Re: [Qemu-devel] [PATCH 15/15] target-s390x: PER: add Breaking-Event-Address register

[Alexander Graf]

On 16.06.15 19:44, Aurelien Jarno wrote:
> On 2015-06-16 18:44, Alexander Graf wrote:
>> On 06/13/15 00:46, Aurelien Jarno wrote:
>>> This patch adds support for PER Breaking-Event-Address register. Like
>>> real hardware, it save the current PSW address when the PSW address is
>>> changed by an instruction. We have to take care of optimizations QEMU
>>> does, a branch to the next instruction is still a branch.
>>>
>>> This register is copied to low core memory when a program exception
>>> happens.
>>>
>>> Cc: Richard Henderson <address@hidden>
>>> Cc: Alexander Graf <address@hidden>
>>> Signed-off-by: Aurelien Jarno <address@hidden>
>>> ---
>>>  target-s390x/cpu.c       |  6 ++++++
>>>  target-s390x/cpu.h       | 12 +++++++-----
>>>  target-s390x/helper.c    |  1 +
>>>  target-s390x/translate.c | 29 +++++++++++++++++++++++------
>>>  4 files changed, 37 insertions(+), 11 deletions(-)
>>>
>>> diff --git a/target-s390x/cpu.c b/target-s390x/cpu.c
>>> index 67579e7..98d2081 100644
>>> --- a/target-s390x/cpu.c
>>> +++ b/target-s390x/cpu.c
>>> @@ -116,6 +116,9 @@ static void s390_cpu_initial_reset(CPUState *s)
>>>      env->cregs[0] = CR0_RESET;
>>>      env->cregs[14] = CR14_RESET;
>>> +    /* architectured initial value for Breaking-Event-Address register */
>>> +    env->gbea = 1;
>>> +
>>>      env->pfault_token = -1UL;
>>>      /* tininess for underflow is detected before rounding */
>>> @@ -145,6 +148,9 @@ static void s390_cpu_full_reset(CPUState *s)
>>>      env->cregs[0] = CR0_RESET;
>>>      env->cregs[14] = CR14_RESET;
>>> +    /* architectured initial value for Breaking-Event-Address register */
>>> +    env->gbea = 1;
>>> +
>>>      env->pfault_token = -1UL;
>>>      /* tininess for underflow is detected before rounding */
>>> diff --git a/target-s390x/cpu.h b/target-s390x/cpu.h
>>> index 61cc5b4..519cef9 100644
>>> --- a/target-s390x/cpu.h
>>> +++ b/target-s390x/cpu.h
>>> @@ -788,14 +788,16 @@ typedef struct LowCore
>>>      uint8_t         pad5[0xf4-0xf0];          /* 0x0f0 */
>>>      uint32_t        external_damage_code;     /* 0x0f4 */
>>>      uint64_t        failing_storage_address;  /* 0x0f8 */
>>> -    uint8_t         pad6[0x120-0x100];        /* 0x100 */
>>> +    uint8_t         pad6[0x110-0x100];        /* 0x100 */
>>> +    uint64_t        per_breaking_event_addr;  /* 0x110 */
>>> +    uint8_t         pad7[0x120-0x118];        /* 0x118 */
>>>      PSW             restart_old_psw;          /* 0x120 */
>>>      PSW             external_old_psw;         /* 0x130 */
>>>      PSW             svc_old_psw;              /* 0x140 */
>>>      PSW             program_old_psw;          /* 0x150 */
>>>      PSW             mcck_old_psw;             /* 0x160 */
>>>      PSW             io_old_psw;               /* 0x170 */
>>> -    uint8_t         pad7[0x1a0-0x180];        /* 0x180 */
>>> +    uint8_t         pad8[0x1a0-0x180];        /* 0x180 */
>>>      PSW             restart_new_psw;          /* 0x1a0 */
>>>      PSW             external_new_psw;         /* 0x1b0 */
>>>      PSW             svc_new_psw;              /* 0x1c0 */
>>> @@ -813,10 +815,10 @@ typedef struct LowCore
>>>      uint64_t        last_update_clock;        /* 0x280 */
>>>      uint64_t        steal_clock;              /* 0x288 */
>>>      PSW             return_mcck_psw;          /* 0x290 */
>>> -    uint8_t         pad8[0xc00-0x2a0];        /* 0x2a0 */
>>> +    uint8_t         pad9[0xc00-0x2a0];        /* 0x2a0 */
>>>      /* System info area */
>>>      uint64_t        save_area[16];            /* 0xc00 */
>>> -    uint8_t         pad9[0xd40-0xc80];        /* 0xc80 */
>>> +    uint8_t         pad10[0xd40-0xc80];       /* 0xc80 */
>>>      uint64_t        kernel_stack;             /* 0xd40 */
>>>      uint64_t        thread_info;              /* 0xd48 */
>>>      uint64_t        async_stack;              /* 0xd50 */
>>> @@ -824,7 +826,7 @@ typedef struct LowCore
>>>      uint64_t        user_asce;                /* 0xd60 */
>>>      uint64_t        panic_stack;              /* 0xd68 */
>>>      uint64_t        user_exec_asce;           /* 0xd70 */
>>> -    uint8_t         pad10[0xdc0-0xd78];       /* 0xd78 */
>>> +    uint8_t         pad11[0xdc0-0xd78];       /* 0xd78 */
>>>      /* SMP info area: defined by DJB */
>>>      uint64_t        clock_comparator;         /* 0xdc0 */
>>> diff --git a/target-s390x/helper.c b/target-s390x/helper.c
>>> index 615cccf..d887006 100644
>>> --- a/target-s390x/helper.c
>>> +++ b/target-s390x/helper.c
>>> @@ -293,6 +293,7 @@ static void do_program_interrupt(CPUS390XState *env)
>>>      lowcore->program_old_psw.addr = cpu_to_be64(env->psw.addr);
>>>      mask = be64_to_cpu(lowcore->program_new_psw.mask);
>>>      addr = be64_to_cpu(lowcore->program_new_psw.addr);
>>> +    lowcore->per_breaking_event_addr = cpu_to_be64(env->gbea);
>>>      cpu_unmap_lowcore(lowcore);
>>> diff --git a/target-s390x/translate.c b/target-s390x/translate.c
>>> index 98e8224..2fde815 100644
>>> --- a/target-s390x/translate.c
>>> +++ b/target-s390x/translate.c
>>> @@ -150,6 +150,7 @@ void s390_cpu_dump_state(CPUState *cs, FILE *f, 
>>> fprintf_function cpu_fprintf,
>>>  static TCGv_i64 psw_addr;
>>>  static TCGv_i64 psw_mask;
>>> +static TCGv_i64 gbea;
>>>  static TCGv_i32 cc_op;
>>>  static TCGv_i64 cc_src;
>>> @@ -173,6 +174,9 @@ void s390x_translate_init(void)
>>>      psw_mask = tcg_global_mem_new_i64(TCG_AREG0,
>>>                                        offsetof(CPUS390XState, psw.mask),
>>>                                        "psw_mask");
>>> +    gbea = tcg_global_mem_new_i64(TCG_AREG0,
>>> +                                  offsetof(CPUS390XState, gbea),
>>> +                                  "gbea");
>>>      cc_op = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUS390XState, 
>>> cc_op),
>>>                                     "cc_op");
>>> @@ -252,14 +256,14 @@ static void update_psw_addr(DisasContext *s)
>>>  static void per_branch(DisasContext *s, bool to_next)
>>>  {
>>>  #ifndef CONFIG_USER_ONLY
>>> +    tcg_gen_movi_i64(gbea, s->pc);
>>
>> This should probably be a call to per_breaking_event(), no?
> 
> Yes, that's possible, but given gbea is reused below instead of
> reloading s->pc, I preferred to make the move more explicit.

You're right, it's probably better to be explicit.

> 
> That said given I have to send a rebased version, I can easily change
> that.

No worries.

> 
>> Also, is there no flag to control this register? I'd assume it to be quite
>> some performance penalty to always store the last branched register.
> 
> No this register is always loaded. This is the value you see in dmesg
> when a user program crashes, but you can also get it via GDB. Quite
> useful in some cases.
> 
> I haven't measured any performance impact, only noise. We are talking
> about writing an immediate to a memory location in the env structure
> (thus very likely with a cache line already allocated), so it's only
> two host instructions more in a TB. That's not a lot given that for
> example every TB starts by loading a value from the env structure and
> doing a test on it.

Ok, let's be correct first and then see what we can do if anyone
complains about performance ;)


Alex