:: commit 746d9146c643f43f3c5257add10ce27ecc6098bb
Make the smp trampoline more UEFI friendly
diff --git a/Makefile b/Makefile
index c1358123..857dbc88 100644
--- a/Makefile
+++ b/Makefile
@@ -72,6 +72,7 @@ toolchain:
gnu-efi:
git clone https://git.code.sf.net/p/gnu-efi/code --branch=3.0.12 --depth=1 $@
+ $(MAKE) -C gnu-efi
ovmf:
mkdir -p ovmf
diff --git a/stage23/Makefile b/stage23/Makefile
index 6c206e3e..e8410438 100644
--- a/stage23/Makefile
+++ b/stage23/Makefile
@@ -5,11 +5,11 @@ ifeq ($(TARGET), bios)
else ifeq ($(TARGET), uefi)
TOOLCHAIN=x86_64-elf
else
- $(error Invalid toolchain)
+ $(error Invalid target)
endif
CC = $(TOOLCHAIN)-gcc
-LD = $(TOOLCHAIN)-gcc
+LD = $(TOOLCHAIN)-ld
OBJCOPY = $(TOOLCHAIN)-objcopy
OBJDUMP = $(TOOLCHAIN)-objdump
READELF = $(TOOLCHAIN)-readelf
@@ -27,7 +27,6 @@ INTERNAL_CFLAGS := \
-fplan9-extensions \
-ffreestanding \
-fno-stack-protector \
- -fno-pic \
-fno-omit-frame-pointer \
-fno-lto \
-Wno-address-of-packed-member \
@@ -42,14 +41,17 @@ INTERNAL_CFLAGS := \
-I. \
-I..
+ifeq ($(TARGET), bios)
+ INTERNAL_CFLAGS += \
+ -fno-pic
+endif
+
ifeq ($(TARGET), uefi)
INTERNAL_CFLAGS += \
-I../gnu-efi/inc \
-I../gnu-efi/inc/x86_64 \
-fpic \
- -fshort-wchar \
- -mno-red-zone \
- -mcmodel=small
+ -mno-red-zone
endif
LDFLAGS = -Os -g
@@ -62,14 +64,9 @@ INTERNAL_LDFLAGS := \
ifeq ($(TARGET), bios)
INTERNAL_LDFLAGS += \
-static \
- -no-pie \
+ -fno-pie \
-lgcc \
-static-libgcc
-else ifeq ($(TARGET), uefi)
- INTERNAL_LDFLAGS += \
- -shared \
- -z nocombreloc \
- -Wl,-Bsymbolic
endif
.PHONY: all clean
@@ -79,19 +76,16 @@ ifeq ($(TARGET), bios)
ASM_FILES := $(shell find -L ./ -type f -name '*.asm' | sort)
endif
OBJ := $(ASM_FILES:.asm=.o) $(C_FILES:.c=.o)
-ifeq ($(TARGET), uefi)
-OBJ += sys/smp_trampoline.o ../gnu-efi/lib/x86_64/efi_stub.o
-endif
HEADER_DEPS := $(C_FILES:.c=.d)
ifeq ($(TARGET), bios)
all: limine_dbg.elf limine.sys stage2.bin stage2.bin.gz
else ifeq ($(TARGET), uefi)
-all: limine_dbg.elf BOOTX64.EFI
+all: BOOTX64.EFI
endif
-BOOTX64.EFI: limine.elf
- $(OBJCOPY) -I elf64-x86-64 -O efi-app-x86_64 limine.elf $@
+BOOTX64.EFI: limine-efi.elf
+ $(OBJCOPY) -j .text -j .sdata -j .data -j .dynamic -j .dynsym -j .rel -j .rela -j .rel.* -j .rela.* -j .reloc --target efi-app-x86_64 --subsystem=10 limine-efi.elf $@
stage2.bin.gz: stage2.bin
gzip -n -9 < stage2.bin > stage2.bin.gz
@@ -100,31 +94,38 @@ stage2.bin: limine.sys
dd if=limine.sys bs=$$(( 0x$$($(READELF) -S limine.elf | grep .stage3.text | sed 's/^.*] //' | awk '{print $$3}' | sed 's/^0*//') - 0x8000 )) count=1 of=$@
limine.map.o: limine_nomap.elf
-ifeq ($(TARGET), bios)
./gensyms.sh $(OBJDUMP) limine_nomap.elf limine
-else ifeq ($(TARGET), uefi)
- ./gensyms64.sh $(OBJDUMP) limine_nomap.elf limine
-endif
limine.sys: limine.elf
$(OBJCOPY) -O binary $< $@
-limine_nomap.elf: $(OBJ) font.o
- $(LD) $(OBJ) font.o $(LDFLAGS) $(INTERNAL_LDFLAGS) -Tlinker_nomap_$(TARGET).ld -o $@
-ifeq ($(TARGET), bios)
- $(LD) $(OBJ) font.o $(LDFLAGS) $(INTERNAL_LDFLAGS) -Wl,--gc-sections -Tlinker_stage2only.ld -o limine_stage2only.elf || \
+limine_nomap.elf: $(OBJ) font.o sys/smp_trampoline.o
+ $(CC) $(OBJ) font.o sys/smp_trampoline.o $(LDFLAGS) $(INTERNAL_LDFLAGS) -Tlinker_nomap.ld -o $@
+ $(CC) $(OBJ) font.o sys/smp_trampoline.o $(LDFLAGS) $(INTERNAL_LDFLAGS) -Wl,--gc-sections -Tlinker_stage2only.ld -o limine_stage2only.elf || \
( echo "This error means that stage 2 was trying to use stage 3 symbols before loading stage 3" && \
false )
-endif
-limine_dbg.elf: $(OBJ) font.o
- $(LD) $(OBJ) font.o $(LDFLAGS) $(INTERNAL_LDFLAGS) -Tlinker_dbg.ld -o $@
+limine_dbg.elf: $(OBJ) font.o sys/smp_trampoline.o
+ $(CC) $(OBJ) font.o sys/smp_trampoline.o $(LDFLAGS) $(INTERNAL_LDFLAGS) -Tlinker_dbg.ld -o $@
+
+sys/smp_trampoline.o: sys/smp_trampoline.bin
+ $(OBJCOPY) -B i8086 -I binary -O default sys/smp_trampoline.bin $@
+
+sys/smp_trampoline.bin: sys/smp_trampoline.real
+ nasm $< -f bin -o $@
-font.o:
+font.o: font.bin
$(OBJCOPY) -B i8086 -I binary -O default font.bin $@
-limine.elf: $(OBJ) font.o limine.map.o
- $(LD) $(OBJ) font.o limine.map.o $(LDFLAGS) $(INTERNAL_LDFLAGS) -Tlinker_$(TARGET).ld -o $@
+limine.elf: $(OBJ) font.o sys/smp_trampoline.o limine.map.o
+ $(CC) $(OBJ) font.o sys/smp_trampoline.o limine.map.o $(LDFLAGS) $(INTERNAL_LDFLAGS) -Tlinker.ld -o $@
+
+limine-efi.elf: $(OBJ) font.o sys/smp_trampoline.o
+ $(LD) -shared -Bsymbolic \
+ -T../gnu-efi/gnuefi/elf_x86_64_efi.lds \
+ ../gnu-efi/x86_64/gnuefi/crt0-efi-x86_64.o \
+ ../gnu-efi/x86_64/gnuefi/libgnuefi.a \
+ $(OBJ) font.o sys/smp_trampoline.o -o $@
-include $(HEADER_DEPS)
@@ -133,13 +134,8 @@ limine.elf: $(OBJ) font.o limine.map.o
%.o: %.S
$(CC) $(CFLAGS) $(INTERNAL_CFLAGS) -c $< -o $@
-ifeq ($(TARGET), bios)
%.o: %.asm
- nasm $< -f elf32 -o $@
-else ifeq ($(TARGET), uefi)
-%.o: %.asm
- nasm $< -f elf64 -o $@
-endif
+ nasm $< -F dwarf -g -f elf32 -o $@
clean:
- rm -f limine.elf limine_nomap.elf limine_stage2only.elf font.o limine.map.o limine.sys stage2.bin stage2.bin.gz BOOTX64.EFI $(OBJ) $(HEADER_DEPS)
+ rm -f limine.elf limine_nomap.elf limine_stage2only.elf font.o limine.map.o limine.sys stage2.bin stage2.bin.gz BOOTX64.EFI sys/smp_trampoline.bin sys/smp_trampoline.o $(OBJ) $(HEADER_DEPS)
diff --git a/stage23/entry.s3.c b/stage23/entry.s3.c
index 46c48577..c03aa345 100644
--- a/stage23/entry.s3.c
+++ b/stage23/entry.s3.c
@@ -32,7 +32,7 @@ EFI_STATUS efi_main(EFI_HANDLE ImageHandle, EFI_SYSTEM_TABLE *SystemTable) {
gBS = SystemTable->BootServices;
gRT = SystemTable->RuntimeServices;
- print("Limine " LIMINE_VERSION "\n\n");
+ print("Limine " LIMINE_VERSION "\n%X\n", print);
volume_create_index();
diff --git a/stage23/gensyms64.sh b/stage23/gensyms64.sh
deleted file mode 100755
index ffad0834..00000000
--- a/stage23/gensyms64.sh
+++ /dev/null
@@ -1,24 +0,0 @@
-#!/bin/sh
-
-set -e
-
-TMP1=$(mktemp)
-TMP2=$(mktemp)
-TMP3=$(mktemp)
-TMP4=$(mktemp)
-
-$1 -t "$2" | sed '/\bd\b/d' | sort > "$TMP1"
-grep "\.text" < "$TMP1" | cut -d' ' -f1 > "$TMP2"
-grep "\.text" < "$TMP1" | awk 'NF{ print $NF }' > "$TMP3"
-
-echo "section .map" > "$TMP4"
-echo "global $3_map" >> "$TMP4"
-echo "$3_map:" >> "$TMP4"
-
-paste -d'$' "$TMP2" "$TMP3" | sed 's/^/dq 0x/g' | sed 's/$/", 0/g' | sed 's/\$/\ndb "/g' >> "$TMP4"
-
-echo "dq 0xffffffffffffffff" >> "$TMP4"
-
-nasm -f elf64 "$TMP4" -o $3.map.o
-
-rm "$TMP1" "$TMP2" "$TMP3" "$TMP4"
diff --git a/stage23/linker_bios.ld b/stage23/linker.ld
similarity index 100%
rename from stage23/linker_bios.ld
rename to stage23/linker.ld
diff --git a/stage23/linker_nomap_bios.ld b/stage23/linker_nomap.ld
similarity index 100%
rename from stage23/linker_nomap_bios.ld
rename to stage23/linker_nomap.ld
diff --git a/stage23/linker_nomap_uefi.ld b/stage23/linker_nomap_uefi.ld
deleted file mode 100644
index cd6878a7..00000000
--- a/stage23/linker_nomap_uefi.ld
+++ /dev/null
@@ -1,37 +0,0 @@
-OUTPUT_FORMAT(elf64-x86-64)
-ENTRY(efi_main)
-
-SECTIONS
-{
- . = 4K;
-
- .text : {
- *(.text*)
- *(.realmode*)
- *(.stage3_entry*)
- }
-
- .rodata : {
- limine_map = .;
- *(.rodata*)
- }
-
- .data : {
- *(.data*)
- }
-
- .bss : {
- *(COMMON)
- *(.bss*)
- }
-
- .reloc : {
- LONG(0);
- LONG(10);
- SHORT(0);
- }
-
- /DISCARD/ : {
- *(*)
- }
-}
diff --git a/stage23/linker_uefi.ld b/stage23/linker_uefi.ld
deleted file mode 100644
index 73a13f8b..00000000
--- a/stage23/linker_uefi.ld
+++ /dev/null
@@ -1,37 +0,0 @@
-OUTPUT_FORMAT(elf64-x86-64)
-ENTRY(efi_main)
-
-SECTIONS
-{
- . = 4K;
-
- .text : {
- *(.text*)
- *(.realmode*)
- *(.stage3_entry*)
- }
-
- .rodata : {
- *(.map*)
- *(.rodata*)
- }
-
- .data : {
- *(.data*)
- }
-
- .bss : {
- *(COMMON)
- *(.bss*)
- }
-
- .reloc : {
- LONG(0);
- LONG(10);
- SHORT(0);
- }
-
- /DISCARD/ : {
- *(*)
- }
-}
diff --git a/stage23/sys/smp.c b/stage23/sys/smp.c
index 322cc719..d7dd100a 100644
--- a/stage23/sys/smp.c
+++ b/stage23/sys/smp.c
@@ -1,6 +1,7 @@
#include <stddef.h>
#include <stdint.h>
#include <stdbool.h>
+#include <lib/libc.h>
#include <lib/acpi.h>
#include <sys/cpu.h>
#include <lib/blib.h>
@@ -9,6 +10,7 @@
#include <sys/lapic.h>
#include <mm/vmm.h>
#include <mm/pmm.h>
+#include <mm/mtrr.h>
struct madt {
struct sdt;
@@ -47,25 +49,50 @@ static void delay(uint32_t cycles) {
inb(0x80);
}
-void smp_trampoline(void);
-extern struct gdtr smp_tpl_gdt;
-struct smp_information *smp_tpl_info_struct;
-uint8_t smp_tpl_booted_flag;
-uint32_t smp_tpl_pagemap;
-uint8_t smp_tpl_target_mode;
+extern symbol _binary_sys_smp_trampoline_bin_start;
+extern symbol _binary_sys_smp_trampoline_bin_end;
+
+struct trampoline_passed_info {
+ uint8_t smp_tpl_booted_flag;
+ uint8_t smp_tpl_target_mode;
+ uint32_t smp_tpl_pagemap;
+ void *mtrr_restore_vector;
+ struct smp_information *smp_tpl_info_struct;
+ struct gdtr smp_tpl_gdt;
+} __attribute__((packed));
static bool smp_start_ap(uint32_t lapic_id, struct gdtr *gdtr,
struct smp_information *info_struct,
bool longmode, bool lv5, uint32_t pagemap,
bool x2apic) {
+ size_t trampoline_size = (size_t)_binary_sys_smp_trampoline_bin_end
+ - (size_t)_binary_sys_smp_trampoline_bin_start;
+
// Prepare the trampoline
- smp_tpl_info_struct = info_struct;
- smp_tpl_booted_flag = 0;
- smp_tpl_pagemap = pagemap;
- smp_tpl_target_mode = ((uint32_t)x2apic << 2)
+ static void *trampoline = NULL;
+ if (trampoline == NULL) {
+ trampoline = conv_mem_alloc_aligned(trampoline_size, 4096);
+
+ memcpy(trampoline, _binary_sys_smp_trampoline_bin_start, trampoline_size);
+ }
+
+ static struct trampoline_passed_info *passed_info = NULL;
+ if (passed_info == NULL) {
+ passed_info = (void *)(((uintptr_t)trampoline + trampoline_size)
+ - sizeof(struct trampoline_passed_info));
+ }
+
+ passed_info->smp_tpl_info_struct = info_struct;
+ passed_info->smp_tpl_booted_flag = 0;
+ passed_info->smp_tpl_pagemap = pagemap;
+ passed_info->smp_tpl_target_mode = ((uint32_t)x2apic << 2)
| ((uint32_t)lv5 << 1)
| (uint32_t)longmode;
- smp_tpl_gdt = *gdtr;
+ passed_info->smp_tpl_gdt = *gdtr;
+ passed_info->mtrr_restore_vector = mtrr_restore;
+ passed_info->smp_tpl_booted_flag = 0;
+
+ asm volatile ("" ::: "memory");
// Send the INIT IPI
if (x2apic) {
@@ -79,14 +106,14 @@ static bool smp_start_ap(uint32_t lapic_id, struct gdtr *gdtr,
// Send the Startup IPI
if (x2apic) {
x2apic_write(LAPIC_REG_ICR0, ((uint64_t)lapic_id << 32) |
- ((size_t)smp_trampoline / 4096) | 0x4600);
+ ((size_t)trampoline / 4096) | 0x4600);
} else {
lapic_write(LAPIC_REG_ICR1, lapic_id << 24);
- lapic_write(LAPIC_REG_ICR0, ((size_t)smp_trampoline / 4096) | 0x4600);
+ lapic_write(LAPIC_REG_ICR0, ((size_t)trampoline / 4096) | 0x4600);
}
for (int i = 0; i < 100; i++) {
- if (locked_read(&smp_tpl_booted_flag) == 1) {
+ if (locked_read(&passed_info->smp_tpl_booted_flag) == 1) {
return true;
}
delay(10000);
diff --git a/stage23/sys/smp_trampoline.asm b/stage23/sys/smp_trampoline.real
similarity index 61%
rename from stage23/sys/smp_trampoline.asm
rename to stage23/sys/smp_trampoline.real
index 06c6558d..c0f31204 100644
--- a/stage23/sys/smp_trampoline.asm
+++ b/stage23/sys/smp_trampoline.real
@@ -1,35 +1,27 @@
-extern smp_tpl_info_struct
-extern smp_tpl_booted_flag
-extern smp_tpl_pagemap
-extern smp_tpl_target_mode
+org 0
-extern mtrr_restore
-
-section .bss
-
-temp_stack:
- resb 1024
- .top:
-
-section .realmode
-
-global smp_trampoline
-align 0x1000
bits 16
smp_trampoline:
cli
cld
- xor ax, ax
- mov ds, ax
+ mov ebx, cs
+ shl ebx, 4
+
+ lgdt [cs:passed_info.gdtr]
- lgdt [smp_tpl_gdt]
+ lea eax, [ebx + .mode32]
+ mov [cs:.farjmp_off], eax
mov eax, cr0
bts eax, 0
mov cr0, eax
+ o32 jmp far [cs:.farjmp]
+
+ .farjmp:
+ .farjmp_off: dd 0
+ .farjmp_seg: dd 0x18
- jmp 0x18:.mode32
bits 32
.mode32:
mov ax, 0x20
@@ -44,7 +36,7 @@ smp_trampoline:
btr eax, 30
mov cr0, eax
- test dword [smp_tpl_target_mode], (1 << 2)
+ test dword [ebx + passed_info.target_mode], (1 << 2)
jz .nox2apic
mov ecx, 0x1b
@@ -54,18 +46,18 @@ smp_trampoline:
wrmsr
.nox2apic:
- mov esp, temp_stack.top
+ lea esp, [ebx + temp_stack.top]
- call mtrr_restore
+ call [ebx + passed_info.mtrr_restore_vector]
- test dword [smp_tpl_target_mode], (1 << 0)
+ test dword [ebx + passed_info.target_mode], (1 << 0)
jz parking32
mov eax, cr4
bts eax, 5
mov cr4, eax
- test dword [smp_tpl_target_mode], (1 << 1)
+ test dword [ebx + passed_info.target_mode], (1 << 1)
jz .no5lv
mov eax, cr4
@@ -73,7 +65,7 @@ smp_trampoline:
mov cr4, eax
.no5lv:
- mov eax, dword [smp_tpl_pagemap]
+ mov eax, dword [ebx + passed_info.pagemap]
mov cr3, eax
mov ecx, 0xc0000080
@@ -85,7 +77,12 @@ smp_trampoline:
bts eax, 31
mov cr0, eax
- jmp 0x28:.mode64
+ mov eax, .mode64
+ add eax, ebx
+ push 0x28
+ push eax
+ retf
+
bits 64
.mode64:
mov ax, 0x30
@@ -97,19 +94,11 @@ smp_trampoline:
jmp parking64
-global smp_tpl_gdt
-align 16
-smp_tpl_gdt:
- dw 0
- dd 0
-
-section .text
-
bits 32
parking32:
- mov edi, dword [smp_tpl_info_struct]
+ mov edi, dword [ebx + passed_info.smp_info_struct]
mov eax, 1
- lock xchg dword [smp_tpl_booted_flag], eax
+ lock xchg dword [ebx + passed_info.booted_flag], eax
xor eax, eax
.loop:
@@ -135,9 +124,10 @@ parking32:
bits 64
parking64:
- mov edi, dword [smp_tpl_info_struct]
+ mov ebx, ebx
+ mov edi, dword [rbx + passed_info.smp_info_struct]
mov eax, 1
- lock xchg dword [smp_tpl_booted_flag], eax
+ lock xchg dword [rbx + passed_info.booted_flag], eax
xor eax, eax
.loop:
@@ -166,3 +156,19 @@ parking64:
xor r14, r14
xor r15, r15
ret
+
+align 16
+temp_stack:
+ times 1024 db 0
+ .top:
+
+align 16
+passed_info:
+ .booted_flag db 0
+ .target_mode db 0
+ .pagemap dd 0
+ .mtrr_restore_vector dd 0
+ .smp_info_struct dd 0
+ .gdtr:
+ dw 0
+ dd 0
