Support riscv64

[giordano]

Fix #55.

I can confirm tests pass locally:

     Testing Running tests...
hardware:
  branch-instructions      user, hypervisor
  branch-misses            user, hypervisor
  cache-misses             user, hypervisor
  cache-references         user, hypervisor
  cpu-cycles               user, hypervisor
  instructions             user, hypervisor
  stalled-cycles-backend   user, hypervisor
  stalled-cycles-frontend  user, hypervisor

software:
  alignment-faults         user, hypervisor
  bpf-output               user, hypervisor
  context-switches         user, hypervisor
  cpu-clock                user, hypervisor
  cpu-migrations           user, hypervisor
  dummy                    user, hypervisor
  emulation-faults         user, hypervisor
  major-faults             user, hypervisor
  minor-faults             user, hypervisor
  page-faults              user, hypervisor
  task-clock               user, hypervisor

cache:
  L1-dcache-load-misses    user, hypervisor
  L1-dcache-loads          user, hypervisor
  L1-dcache-store-misses   user, hypervisor
  L1-dcache-stores         user, hypervisor
  L1-icache-load-misses    user, hypervisor
  L1-icache-loads          user, hypervisor
  LLC-load-misses          user, hypervisor
  LLC-loads                user, hypervisor
  LLC-store-misses         user, hypervisor
  LLC-stores               user, hypervisor
  dTLB-load-misses         user, hypervisor
  dTLB-loads               user, hypervisor
  iTLB-load-misses         user, hypervisor
  iTLB-loads               user, hypervisor
Test Summary: | Pass  Total   Time
LinuxPerf     |   37     37  17.1s
     Testing LinuxPerf tests passed

Question: I presume it'd be nicer to add the llvmcall in

LinuxPerf.jl/src/LinuxPerf.jl

Lines 359 to 384 in e7dfcbd

	@inline function prctl(op)
	if SYS_prctl == -1
	res = ccall(:prctl, Cint, (Cint, Cint...), op)
	return Base.systemerror(:prctl, res < 0)
	elseif Sys.ARCH == :x86_64
	res = Base.llvmcall("""%val = call i64 asm sideeffect "syscall", "={rax},{rax},{rdi},~{rcx},~{r11},~{memory}"(i64 %0, i64 %1)
	ret i64 %val""", Int64, Tuple{Int64, Int64}, SYS_prctl, Int64(op))
	return (res >= 0) ? nothing : throw(Base.SystemError("prctl", -res, nothing))
	elseif Sys.ARCH == :i686
	res = Base.llvmcall("""%val = call i32 asm sideeffect "int \$\$0x80", "={eax},{eax},{ebx},~{memory}"(i32 %0, i32 %1)
	ret i32 %val""", Int32, Tuple{Int32, Int32}, SYS_prctl, Int32(op))
	return (res >= 0) ? nothing : throw(Base.SystemError("prctl", -res, nothing))
	elseif Sys.ARCH == :aarch64
	res = Base.llvmcall("""%val = call i64 asm sideeffect "svc #0", "={x0},{x8},{x0},~{memory}"(i64 %0, i64 %1)
	ret i64 %val""", Int64, Tuple{Int64, Int64}, SYS_prctl, Int64(op))
	return (res >= 0) ? nothing : throw(Base.SystemError("prctl", -res, nothing))
	elseif Sys.ARCH == :arm
	res = Base.llvmcall("""%val = call i32 asm sideeffect "swi 0", "={r0},{r7},{r0},~{memory}"(i32 %0, i32 %1)
	ret i32 %val""", Int32, Tuple{Int32, Int32}, SYS_prctl, Int32(op))
	return (res >= 0) ? nothing : throw(Base.SystemError("prctl", -res, nothing))
	else
	# syscall is lower overhead than calling libc's prctl
	res = ccall(:syscall, Cint, (Clong, Clong...), SYS_prctl, op)
	return Base.systemerror(:prctl, res < 0)
	end
	end

?

[Zentrik]

Question: I presume it'd be nicer to add the llvmcall in

Yes, if you want a slightly lower overhead.

[giordano]

Oh, this is a nicer table as a reference, it has all architectures: https://gpages.juszkiewicz.com.pl/syscalls-table/syscalls.html

[Zentrik]

Lovely

[Zentrik]

This and risc use the generic syscall table in the kernel btw

[giordano]

I did find https://github.com/torvalds/linux/blob/b62cef9a5c673f1b8083159f5dc03c1c5daced2f/arch/arm64/tools/syscall_64.tbl (while I didn't find any symlink at all for riscv64), I was a bit hesitant to link here a generic table, but I guess that's what it is. Ok, I'll update the reference later.