[Qemu-devel] softmmu_header: ldb_kernel vs. ldsb_kernel

[Qemu-devel] softmmu_header: ldb_kernel vs. ldsb_kernel

[Clemens Kolbitsch]

hi everyone!
i have a (maybe rather tricky) question:

why do you define 2 different inline load-functions in softmmu_header:

static inline int glue(glue(lds, SUFFIX), MEMSUFFIX)(target_ulong ptr)

vs.

static inline RES_TYPE glue(glue(ld, USUFFIX), MEMSUFFIX)(target_ulong ptr)

??

Obviously this is for signed/unsigned access... but why do you need it? In 
case there is a TLB miss, both functions call the same function (e.g. 
__ldl_mmu) --> the return value cannot differ that much.

The only difference I see (that really matters) is how the bytes are copied to 
the result-pointer (i.e. using movzbl vs. movsbl)... but that's it.

If there is some deeper reason behind that - could you please point that out 
to me? And if there is such a thing, why is it not necessary for storing 
(e.g. stb_kernel)??

Thanks!!
Clemens

Re: [Qemu-devel] softmmu_header: ldb_kernel vs. ldsb_kernel

[Thiemo Seufer]

Clemens Kolbitsch wrote:
> hi everyone!
> i have a (maybe rather tricky) question:
> 
> why do you define 2 different inline load-functions in softmmu_header:
> 
> static inline int glue(glue(lds, SUFFIX), MEMSUFFIX)(target_ulong ptr)
> 
> vs.
> 
> static inline RES_TYPE glue(glue(ld, USUFFIX), MEMSUFFIX)(target_ulong ptr)
> 
> ??
> 
> Obviously this is for signed/unsigned access... but why do you need it? In 
> case there is a TLB miss, both functions call the same function (e.g. 
> __ldl_mmu) --> the return value cannot differ that much.
> 
> The only difference I see (that really matters) is how the bytes are copied to 
> the result-pointer (i.e. using movzbl vs. movsbl)... but that's it.

It is a cast. The generic C version for the other architectures makes
this more obvious.

> If there is some deeper reason behind that - could you please point that out 
> to me? And if there is such a thing, why is it not necessary for storing 
> (e.g. stb_kernel)??

A load (sign-)extends a value to register size, a store doesn't.

That's why you have e.g. in the MIPS instruction set LB, LBU and SB
but no SBU, it would do the same as SB.


Thiemo

Re: [Qemu-devel] softmmu_header: ldb_kernel vs. ldsb_kernel

[Clemens Kolbitsch]

> > The only difference I see (that really matters) is how the bytes are
> > copied to the result-pointer (i.e. using movzbl vs. movsbl)... but that's
> > it.
>
> It is a cast. The generic C version for the other architectures makes
> this more obvious.
>
> > If there is some deeper reason behind that - could you please point that
> > out to me? And if there is such a thing, why is it not necessary for
> > storing (e.g. stb_kernel)??
>
> A load (sign-)extends a value to register size, a store doesn't.
>
> That's why you have e.g. in the MIPS instruction set LB, LBU and SB
> but no SBU, it would do the same as SB.

all right .. thanks for the fast answer :-)