[Qemu-devel] TCG debugging improvements

[Qemu-devel] TCG debugging improvements

[Paul Brook]

I'd just applied changes that enforce type size checking for TCG variables 
when DEBUG_TCGV is defined. This should help avoid subtle bugs due to 
64/32-bit mismatches.

I fixed a few bugs in the process, but it's mostly mechanical changes.  TCGv 
is target_ulong sized, so the majority of code remains unchanged.  Explicit 
sizes are typically only required when working with different sized ops, or 
floating point.

One of the main side-effects of this is that helper functions now need to be 
declared properly.  Previously we sort-of did this, but not consistently 
enough to enforce types.  With the exception of the SSE/MMX ops (which are 
table driven and use messy typecasts) we should be able to detect 
inconsistencies between the code generator and the helper function 
implementations.

There should be no functional changes.  I've tested what I can, but please 
test your favourite host/target.

Paul

Re: [Qemu-devel] TCG debugging improvements

[Blue Swirl]

On 11/17/08, Paul Brook <paul@codesourcery.com> wrote:
> I'd just applied changes that enforce type size checking for TCG variables
>  when DEBUG_TCGV is defined. This should help avoid subtle bugs due to
>  64/32-bit mismatches.
>
>  I fixed a few bugs in the process, but it's mostly mechanical changes.  TCGv
>  is target_ulong sized, so the majority of code remains unchanged.  Explicit
>  sizes are typically only required when working with different sized ops, or
>  floating point.
>
>  One of the main side-effects of this is that helper functions now need to be
>  declared properly.  Previously we sort-of did this, but not consistently
>  enough to enforce types.  With the exception of the SSE/MMX ops (which are
>  table driven and use messy typecasts) we should be able to detect
>  inconsistencies between the code generator and the helper function
>  implementations.
>
>  There should be no functional changes.  I've tested what I can, but please
>  test your favourite host/target.

PPC is broken. This used to print something, now it doesn't:
./qemu-system-ppc -L ./pc-bios/ -hda /dev/null

Sparc32 and Sparc64 are fine, thanks for the fixes.

Re: [Qemu-devel] TCG debugging improvements

[Andreas Färber]

Am 17.11.2008 um 21:08 schrieb Blue Swirl:

> On 11/17/08, Paul Brook <paul@codesourcery.com> wrote:
>> I'd just applied changes that enforce type size checking for TCG  
>> variables
>> when DEBUG_TCGV is defined. This should help avoid subtle bugs due to
>> 64/32-bit mismatches.
>>
>> I fixed a few bugs in the process, but it's mostly mechanical  
>> changes.  TCGv
>> is target_ulong sized, so the majority of code remains unchanged.   
>> Explicit
>> sizes are typically only required when working with different sized  
>> ops, or
>> floating point.
>>
>> One of the main side-effects of this is that helper functions now  
>> need to be
>> declared properly.  Previously we sort-of did this, but not  
>> consistently
>> enough to enforce types.  With the exception of the SSE/MMX ops  
>> (which are
>> table driven and use messy typecasts) we should be able to detect
>> inconsistencies between the code generator and the helper function
>> implementations.
>>
>> There should be no functional changes.  I've tested what I can, but  
>> please
>> test your favourite host/target.
>
> PPC is broken. This used to print something, now it doesn't:
> ./qemu-system-ppc -L ./pc-bios/ -hda /dev/null

ppc64 mixed tl and i64 in some places, possibly also ppc tl and i32  
for SPE, maybe that's why?

Re: [Qemu-devel] TCG debugging improvements

[Aurelien Jarno]

Blue Swirl a écrit :
> On 11/17/08, Paul Brook <paul@codesourcery.com> wrote:
>> I'd just applied changes that enforce type size checking for TCG variables
>>  when DEBUG_TCGV is defined. This should help avoid subtle bugs due to
>>  64/32-bit mismatches.
>>
>>  I fixed a few bugs in the process, but it's mostly mechanical changes.  TCGv
>>  is target_ulong sized, so the majority of code remains unchanged.  Explicit
>>  sizes are typically only required when working with different sized ops, or
>>  floating point.
>>
>>  One of the main side-effects of this is that helper functions now need to be
>>  declared properly.  Previously we sort-of did this, but not consistently
>>  enough to enforce types.  With the exception of the SSE/MMX ops (which are
>>  table driven and use messy typecasts) we should be able to detect
>>  inconsistencies between the code generator and the helper function
>>  implementations.
>>
>>  There should be no functional changes.  I've tested what I can, but please
>>  test your favourite host/target.
> 
> PPC is broken. This used to print something, now it doesn't:
> ./qemu-system-ppc -L ./pc-bios/ -hda /dev/null
> 

That should be fixed in the SVN.

-- 
  .''`.  Aurelien Jarno	            | GPG: 1024D/F1BCDB73
 : :' :  Debian developer           | Electrical Engineer
 `. `'   aurel32@debian.org         | aurelien@aurel32.net
   `-    people.debian.org/~aurel32 | www.aurel32.net