Re: [Qemu-devel] [PATCH v1] s390x/kvm: fix and cleanup storing CPU status

[Richard Henderson]

On 09/22/2017 08:18 AM, David Hildenbrand wrote:
> On 22.09.2017 15:12, Richard Henderson wrote:
>> I don't believe you need to use QEMU_PACKED.  Unless the "addr" that's passed
>> in can really be arbitrary?  (If so, the memory returned by
>> cpu_physical_memory_map will be unaligned and so we do need the PACKED to 
>> force
>> the compiler to generate unaligned stores into the structure.)
> 
> The address is either 0x1200 or it is masked with 0x7e00u.
> 
> So most probably this should be fine without QEMU_PACKED, right?

Yes.


r~

Re: [Qemu-devel] [PATCH v1] s390x/kvm: fix and cleanup storing CPU status

[Christian Borntraeger]

On 09/21/2017 11:30 PM, David Hildenbrand wrote:
> env->psa is a 64bit value, while we copy 4 bytes into the save area,
> resulting always in 0 getting stored.
> 
> Let's try to reduce such errors by using a proper structure. While at
> it, use correct cpu->be conversion (and get_psw_mask()), as we will be
> reusing this code for TCG soon.

Interesting, I would have expected it to break kdump, but this did work.
So it probably works differently.

Anyway this looks sane, no obvious bug.


> 
> Signed-off-by: David Hildenbrand 
> ---
> 
> Found with my fancy SIGP kvm-unit-tests I am currently working on. And
> looks like there is even more ... stay tuned :)
> 
>  target/s390x/kvm.c | 63 
> +-
>  1 file changed, 43 insertions(+), 20 deletions(-)
> 
> diff --git a/target/s390x/kvm.c b/target/s390x/kvm.c
> index 720cb1d833..21c22f48b9 100644
> --- a/target/s390x/kvm.c
> +++ b/target/s390x/kvm.c
> @@ -1548,22 +1548,38 @@ static int do_store_adtl_status(S390CPU *cpu, hwaddr 
> addr, hwaddr len)
>  return 0;
>  }
> 
> +struct sigp_save_area {
> +uint64_tfprs[16];   /* 0x */
> +uint64_tgrs[16];/* 0x0080 */
> +PSW psw;/* 0x0100 */
> +uint8_t pad_0x0110[0x0118 - 0x0110];/* 0x0110 */
> +uint32_tprefix; /* 0x0118 */
> +uint32_tfpc;/* 0x011c */
> +uint8_t pad_0x0120[0x0124 - 0x0120];/* 0x0120 */

> +uint32_ttodpr;  /* 0x0124 */
> +uint64_tcputm;  /* 0x0128 */
> +uint64_tckc;/* 0x0130 */
> +uint8_t pad_0x0138[0x0140 - 0x0138];/* 0x0138 */
> +uint32_tars[16];/* 0x0140 */
> +uint64_tcrs[16];/* 0x0384 */
> +} QEMU_PACKED;
> +
>  #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, 
> floating_pt_save_area)
> -#define SAVE_AREA_SIZE 512
>  static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
>  {
>  static const uint8_t ar_id = 1;
> -uint64_t ckc = cpu->env.ckc >> 8;
> -void *mem;
> +struct sigp_save_area *sa;
> +hwaddr len = sizeof(*sa);
>  int i;
> -hwaddr len = SAVE_AREA_SIZE;
> 
> -mem = cpu_physical_memory_map(addr, , 1);
> -if (!mem) {
> +QEMU_BUILD_BUG_ON(sizeof(*sa) != 512);
> +
> +sa = cpu_physical_memory_map(addr, , 1);
> +if (!sa) {
>  return -EFAULT;
>  }
> -if (len != SAVE_AREA_SIZE) {
> -cpu_physical_memory_unmap(mem, len, 1, 0);
> +if (len != sizeof(*sa)) {
> +cpu_physical_memory_unmap(sa, len, 1, 0);
>  return -EFAULT;
>  }
> 
> @@ -1571,19 +1587,26 @@ static int kvm_s390_store_status(S390CPU *cpu, hwaddr 
> addr, bool store_arch)
>  cpu_physical_memory_write(offsetof(LowCore, ar_access_id), _id, 
> 1);
>  }
>  for (i = 0; i < 16; ++i) {
> -*((uint64_t *)mem + i) = get_freg(>env, i)->ll;
> -}
> -memcpy(mem + 128, >env.regs, 128);
> -memcpy(mem + 256, >env.psw, 16);
> -memcpy(mem + 280, >env.psa, 4);
> -memcpy(mem + 284, >env.fpc, 4);
> -memcpy(mem + 292, >env.todpr, 4);
> -memcpy(mem + 296, >env.cputm, 8);
> -memcpy(mem + 304, , 8);
> -memcpy(mem + 320, >env.aregs, 64);
> -memcpy(mem + 384, >env.cregs, 128);
> +sa->fprs[i] = cpu_to_be64(get_freg(>env, i)->ll);
> +}
> +for (i = 0; i < 16; ++i) {
> +sa->grs[i] = cpu_to_be64(cpu->env.regs[i]);
> +}
> +sa->psw.addr = cpu_to_be64(cpu->env.psw.addr);
> +sa->psw.mask = cpu_to_be64(get_psw_mask(>env));
> +sa->prefix = cpu_to_be32(cpu->env.psa);
> +sa->fpc = cpu_to_be32(cpu->env.fpc);
> +sa->todpr = cpu_to_be32(cpu->env.todpr);
> +sa->cputm = cpu_to_be64(cpu->env.cputm);
> +sa->ckc = cpu_to_be64(cpu->env.ckc >> 8);
> +for (i = 0; i < 16; ++i) {
> +sa->ars[i] = cpu_to_be32(cpu->env.aregs[i]);
> +}
> +for (i = 0; i < 16; ++i) {
> +sa->ars[i] = cpu_to_be64(cpu->env.cregs[i]);
> +}
> 
> -cpu_physical_memory_unmap(mem, len, 1, len);
> +cpu_physical_memory_unmap(sa, len, 1, len);
> 
>  return 0;
>  }
> 

Re: [Qemu-devel] [PATCH v1] s390x/kvm: fix and cleanup storing CPU status

[David Hildenbrand]

On 22.09.2017 15:12, Richard Henderson wrote:
> On 09/21/2017 04:30 PM, David Hildenbrand wrote:
>> +struct sigp_save_area {
>> +uint64_tfprs[16];   /* 0x */
>> +uint64_tgrs[16];/* 0x0080 */
>> +PSW psw;/* 0x0100 */
>> +uint8_t pad_0x0110[0x0118 - 0x0110];/* 0x0110 */
>> +uint32_tprefix; /* 0x0118 */
>> +uint32_tfpc;/* 0x011c */
>> +uint8_t pad_0x0120[0x0124 - 0x0120];/* 0x0120 */
>> +uint32_ttodpr;  /* 0x0124 */
>> +uint64_tcputm;  /* 0x0128 */
>> +uint64_tckc;/* 0x0130 */
>> +uint8_t pad_0x0138[0x0140 - 0x0138];/* 0x0138 */
>> +uint32_tars[16];/* 0x0140 */
>> +uint64_tcrs[16];/* 0x0384 */
>> +} QEMU_PACKED;
> 
> I don't believe you need to use QEMU_PACKED.  Unless the "addr" that's passed
> in can really be arbitrary?  (If so, the memory returned by
> cpu_physical_memory_map will be unaligned and so we do need the PACKED to 
> force
> the compiler to generate unaligned stores into the structure.)

The address is either 0x1200 or it is masked with 0x7e00u.

So most probably this should be fine without QEMU_PACKED, right?

> 
>> QEMU_BUILD_BUG_ON(sizeof(*sa) != 512);
> 
> Put this right next to the structure?
> 

Wasn't aware that this can be moved outside of a function. Thanks!

Thanks!

> 
> r~
> 


-- 

Thanks,

David

Re: [Qemu-devel] [PATCH v1] s390x/kvm: fix and cleanup storing CPU status

[Richard Henderson]

On 09/21/2017 04:30 PM, David Hildenbrand wrote:
> +struct sigp_save_area {
> +uint64_tfprs[16];   /* 0x */
> +uint64_tgrs[16];/* 0x0080 */
> +PSW psw;/* 0x0100 */
> +uint8_t pad_0x0110[0x0118 - 0x0110];/* 0x0110 */
> +uint32_tprefix; /* 0x0118 */
> +uint32_tfpc;/* 0x011c */
> +uint8_t pad_0x0120[0x0124 - 0x0120];/* 0x0120 */
> +uint32_ttodpr;  /* 0x0124 */
> +uint64_tcputm;  /* 0x0128 */
> +uint64_tckc;/* 0x0130 */
> +uint8_t pad_0x0138[0x0140 - 0x0138];/* 0x0138 */
> +uint32_tars[16];/* 0x0140 */
> +uint64_tcrs[16];/* 0x0384 */
> +} QEMU_PACKED;

I don't believe you need to use QEMU_PACKED.  Unless the "addr" that's passed
in can really be arbitrary?  (If so, the memory returned by
cpu_physical_memory_map will be unaligned and so we do need the PACKED to force
the compiler to generate unaligned stores into the structure.)

> QEMU_BUILD_BUG_ON(sizeof(*sa) != 512);

Put this right next to the structure?


r~