arch/x86: Add support for cooperative multitasking on x86_64

This commit extends the cooperative multitasking functionality to
support the x86_64 architecture. Previously, cooperative multitasking
was only available for x86_32, as indicated by the error directive in
thread.c.

A new thread_switch_64.S file has been added to implement 64-bit
register handling for thread switching, and thread_switch.S has been
renamed to thread_switch_32.S accordingly.

Change-Id: I14ed625160a62e42d800757d30397c6c85f943b4
Signed-off-by: Jeremy Compostella <jeremy.compostella@intel.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/88216
Reviewed-by: Shuo Liu <shuo.liu@intel.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>

src/arch/x86/Makefile.mk

@@ -159,7 +159,10 @@ romstage-y += postcar_loader.c
 romstage-$(CONFIG_COLLECT_TIMESTAMPS_TSC) += timestamp.c
 romstage-$(CONFIG_HAVE_CF9_RESET) += cf9_reset.c
 romstage-$(CONFIG_COOP_MULTITASKING) += thread.c
-romstage-$(CONFIG_COOP_MULTITASKING) += thread_switch.S
+ifeq ($(CONFIG_COOP_MULTITASKING),y)
+romstage-$(CONFIG_ARCH_ROMSTAGE_X86_32) += thread_switch_32.S
+romstage-$(CONFIG_ARCH_ROMSTAGE_X86_64) += thread_switch_64.S
+endif # CONFIG_COOP_MULTITASKING
 romstage-y += car.ld
 
 romstage-srcs += $(wildcard $(src)/mainboard/$(MAINBOARDDIR)/romstage.c)
@@ -248,7 +251,10 @@ ramstage-y += rdrand.c
 ramstage-$(CONFIG_GENERATE_SMBIOS_TABLES) += smbios.c
 ramstage-y += tables.c
 ramstage-$(CONFIG_COOP_MULTITASKING) += thread.c
-ramstage-$(CONFIG_COOP_MULTITASKING) += thread_switch.S
+ifeq ($(CONFIG_COOP_MULTITASKING),y)
+ramstage-$(CONFIG_ARCH_RAMSTAGE_X86_32) += thread_switch_32.S
+ramstage-$(CONFIG_ARCH_RAMSTAGE_X86_64) += thread_switch_64.S
+endif # CONFIG_COOP_MULTITASKING
 ramstage-$(CONFIG_COLLECT_TIMESTAMPS_TSC) += timestamp.c
 ramstage-$(CONFIG_HAVE_ACPI_RESUME) += wakeup.S
 ramstage-$(CONFIG_HAVE_CF9_RESET) += cf9_reset.c

src/arch/x86/thread.c

@@ -3,11 +3,26 @@
 #include <thread.h>
 
 #if ENV_X86_64
-#error COOP_MULTITASKING does not currently support x86_64
-#endif
-
+struct saved_regs {
+	uint64_t r15; /* Offset 0x00 */
+	uint64_t r14; /* Offset 0x08 */
+	uint64_t r13; /* Offset 0x10 */
+	uint64_t r12; /* Offset 0x18 */
+	uint64_t r11; /* Offset 0x20 */
+	uint64_t r10; /* Offset 0x28 */
+	uint64_t r9;  /* Offset 0x30 */
+	uint64_t r8;  /* Offset 0x38 */
+	uint64_t rdi; /* Offset 0x40 */
+	uint64_t rsi; /* Offset 0x48 */
+	uint64_t rbp; /* Offset 0x50 */
+	uint64_t rbx; /* Offset 0x58 */
+	uint64_t rdx; /* Offset 0x60 */
+	uint64_t rcx; /* Offset 0x68 */
+	uint64_t rax; /* Offset 0x70 */
+};
+#else
 /* The stack frame looks like the following after a pushad instruction. */
-struct pushad_regs {
+struct saved_regs {
 	uint32_t edi; /* Offset 0x00 */
 	uint32_t esi; /* Offset 0x04 */
 	uint32_t ebp; /* Offset 0x08 */
@@ -17,6 +32,7 @@ struct pushad_regs {
 	uint32_t ecx; /* Offset 0x18 */
 	uint32_t eax; /* Offset 0x1c */
 };
+#endif	/* ENV_X86_64 */
 
 static inline uintptr_t push_stack(uintptr_t cur_stack, uintptr_t value)
 {
@@ -39,7 +55,7 @@ void arch_prepare_thread(struct thread *t,
 	stack = push_stack(stack, (uintptr_t)0);
 	stack = push_stack(stack, (uintptr_t)thread_entry);
 	/* Make room for the registers. Ignore initial values. */
-	stack -= sizeof(struct pushad_regs);
+	stack -= sizeof(struct saved_regs);
 
 	t->stack_current = stack;
 }

src/arch/x86/thread_switch_64.S

@@ -0,0 +1,86 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+
+/*
+ * Stack layout after pushing all the registers.
+ * +------------+
+ * |  ret addr  | <-- esp + 0x78
+ * +------------+
+ * |    rax     | <-- esp + 0x70
+ * +------------+
+ * |    rcx     | <-- esp + 0x68
+ * +------------+
+ * |    rdx     | <-- esp + 0x60
+ * +------------+
+ * |    rbx     | <-- esp + 0x58
+ * +------------+
+ * |    rbp     | <-- esp + 0x50
+ * +------------+
+ * |    rsi     | <-- esp + 0x48
+ * +------------+
+ * |    rdi     | <-- esp + 0x40
+ * +------------+
+ * |    r8      | <-- esp + 0x38
+ * +------------+
+ * |    r9      | <-- esp + 0x30
+ * +------------+
+ * |    r10     | <-- esp + 0x28
+ * +------------+
+ * |    r11     | <-- esp + 0x20
+ * +------------+
+ * |    r12     | <-- esp + 0x18
+ * +------------+
+ * |    r13     | <-- esp + 0x10
+ * +------------+
+ * |    r14     | <-- esp + 0x08
+ * +------------+
+ * |    r15     | <-- esp + 0x00
+ * +------------+
+ */
+.global switch_to_thread
+switch_to_thread:
+	/* Backup registers. */
+	push	%rax
+	push	%rcx
+	push	%rdx
+	push	%rbx
+
+	push	%rbp
+	push	%rsi
+	push	%rdi
+
+	push	%r8
+	push	%r9
+	push	%r10
+	push	%r11
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+
+	/* Save the current stack */
+	mov	%rsp, (%rsi)
+
+	/* Switch to the new stack. */
+	mov	%rdi, %rsp
+
+	/* Restore registers. */
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%r11
+	pop	%r10
+	pop	%r9
+	pop	%r8
+
+	pop	%rdi
+	pop	%rsi
+	pop	%rbp
+
+	pop	%rbx
+	pop	%rdx
+	pop	%rcx
+	pop	%rax
+
+	/* Done. */
+	ret