According to the architecture, SET PREFIX must try to access the new
prefix area and recognize an addressing exception if the area is not
accessible.
For qemu this check prevents a crash in cpu_map_lowcore after an
inaccessible prefix area has been set.

Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
---


Is there a stricter check to see if the memory is accessible?


 target/s390x/tcg/misc_helper.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/target/s390x/tcg/misc_helper.c b/target/s390x/tcg/misc_helper.c
index aab9c47747..c8447b36fc 100644
--- a/target/s390x/tcg/misc_helper.c
+++ b/target/s390x/tcg/misc_helper.c
@@ -158,6 +158,10 @@ void HELPER(spx)(CPUS390XState *env, uint64_t a1)
     if (prefix == old_prefix) {
         return;
     }
+    if (!mmu_absolute_addr_valid(prefix, true) ||
+        !mmu_absolute_addr_valid(prefix + TARGET_PAGE_SIZE, true)) {
+        tcg_s390_program_interrupt(env, PGM_ADDRESSING, GETPC());
+    }
 
     env->psa = prefix;
     HELPER_LOG("prefix: %#x\n", prefix);

base-commit: 3a821c52e1a30ecd9a436f2c67cc66b5628c829f
-- 
2.36.1

On 27.06.22 15:12, Janis Schoetterl-Glausch wrote:
> According to the architecture, SET PREFIX must try to access the new
> prefix area and recognize an addressing exception if the area is not
> accessible.
> For qemu this check prevents a crash in cpu_map_lowcore after an
> inaccessible prefix area has been set.

I don't think that's possible. Our memory increments are 1 MiB and one
would have to cross a 1~MiB range with the second page to trigger that.
IIRC that's impossible with SPX address alignment requirements?


> 
> Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
> ---
> 
> 
> Is there a stricter check to see if the memory is accessible?
> 
> 
>  target/s390x/tcg/misc_helper.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/target/s390x/tcg/misc_helper.c b/target/s390x/tcg/misc_helper.c
> index aab9c47747..c8447b36fc 100644
> --- a/target/s390x/tcg/misc_helper.c
> +++ b/target/s390x/tcg/misc_helper.c
> @@ -158,6 +158,10 @@ void HELPER(spx)(CPUS390XState *env, uint64_t a1)
>      if (prefix == old_prefix) {
>          return;
>      }
> +    if (!mmu_absolute_addr_valid(prefix, true) ||
> +        !mmu_absolute_addr_valid(prefix + TARGET_PAGE_SIZE, true)) {
> +        tcg_s390_program_interrupt(env, PGM_ADDRESSING, GETPC());
> +    }
>  
>      env->psa = prefix;
>      HELPER_LOG("prefix: %#x\n", prefix);
> 
> base-commit: 3a821c52e1a30ecd9a436f2c67cc66b5628c829f


-- 
Thanks,

David / dhildenb

On 6/27/22 18:27, David Hildenbrand wrote:
> On 27.06.22 15:12, Janis Schoetterl-Glausch wrote:
>> According to the architecture, SET PREFIX must try to access the new
>> prefix area and recognize an addressing exception if the area is not
>> accessible.
>> For qemu this check prevents a crash in cpu_map_lowcore after an
>> inaccessible prefix area has been set.
> 
> I don't think that's possible. Our memory increments are 1 MiB and one
> would have to cross a 1~MiB range with the second page to trigger that.
> IIRC that's impossible with SPX address alignment requirements?
> 
Are you saying that checking the first page is sufficient?
I'm not sure that this the case for the architecture in general,
but I guess it is true for tcg. Do you want me to remove the second check?
> 
>>
>> Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
>> ---
>>
>>
>> Is there a stricter check to see if the memory is accessible?
>>
>>
>>  target/s390x/tcg/misc_helper.c | 4 ++++
>>  1 file changed, 4 insertions(+)
>>
>> diff --git a/target/s390x/tcg/misc_helper.c b/target/s390x/tcg/misc_helper.c
>> index aab9c47747..c8447b36fc 100644
>> --- a/target/s390x/tcg/misc_helper.c
>> +++ b/target/s390x/tcg/misc_helper.c
>> @@ -158,6 +158,10 @@ void HELPER(spx)(CPUS390XState *env, uint64_t a1)
>>      if (prefix == old_prefix) {
>>          return;
>>      }
>> +    if (!mmu_absolute_addr_valid(prefix, true) ||
>> +        !mmu_absolute_addr_valid(prefix + TARGET_PAGE_SIZE, true)) {
>> +        tcg_s390_program_interrupt(env, PGM_ADDRESSING, GETPC());
>> +    }
>>  
>>      env->psa = prefix;
>>      HELPER_LOG("prefix: %#x\n", prefix);
>>
>> base-commit: 3a821c52e1a30ecd9a436f2c67cc66b5628c829f
> 
>

On 27.06.22 19:06, Janis Schoetterl-Glausch wrote:
> On 6/27/22 18:27, David Hildenbrand wrote:
>> On 27.06.22 15:12, Janis Schoetterl-Glausch wrote:
>>> According to the architecture, SET PREFIX must try to access the new
>>> prefix area and recognize an addressing exception if the area is not
>>> accessible.
>>> For qemu this check prevents a crash in cpu_map_lowcore after an
>>> inaccessible prefix area has been set.
>>
>> I don't think that's possible. Our memory increments are 1 MiB and one
>> would have to cross a 1~MiB range with the second page to trigger that.
>> IIRC that's impossible with SPX address alignment requirements?
>>
> Are you saying that checking the first page is sufficient?

Yes, unless I'm wrong. :)

/* Due to alignment and QEMU memory sizes, it's sufficient to check the
first page only. */

> I'm not sure that this the case for the architecture in general,
> but I guess it is true for tcg. Do you want me to remove the second check?

Should have been clearer (and I somehow thought we'd be checking the
first page already :))

Reviewed-by: David Hildenbrand <david@redhat.com>

-- 
Thanks,

David / dhildenb