interrupt: nmi work

kernel-src/arch/x86-64/cpu.c

@@ -9,6 +9,7 @@
 #include <arch/gdt.h>
 #include <kernel/dpc.h>
 #include <arch/context.h>
+#include <arch/prof.h>
 
 #define EFER_SYSCALLENABLE 1
 
@@ -45,6 +46,27 @@ static void x87isr(isr_t *self, context_t *ctx) {
 	}
 }
 
+void arch_nmi_isr(context_t *ctx) {
+	// we are in an NMI context, running on IST0 
+	// first load the proper gsbase to get access to cpu-local variables. 
+	// the pointer to cpu's data is at the beggining of the stack so after the context struct.
+	uint64_t old_gsbase = rdmsr(MSR_GSBASE);
+	wrmsr(MSR_GSBASE, *(uint64_t *)(ctx + 1));
+
+	bool handled = false;
+
+#ifdef ENABLE_PROFILING
+	if (arch_profiling_irq(ctx))
+		handled = true;
+#endif
+
+	if (!handled)
+		_panic("Unhandled NMI", ctx);
+
+	// restore the old gsbase
+	wrmsr(MSR_GSBASE, old_gsbase);
+}
+
 #define TOPOLOGY_TYPE_THREAD 1
 #define TOPOLOGY_TYPE_CORE   2
 #define TOPOLOGY_TYPE_PACKAGE 0xffffffff
@@ -237,6 +259,15 @@ void arch_cpu_init() {
 	}
 
 	current_cpu()->topology_node = topology_nodes[topology_depth - 1];
+
+	// allocate ists
+	uint64_t *ist1 = vmm_map(NULL, PAGE_SIZE * 4, VMM_FLAGS_ALLOCATE, ARCH_MMU_FLAGS_READ | ARCH_MMU_FLAGS_WRITE | ARCH_MMU_FLAGS_NOEXEC, NULL);
+	__assert(ist1);
+
+	// the cpu automatically aligns to 16 bytes when pushing the interrupt frame, so save ourselves the trouble:
+	ist1 += PAGE_SIZE * 4 / sizeof(uint64_t) - 2;
+	*ist1 = (uint64_t)current_cpu();
+	current_cpu()->ist.ist1 = (uint64_t)ist1;
 }
 
 INIT_ROUTINE_DEFINE(cpu, INIT_ROUTINE_FLAGS_NONE, arch_cpu_init, acpi_early);

kernel-src/arch/x86-64/idt.c

@@ -29,16 +29,18 @@ static idtr_t idtr = {
 	.offset = (uint64_t)idt
 };
 
+void asm_nmi_handler();
 extern const uint64_t isr_table[256];
 
 void arch_idt_setup() {
 	for (int i = 0; i < 256; ++i) {
-		idt[i].offset = isr_table[i] & 0xffff;
+		uint64_t addr = i == 2 ? (uint64_t)asm_nmi_handler : isr_table[i];
+		idt[i].offset = addr & 0xffff;
 		idt[i].segment = 0x8;
-		idt[i].ist = 0;
+		idt[i].ist = i == 2 ? 1 : 0;
 		idt[i].flags = 0x8e;
-		idt[i].offset2 = (isr_table[i] >> 16) & 0xffff;
-		idt[i].offset3 = (isr_table[i] >> 32) & 0xffffffff;
+		idt[i].offset2 = (addr >> 16) & 0xffff;
+		idt[i].offset3 = (addr >> 32) & 0xffffffff;
 	}
 }
 

kernel-src/arch/x86-64/isr.asm

@@ -57,6 +57,9 @@ isr_common:
 	mov rdi, rbp ; vec
 	mov rsi, rsp ; ctx
 
+	; set this as the end of the stack frame
+	xor rbp, rbp
+
 	cld
 	extern interrupt_isr
 	call interrupt_isr
@@ -93,6 +96,78 @@ isr_common:
 	.notneeded2:
 	o64 iret
 
+global asm_nmi_handler
+asm_nmi_handler:
+	sub rsp, 8 ; error code, unused for nmi
+	push rbp
+	sub rsp, 8 ; vec number, unused for nmi
+	push rsi
+	push rdi
+	push r15
+	push r14
+	push r13
+	push r12
+	push r11
+	push r10
+	push r9
+	push r8
+	push rdx
+	push rcx
+	push rbx
+	push rax
+	mov rax, ds
+	push rax
+	mov rax, es
+	push rax
+	mov rax, fs
+	push rax
+	mov rax, gs
+	push rax
+	mov rdi,cr2
+	push rdi
+
+	; set up kernel segments
+	mov rdi, 0x10
+	mov ds, rdi
+	mov es, rdi
+
+	; save ctx pointer
+	mov rdi, rsp
+
+	; set this as the end of the stack frame
+	xor rbp, rbp
+
+	cld
+	extern arch_nmi_isr
+	call arch_nmi_isr
+	cli
+
+	; pop context
+	add rsp, 24 ; cr2 gs and fs are not popped.
+	pop rax
+	mov es,rax
+	pop rax
+	mov ds,rax
+	pop rax
+	pop rbx
+	pop rcx
+	pop rdx
+	pop r8
+	pop r9
+	pop r10
+	pop r11
+	pop r12
+	pop r13
+	pop r14
+	pop r15
+	pop rdi
+	pop rsi
+	add rsp, 8 ; vector num, unused for nmi
+	pop rbp
+	add rsp, 8 ; error code, unused for nmi
+
+	o64 iret
+
 section .rodata
 global isr_table
 isr_table:

kernel-src/arch/x86-64/syscall.asm

@@ -273,6 +273,9 @@ arch_syscall_entry:
 	mov rdx, rsi
 	mov rsi, rdi
 
+	; set this as the end of the stack frame
+	xor rbp, rbp
+
 	; call logging function
 
 	sti