diff --git a/arch/x86/config/kernel32.ld b/arch/x86/config/kernel32.ld
index 38e8fa3..668ec85 100644
--- a/arch/x86/config/kernel32.ld
+++ b/arch/x86/config/kernel32.ld
@@ -34,6 +34,11 @@ SECTIONS {
 
 	.text ALIGN(4K) : AT(ADDR(.text) - KERNBASE) {
 		_text_start = .;
+		/* put all ISRs into one contiguous region so the
+		 * stack unwinder knows when to stop unwinding */
+		_isr_start = .;
+		KEEP(*(.text.isr))
+		_isr_end = .;
 		*(.text .text.* .gnu.linkonce.t.*)
 		_text_end = .;
 	}
diff --git a/arch/x86/config/kernel64.ld b/arch/x86/config/kernel64.ld
index 2e22a76..f0f18bd 100644
--- a/arch/x86/config/kernel64.ld
+++ b/arch/x86/config/kernel64.ld
@@ -34,6 +34,11 @@ SECTIONS {
 
 	.text ALIGN(PAGE_SIZE) : AT(ADDR(.text) - KERNBASE) {
 		_text_start = .;
+		/* put all ISRs into one contiguous region so the
+		 * stack unwinder knows when to stop unwinding */
+		_isr_start = .;
+		KEEP(*(.text.isr))
+		_isr_end = .;
 		*(.text .text.* .gnu.linkonce.t.*)
 		_text_end = .;
 	}
diff --git a/arch/x86/config/toolchain.cmake b/arch/x86/config/toolchain.cmake
index e05fe5b..349328c 100644
--- a/arch/x86/config/toolchain.cmake
+++ b/arch/x86/config/toolchain.cmake
@@ -26,7 +26,8 @@ set(CMAKE_RANLIB        ${TOOLCHAIN_PATH}/${CMAKE_EXECUTABLE_PREFIX}ranlib${CMAK
 set(CMAKE_SZE           ${TOOLCHAIN_PATH}/${CMAKE_EXECUTABLE_PREFIX}szr${CMAKE_EXECUTABLE_SUFFIX} CACHE INTERNAL "")
 set(CMAKE_STRIP         ${TOOLCHAIN_PATH}/${CMAKE_EXECUTABLE_PREFIX}strip${CMAKE_EXECUTABLE_SUFFIX} CACHE INTERNAL "")
 
-set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${_toolchain_common_flags} -mcmodel=kernel -mno-red-zone -mno-mmx -mno-sse -msoft-float")
+# XXX ktrace_print() currently relies on -fno-omit-frame-pointer
+set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${_toolchain_common_flags} -mcmodel=kernel -mno-red-zone -mno-mmx -mno-sse -msoft-float -fno-omit-frame-pointer")
 set(CMAKE_ASM_FLAGS "${CMAKE_ASM_FLAGS} ${_toolchain_common_flags}")
 set(CMAKE_C_COMPILER_TARGET ${_toolchain_triple})
 set(CMAKE_ASM_COMPILER_TARGET ${_toolchain_triple})
diff --git a/arch/x86/sys/amd64/irq.S b/arch/x86/sys/amd64/irq.S
index 6c637e0..20cbc5d 100644
--- a/arch/x86/sys/amd64/irq.S
+++ b/arch/x86/sys/amd64/irq.S
@@ -18,13 +18,14 @@
 	.extern irq_table
 
 	/* there is probably a fancy CPU feature for this, but idk */
-irq_count:
+L_DATA(irq_count)
 	.long 0
+L_END(irq_count)
 
 	.text
 
 /* bool in_irq(void); */
-ASM_ENTRY(in_irq)
+ENTRY(in_irq)
 	movabsq	$irq_count, %rdx
 	xor	%eax, %eax
 	mov	%eax, %ecx
@@ -32,10 +33,12 @@ ASM_ENTRY(in_irq)
 	testl	%ecx, (%rdx)
 	setne	%al
 	retq
-ASM_END(in_irq)
+END(in_irq)
+
+	.section .text.isr
 
 .macro gen_irq num
-ASM_ENTRY(_x86_isr_irq\num )
+ENTRY(_x86_isr_irq\num )
 	push	%rax
 	push	%rcx
 	push	%rdx
@@ -53,7 +56,7 @@ ASM_ENTRY(_x86_isr_irq\num )
 	movabsq	$(irq_table + \num * 8), %rax
 	callq	*%rax
 	jmp	leave_irq
-ASM_END(_x86_isr_irq\num )
+END(_x86_isr_irq\num )
 .endm
 
 	gen_irq 0
@@ -73,8 +76,7 @@ ASM_END(_x86_isr_irq\num )
 	gen_irq 14
 	gen_irq 15
 
-	.align 4
-leave_irq:
+L_ENTRY(leave_irq)
 	movabsq	$irq_count, %rax
 	decl	(%rax)
 	pop	%r11
@@ -87,5 +89,4 @@ leave_irq:
 	pop	%rcx
 	pop	%rax
 	iretq
-
-	.size leave_irq, . - leave_irq
+L_END(leave_irq)
diff --git a/arch/x86/sys/amd64/ktrace.c b/arch/x86/sys/amd64/ktrace.c
index 42e5c13..62249b3 100644
--- a/arch/x86/sys/amd64/ktrace.c
+++ b/arch/x86/sys/amd64/ktrace.c
@@ -1,18 +1,29 @@
 /* Copyright (C) 2021 fef <owo@fef.moe>.  All rights reserved. */
 
-#include <arch/vmparam.h>
-
 #include <gay/kprintf.h>
 #include <gay/ktrace.h>
+#include <gay/linker.h>
 
-void ktrace_print(void)
+__naked void ktrace_print(void)
 {
-	void **rbp;
-	__asm__ volatile("movq (%%rbp), %0" : "=r"(rbp));
+	__asm__ volatile(
+"	movq	%rbp, %rdi		\n"
+"	jmp	ktrace_print_from	\n"
+	);
+}
 
+void ktrace_print_from(void *frame)
+{
+	void **rbp = (void **)frame;
 	kprintf("Stack trace:\n");
-	while (rbp >= (void **)KERNBASE) {
+
+	/* XXX Rather than spitting out raw addresses, parse the kernel image's
+	 *     ELF sections to figure out what the address actually belongs to */
+	while (rbp >= (void **)image_start && rbp < (void **)image_end) {
+		/* caller return address is immediately above the stack frame */
 		kprintf("  %p\n", rbp[1]);
+		if (rbp[1] >= isr_start && rbp[1] < isr_end)
+			break;
 		rbp = *rbp;
 	}
 }
diff --git a/arch/x86/sys/amd64/trap.S b/arch/x86/sys/amd64/trap.S
index 3e7300c..e72d955 100644
--- a/arch/x86/sys/amd64/trap.S
+++ b/arch/x86/sys/amd64/trap.S
@@ -4,7 +4,7 @@
 
 #include <asm/common.h>
 
-	.text
+	.section .text.isr
 
 /*
  * push all registers except %rsi to the stack, and store a pointer to the
@@ -64,7 +64,7 @@
 
 .macro gen_isr_noerror name
 	.extern x86_isr_\name
-ASM_ENTRY(_x86_isr_\name )
+ENTRY(_x86_isr_\name )
 	cld
 	pushq	%rsi
 	prepare_trap_entry
@@ -74,7 +74,7 @@ ASM_ENTRY(_x86_isr_\name )
 	prepare_trap_leave
 	popq	%rsi
 	iretq
-ASM_END(_x86_isr_\name )
+END(_x86_isr_\name )
 .endm
 
 /*
@@ -85,7 +85,7 @@ ASM_END(_x86_isr_\name )
  */
 .macro gen_isr_error name
 	.extern x86_isr_\name
-ASM_ENTRY(_x86_isr_\name )
+ENTRY(_x86_isr_\name )
 	cld
 
 	/*
@@ -103,7 +103,7 @@ ASM_ENTRY(_x86_isr_\name )
 	prepare_trap_leave
 	popq	%rsi
 	iretq
-ASM_END(_x86_isr_\name )
+END(_x86_isr_\name )
 .endm
 
 gen_isr_noerror divide_error
diff --git a/arch/x86/sys/i386/CMakeLists.txt b/arch/x86/sys/i386/CMakeLists.txt
index 0ee3c06..9068a0e 100644
--- a/arch/x86/sys/i386/CMakeLists.txt
+++ b/arch/x86/sys/i386/CMakeLists.txt
@@ -3,6 +3,7 @@
 target_sources(gay_arch PRIVATE
     idt.S
     irq.S
+    ktrace.c
     switch.S
     systm.c
     trap.S
diff --git a/arch/x86/sys/i386/irq.S b/arch/x86/sys/i386/irq.S
index 025d5e1..0706c04 100644
--- a/arch/x86/sys/i386/irq.S
+++ b/arch/x86/sys/i386/irq.S
@@ -12,21 +12,24 @@
 	.extern irq_table
 
 	/* there is probably a fancy CPU feature for this, but idk */
-irq_count:
+L_DATA(irq_count)
 	.byte 0
+L_END(irq_count)
 
 	.text
 
 /* bool in_irq(void); */
-ASM_ENTRY(in_irq)
+ENTRY(in_irq)
 	xor	%eax, %eax
 	cmpb	$0, irq_count
 	setne	%al
 	ret
-ASM_END(in_irq)
+END(in_irq)
+
+	.section .text.isr
 
 .macro gen_irq num
-ASM_ENTRY(_x86_isr_irq\num )
+ENTRY(_x86_isr_irq\num )
 	push	%eax
 	push	%ecx
 	push	%edx
@@ -39,7 +42,7 @@ ASM_ENTRY(_x86_isr_irq\num )
 	add	$4, %esp
 #endif
 	jmp	leave_irq
-ASM_END(_x86_isr_irq\num )
+END(_x86_isr_irq\num )
 .endm
 
 	gen_irq 0
@@ -59,12 +62,10 @@ ASM_END(_x86_isr_irq\num )
 	gen_irq 14
 	gen_irq 15
 
-	.align 4
-leave_irq:
+L_ENTRY(leave_irq)
 	decb	irq_count
 	pop	%edx
 	pop	%ecx
 	pop	%eax
 	iret
-
-	.size leave_irq, . - leave_irq
+L_END(leave_irq)
diff --git a/arch/x86/sys/i386/ktrace.c b/arch/x86/sys/i386/ktrace.c
new file mode 100644
index 0000000..6d29cea
--- /dev/null
+++ b/arch/x86/sys/i386/ktrace.c
@@ -0,0 +1,30 @@
+/* Copyright (C) 2021 fef <owo@fef.moe>.  All rights reserved. */
+
+#include <gay/kprintf.h>
+#include <gay/ktrace.h>
+#include <gay/linker.h>
+
+__naked void ktrace_print(void)
+{
+	__asm__ volatile(
+"	pushl	%ebp			\n"
+"	call	ktrace_print_from	\n"
+"	ret				\n"
+	);
+}
+
+void ktrace_print_from(void *frame)
+{
+	void **ebp = (void **)frame;
+	kprintf("Stack trace:\n");
+
+	/* XXX Rather than spitting out raw addresses, parse the kernel image's
+	 *     ELF sections to figure out what the address actually belongs to */
+	while (ebp >= (void **)image_start && ebp < (void **)image_end) {
+		/* caller return address is immediately above the stack frame */
+		kprintf("  %p\n", ebp[1]);
+		if (rbp[1] >= isr_start && rbp[1] < isr_end)
+			break;
+		rbp = *rbp;
+	}
+}
diff --git a/arch/x86/sys/i386/trap.S b/arch/x86/sys/i386/trap.S
index 754c508..c3378ab 100644
--- a/arch/x86/sys/i386/trap.S
+++ b/arch/x86/sys/i386/trap.S
@@ -4,7 +4,7 @@
 
 #include <asm/common.h>
 
-	.text
+	.section .text.isr
 
 /*
  * Low level trap entry points, this is what gets put into the IDT.
@@ -19,7 +19,7 @@
 
 .macro gen_isr_noerror name
 	.extern x86_isr_\name
-ASM_ENTRY(_x86_isr_\name )
+ENTRY(_x86_isr_\name )
 	cld
 	pushal
 	mov	%esp, %eax
@@ -29,7 +29,7 @@ ASM_ENTRY(_x86_isr_\name )
 	add	$4, %esp
 	popal
 	iretl
-ASM_END(_x86_isr_\name )
+END(_x86_isr_\name )
 .endm
 
 /*
@@ -43,7 +43,7 @@ ASM_END(_x86_isr_\name )
  */
 .macro gen_isr_error name
 	.extern x86_isr_\name
-ASM_ENTRY(_x86_isr_\name )
+ENTRY(_x86_isr_\name )
 	cld
 	pushal
 	mov	%esp, %eax
@@ -63,7 +63,7 @@ ASM_ENTRY(_x86_isr_\name )
 	popal
 	add	$4, %esp /* "pop" the hardware error code from the stack */
 	iretl
-ASM_END(_x86_isr_\name )
+END(_x86_isr_\name )
 .endm
 
 gen_isr_noerror divide_error
diff --git a/include/gay/cdefs.h b/include/gay/cdefs.h
index 28e2061..94adf23 100644
--- a/include/gay/cdefs.h
+++ b/include/gay/cdefs.h
@@ -51,6 +51,8 @@
  */
 #define __alloc_size(argn) __attribute__(( alloc_size(argn) ))
 
+#define __naked __attribute(( naked ))
+
 #define __noreturn __attribute__(( noreturn ))
 
 /**
@@ -81,7 +83,7 @@
 #define __pure2 __attribute__(( const ))
 
 /** @brief Put the annotated symbol in a specific section. */
-#define __section(name) __attribute__(( section(#name) ))
+#define __section(name) __attribute__(( section(name) ))
 
 /** @brief Mark the symbol as used, even if it really isn't. */
 #define __used __attribute__(( used ))
diff --git a/include/gay/ktrace.h b/include/gay/ktrace.h
index 2c4778c..4d8e273 100644
--- a/include/gay/ktrace.h
+++ b/include/gay/ktrace.h
@@ -3,4 +3,9 @@
 #pragma once
 
 /** @brief Print a full stack trace to the kernel log, starting from the caller. */
+__asmlink
 void ktrace_print(void);
+
+/** @brief Print a full stack trace to the kernel log, starting from `frame`. */
+__asmlink
+void ktrace_print_from(void *frame);