qemu-e2k/target/ppc/machine.c

801 lines
23 KiB
C
Raw Normal View History

#include "qemu/osdep.h"
#include "qemu-common.h"
#include "cpu.h"
#include "exec/exec-all.h"
#include "hw/hw.h"
#include "hw/boards.h"
#include "sysemu/kvm.h"
#include "helper_regs.h"
#include "mmu-hash64.h"
#include "migration/cpu.h"
ppc: Rework CPU compatibility testing across migration Migrating between different CPU versions is a bit complicated for ppc. A long time ago, we ensured identical CPU versions at either end by checking the PVR had the same value. However, this breaks under KVM HV, because we always have to use the host's PVR - it's not virtualized. That would mean we couldn't migrate between hosts with different PVRs, even if the CPUs are close enough to compatible in practice (sometimes identical cores with different surrounding logic have different PVRs, so this happens in practice quite often). So, we removed the PVR check, but instead checked that several flags indicating supported instructions matched. This turns out to be a bad idea, because those instruction masks are not architected information, but essentially a TCG implementation detail. So changes to qemu internal CPU modelling can break migration - this happened between qemu-2.6 and qemu-2.7. That was addressed by 146c11f1 "target-ppc: Allow eventual removal of old migration mistakes". Now, verification of CPU compatibility across a migration basically doesn't happen. We simply ignore the PVR of the incoming migration, and hope the cpu on the destination is close enough to work. Now that we've cleaned up handling of processor compatibility modes for pseries machine type, we can do better. For new machine types (pseries-2.10+) We allow migration if: * The source and destination PVRs are for the same type of CPU, as determined by CPU class's pvr_match function OR * When the source was in a compatibility mode, and the destination CPU supports the same compatibility mode For older machine types we retain the existing behaviour - current CAS code will usually set a compat mode which would break backwards migration if we made them use the new behaviour. [Fixed from an earlier version by Greg Kurz]. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Greg Kurz <groug@kaod.org> Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com> Tested-by: Andrea Bolognani <abologna@redhat.com>
2017-06-02 04:26:11 +02:00
#include "qapi/error.h"
target/ppc: 'PVR != host PVR' in KVM_SET_SREGS workaround Commit d5fc133eed ("ppc: Rework CPU compatibility testing across migration") changed the way cpu_post_load behaves with the PVR setting, causing an unexpected bug in KVM-HV migrations between hosts that are compatible (POWER8 and POWER8E, for example). Even with pvr_match() returning true, the guest freezes right after cpu_post_load. The reason is that the guest kernel can't handle a different PVR value other that the running host in KVM_SET_SREGS. In [1] it was discussed the possibility of a new KVM capability that would indicate that the guest kernel can handle a different PVR in KVM_SET_SREGS. Even if such feature is implemented, there is still the problem with older kernels that will not have this capability and will fail to migrate. This patch implements a workaround for that scenario. If running with KVM, check if the guest kernel does not have the capability (named here as 'cap_ppc_pvr_compat'). If it doesn't, calls kvmppc_is_pr() to see if the guest is running in KVM-HV. If all this happens, set env->spr[SPR_PVR] to the same value as the current host PVR. This ensures that we allow migrations with 'close enough' PVRs to still work in KVM-HV but also makes the code ready for this new KVM capability when it is done. A new function called 'kvmppc_pvr_workaround_required' was created to encapsulate the conditions said above and to avoid calling too many kvm.c internals inside cpu_post_load. [1] https://lists.gnu.org/archive/html/qemu-ppc/2017-06/msg00503.html Signed-off-by: Daniel Henrique Barboza <danielhb@linux.vnet.ibm.com> [dwg: Fix for the case of using TCG on a PPC host] Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2017-08-09 22:43:46 +02:00
#include "kvm_ppc.h"
static int cpu_load_old(QEMUFile *f, void *opaque, int version_id)
{
PowerPCCPU *cpu = opaque;
CPUPPCState *env = &cpu->env;
unsigned int i, j;
target_ulong sdr1;
uint32_t fpscr;
#if defined(TARGET_PPC64)
int32_t slb_nr;
#endif
target_ulong xer;
for (i = 0; i < 32; i++)
qemu_get_betls(f, &env->gpr[i]);
#if !defined(TARGET_PPC64)
for (i = 0; i < 32; i++)
qemu_get_betls(f, &env->gprh[i]);
#endif
qemu_get_betls(f, &env->lr);
qemu_get_betls(f, &env->ctr);
for (i = 0; i < 8; i++)
qemu_get_be32s(f, &env->crf[i]);
qemu_get_betls(f, &xer);
cpu_write_xer(env, xer);
qemu_get_betls(f, &env->reserve_addr);
qemu_get_betls(f, &env->msr);
for (i = 0; i < 4; i++)
qemu_get_betls(f, &env->tgpr[i]);
for (i = 0; i < 32; i++) {
union {
float64 d;
uint64_t l;
} u;
u.l = qemu_get_be64(f);
target/ppc: move FP and VMX registers into aligned vsr register array The VSX register array is a block of 64 128-bit registers where the first 32 registers consist of the existing 64-bit FP registers extended to 128-bit using new VSR registers, and the last 32 registers are the VMX 128-bit registers as show below: 64-bit 64-bit +--------------------+--------------------+ | FP0 | | VSR0 +--------------------+--------------------+ | FP1 | | VSR1 +--------------------+--------------------+ | ... | ... | ... +--------------------+--------------------+ | FP30 | | VSR30 +--------------------+--------------------+ | FP31 | | VSR31 +--------------------+--------------------+ | VMX0 | VSR32 +-----------------------------------------+ | VMX1 | VSR33 +-----------------------------------------+ | ... | ... +-----------------------------------------+ | VMX30 | VSR62 +-----------------------------------------+ | VMX31 | VSR63 +-----------------------------------------+ In order to allow for future conversion of VSX instructions to use TCG vector operations, recreate the same layout using an aligned version of the existing vsr register array. Since the old fpr and avr register arrays are removed, the existing callers must also be updated to use the correct offset in the vsr register array. This also includes switching the relevant VMState fields over to using subarrays to make sure that migration is preserved. Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Acked-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-01-02 10:14:22 +01:00
*cpu_fpr_ptr(env, i) = u.d;
}
qemu_get_be32s(f, &fpscr);
env->fpscr = fpscr;
qemu_get_sbe32s(f, &env->access_type);
#if defined(TARGET_PPC64)
qemu_get_betls(f, &env->spr[SPR_ASR]);
qemu_get_sbe32s(f, &slb_nr);
#endif
qemu_get_betls(f, &sdr1);
for (i = 0; i < 32; i++)
qemu_get_betls(f, &env->sr[i]);
for (i = 0; i < 2; i++)
for (j = 0; j < 8; j++)
qemu_get_betls(f, &env->DBAT[i][j]);
for (i = 0; i < 2; i++)
for (j = 0; j < 8; j++)
qemu_get_betls(f, &env->IBAT[i][j]);
qemu_get_sbe32s(f, &env->nb_tlb);
qemu_get_sbe32s(f, &env->tlb_per_way);
qemu_get_sbe32s(f, &env->nb_ways);
qemu_get_sbe32s(f, &env->last_way);
qemu_get_sbe32s(f, &env->id_tlbs);
qemu_get_sbe32s(f, &env->nb_pids);
if (env->tlb.tlb6) {
// XXX assumes 6xx
for (i = 0; i < env->nb_tlb; i++) {
qemu_get_betls(f, &env->tlb.tlb6[i].pte0);
qemu_get_betls(f, &env->tlb.tlb6[i].pte1);
qemu_get_betls(f, &env->tlb.tlb6[i].EPN);
}
}
for (i = 0; i < 4; i++)
qemu_get_betls(f, &env->pb[i]);
for (i = 0; i < 1024; i++)
qemu_get_betls(f, &env->spr[i]);
if (!cpu->vhyp) {
ppc_store_sdr1(env, sdr1);
}
qemu_get_be32s(f, &env->vscr);
qemu_get_be64s(f, &env->spe_acc);
qemu_get_be32s(f, &env->spe_fscr);
qemu_get_betls(f, &env->msr_mask);
qemu_get_be32s(f, &env->flags);
qemu_get_sbe32s(f, &env->error_code);
qemu_get_be32s(f, &env->pending_interrupts);
qemu_get_be32s(f, &env->irq_input_state);
for (i = 0; i < POWERPC_EXCP_NB; i++)
qemu_get_betls(f, &env->excp_vectors[i]);
qemu_get_betls(f, &env->excp_prefix);
qemu_get_betls(f, &env->ivor_mask);
qemu_get_betls(f, &env->ivpr_mask);
qemu_get_betls(f, &env->hreset_vector);
qemu_get_betls(f, &env->nip);
qemu_get_betls(f, &env->hflags);
qemu_get_betls(f, &env->hflags_nmsr);
qemu_get_sbe32(f); /* Discard unused mmu_idx */
qemu_get_sbe32(f); /* Discard unused power_mode */
/* Recompute mmu indices */
hreg_compute_mem_idx(env);
return 0;
}
static int get_avr(QEMUFile *f, void *pv, size_t size,
const VMStateField *field)
{
ppc_avr_t *v = pv;
v->u64[0] = qemu_get_be64(f);
v->u64[1] = qemu_get_be64(f);
return 0;
}
static int put_avr(QEMUFile *f, void *pv, size_t size,
const VMStateField *field, QJSON *vmdesc)
{
ppc_avr_t *v = pv;
qemu_put_be64(f, v->u64[0]);
qemu_put_be64(f, v->u64[1]);
return 0;
}
static const VMStateInfo vmstate_info_avr = {
.name = "avr",
.get = get_avr,
.put = put_avr,
};
#define VMSTATE_AVR_ARRAY_V(_f, _s, _n, _v) \
target/ppc: move FP and VMX registers into aligned vsr register array The VSX register array is a block of 64 128-bit registers where the first 32 registers consist of the existing 64-bit FP registers extended to 128-bit using new VSR registers, and the last 32 registers are the VMX 128-bit registers as show below: 64-bit 64-bit +--------------------+--------------------+ | FP0 | | VSR0 +--------------------+--------------------+ | FP1 | | VSR1 +--------------------+--------------------+ | ... | ... | ... +--------------------+--------------------+ | FP30 | | VSR30 +--------------------+--------------------+ | FP31 | | VSR31 +--------------------+--------------------+ | VMX0 | VSR32 +-----------------------------------------+ | VMX1 | VSR33 +-----------------------------------------+ | ... | ... +-----------------------------------------+ | VMX30 | VSR62 +-----------------------------------------+ | VMX31 | VSR63 +-----------------------------------------+ In order to allow for future conversion of VSX instructions to use TCG vector operations, recreate the same layout using an aligned version of the existing vsr register array. Since the old fpr and avr register arrays are removed, the existing callers must also be updated to use the correct offset in the vsr register array. This also includes switching the relevant VMState fields over to using subarrays to make sure that migration is preserved. Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Acked-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-01-02 10:14:22 +01:00
VMSTATE_SUB_ARRAY(_f, _s, 32, _n, _v, vmstate_info_avr, ppc_avr_t)
#define VMSTATE_AVR_ARRAY(_f, _s, _n) \
VMSTATE_AVR_ARRAY_V(_f, _s, _n, 0)
target/ppc: move FP and VMX registers into aligned vsr register array The VSX register array is a block of 64 128-bit registers where the first 32 registers consist of the existing 64-bit FP registers extended to 128-bit using new VSR registers, and the last 32 registers are the VMX 128-bit registers as show below: 64-bit 64-bit +--------------------+--------------------+ | FP0 | | VSR0 +--------------------+--------------------+ | FP1 | | VSR1 +--------------------+--------------------+ | ... | ... | ... +--------------------+--------------------+ | FP30 | | VSR30 +--------------------+--------------------+ | FP31 | | VSR31 +--------------------+--------------------+ | VMX0 | VSR32 +-----------------------------------------+ | VMX1 | VSR33 +-----------------------------------------+ | ... | ... +-----------------------------------------+ | VMX30 | VSR62 +-----------------------------------------+ | VMX31 | VSR63 +-----------------------------------------+ In order to allow for future conversion of VSX instructions to use TCG vector operations, recreate the same layout using an aligned version of the existing vsr register array. Since the old fpr and avr register arrays are removed, the existing callers must also be updated to use the correct offset in the vsr register array. This also includes switching the relevant VMState fields over to using subarrays to make sure that migration is preserved. Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Acked-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-01-02 10:14:22 +01:00
static int get_fpr(QEMUFile *f, void *pv, size_t size,
const VMStateField *field)
{
ppc_vsr_t *v = pv;
v->u64[0] = qemu_get_be64(f);
return 0;
}
static int put_fpr(QEMUFile *f, void *pv, size_t size,
const VMStateField *field, QJSON *vmdesc)
{
ppc_vsr_t *v = pv;
qemu_put_be64(f, v->u64[0]);
return 0;
}
static const VMStateInfo vmstate_info_fpr = {
.name = "fpr",
.get = get_fpr,
.put = put_fpr,
};
#define VMSTATE_FPR_ARRAY_V(_f, _s, _n, _v) \
VMSTATE_SUB_ARRAY(_f, _s, 0, _n, _v, vmstate_info_fpr, ppc_vsr_t)
#define VMSTATE_FPR_ARRAY(_f, _s, _n) \
VMSTATE_FPR_ARRAY_V(_f, _s, _n, 0)
static int get_vsr(QEMUFile *f, void *pv, size_t size,
const VMStateField *field)
{
ppc_vsr_t *v = pv;
v->u64[1] = qemu_get_be64(f);
return 0;
}
static int put_vsr(QEMUFile *f, void *pv, size_t size,
const VMStateField *field, QJSON *vmdesc)
{
ppc_vsr_t *v = pv;
qemu_put_be64(f, v->u64[1]);
return 0;
}
static const VMStateInfo vmstate_info_vsr = {
.name = "vsr",
.get = get_vsr,
.put = put_vsr,
};
#define VMSTATE_VSR_ARRAY_V(_f, _s, _n, _v) \
VMSTATE_SUB_ARRAY(_f, _s, 0, _n, _v, vmstate_info_vsr, ppc_vsr_t)
#define VMSTATE_VSR_ARRAY(_f, _s, _n) \
VMSTATE_VSR_ARRAY_V(_f, _s, _n, 0)
static bool cpu_pre_2_8_migration(void *opaque, int version_id)
{
PowerPCCPU *cpu = opaque;
return cpu->pre_2_8_migration;
}
#if defined(TARGET_PPC64)
static bool cpu_pre_3_0_migration(void *opaque, int version_id)
{
PowerPCCPU *cpu = opaque;
return cpu->pre_3_0_migration;
}
#endif
static int cpu_pre_save(void *opaque)
{
PowerPCCPU *cpu = opaque;
CPUPPCState *env = &cpu->env;
int i;
uint64_t insns_compat_mask =
PPC_INSNS_BASE | PPC_ISEL | PPC_STRING | PPC_MFTB
| PPC_FLOAT | PPC_FLOAT_FSEL | PPC_FLOAT_FRES
| PPC_FLOAT_FSQRT | PPC_FLOAT_FRSQRTE | PPC_FLOAT_FRSQRTES
| PPC_FLOAT_STFIWX | PPC_FLOAT_EXT
| PPC_CACHE | PPC_CACHE_ICBI | PPC_CACHE_DCBZ
| PPC_MEM_SYNC | PPC_MEM_EIEIO | PPC_MEM_TLBIE | PPC_MEM_TLBSYNC
| PPC_64B | PPC_64BX | PPC_ALTIVEC
| PPC_SEGMENT_64B | PPC_SLBI | PPC_POPCNTB | PPC_POPCNTWD;
uint64_t insns_compat_mask2 = PPC2_VSX | PPC2_VSX207 | PPC2_DFP | PPC2_DBRX
| PPC2_PERM_ISA206 | PPC2_DIVE_ISA206
| PPC2_ATOMIC_ISA206 | PPC2_FP_CVT_ISA206
| PPC2_FP_TST_ISA206 | PPC2_BCTAR_ISA207
| PPC2_LSQ_ISA207 | PPC2_ALTIVEC_207
| PPC2_ISA205 | PPC2_ISA207S | PPC2_FP_CVT_S64 | PPC2_TM;
env->spr[SPR_LR] = env->lr;
env->spr[SPR_CTR] = env->ctr;
env->spr[SPR_XER] = cpu_read_xer(env);
#if defined(TARGET_PPC64)
env->spr[SPR_CFAR] = env->cfar;
#endif
env->spr[SPR_BOOKE_SPEFSCR] = env->spe_fscr;
for (i = 0; (i < 4) && (i < env->nb_BATs); i++) {
env->spr[SPR_DBAT0U + 2*i] = env->DBAT[0][i];
env->spr[SPR_DBAT0U + 2*i + 1] = env->DBAT[1][i];
env->spr[SPR_IBAT0U + 2*i] = env->IBAT[0][i];
env->spr[SPR_IBAT0U + 2*i + 1] = env->IBAT[1][i];
}
for (i = 0; (i < 4) && ((i+4) < env->nb_BATs); i++) {
env->spr[SPR_DBAT4U + 2*i] = env->DBAT[0][i+4];
env->spr[SPR_DBAT4U + 2*i + 1] = env->DBAT[1][i+4];
env->spr[SPR_IBAT4U + 2*i] = env->IBAT[0][i+4];
env->spr[SPR_IBAT4U + 2*i + 1] = env->IBAT[1][i+4];
}
/* Hacks for migration compatibility between 2.6, 2.7 & 2.8 */
if (cpu->pre_2_8_migration) {
/* Mask out bits that got added to msr_mask since the versions
* which stupidly included it in the migration stream. */
target_ulong metamask = 0
#if defined(TARGET_PPC64)
| (1ULL << MSR_TS0)
| (1ULL << MSR_TS1)
#endif
;
cpu->mig_msr_mask = env->msr_mask & ~metamask;
cpu->mig_insns_flags = env->insns_flags & insns_compat_mask;
/* CPU models supported by old machines all have PPC_MEM_TLBIE,
* so we set it unconditionally to allow backward migration from
* a POWER9 host to a POWER8 host.
*/
cpu->mig_insns_flags |= PPC_MEM_TLBIE;
cpu->mig_insns_flags2 = env->insns_flags2 & insns_compat_mask2;
cpu->mig_nb_BATs = env->nb_BATs;
}
if (cpu->pre_3_0_migration) {
if (cpu->hash64_opts) {
cpu->mig_slb_nr = cpu->hash64_opts->slb_size;
}
}
return 0;
}
ppc: Rework CPU compatibility testing across migration Migrating between different CPU versions is a bit complicated for ppc. A long time ago, we ensured identical CPU versions at either end by checking the PVR had the same value. However, this breaks under KVM HV, because we always have to use the host's PVR - it's not virtualized. That would mean we couldn't migrate between hosts with different PVRs, even if the CPUs are close enough to compatible in practice (sometimes identical cores with different surrounding logic have different PVRs, so this happens in practice quite often). So, we removed the PVR check, but instead checked that several flags indicating supported instructions matched. This turns out to be a bad idea, because those instruction masks are not architected information, but essentially a TCG implementation detail. So changes to qemu internal CPU modelling can break migration - this happened between qemu-2.6 and qemu-2.7. That was addressed by 146c11f1 "target-ppc: Allow eventual removal of old migration mistakes". Now, verification of CPU compatibility across a migration basically doesn't happen. We simply ignore the PVR of the incoming migration, and hope the cpu on the destination is close enough to work. Now that we've cleaned up handling of processor compatibility modes for pseries machine type, we can do better. For new machine types (pseries-2.10+) We allow migration if: * The source and destination PVRs are for the same type of CPU, as determined by CPU class's pvr_match function OR * When the source was in a compatibility mode, and the destination CPU supports the same compatibility mode For older machine types we retain the existing behaviour - current CAS code will usually set a compat mode which would break backwards migration if we made them use the new behaviour. [Fixed from an earlier version by Greg Kurz]. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Greg Kurz <groug@kaod.org> Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com> Tested-by: Andrea Bolognani <abologna@redhat.com>
2017-06-02 04:26:11 +02:00
/*
* Determine if a given PVR is a "close enough" match to the CPU
* object. For TCG and KVM PR it would probably be sufficient to
* require an exact PVR match. However for KVM HV the user is
* restricted to a PVR exactly matching the host CPU. The correct way
* to handle this is to put the guest into an architected
* compatibility mode. However, to allow a more forgiving transition
* and migration from before this was widely done, we allow migration
* between sufficiently similar PVRs, as determined by the CPU class's
* pvr_match() hook.
*/
static bool pvr_match(PowerPCCPU *cpu, uint32_t pvr)
{
PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
if (pvr == pcc->pvr) {
return true;
}
return pcc->pvr_match(pcc, pvr);
}
static int cpu_post_load(void *opaque, int version_id)
{
PowerPCCPU *cpu = opaque;
CPUPPCState *env = &cpu->env;
int i;
target_ulong msr;
/*
ppc: Rework CPU compatibility testing across migration Migrating between different CPU versions is a bit complicated for ppc. A long time ago, we ensured identical CPU versions at either end by checking the PVR had the same value. However, this breaks under KVM HV, because we always have to use the host's PVR - it's not virtualized. That would mean we couldn't migrate between hosts with different PVRs, even if the CPUs are close enough to compatible in practice (sometimes identical cores with different surrounding logic have different PVRs, so this happens in practice quite often). So, we removed the PVR check, but instead checked that several flags indicating supported instructions matched. This turns out to be a bad idea, because those instruction masks are not architected information, but essentially a TCG implementation detail. So changes to qemu internal CPU modelling can break migration - this happened between qemu-2.6 and qemu-2.7. That was addressed by 146c11f1 "target-ppc: Allow eventual removal of old migration mistakes". Now, verification of CPU compatibility across a migration basically doesn't happen. We simply ignore the PVR of the incoming migration, and hope the cpu on the destination is close enough to work. Now that we've cleaned up handling of processor compatibility modes for pseries machine type, we can do better. For new machine types (pseries-2.10+) We allow migration if: * The source and destination PVRs are for the same type of CPU, as determined by CPU class's pvr_match function OR * When the source was in a compatibility mode, and the destination CPU supports the same compatibility mode For older machine types we retain the existing behaviour - current CAS code will usually set a compat mode which would break backwards migration if we made them use the new behaviour. [Fixed from an earlier version by Greg Kurz]. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Greg Kurz <groug@kaod.org> Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com> Tested-by: Andrea Bolognani <abologna@redhat.com>
2017-06-02 04:26:11 +02:00
* If we're operating in compat mode, we should be ok as long as
* the destination supports the same compatiblity mode.
*
* Otherwise, however, we require that the destination has exactly
* the same CPU model as the source.
*/
ppc: Rework CPU compatibility testing across migration Migrating between different CPU versions is a bit complicated for ppc. A long time ago, we ensured identical CPU versions at either end by checking the PVR had the same value. However, this breaks under KVM HV, because we always have to use the host's PVR - it's not virtualized. That would mean we couldn't migrate between hosts with different PVRs, even if the CPUs are close enough to compatible in practice (sometimes identical cores with different surrounding logic have different PVRs, so this happens in practice quite often). So, we removed the PVR check, but instead checked that several flags indicating supported instructions matched. This turns out to be a bad idea, because those instruction masks are not architected information, but essentially a TCG implementation detail. So changes to qemu internal CPU modelling can break migration - this happened between qemu-2.6 and qemu-2.7. That was addressed by 146c11f1 "target-ppc: Allow eventual removal of old migration mistakes". Now, verification of CPU compatibility across a migration basically doesn't happen. We simply ignore the PVR of the incoming migration, and hope the cpu on the destination is close enough to work. Now that we've cleaned up handling of processor compatibility modes for pseries machine type, we can do better. For new machine types (pseries-2.10+) We allow migration if: * The source and destination PVRs are for the same type of CPU, as determined by CPU class's pvr_match function OR * When the source was in a compatibility mode, and the destination CPU supports the same compatibility mode For older machine types we retain the existing behaviour - current CAS code will usually set a compat mode which would break backwards migration if we made them use the new behaviour. [Fixed from an earlier version by Greg Kurz]. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Greg Kurz <groug@kaod.org> Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com> Tested-by: Andrea Bolognani <abologna@redhat.com>
2017-06-02 04:26:11 +02:00
#if defined(TARGET_PPC64)
if (cpu->compat_pvr) {
uint32_t compat_pvr = cpu->compat_pvr;
ppc: Rework CPU compatibility testing across migration Migrating between different CPU versions is a bit complicated for ppc. A long time ago, we ensured identical CPU versions at either end by checking the PVR had the same value. However, this breaks under KVM HV, because we always have to use the host's PVR - it's not virtualized. That would mean we couldn't migrate between hosts with different PVRs, even if the CPUs are close enough to compatible in practice (sometimes identical cores with different surrounding logic have different PVRs, so this happens in practice quite often). So, we removed the PVR check, but instead checked that several flags indicating supported instructions matched. This turns out to be a bad idea, because those instruction masks are not architected information, but essentially a TCG implementation detail. So changes to qemu internal CPU modelling can break migration - this happened between qemu-2.6 and qemu-2.7. That was addressed by 146c11f1 "target-ppc: Allow eventual removal of old migration mistakes". Now, verification of CPU compatibility across a migration basically doesn't happen. We simply ignore the PVR of the incoming migration, and hope the cpu on the destination is close enough to work. Now that we've cleaned up handling of processor compatibility modes for pseries machine type, we can do better. For new machine types (pseries-2.10+) We allow migration if: * The source and destination PVRs are for the same type of CPU, as determined by CPU class's pvr_match function OR * When the source was in a compatibility mode, and the destination CPU supports the same compatibility mode For older machine types we retain the existing behaviour - current CAS code will usually set a compat mode which would break backwards migration if we made them use the new behaviour. [Fixed from an earlier version by Greg Kurz]. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Greg Kurz <groug@kaod.org> Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com> Tested-by: Andrea Bolognani <abologna@redhat.com>
2017-06-02 04:26:11 +02:00
Error *local_err = NULL;
cpu->compat_pvr = 0;
ppc_set_compat(cpu, compat_pvr, &local_err);
ppc: Rework CPU compatibility testing across migration Migrating between different CPU versions is a bit complicated for ppc. A long time ago, we ensured identical CPU versions at either end by checking the PVR had the same value. However, this breaks under KVM HV, because we always have to use the host's PVR - it's not virtualized. That would mean we couldn't migrate between hosts with different PVRs, even if the CPUs are close enough to compatible in practice (sometimes identical cores with different surrounding logic have different PVRs, so this happens in practice quite often). So, we removed the PVR check, but instead checked that several flags indicating supported instructions matched. This turns out to be a bad idea, because those instruction masks are not architected information, but essentially a TCG implementation detail. So changes to qemu internal CPU modelling can break migration - this happened between qemu-2.6 and qemu-2.7. That was addressed by 146c11f1 "target-ppc: Allow eventual removal of old migration mistakes". Now, verification of CPU compatibility across a migration basically doesn't happen. We simply ignore the PVR of the incoming migration, and hope the cpu on the destination is close enough to work. Now that we've cleaned up handling of processor compatibility modes for pseries machine type, we can do better. For new machine types (pseries-2.10+) We allow migration if: * The source and destination PVRs are for the same type of CPU, as determined by CPU class's pvr_match function OR * When the source was in a compatibility mode, and the destination CPU supports the same compatibility mode For older machine types we retain the existing behaviour - current CAS code will usually set a compat mode which would break backwards migration if we made them use the new behaviour. [Fixed from an earlier version by Greg Kurz]. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Greg Kurz <groug@kaod.org> Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com> Tested-by: Andrea Bolognani <abologna@redhat.com>
2017-06-02 04:26:11 +02:00
if (local_err) {
error_report_err(local_err);
return -1;
}
} else
#endif
{
if (!pvr_match(cpu, env->spr[SPR_PVR])) {
return -1;
}
}
target/ppc: 'PVR != host PVR' in KVM_SET_SREGS workaround Commit d5fc133eed ("ppc: Rework CPU compatibility testing across migration") changed the way cpu_post_load behaves with the PVR setting, causing an unexpected bug in KVM-HV migrations between hosts that are compatible (POWER8 and POWER8E, for example). Even with pvr_match() returning true, the guest freezes right after cpu_post_load. The reason is that the guest kernel can't handle a different PVR value other that the running host in KVM_SET_SREGS. In [1] it was discussed the possibility of a new KVM capability that would indicate that the guest kernel can handle a different PVR in KVM_SET_SREGS. Even if such feature is implemented, there is still the problem with older kernels that will not have this capability and will fail to migrate. This patch implements a workaround for that scenario. If running with KVM, check if the guest kernel does not have the capability (named here as 'cap_ppc_pvr_compat'). If it doesn't, calls kvmppc_is_pr() to see if the guest is running in KVM-HV. If all this happens, set env->spr[SPR_PVR] to the same value as the current host PVR. This ensures that we allow migrations with 'close enough' PVRs to still work in KVM-HV but also makes the code ready for this new KVM capability when it is done. A new function called 'kvmppc_pvr_workaround_required' was created to encapsulate the conditions said above and to avoid calling too many kvm.c internals inside cpu_post_load. [1] https://lists.gnu.org/archive/html/qemu-ppc/2017-06/msg00503.html Signed-off-by: Daniel Henrique Barboza <danielhb@linux.vnet.ibm.com> [dwg: Fix for the case of using TCG on a PPC host] Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2017-08-09 22:43:46 +02:00
/*
* If we're running with KVM HV, there is a chance that the guest
* is running with KVM HV and its kernel does not have the
* capability of dealing with a different PVR other than this
* exact host PVR in KVM_SET_SREGS. If that happens, the
* guest freezes after migration.
*
* The function kvmppc_pvr_workaround_required does this verification
* by first checking if the kernel has the cap, returning true immediately
* if that is the case. Otherwise, it checks if we're running in KVM PR.
* If the guest kernel does not have the cap and we're not running KVM-PR
* (so, it is running KVM-HV), we need to ensure that KVM_SET_SREGS will
* receive the PVR it expects as a workaround.
*
*/
#if defined(CONFIG_KVM)
if (kvmppc_pvr_workaround_required(cpu)) {
env->spr[SPR_PVR] = env->spr_cb[SPR_PVR].default_value;
}
#endif
env->lr = env->spr[SPR_LR];
env->ctr = env->spr[SPR_CTR];
cpu_write_xer(env, env->spr[SPR_XER]);
#if defined(TARGET_PPC64)
env->cfar = env->spr[SPR_CFAR];
#endif
env->spe_fscr = env->spr[SPR_BOOKE_SPEFSCR];
for (i = 0; (i < 4) && (i < env->nb_BATs); i++) {
env->DBAT[0][i] = env->spr[SPR_DBAT0U + 2*i];
env->DBAT[1][i] = env->spr[SPR_DBAT0U + 2*i + 1];
env->IBAT[0][i] = env->spr[SPR_IBAT0U + 2*i];
env->IBAT[1][i] = env->spr[SPR_IBAT0U + 2*i + 1];
}
for (i = 0; (i < 4) && ((i+4) < env->nb_BATs); i++) {
env->DBAT[0][i+4] = env->spr[SPR_DBAT4U + 2*i];
env->DBAT[1][i+4] = env->spr[SPR_DBAT4U + 2*i + 1];
env->IBAT[0][i+4] = env->spr[SPR_IBAT4U + 2*i];
env->IBAT[1][i+4] = env->spr[SPR_IBAT4U + 2*i + 1];
}
if (!cpu->vhyp) {
ppc_store_sdr1(env, env->spr[SPR_SDR1]);
}
/* Invalidate all supported msr bits except MSR_TGPR/MSR_HVB before restoring */
msr = env->msr;
env->msr ^= env->msr_mask & ~((1ULL << MSR_TGPR) | MSR_HVB);
ppc_store_msr(env, msr);
hreg_compute_mem_idx(env);
return 0;
}
static bool fpu_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
return (cpu->env.insns_flags & PPC_FLOAT);
}
static const VMStateDescription vmstate_fpu = {
.name = "cpu/fpu",
.version_id = 1,
.minimum_version_id = 1,
.needed = fpu_needed,
.fields = (VMStateField[]) {
target/ppc: move FP and VMX registers into aligned vsr register array The VSX register array is a block of 64 128-bit registers where the first 32 registers consist of the existing 64-bit FP registers extended to 128-bit using new VSR registers, and the last 32 registers are the VMX 128-bit registers as show below: 64-bit 64-bit +--------------------+--------------------+ | FP0 | | VSR0 +--------------------+--------------------+ | FP1 | | VSR1 +--------------------+--------------------+ | ... | ... | ... +--------------------+--------------------+ | FP30 | | VSR30 +--------------------+--------------------+ | FP31 | | VSR31 +--------------------+--------------------+ | VMX0 | VSR32 +-----------------------------------------+ | VMX1 | VSR33 +-----------------------------------------+ | ... | ... +-----------------------------------------+ | VMX30 | VSR62 +-----------------------------------------+ | VMX31 | VSR63 +-----------------------------------------+ In order to allow for future conversion of VSX instructions to use TCG vector operations, recreate the same layout using an aligned version of the existing vsr register array. Since the old fpr and avr register arrays are removed, the existing callers must also be updated to use the correct offset in the vsr register array. This also includes switching the relevant VMState fields over to using subarrays to make sure that migration is preserved. Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Acked-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-01-02 10:14:22 +01:00
VMSTATE_FPR_ARRAY(env.vsr, PowerPCCPU, 32),
VMSTATE_UINTTL(env.fpscr, PowerPCCPU),
VMSTATE_END_OF_LIST()
},
};
static bool altivec_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
return (cpu->env.insns_flags & PPC_ALTIVEC);
}
static const VMStateDescription vmstate_altivec = {
.name = "cpu/altivec",
.version_id = 1,
.minimum_version_id = 1,
.needed = altivec_needed,
.fields = (VMStateField[]) {
target/ppc: move FP and VMX registers into aligned vsr register array The VSX register array is a block of 64 128-bit registers where the first 32 registers consist of the existing 64-bit FP registers extended to 128-bit using new VSR registers, and the last 32 registers are the VMX 128-bit registers as show below: 64-bit 64-bit +--------------------+--------------------+ | FP0 | | VSR0 +--------------------+--------------------+ | FP1 | | VSR1 +--------------------+--------------------+ | ... | ... | ... +--------------------+--------------------+ | FP30 | | VSR30 +--------------------+--------------------+ | FP31 | | VSR31 +--------------------+--------------------+ | VMX0 | VSR32 +-----------------------------------------+ | VMX1 | VSR33 +-----------------------------------------+ | ... | ... +-----------------------------------------+ | VMX30 | VSR62 +-----------------------------------------+ | VMX31 | VSR63 +-----------------------------------------+ In order to allow for future conversion of VSX instructions to use TCG vector operations, recreate the same layout using an aligned version of the existing vsr register array. Since the old fpr and avr register arrays are removed, the existing callers must also be updated to use the correct offset in the vsr register array. This also includes switching the relevant VMState fields over to using subarrays to make sure that migration is preserved. Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Acked-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-01-02 10:14:22 +01:00
VMSTATE_AVR_ARRAY(env.vsr, PowerPCCPU, 32),
VMSTATE_UINT32(env.vscr, PowerPCCPU),
VMSTATE_END_OF_LIST()
},
};
static bool vsx_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
return (cpu->env.insns_flags2 & PPC2_VSX);
}
static const VMStateDescription vmstate_vsx = {
.name = "cpu/vsx",
.version_id = 1,
.minimum_version_id = 1,
.needed = vsx_needed,
.fields = (VMStateField[]) {
target/ppc: move FP and VMX registers into aligned vsr register array The VSX register array is a block of 64 128-bit registers where the first 32 registers consist of the existing 64-bit FP registers extended to 128-bit using new VSR registers, and the last 32 registers are the VMX 128-bit registers as show below: 64-bit 64-bit +--------------------+--------------------+ | FP0 | | VSR0 +--------------------+--------------------+ | FP1 | | VSR1 +--------------------+--------------------+ | ... | ... | ... +--------------------+--------------------+ | FP30 | | VSR30 +--------------------+--------------------+ | FP31 | | VSR31 +--------------------+--------------------+ | VMX0 | VSR32 +-----------------------------------------+ | VMX1 | VSR33 +-----------------------------------------+ | ... | ... +-----------------------------------------+ | VMX30 | VSR62 +-----------------------------------------+ | VMX31 | VSR63 +-----------------------------------------+ In order to allow for future conversion of VSX instructions to use TCG vector operations, recreate the same layout using an aligned version of the existing vsr register array. Since the old fpr and avr register arrays are removed, the existing callers must also be updated to use the correct offset in the vsr register array. This also includes switching the relevant VMState fields over to using subarrays to make sure that migration is preserved. Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Acked-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2019-01-02 10:14:22 +01:00
VMSTATE_VSR_ARRAY(env.vsr, PowerPCCPU, 32),
VMSTATE_END_OF_LIST()
},
};
#ifdef TARGET_PPC64
/* Transactional memory state */
static bool tm_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
CPUPPCState *env = &cpu->env;
return msr_ts;
}
static const VMStateDescription vmstate_tm = {
.name = "cpu/tm",
.version_id = 1,
.minimum_version_id = 1,
.minimum_version_id_old = 1,
.needed = tm_needed,
.fields = (VMStateField []) {
VMSTATE_UINTTL_ARRAY(env.tm_gpr, PowerPCCPU, 32),
VMSTATE_AVR_ARRAY(env.tm_vsr, PowerPCCPU, 64),
VMSTATE_UINT64(env.tm_cr, PowerPCCPU),
VMSTATE_UINT64(env.tm_lr, PowerPCCPU),
VMSTATE_UINT64(env.tm_ctr, PowerPCCPU),
VMSTATE_UINT64(env.tm_fpscr, PowerPCCPU),
VMSTATE_UINT64(env.tm_amr, PowerPCCPU),
VMSTATE_UINT64(env.tm_ppr, PowerPCCPU),
VMSTATE_UINT64(env.tm_vrsave, PowerPCCPU),
VMSTATE_UINT32(env.tm_vscr, PowerPCCPU),
VMSTATE_UINT64(env.tm_dscr, PowerPCCPU),
VMSTATE_UINT64(env.tm_tar, PowerPCCPU),
VMSTATE_END_OF_LIST()
},
};
#endif
static bool sr_needed(void *opaque)
{
#ifdef TARGET_PPC64
PowerPCCPU *cpu = opaque;
return !(cpu->env.mmu_model & POWERPC_MMU_64);
#else
return true;
#endif
}
static const VMStateDescription vmstate_sr = {
.name = "cpu/sr",
.version_id = 1,
.minimum_version_id = 1,
.needed = sr_needed,
.fields = (VMStateField[]) {
VMSTATE_UINTTL_ARRAY(env.sr, PowerPCCPU, 32),
VMSTATE_END_OF_LIST()
},
};
#ifdef TARGET_PPC64
static int get_slbe(QEMUFile *f, void *pv, size_t size,
const VMStateField *field)
{
ppc_slb_t *v = pv;
v->esid = qemu_get_be64(f);
v->vsid = qemu_get_be64(f);
return 0;
}
static int put_slbe(QEMUFile *f, void *pv, size_t size,
const VMStateField *field, QJSON *vmdesc)
{
ppc_slb_t *v = pv;
qemu_put_be64(f, v->esid);
qemu_put_be64(f, v->vsid);
return 0;
}
static const VMStateInfo vmstate_info_slbe = {
.name = "slbe",
.get = get_slbe,
.put = put_slbe,
};
#define VMSTATE_SLB_ARRAY_V(_f, _s, _n, _v) \
VMSTATE_ARRAY(_f, _s, _n, _v, vmstate_info_slbe, ppc_slb_t)
#define VMSTATE_SLB_ARRAY(_f, _s, _n) \
VMSTATE_SLB_ARRAY_V(_f, _s, _n, 0)
static bool slb_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
/* We don't support any of the old segment table based 64-bit CPUs */
return (cpu->env.mmu_model & POWERPC_MMU_64);
}
static int slb_post_load(void *opaque, int version_id)
{
PowerPCCPU *cpu = opaque;
CPUPPCState *env = &cpu->env;
int i;
/* We've pulled in the raw esid and vsid values from the migration
* stream, but we need to recompute the page size pointers */
for (i = 0; i < cpu->hash64_opts->slb_size; i++) {
if (ppc_store_slb(cpu, i, env->slb[i].esid, env->slb[i].vsid) < 0) {
/* Migration source had bad values in its SLB */
return -1;
}
}
return 0;
}
static const VMStateDescription vmstate_slb = {
.name = "cpu/slb",
.version_id = 1,
.minimum_version_id = 1,
.needed = slb_needed,
.post_load = slb_post_load,
.fields = (VMStateField[]) {
VMSTATE_INT32_TEST(mig_slb_nr, PowerPCCPU, cpu_pre_3_0_migration),
VMSTATE_SLB_ARRAY(env.slb, PowerPCCPU, MAX_SLB_ENTRIES),
VMSTATE_END_OF_LIST()
}
};
#endif /* TARGET_PPC64 */
static const VMStateDescription vmstate_tlb6xx_entry = {
.name = "cpu/tlb6xx_entry",
.version_id = 1,
.minimum_version_id = 1,
.fields = (VMStateField[]) {
VMSTATE_UINTTL(pte0, ppc6xx_tlb_t),
VMSTATE_UINTTL(pte1, ppc6xx_tlb_t),
VMSTATE_UINTTL(EPN, ppc6xx_tlb_t),
VMSTATE_END_OF_LIST()
},
};
static bool tlb6xx_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
CPUPPCState *env = &cpu->env;
return env->nb_tlb && (env->tlb_type == TLB_6XX);
}
static const VMStateDescription vmstate_tlb6xx = {
.name = "cpu/tlb6xx",
.version_id = 1,
.minimum_version_id = 1,
.needed = tlb6xx_needed,
.fields = (VMStateField[]) {
VMSTATE_INT32_EQUAL(env.nb_tlb, PowerPCCPU, NULL),
VMSTATE_STRUCT_VARRAY_POINTER_INT32(env.tlb.tlb6, PowerPCCPU,
env.nb_tlb,
vmstate_tlb6xx_entry,
ppc6xx_tlb_t),
VMSTATE_UINTTL_ARRAY(env.tgpr, PowerPCCPU, 4),
VMSTATE_END_OF_LIST()
}
};
static const VMStateDescription vmstate_tlbemb_entry = {
.name = "cpu/tlbemb_entry",
.version_id = 1,
.minimum_version_id = 1,
.fields = (VMStateField[]) {
VMSTATE_UINT64(RPN, ppcemb_tlb_t),
VMSTATE_UINTTL(EPN, ppcemb_tlb_t),
VMSTATE_UINTTL(PID, ppcemb_tlb_t),
VMSTATE_UINTTL(size, ppcemb_tlb_t),
VMSTATE_UINT32(prot, ppcemb_tlb_t),
VMSTATE_UINT32(attr, ppcemb_tlb_t),
VMSTATE_END_OF_LIST()
},
};
static bool tlbemb_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
CPUPPCState *env = &cpu->env;
return env->nb_tlb && (env->tlb_type == TLB_EMB);
}
static bool pbr403_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
uint32_t pvr = cpu->env.spr[SPR_PVR];
return (pvr & 0xffff0000) == 0x00200000;
}
static const VMStateDescription vmstate_pbr403 = {
.name = "cpu/pbr403",
.version_id = 1,
.minimum_version_id = 1,
.needed = pbr403_needed,
.fields = (VMStateField[]) {
VMSTATE_UINTTL_ARRAY(env.pb, PowerPCCPU, 4),
VMSTATE_END_OF_LIST()
},
};
static const VMStateDescription vmstate_tlbemb = {
.name = "cpu/tlb6xx",
.version_id = 1,
.minimum_version_id = 1,
.needed = tlbemb_needed,
.fields = (VMStateField[]) {
VMSTATE_INT32_EQUAL(env.nb_tlb, PowerPCCPU, NULL),
VMSTATE_STRUCT_VARRAY_POINTER_INT32(env.tlb.tlbe, PowerPCCPU,
env.nb_tlb,
vmstate_tlbemb_entry,
ppcemb_tlb_t),
/* 403 protection registers */
VMSTATE_END_OF_LIST()
},
.subsections = (const VMStateDescription*[]) {
&vmstate_pbr403,
NULL
}
};
static const VMStateDescription vmstate_tlbmas_entry = {
.name = "cpu/tlbmas_entry",
.version_id = 1,
.minimum_version_id = 1,
.fields = (VMStateField[]) {
VMSTATE_UINT32(mas8, ppcmas_tlb_t),
VMSTATE_UINT32(mas1, ppcmas_tlb_t),
VMSTATE_UINT64(mas2, ppcmas_tlb_t),
VMSTATE_UINT64(mas7_3, ppcmas_tlb_t),
VMSTATE_END_OF_LIST()
},
};
static bool tlbmas_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
CPUPPCState *env = &cpu->env;
return env->nb_tlb && (env->tlb_type == TLB_MAS);
}
static const VMStateDescription vmstate_tlbmas = {
.name = "cpu/tlbmas",
.version_id = 1,
.minimum_version_id = 1,
.needed = tlbmas_needed,
.fields = (VMStateField[]) {
VMSTATE_INT32_EQUAL(env.nb_tlb, PowerPCCPU, NULL),
VMSTATE_STRUCT_VARRAY_POINTER_INT32(env.tlb.tlbm, PowerPCCPU,
env.nb_tlb,
vmstate_tlbmas_entry,
ppcmas_tlb_t),
VMSTATE_END_OF_LIST()
}
};
ppc: Rework CPU compatibility testing across migration Migrating between different CPU versions is a bit complicated for ppc. A long time ago, we ensured identical CPU versions at either end by checking the PVR had the same value. However, this breaks under KVM HV, because we always have to use the host's PVR - it's not virtualized. That would mean we couldn't migrate between hosts with different PVRs, even if the CPUs are close enough to compatible in practice (sometimes identical cores with different surrounding logic have different PVRs, so this happens in practice quite often). So, we removed the PVR check, but instead checked that several flags indicating supported instructions matched. This turns out to be a bad idea, because those instruction masks are not architected information, but essentially a TCG implementation detail. So changes to qemu internal CPU modelling can break migration - this happened between qemu-2.6 and qemu-2.7. That was addressed by 146c11f1 "target-ppc: Allow eventual removal of old migration mistakes". Now, verification of CPU compatibility across a migration basically doesn't happen. We simply ignore the PVR of the incoming migration, and hope the cpu on the destination is close enough to work. Now that we've cleaned up handling of processor compatibility modes for pseries machine type, we can do better. For new machine types (pseries-2.10+) We allow migration if: * The source and destination PVRs are for the same type of CPU, as determined by CPU class's pvr_match function OR * When the source was in a compatibility mode, and the destination CPU supports the same compatibility mode For older machine types we retain the existing behaviour - current CAS code will usually set a compat mode which would break backwards migration if we made them use the new behaviour. [Fixed from an earlier version by Greg Kurz]. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Greg Kurz <groug@kaod.org> Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com> Tested-by: Andrea Bolognani <abologna@redhat.com>
2017-06-02 04:26:11 +02:00
static bool compat_needed(void *opaque)
{
PowerPCCPU *cpu = opaque;
assert(!(cpu->compat_pvr && !cpu->vhyp));
return !cpu->pre_2_10_migration && cpu->compat_pvr != 0;
}
static const VMStateDescription vmstate_compat = {
.name = "cpu/compat",
.version_id = 1,
.minimum_version_id = 1,
.needed = compat_needed,
.fields = (VMStateField[]) {
VMSTATE_UINT32(compat_pvr, PowerPCCPU),
VMSTATE_END_OF_LIST()
}
};
const VMStateDescription vmstate_ppc_cpu = {
.name = "cpu",
.version_id = 5,
.minimum_version_id = 5,
.minimum_version_id_old = 4,
.load_state_old = cpu_load_old,
.pre_save = cpu_pre_save,
.post_load = cpu_post_load,
.fields = (VMStateField[]) {
VMSTATE_UNUSED(sizeof(target_ulong)), /* was _EQUAL(env.spr[SPR_PVR]) */
/* User mode architected state */
VMSTATE_UINTTL_ARRAY(env.gpr, PowerPCCPU, 32),
#if !defined(TARGET_PPC64)
VMSTATE_UINTTL_ARRAY(env.gprh, PowerPCCPU, 32),
#endif
VMSTATE_UINT32_ARRAY(env.crf, PowerPCCPU, 8),
VMSTATE_UINTTL(env.nip, PowerPCCPU),
/* SPRs */
VMSTATE_UINTTL_ARRAY(env.spr, PowerPCCPU, 1024),
VMSTATE_UINT64(env.spe_acc, PowerPCCPU),
/* Reservation */
VMSTATE_UINTTL(env.reserve_addr, PowerPCCPU),
/* Supervisor mode architected state */
VMSTATE_UINTTL(env.msr, PowerPCCPU),
/* Internal state */
VMSTATE_UINTTL(env.hflags_nmsr, PowerPCCPU),
/* FIXME: access_type? */
/* Sanity checking */
VMSTATE_UINTTL_TEST(mig_msr_mask, PowerPCCPU, cpu_pre_2_8_migration),
VMSTATE_UINT64_TEST(mig_insns_flags, PowerPCCPU, cpu_pre_2_8_migration),
VMSTATE_UINT64_TEST(mig_insns_flags2, PowerPCCPU,
cpu_pre_2_8_migration),
VMSTATE_UINT32_TEST(mig_nb_BATs, PowerPCCPU, cpu_pre_2_8_migration),
VMSTATE_END_OF_LIST()
},
.subsections = (const VMStateDescription*[]) {
&vmstate_fpu,
&vmstate_altivec,
&vmstate_vsx,
&vmstate_sr,
#ifdef TARGET_PPC64
&vmstate_tm,
&vmstate_slb,
#endif /* TARGET_PPC64 */
&vmstate_tlb6xx,
&vmstate_tlbemb,
&vmstate_tlbmas,
ppc: Rework CPU compatibility testing across migration Migrating between different CPU versions is a bit complicated for ppc. A long time ago, we ensured identical CPU versions at either end by checking the PVR had the same value. However, this breaks under KVM HV, because we always have to use the host's PVR - it's not virtualized. That would mean we couldn't migrate between hosts with different PVRs, even if the CPUs are close enough to compatible in practice (sometimes identical cores with different surrounding logic have different PVRs, so this happens in practice quite often). So, we removed the PVR check, but instead checked that several flags indicating supported instructions matched. This turns out to be a bad idea, because those instruction masks are not architected information, but essentially a TCG implementation detail. So changes to qemu internal CPU modelling can break migration - this happened between qemu-2.6 and qemu-2.7. That was addressed by 146c11f1 "target-ppc: Allow eventual removal of old migration mistakes". Now, verification of CPU compatibility across a migration basically doesn't happen. We simply ignore the PVR of the incoming migration, and hope the cpu on the destination is close enough to work. Now that we've cleaned up handling of processor compatibility modes for pseries machine type, we can do better. For new machine types (pseries-2.10+) We allow migration if: * The source and destination PVRs are for the same type of CPU, as determined by CPU class's pvr_match function OR * When the source was in a compatibility mode, and the destination CPU supports the same compatibility mode For older machine types we retain the existing behaviour - current CAS code will usually set a compat mode which would break backwards migration if we made them use the new behaviour. [Fixed from an earlier version by Greg Kurz]. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Greg Kurz <groug@kaod.org> Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com> Tested-by: Andrea Bolognani <abologna@redhat.com>
2017-06-02 04:26:11 +02:00
&vmstate_compat,
NULL
}
};