Re: netbsd32_{,u}int64 in sys/types.h for compat/sys/siginfo.h

[christos%astron.com@localhost (Christos Zoulas)]

In article <qr9j2b$17nj$1%blaine.gmane.org@localhost>,
Christos Zoulas <christos%astron.com@localhost> wrote:
>In article <468095c0-b973-7f23-1cfa-3dc19e3b8ddb%gmail.com@localhost>,
>Rin Okuyama  <rokuyama.rk%gmail.com@localhost> wrote:
>>On 2019/11/22 10:56, Christos Zoulas wrote:
>>> In article <679493cf-3e85-f56d-85e4-dfaf7958a379%gmail.com@localhost>,
>>> Rin Okuyama  <rokuyama.rk%gmail.com@localhost> wrote:
>>...
>>>> This was my misunderstanding. These codes are used when tracer is 64-bit
>>>> and traced is 32-bit. Don't know whether this is useful though...
>>> 
>>> Yes, and someone broke them because all the ptrace 64->32 calls for
>>> registers seem to return 0 now. Was that code changed recently?
>>...
>>
>>I fixed it:
>>http://mail-index.netbsd.org/source-changes/2019/11/22/msg111068.html
>>
>>Now, gdb/amd64 seems to work for i386 binaries to some extent, at least.
>
>Thanks! I think that the gdb on head should working for i386 binaries
>on amd64. I am installing a new kernel and userland and I will test
>shortly.

And it works fine for both static and dynamic binaries. We could also add
kernel debugging support for i386 kernels, but that needs changes in
the i386 header files:

1. Everywhere s/#include <machine/#include <i386/
   I think this is a good change anyway for all MD headers because they
   expect to load their own arch stuff. MI stuff needs to stay "<machine"
2. change __vaddr_t in segments.h ld_descriptor to uint32_t
   That should have no impact and it is a good change because it is a
   packed structure and needs to be fixed size.
3. Add some fixed field sizes in struct pcb instead of pointers with
   #ifdef __x86_64__ (or copy struct pcb into gdb like kamil did). This
   is just ugly, but I think it is better than copying the struct. I
   could add a comment explaining why.

Is it worth it?

chritod