early. > > > > As I said above, I think we would be even more inconsistent w.r.t. > > > HWCAPs if we require at least one early AArch32-capable CPU, otherwise > > > don't expose 32-bit at all. I don't see what we gain by allowing all > > > 32-bit CPUs to come in late, other than maybe saving an entry in the > > > cpufeature array. > > > > It's a combination of there not being a good reason to prevent the > > late-onlining and not gaining anything from the additional feature (I've > > already shown why it doesn't help with the vast majority of callsites). > > I underlined above, this is not about preventing late onlining, only > preventing late 32-bit support. Late AArch32-capable CPUs will be > onlined just fine, only that if we haven't got any prior 32-bit CPU, we > no longer report the feature and the sysfs mask. Ok. Then we're in agreement about not preventing late-onlining. The problem then is that the existing 32-bit EL0 capability is a SYSTEM cap so even with your diff, we still have an issue if you boot on the CPUs that support 32-bit and then try to online a 64-bit-only core (it will fail). So I think we do need my changes to the existing cap, but perhaps we could return false from system_supports_32bit_el0() until we've actually seen a 32-bit capable core. That way you would keep the existing behaviour on TX2, and we wouldn't get any unusual late-onlining failures. I've hacked something together that seems to work, so I'll clean it up and post it tomorrow. I've spotted a couple of pre-existing issues at the same time, so I need to fix those first (WEAK_LOCAL_CPU_FEATURE doesn't set the cap for late CPUs and failed onlining causes RCU stalls). Will
