Fix syscall numbers for non-x86_64 architectures

[topolarity]

This gets LinuxPerf.jl working for me on aarch64-linux:

julia> @pstats nothing
┌ Warning: LinuxPerf.EventTypeExt(hw:branches, false, 0x0000000000000006) not supported, skipping
└ @ LinuxPerf ~/.julia/dev/LinuxPerf/src/LinuxPerf.jl:299
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
┌ cpu-cycles               2.50e+02  100.0%  #  0.0 cycles per ns
│ stalled-cycles-frontend  2.18e+02  100.0%  # 87.2% of cycles
└ stalled-cycles-backend   2.30e+01  100.0%  #  9.2% of cycles
┌ instructions             1.80e+01  100.0%  #  0.1 insns per cycle
└ branch-misses            2.00e+00  100.0%
┌ task-clock               1.64e+05  100.0%  # 164.5 μs
│ context-switches         0.00e+00  100.0%
│ cpu-migrations           0.00e+00  100.0%
└ page-faults              0.00e+00  100.0%
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Fixes #39

(note: the warning is due to a bug in that hasn't been fixed yet in my kernel:
https://patchwork.kernel.org/project/linux-arm-kernel/patch/20230403091547.441550-1-peternewman@google.com/)

[Zentrik]

Do you know how much overhead this adds? Does it get compiled out, if so can we just ccall prctl in this branch.

[topolarity]

It gets compiled out 👍

julia> @code_typed LinuxPerf.disable_all!()
CodeInfo(
1 ─      nothing::Nothing
│   %2 = $(Expr(:foreigncall, :(:syscall), Int32, svec(Int64, Int64), 1, :(:ccall), 167, 31, 31, 167))::Int32
│   %3 = Core.sext_int(Core.Int64, %2)::Int64
│   %4 = Base.slt_int(%3, 0)::Bool
└──      goto #3 if not %4
2 ─      invoke Core.kwcall($(QuoteNode((extrainfo = nothing,)))::@NamedTuple{extrainfo::Nothing}, Base.systemerror::typeof(systemerror), :prctl::Symbol)::Union{}
└──      unreachable
3 ─      goto #4
4 ─      goto #5
5 ─      return nothing
) => Nothing

You can see that it goes straight into the :foreigncall there on the second line, which is the ccall

[Zentrik]

Seems sensible to then just ccall prctl instead of erroring.

[topolarity]

Maybe.. I'd prefer the error so that an issue gets raised against the repo and then this gets fixed

[Zentrik]

I mean we do

if Sys_prctl == -1
    ccall(:prctl, ...)
else
    ccall(:syscall,...)

Unless some machines don't have prctl as a syscall, this should always work right?

[topolarity]

@vchuravy It'd be nice to get a minor release with this + #38 + #42, when time permits