Re: [Qemu-devel] [PATCH 3/3 v2] Enable kvm emulated watchdog

[Alexander Graf]

Am 11.01.2013 um 07:42 schrieb Bhushan Bharat-R65777 <address@hidden>:

> 
> 
>> -----Original Message-----
>> From: Alexander Graf [mailto:address@hidden
>> Sent: Thursday, January 10, 2013 9:07 PM
>> To: Bhushan Bharat-R65777
>> Cc: address@hidden; address@hidden; Bhushan Bharat-R65777
>> Subject: Re: [PATCH 3/3 v2] Enable kvm emulated watchdog
>> 
>> 
>> On 28.12.2012, at 06:16, Bharat Bhushan wrote:
>> 
>>> Enable the KVM emulated watchdog if KVM supports (use the capability
>>> enablement in watchdog handler). Also watchdog exit
>>> (KVM_EXIT_WATCHDOG) handling is added.
>>> Watchdog state machine is cleared whenever VM state changes to running.
>>> This is to handle the cases like return from debug halt etc.
>>> 
>>> Signed-off-by: Bharat Bhushan <address@hidden>
>>> ---
>>> v2:
>>> - access cap_* from target_ppc/kvm.c only.
>>> - Added wrapper functions in target_ppc/kvm.c for
>>>   enable_watchdog and tsr_sregs synchronization.
>>> - Incorporated other Review comments
>>> 
>>> hw/ppc.h             |    1 +
>>> hw/ppc_booke.c       |   36 +++++++++++++++++++++++++++++++-
>>> target-ppc/kvm.c     |   56 
>>> ++++++++++++++++++++++++++++++++++++++++++++++++++
>>> target-ppc/kvm_ppc.h |   11 +++++++++
>>> 4 files changed, 103 insertions(+), 1 deletions(-)
>>> 
>>> diff --git a/hw/ppc.h b/hw/ppc.h
>>> index 2f3ea27..6ad9e1f 100644
>>> --- a/hw/ppc.h
>>> +++ b/hw/ppc.h
>>> @@ -90,3 +90,4 @@ enum {
>>> 
>>> /* ppc_booke.c */
>>> void ppc_booke_timers_init(CPUPPCState *env, uint32_t freq, uint32_t
>>> flags);
>>> +void ppc_booke_watchdog_clear_tsr(CPUPPCState *env);
>>> diff --git a/hw/ppc_booke.c b/hw/ppc_booke.c index 837a5b6..7273259
>>> 100644
>>> --- a/hw/ppc_booke.c
>>> +++ b/hw/ppc_booke.c
>>> @@ -28,7 +28,7 @@
>>> #include "nvram.h"
>>> #include "qemu-log.h"
>>> #include "loader.h"
>>> -
>>> +#include "kvm_ppc.h"
>>> 
>>> /* Timer Control Register */
>>> 
>>> @@ -203,6 +203,11 @@ static void booke_wdt_cb(void *opaque)
>>>                             booke_timer->wdt_timer); }
>>> 
>>> +void ppc_booke_watchdog_clear_tsr(CPUPPCState *env) {
>>> +    env->spr[SPR_BOOKE_TSR] &= ~(TSR_ENW | TSR_WIS | TSR_WRS_MASK); }
>>> +
>>> void store_booke_tsr(CPUPPCState *env, target_ulong val) {
>>>    env->spr[SPR_BOOKE_TSR] &= ~val;
>>> @@ -241,10 +246,27 @@ static void ppc_booke_timer_reset_handle(void *opaque)
>>>    booke_update_irq(env);
>>> }
>>> 
>>> +static void cpu_state_change_handler(void *opaque, int running,
>>> +RunState state) {
>> 
>> Needs a comment when this happens
>> 
>>> +    CPUPPCState *env = opaque;
>>> +
>>> +    if (!running) {
>>> +        return;
>>> +    }
>>> +
>>> +    /*
>>> +     * Clear watchdog interrupt condition by clearing TSR.
>>> +     */
>>> +    ppc_booke_watchdog_clear_tsr(env);
>>> +
>>> +    kvmppc_synch_sregs_tsr(env);
>> 
>> kvmppc_sync_tsr. Also please add the sync to store_booke_tsr(). Then here, 
>> you
>> can just do
>> 
>>  store_booke_tsr(TSR_ENW | TSR_WIS | TSR_WRS_MASK);
>> 
>>> +}
>>> +
>>> void ppc_booke_timers_init(CPUPPCState *env, uint32_t freq, uint32_t
>>> flags) {
>>>    ppc_tb_t *tb_env;
>>>    booke_timer_t *booke_timer;
>>> +    int ret = 0;
>>> 
>>>    tb_env      = g_malloc0(sizeof(ppc_tb_t));
>>>    booke_timer = g_malloc0(sizeof(booke_timer_t)); @@ -262,5 +284,17
>>> @@ void ppc_booke_timers_init(CPUPPCState *env, uint32_t freq, uint32_t 
>>> flags)
>>>    booke_timer->wdt_timer =
>>>        qemu_new_timer_ns(vm_clock, &booke_wdt_cb, env);
>>> 
>>> +    ret = kvmppc_booke_watchdog_enable(env);
>>> +
>>> +    if (ret) {
>>> +        /* TODO: Start the QEMU emulated watchdog if not running on KVM.
>>> +         * Also start the QEMU emulated watchdog if KVM does not support
>>> +         * emulated watchdog or somehow it is not enabled (supported but
>>> +         * not enabled is though some bug and requires debugging :)).
>>> +         */
>>> +    }
>>> +
>>> +    qemu_add_vm_change_state_handler(cpu_state_change_handler, env);
>>> +
>>>    qemu_register_reset(ppc_booke_timer_reset_handle, env); } diff
>>> --git a/target-ppc/kvm.c b/target-ppc/kvm.c index 3f5df57..6828afa
>>> 100644
>>> --- a/target-ppc/kvm.c
>>> +++ b/target-ppc/kvm.c
>>> @@ -32,10 +32,12 @@
>>> #include "device_tree.h"
>>> #include "hw/sysbus.h"
>>> #include "hw/spapr.h"
>>> +#include "hw/watchdog.h"
>>> 
>>> #include "hw/sysbus.h"
>>> #include "hw/spapr.h"
>>> #include "hw/spapr_vio.h"
>>> +#include "hw/ppc.h"
>>> 
>>> //#define DEBUG_KVM
>>> 
>>> @@ -61,6 +63,7 @@ static int cap_ppc_smt; static int cap_ppc_rma;
>>> static int cap_spapr_tce; static int cap_hior;
>>> +static int cap_ppc_watchdog;
>>> 
>>> /* XXX We have a race condition where we actually have a level triggered
>>> *     interrupt, but the infrastructure can't expose that yet, so the guest
>>> @@ -90,6 +93,7 @@ int kvm_arch_init(KVMState *s)
>>>    cap_ppc_rma = kvm_check_extension(s, KVM_CAP_PPC_RMA);
>>>    cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
>>>    cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
>>> +    cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
>>> 
>>>    if (!cap_interrupt_level) {
>>>        fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the 
>>> "
>>> @@ -823,6 +827,12 @@ int kvm_arch_handle_exit(CPUPPCState *env, struct 
>>> kvm_run
>> *run)
>>>        ret = 0;
>>>        break;
>>> #endif
>>> +    case KVM_EXIT_WATCHDOG:
>>> +        dprintf("handle watchdog expiry\n");
>>> +        watchdog_perform_action();
>>> +        ret = 0;
>>> +        break;
>>> +
>>>    default:
>>>        fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
>>>        ret = -1;
>>> @@ -1239,3 +1249,49 @@ int kvm_arch_on_sigbus(int code, void *addr)
>>> {
>>>    return 1;
>>> }
>>> +
>>> +int kvmppc_synch_sregs_tsr(CPUPPCState *env)
>>> +{
>>> +    struct kvm_sregs sregs;
>>> +    int ret;
>>> +
>>> +    if (!kvm_enabled() || !cap_booke_sregs ||  !cap_ppc_watchdog) {
>>> +        return -1;
>>> +    }
>>> +
>>> +    ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
>>> +
>>> +    env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
>>> +    ppc_booke_watchdog_clear_tsr(env);
>> 
>> You're getting _and_ setting it in the same function? This is nonsense. 
>> Please
>> split it into a get and a set function if you need it.
>> 
>>> +    sregs.u.e.tsr = env->spr[SPR_BOOKE_TSR];
>>> +    sregs.u.e.update_special = KVM_SREGS_E_UPDATE_TSR;
>>> +
>>> +    ret = kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs);
>>> +
>>> +    return ret;
>>> +}
>>> +
>>> +int kvmppc_booke_watchdog_enable(CPUPPCState *env)
>>> +{
>>> +    int ret;
>>> +    struct kvm_enable_cap encap = {};
>>> +
>>> +    if (!kvm_enabled()) {
>>> +        return -1;
>>> +    }
>>> +
>>> +    if (!cap_ppc_watchdog) {
>>> +        printf("warning: KVM does not support watchdog");
>>> +        return -1;
>>> +    }
>>> +
>>> +    encap.cap = KVM_CAP_PPC_BOOKE_WATCHDOG;
>>> +    ret = kvm_vcpu_ioctl(env, KVM_ENABLE_CAP, &encap);
>>> +    if (ret < 0) {
>>> +        fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG:
>> %s\n",
>>> +                __func__, strerror(-ret));
>>> +        return ret;
>>> +    }
>>> +
>>> +    return ret;
>>> +}
>>> diff --git a/target-ppc/kvm_ppc.h b/target-ppc/kvm_ppc.h
>>> index baad6eb..4ccc9fa 100644
>>> --- a/target-ppc/kvm_ppc.h
>>> +++ b/target-ppc/kvm_ppc.h
>>> @@ -23,6 +23,8 @@ int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, 
>>> int
>> buf_len);
>>> int kvmppc_set_interrupt(CPUPPCState *env, int irq, int level);
>>> void kvmppc_set_papr(CPUPPCState *env);
>>> int kvmppc_smt_threads(void);
>>> +int kvmppc_synch_sregs_tsr(CPUPPCState *cenv);
>>> +int kvmppc_booke_watchdog_enable(CPUPPCState *env);
>>> #ifndef CONFIG_USER_ONLY
>> 
>> Missing code for !CONFIG_KVM. Please always check compile your code on x86.
> 
> Alex is not the below code covers !CONFIG_KVM  ...
> 
>> 
>> 
>> Alex
>> 
>>> off_t kvmppc_alloc_rma(const char *name, MemoryRegion *sysmem);
>>> void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t window_size, int 
>>> *pfd);
>>> @@ -79,6 +81,15 @@ static inline int kvmppc_smt_threads(void)
>>>    return 1;
>>> }
>>> 
>>> +static inline int kvmppc_synch_sregs_tsr(CPUPPCState *env)
>>> +{
>>> +    return -1;
>>> +}
>>> +
>>> +static inline int kvmppc_booke_watchdog_enable(CPUPPCState *env)
>>> +{
>>> +    return -1;
>>> +}
> 
> I mean this one

Hrm. Somehow I didn't see those lines during review :). Yes, it's fine - sorry.

Alex

> 
> Thanks
> -Bharat
> 
>>> #ifndef CONFIG_USER_ONLY
>>> static inline off_t kvmppc_alloc_rma(const char *name, MemoryRegion *sysmem)
>>> {
>>> --
>>> 1.7.0.4
>>> 
>>> 
>> 
> 
>