On Mon, 23 Nov 2020 16:11:30 +1100 David Gibson <da...@gibson.dropbear.id.au> wrote:
> On Sat, Nov 21, 2020 at 12:42:03AM +0100, Greg Kurz wrote: > > When it comes to resetting the compat mode of the vCPUS, there are > > two situations to consider: > > (1) machine reset should set the compat mode back to the machine default, > > ie. spapr->max_compat_pvr > > (2) hot plugged vCPUs should set their compat mode to mach the boot vCPU, > > ie. POWERPC_CPU(first_cpu)->compat_pvr > > > > This is currently handled in two separate places: globally for all vCPUs > > from the machine reset code for (1) and for each thread of a core from > > the hotplug path for (2). > > > > Since the machine reset code already resets all vCPUs, starting with boot > > vCPU, consolidate the logic in spapr_reset_vcpu(). Special case the boot > > vCPU so that it resets its compat mode back to the machine default. Any > > other vCPU just need to match the compat mode of the boot vCPU. > > > > Failing to set the compat mode during machine reset is a fatal error, > > but not for hot plugged vCPUs. This is arguable because if we've been > > able to set the boot vCPU compat mode at CAS or during machine reset, > > it should definitely not fail for other vCPUs. Since spapr_reset_vcpu() > > already has a fatal error path for kvm_check_mmu() failures, do the > > same for ppc_set_compat(). > > > > This gets rid of an error path in spapr_core_plug(). It will allow > > further simplifications. > > > > Signed-off-by: Greg Kurz <gr...@kaod.org> > > --- > > hw/ppc/spapr.c | 16 ---------------- > > hw/ppc/spapr_cpu_core.c | 13 +++++++++++++ > > 2 files changed, 13 insertions(+), 16 deletions(-) > > > > diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c > > index f58f77389e8e..da7586f548df 100644 > > --- a/hw/ppc/spapr.c > > +++ b/hw/ppc/spapr.c > > @@ -1606,8 +1606,6 @@ static void spapr_machine_reset(MachineState *machine) > > spapr_ovec_cleanup(spapr->ov5_cas); > > spapr->ov5_cas = spapr_ovec_new(); > > > > - ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal); > > - > > /* > > * This is fixing some of the default configuration of the XIVE > > * devices. To be called after the reset of the machine devices. > > @@ -3785,20 +3783,6 @@ static void spapr_core_plug(HotplugHandler > > *hotplug_dev, DeviceState *dev, > > > > core_slot->cpu = OBJECT(dev); > > > > - /* > > - * Set compatibility mode to match the boot CPU, which was either set > > - * by the machine reset code or by CAS. > > - */ > > - if (hotplugged) { > > - for (i = 0; i < cc->nr_threads; i++) { > > - if (ppc_set_compat(core->threads[i], > > - POWERPC_CPU(first_cpu)->compat_pvr, > > - errp) < 0) { > > - return; > > - } > > - } > > - } > > - > > if (smc->pre_2_10_has_unused_icps) { > > for (i = 0; i < cc->nr_threads; i++) { > > cs = CPU(core->threads[i]); > > diff --git a/hw/ppc/spapr_cpu_core.c b/hw/ppc/spapr_cpu_core.c > > index 2f7dc3c23ded..17741a3fb77f 100644 > > --- a/hw/ppc/spapr_cpu_core.c > > +++ b/hw/ppc/spapr_cpu_core.c > > @@ -27,6 +27,7 @@ > > > > static void spapr_reset_vcpu(PowerPCCPU *cpu) > > { > > + PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu); > > CPUState *cs = CPU(cpu); > > CPUPPCState *env = &cpu->env; > > PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu); > > @@ -69,6 +70,18 @@ static void spapr_reset_vcpu(PowerPCCPU *cpu) > > kvm_check_mmu(cpu, &error_fatal); > > > > spapr_irq_cpu_intc_reset(spapr, cpu); > > + > > + /* > > + * The boot CPU is only reset during machine reset : reset its > > + * compatibility mode to the machine default. For other CPUs, > > + * either cold plugged or hot plugged, set the compatibility mode > > + * to match the boot CPU, which was either set by the machine reset > > + * code or by CAS. > > + */ > > + ppc_set_compat(cpu, > > + cpu == first_ppc_cpu ? > > + spapr->max_compat_pvr : first_ppc_cpu->compat_pvr, > > + &error_fatal); > > This assumes that when it is called for a non-boot CPU, it has already > been called for the boot CPU.. Are we certain that's guaranteed by > the sequence of reset calls during a full machine reset? > This happens to be the case. Basically because the boot CPU core is created (including registering its reset handler) first and qemu_devices_reset() calls handlers in the same order they were registered. > > } > > > > void spapr_cpu_set_entry_state(PowerPCCPU *cpu, target_ulong nip, >
pgpL0kZEpnRe9.pgp
Description: OpenPGP digital signature