Re: [RFC 1/3] [gdb] Call gdbarch_get_syscall_number less often

[Simon Marchi <simark@simark.ca>]

On 2023-11-20 11:12, Simon Marchi wrote:
> I like this change, since (if we do things right, i.e. invalidate
> syscall_number at the right plcaes), lp->syscall_number is more
> trustworthy than gdbarch_get_syscall_number.
> 
> But, this overlaps a bit with your other patch "Use
> PTRACE_GET_SYSCALL_INFO for syscall number".  I'm not sure how you would
> reconcile the two patches, but I think the general approach (since we
> don't have a one size fits all solution) should be to try the most
> trustworthy things first.  So that would be:
> 
>  1. PTRACE_GET_SYSCALL_INFO, if available
>  2. lp->syscall, if available
>  3. gdbarch_get_syscall_number
> 
> In that other patch, you wrap the gdbarch_get_syscall_number call to try
> PTRACE_GET_SYSCALL_INFO before gdbarch_get_syscall_number, so depending
> on how the two patches are merged, the code might not match the order
> above.

So I hadn't read your commit message fully for that other patch, but
John' reply made me realize that PTRACE_GET_SYSCALL_INFO doesn't give
the syscall number for exit... that's a bummer.

Simon