From 94e3e3652f9f04810126ee754fa9a788289e2897 Mon Sep 17 00:00:00 2001
From: Prefetch
Date: Mon, 24 Jul 2023 19:21:30 +0200
Subject: Reduce total code size by 53 bytes (big deal, right?)

---
 src/thread_create.asm | 52 +++++++++++++++++++++++++++++++--------------------
 1 file changed, 32 insertions(+), 20 deletions(-)

(limited to 'src/thread_create.asm')

diff --git a/src/thread_create.asm b/src/thread_create.asm
index 9a6fe78..8dc8813 100644
--- a/src/thread_create.asm
+++ b/src/thread_create.asm
@@ -72,12 +72,16 @@ linen_thread_create:
 		; Callee-save registers
 		push rbx
 
+		; It's handy to have a register that's 0 for a while
+		xor ecx, ecx
+
 		;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 		;;;; Check validity of arguments ;;;;
 		;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
-		; Return EINVAL if any argument is NULL
-		mov eax, -22 ; (EINVAL = -22)
+		; Return EINVAL (-22) if any argument is NULL
+		lea eax, [rcx - 22] ; mov eax, -22
+
 		test rdi, rdi
 		jz create_return ; Nowhere to store the thread handle
 		test rsi, rsi
@@ -91,9 +95,9 @@ linen_thread_create:
 		;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
 		; Save these registers: we'll clobber them for the mmap call
-		mov rbx, rdi
-		push rdx
-		push rsi
+		push rdx ; argument
+		push rsi ; function
+		push rdi ; thread handle destination
 
 		; The mmap system call does many things, in this case allocate memory.
 		; See: man 2 mmap
@@ -103,25 +107,26 @@ linen_thread_create:
 		; mmap: rsi = length: size of buffer to allocate
 		mov esi, (STACK_SIZE + GUARD_PAGE)
 		; mmap: rdx = prot: mprotect-style access permissions
-		mov edx, (PROT_WRITE | PROT_READ)
+		lea edx, [rcx + 3] ; mov edx, (PROT_READ | PROT_WRITE)
 		; mmap: r10 = flags: configuration flags for mapping:
 		; - MAP_ANONYMOUS: there is no file backing this buffer
 		; - MAP_PRIVATE:   only this process can see thread's stack
 		; - MAP_STACK:     no-op; inform kernel that this is a stack
-		mov r10, (MAP_ANONYMOUS | MAP_PRIVATE | MAP_STACK)
+		mov r10d, (MAP_ANONYMOUS | MAP_PRIVATE | MAP_STACK)
 		; mmap: r8 = fd: ignored for MAP_ANONYMOUS, recommended -1
-		mov r8, -1
+		lea r8, [rcx - 1] ; mov r8, -1
 		; mmap: r9 = offset: should be 0 when MAP_ANONYMOUS is used
 		xor r9, r9
 		; mmap: rax = system call ID
-		mov eax, SYS_MMAP
-		; mmap: rax = mmap(rdi, rsi, rdx, r10, r8, 9)
+		lea eax, [rcx + SYS_MMAP] ; mov eax, SYS_MMAP
+		; mmap: rax = mmap(rdi, rsi, rdx, r10, (r8), (r9))
 		syscall
 
 		; Pop these now before we start branching. Those registers
 		; won't be used by the next system calls, so they're safe.
-		pop r8 ; function
-		pop r9 ; argument
+		pop rbx ; thread handle destination
+		pop r8  ; function
+		pop r9  ; argument
 
 		; Check result of mmap: negative means failure,
 		; otherwise rax is the address of the new mapping.
@@ -145,12 +150,13 @@ linen_thread_create:
 		; mprotect: rdx = prot: access permissions; zero for none
 		xor edx, edx
 		; mprotect: rax = system call ID
-		mov eax, SYS_MPROTECT
+		xor eax, eax
+		mov al, SYS_MPROTECT
 		; mprotect: rax = mprotect(rdi, rsi, rdx)
 		syscall
 
 		; Check result of mprotect: nonzero means failure
-		test rax, rax
+		test eax, eax
 		jnz create_return
 
 		;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
@@ -162,7 +168,7 @@ linen_thread_create:
 		; once in the parent (0 if success) and once in the child (the TID).
 		; See: man 2 clone
 
-		; clone: rsi = stack
+		; clone: rsi = stack: pointer for child's initial rsp
 		; Currently rdi points to the lowest byte of the stack area.
 		; Again, stacks grow downward, so we calculate the address of
 		; the top qword to use as the child thread's starting point.
@@ -194,14 +200,14 @@ linen_thread_create:
 		; clone: r8 = tls: ignored unless CLONE_SETTLS is used
 
 		; clone: rax = system call ID
-		mov eax, SYS_CLONE
+		mov al, SYS_CLONE
 		; clone: rax = clone(rdi, rsi, (rdx), r10, (r8));
 		syscall
 
 		; Ideally, both parent and new-born child are executing this code now.
 
 		; Check result of clone:
-		test rax, rax
+		test eax, eax
 		js create_return   ; Negative means failure
 		jnz create_success ; Positive means we're in the parent thread
 		; Zero means we're in the child thread
@@ -220,10 +226,16 @@ linen_thread_create:
 		; Once done, leave function's return value lying around
 		push rax
 
-		; Exit the thread with return value 0
+		; Exit the thread with status 0 using the exit system call.
+		; See: man 2 exit
+
+		; exit: rdi = status to report
 		xor edi, edi
-		mov rax, SYS_EXIT
-		syscall ; (never returns)
+		; exit: rax = system call ID
+		xor eax, eax
+		mov al, SYS_EXIT
+		; exit: call never returns
+		syscall
 
 		;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 		;;;; Clean up in parent thread ;;;;
-- 
cgit v1.2.3