Re: [HPPAQEMU-Devel] A few small patches as my self introduction

[Stuart B. <sd...@nt...>]

On Fri, Apr 20, 2007 at 07:28:59PM +0900, Tomonori Yamane wrote:
> I am really happy to know that someone else started to work
> on PA-RISC emulation on QEMU.

Thanks!  Seems quite a few people are interested in PA-RISC emulation. :-)

> I have developed HP-UX user mode emulation
> and found my interpreter is slow
> and I am concerned the legality of running HP-UX application
> on an emulated system. ( I need to copy HP-UX shared libraries
> to run HP-UX applications.)

That's quite impressive!  May I ask what the host system was?

> Aside from the legality, I am interested in the fast emulator.
> 
> Therefore I would like to know what I can do for your project.
> 
> I attached several patches as my self introduction.
> I hope they may be some help.

The patches you attached are exactly the sort of thing that's needed.

However, I think we're now at the point where we need to start testing
small binaries (such as /bin/true) and start debugging.  I've found that
load ops are crashing, and I'm not sure why...

> (1) First one (cpu.h) is a reorder of the PSW flags according to the 
> PA-RISC 1.1 Architecture and Instruction Set Reference Manual.
> (This is just bitwise reverse.)

*Blush* -- I think the PSW flags were one of the first things I did...
I obviously got confused by the big endian bit enumeration that's used
throughout the PA docs.  Committed -- thanks for spotting this!

> (2) The second one (op.c ) is corresponding modifications of 
> ssm/rsm/mtsm

Committed.

> (3) The third one (op.c) fixes idcor / dcor instruction from misuse 
> of  'and' operator.

Comitted (and simplified).

> (4) The fourth one (op.c) fixes op_shrpw_cc() in case of  32bit left shift.
>      ( I have stacked this problem for 2 months and I don't want 
> you stack the same problem.)
> 
> 0x12345678 << 32  => 0 on HP-UX
> 0x12345678 << 32 => 0x12345678 on Linux/Pentium (No optimize)
> 0x12345678 << 32 => 0 on Linux/Pentium (Optimized)

Ouch...  C99 states that the result is undefined.  I've added the
missing FORCE_RET() and committed this fix.  Thanks.

(BTW, I noticed that your email client damaged the patch in a few
places.  Sending patches as attachments should avoid the problem.)

Many thanks again for your patches.  I'm looking forward to any more you
may send. :-)

Cheers,
-- 
Stuart Brady