[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH 1/6] add support for booting from grub with x86_64
|
From: |
Samuel Thibault |
|
Subject: |
Re: [PATCH 1/6] add support for booting from grub with x86_64 |
|
Date: |
Sat, 27 Aug 2022 20:38:08 +0200 |
|
User-agent: |
NeoMutt/20170609 (1.8.3) |
Samuel Thibault, le sam. 27 août 2022 20:26:10 +0200, a ecrit:
> I can't manage to make this boot at all. Is there a particular configure
> option to pass? I have let it just use the linux 64bit compiler, I have
> tried with --host=x86_64-gnu, in all cases after pressing enter at the
> grub menu it loads the kernel but just reboots immediately after that,
> even if I have put a jmp boot_entry at boot_entry, i.e. it's really the
> load that is a problem and not any gnumach code.
Actually, building with 32bit without your changes brings me the same
issue, oh joy...
Samuel
> Luca Dariz, le sam. 05 févr. 2022 18:51:24 +0100, a ecrit:
> > * configure: compile for native x86_64 by default instead of xen
> > * x86_64/Makefrag.am: introduce KERNEL_MAP_BASE to reuse the constant
> > in both code and linker script
> > * x86_64/ldscript: use a .boot section for the very first operations,
> > until we reach long mode. This section is not really allocated, so
> > it doesn't need to be freed later. The vm system is later
> > initialized starting from .text and not including .boot
> > * link kernel at 0x4000000 as the xen version, higher values causes
> > linker errors
> > * we can't use full segmentation in long mode, so we need to create a
> > temporary mapping during early boot to be able to jump to high
> > addresses
> > * build direct map for first 4G in boothdr, it seems required by Linux
> > drivers
> > * add INTEL_PTE_PS bit definition to enable 2MB pages during bootstrap
> > * ensure write bit is set in PDP entry access rights. This only
> > applies to PAE-enabled kernels, mandatory for x86_64. On xen
> > platform it seems to be handled differently
> >
> > Signed-off-by: Luca Dariz <luca@orpolo.org>
> > ---
> > configure.ac | 3 +-
> > i386/configfrag.ac | 2 +
> > i386/i386/i386asm.sym | 1 +
> > i386/i386/vm_param.h | 2 +-
> > i386/intel/pmap.c | 4 +-
> > i386/intel/pmap.h | 1 +
> > x86_64/Makefrag.am | 18 +++-
> > x86_64/boothdr.S | 238 ++++++++++++++++++++++++++++++++++++++++++
> > x86_64/ldscript | 28 +++--
> > 9 files changed, 281 insertions(+), 16 deletions(-)
> > create mode 100644 x86_64/boothdr.S
> >
> > diff --git a/configure.ac b/configure.ac
> > index 019842db..3aaa935c 100644
> > --- a/configure.ac
> > +++ b/configure.ac
> > @@ -56,8 +56,7 @@ case $host_platform:$host_cpu in
> > default:i?86)
> > host_platform=at;;
> > default:x86_64)]
> > - AC_MSG_WARN([Platform set to Xen by default, this can not boot on
> > non-Xen systems, you currently need a 32bit build for that.])
> > - [host_platform=xen;;
> > + [host_platform=at;;
> > at:i?86 | xen:i?86 | at:x86_64 | xen:x86_64)
> > :;;
> > *)]
> > diff --git a/i386/configfrag.ac b/i386/configfrag.ac
> > index f697e277..f07a98ca 100644
> > --- a/i386/configfrag.ac
> > +++ b/i386/configfrag.ac
> > @@ -106,6 +106,8 @@ AC_ARG_ENABLE([apic],
> > enable_pae=${enable_pae-yes};;
> > *:i?86)
> > :;;
> > + *:x86_64)
> > + enable_pae=${enable_pae-yes};;
> > *)
> > if [ x"$enable_pae" = xyes ]; then]
> > AC_MSG_ERROR([can only enable the `PAE' feature on ix86.])
> > diff --git a/i386/i386/i386asm.sym b/i386/i386/i386asm.sym
> > index 0662aea0..9e1d13d7 100644
> > --- a/i386/i386/i386asm.sym
> > +++ b/i386/i386/i386asm.sym
> > @@ -122,6 +122,7 @@ expr sizeof(pt_entry_t)
> > PTE_SIZE
> > expr INTEL_PTE_PFN PTE_PFN
> > expr INTEL_PTE_VALID PTE_V
> > expr INTEL_PTE_WRITE PTE_W
> > +expr INTEL_PTE_PS PTE_S
> > expr ~INTEL_PTE_VALID PTE_INVALID
> > expr NPTES PTES_PER_PAGE
> > expr INTEL_PTE_VALID|INTEL_PTE_WRITE INTEL_PTE_KERNEL
> > diff --git a/i386/i386/vm_param.h b/i386/i386/vm_param.h
> > index edd9522c..314fdb35 100644
> > --- a/i386/i386/vm_param.h
> > +++ b/i386/i386/vm_param.h
> > @@ -36,7 +36,7 @@
> > * for better trace support in kdb; the _START symbol has to be offset by
> > the
> > * same amount. */
> > #ifdef __x86_64__
> > -#define VM_MIN_KERNEL_ADDRESS 0x40000000UL
> > +#define VM_MIN_KERNEL_ADDRESS KERNEL_MAP_BASE
> > #else
> > #define VM_MIN_KERNEL_ADDRESS 0xC0000000UL
> > #endif
> > diff --git a/i386/intel/pmap.c b/i386/intel/pmap.c
> > index 3bf00659..d0bd3b5d 100644
> > --- a/i386/intel/pmap.c
> > +++ b/i386/intel/pmap.c
> > @@ -655,7 +655,7 @@ void pmap_bootstrap(void)
> > pa_to_pte(_kvtophys((void *) kernel_page_dir
> > + i * INTEL_PGBYTES))
> > | INTEL_PTE_VALID
> > -#ifdef MACH_PV_PAGETABLES
> > +#if !defined(MACH_HYP) || defined(MACH_PV_PAGETABLES)
> > | INTEL_PTE_WRITE
> > #endif
> > );
> > @@ -1297,7 +1297,7 @@ pmap_t pmap_create(vm_size_t size)
> > WRITE_PTE(&p->pdpbase[i],
> > pa_to_pte(kvtophys((vm_offset_t) page_dir[i]))
> > | INTEL_PTE_VALID
> > -#ifdef MACH_PV_PAGETABLES
> > +#if !defined(MACH_HYP) || defined(MACH_PV_PAGETABLES)
> > | INTEL_PTE_WRITE
> > #endif
> > );
> > diff --git a/i386/intel/pmap.h b/i386/intel/pmap.h
> > index f24b3a71..b93c4ad4 100644
> > --- a/i386/intel/pmap.h
> > +++ b/i386/intel/pmap.h
> > @@ -148,6 +148,7 @@ typedef phys_addr_t pt_entry_t;
> > #define INTEL_PTE_NCACHE 0x00000010
> > #define INTEL_PTE_REF 0x00000020
> > #define INTEL_PTE_MOD 0x00000040
> > +#define INTEL_PTE_PS 0x00000080
> > #ifdef MACH_PV_PAGETABLES
> > /* Not supported */
> > #define INTEL_PTE_GLOBAL 0x00000000
> > diff --git a/x86_64/Makefrag.am b/x86_64/Makefrag.am
> > index 40b50bc9..5da734de 100644
> > --- a/x86_64/Makefrag.am
> > +++ b/x86_64/Makefrag.am
> > @@ -207,11 +207,27 @@ nodist_libkernel_a_SOURCES += \
> >
> > EXTRA_DIST += \
> > x86_64/ldscript
> > +
> > if PLATFORM_at
> > +# This should probably be 0xffffffff80000000 for mcmodel=kernel, but let's
> > try
> > +# to stay in the first 8G first, otherwise we have to fix the pmap module
> > to
> > +# actually use the l4 page level
> > +#KERNEL_MAP_BASE=0x100000000
> > +# but for nor try with < 4G, otherwise we have linker errors
> > +KERNEL_MAP_BASE=0x40000000
> > gnumach_LINKFLAGS += \
> > --defsym _START_MAP=$(_START_MAP) \
> > - --defsym _START=_START_MAP+0x40000000 \
> > + --defsym _START=_START_MAP \
> > + --defsym KERNEL_MAP_BASE=$(KERNEL_MAP_BASE) \
> > -T '$(srcdir)'/x86_64/ldscript
> > +
> > +AM_CFLAGS += -D_START_MAP=$(_START_MAP) \
> > + -DKERNEL_MAP_BASE=$(KERNEL_MAP_BASE)
> > +AM_CCASFLAGS += -D_START_MAP=$(_START_MAP) \
> > + -DKERNEL_MAP_BASE=$(KERNEL_MAP_BASE)
> > +
> > +AM_CCASFLAGS += \
> > + -Ii386
> > endif
> >
> > AM_CPPFLAGS += \
> > diff --git a/x86_64/boothdr.S b/x86_64/boothdr.S
> > new file mode 100644
> > index 00000000..12fc7ca2
> > --- /dev/null
> > +++ b/x86_64/boothdr.S
> > @@ -0,0 +1,238 @@
> > +/*
> > + * Copyright (C) 2022 Free Software Foundation
> > + *
> > + * This program is free software ; you can redistribute it and/or modify
> > + * it under the terms of the GNU General Public License as published by
> > + * the Free Software Foundation ; either version 2 of the License, or
> > + * (at your option) any later version.
> > + *
> > + * This program is distributed in the hope that it will be useful,
> > + * but WITHOUT ANY WARRANTY ; without even the implied warranty of
> > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > + * GNU General Public License for more details.
> > + *
> > + * You should have received a copy of the GNU General Public License
> > + * along with the program ; if not, write to the Free Software
> > + * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
> > + */
> > +
> > +#include <mach/machine/asm.h>
> > +
> > +#include <i386/i386asm.h>
> > +#include <i386/i386/proc_reg.h>
> > +#include <i386/i386/seg.h>
> > + /*
> > + * This section will be put first into .boot. See also x86_64/ldscript.
> > + */
> > + .section .boot.text,"ax"
> > + .globl boot_start
> > +
> > + /* We should never be entered this way. */
> > + .code32
> > +boot_start:
> > + jmp boot_entry
> > +
> > + /* MultiBoot header - see multiboot.h. */
> > +#define MULTIBOOT_MAGIC 0x1BADB002
> > +#ifdef __ELF__
> > +#define MULTIBOOT_FLAGS 0x00000003
> > +#else /* __ELF__ */
> > +#define MULTIBOOT_FLAGS 0x00010003
> > +#endif /* __ELF__ */
> > + P2ALIGN(2)
> > +boot_hdr:
> > + .long MULTIBOOT_MAGIC
> > + .long MULTIBOOT_FLAGS
> > + /*
> > + * The next item here is the checksum.
> > + * XX this works OK until we need at least the 30th bit.
> > + */
> > + .long - (MULTIBOOT_MAGIC+MULTIBOOT_FLAGS)
> > +#ifndef __ELF__ /* a.out kludge */
> > + .long boot_hdr /* header_addr */
> > + .long _start /* load_addr */
> > + .long _edata /* load_end_addr */
> > + .long _end /* bss_end_addr */
> > + .long boot_entry /* entry */
> > +#endif /* __ELF__ */
> > +
> > +boot_entry:
> > + /*
> > + * Prepare minimal page mapping to jump to 64 bit and to C code.
> > + * The first 4GB is identity mapped, and the first 2GB are re-mapped
> > + * to high addresses at KERNEL_MAP_BASE
> > + */
> > +
> > + movl $p3table,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p4table)
> > + /*
> > + * Fill 4 entries in L3 table to cover the whole 32-bit 4GB address
> > + * space. Part of it might be remapped later if the kernel is mapped
> > + * below 4G.
> > + */
> > + movl $p2table,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p3table)
> > + movl $p2table1,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p3table + 8)
> > + movl $p2table2,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p3table + 16)
> > + movl $p2table3,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p3table + 24)
> > + /* point each page table level two entry to a page */
> > + mov $0,%ecx
> > +.map_p2_table:
> > + mov $0x200000,%eax // 2MiB page, should be always available
> > + mul %ecx
> > + or $(PTE_V|PTE_W|PTE_S),%eax // enable 2MiB page instead of 4k
> > + mov %eax,p2table(,%ecx,8)
> > + inc %ecx
> > + cmp $2048,%ecx // 512 entries per table, map 4 L2 tables
> > + jne .map_p2_table
> > +
> > + /*
> > + * KERNEL_MAP_BASE must me aligned to 2GB.
> > + * Depending on kernel starting address, we might need to add another
> > + * entry in the L4 table (controlling 512 GB chunks). In any case, we
> > + * add two entries in L3 table to make sure we map 2GB for the kernel.
> > + * Note that this may override part of the mapping create above.
> > + */
> > +.kernel_map:
> > +#if KERNEL_MAP_BASE >= (1U << 39)
> > + movl $p3ktable,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p4table + (8 * ((KERNEL_MAP_BASE >> 39) & 0x1FF))) //
> > select 512G block
> > + movl $p2ktable1,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p3ktable + (8 * ((KERNEL_MAP_BASE >> 30) & 0x1FF) )) //
> > select first 1G block
> > + movl $p2ktable2,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p3ktable + (8 * (((KERNEL_MAP_BASE >> 30) & 0x1FF) + 1)
> > )) // select second 1G block
> > +#else
> > + movl $p2ktable1,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p3table + (8 * ((KERNEL_MAP_BASE >> 30) & 0x1FF) )) //
> > select first 1G block
> > + movl $p2ktable2,%eax
> > + or $(PTE_V|PTE_W),%eax
> > + movl %eax,(p3table + (8 * (((KERNEL_MAP_BASE >> 30) & 0x1FF) + 1) ))
> > // select second 1G block
> > +#endif
> > +
> > + mov $0,%ecx
> > +.map_p2k_table:
> > + mov $0x200000,%eax // 2MiB page, should be always available
> > + mul %ecx
> > + or $(PTE_V|PTE_W|PTE_S),%eax // enable 2MiB page instead of 4K
> > + mov %eax,p2ktable1(,%ecx,8)
> > + inc %ecx
> > + cmp $1024,%ecx // 512 entries per table, map 2 L2 tables
> > + jne .map_p2k_table
> > +
> > +switch64:
> > + /*
> > + * Jump to 64 bit mode, we have to
> > + * - enable PAE
> > + * - enable long mode
> > + * - enable paging and load the tables filled above in CR3
> > + * - jump to a 64-bit code segment
> > + */
> > + mov %cr4,%eax
> > + or $CR4_PAE,%eax
> > + mov %eax,%cr4
> > + mov $0xC0000080,%ecx // select EFER register
> > + rdmsr
> > + or $(1 << 8),%eax // long mode enable bit
> > + wrmsr
> > + mov $p4table,%eax
> > + mov %eax,%cr3
> > + mov %cr0,%eax
> > + or $CR0_PG,%eax
> > + or $CR0_WP,%eax
> > + mov %eax,%cr0
> > +
> > + lgdt gdt64pointer
> > + movw $0,%ax
> > + movw %ax,%fs
> > + movw %ax,%gs
> > + movw $16,%ax
> > + movw %ax,%ds
> > + movw %ax,%es
> > + movw %ax,%ss
> > + ljmp $8,$boot_entry64
> > +
> > + .code64
> > +
> > + /* why do we need this? it seems overwritten by linker */
> > + .globl _start
> > +_start:
> > +
> > +boot_entry64:
> > + /* Switch to our own interrupt stack. */
> > + movq $(_intstack+INTSTACK_SIZE),%rax
> > + andq $(~15),%rax
> > + movq %rax,%rsp
> > +
> > + /* Reset EFLAGS to a known state. */
> > + pushq $0
> > + popf
> > + /* save multiboot info for later */
> > + movq %rbx,%r8
> > +
> > + /* Fix ifunc entries */
> > + movq $__rela_iplt_start,%rsi
> > + movq $__rela_iplt_end,%rdi
> > +iplt_cont:
> > + cmpq %rdi,%rsi
> > + jae iplt_done
> > + movq (%rsi),%rbx /* r_offset */
> > + movb 4(%rsi),%al /* info */
> > + cmpb $42,%al /* IRELATIVE */
> > + jnz iplt_next
> > + call *(%ebx) /* call ifunc */
> > + movq %rax,(%rbx) /* fixed address */
> > +iplt_next:
> > + addq $8,%rsi
> > + jmp iplt_cont
> > +iplt_done:
> > +
> > + /* restore multiboot info */
> > + movq %r8,%rdi
> > + /* Jump into C code. */
> > + call EXT(c_boot_entry)
> > + /* not reached */
> > + nop
> > +
> > + .section .boot.data
> > + .comm _intstack,INTSTACK_SIZE
> > +
> > + .code32
> > + .section .boot.data
> > + .align 4096
> > +#define SEG_ACCESS_OFS 40
> > +#define SEG_GRANULARITY_OFS 52
> > +gdt64:
> > + .quad 0
> > +gdt64code:
> > + .quad (ACC_P << SEG_ACCESS_OFS) | (ACC_CODE_R <<
> > SEG_ACCESS_OFS) | (SZ_64 << SEG_GRANULARITY_OFS)
> > +gdt64data:
> > + .quad (ACC_P << SEG_ACCESS_OFS) | (ACC_DATA_W <<
> > SEG_ACCESS_OFS)
> > +gdt64end:
> > + .skip (4096 - (gdt64end - gdt64))
> > +gdt64pointer:
> > + .word gdt64end - gdt64 - 1
> > + .quad gdt64
> > +
> > + .section .boot.data
> > + .align 4096
> > +p4table: .space 4096
> > +p3table: .space 4096
> > +p2table: .space 4096
> > +p2table1: .space 4096
> > +p2table2: .space 4096
> > +p2table3: .space 4096
> > +p3ktable: .space 4096
> > +p2ktable1: .space 4096
> > +p2ktable2: .space 4096
> > diff --git a/x86_64/ldscript b/x86_64/ldscript
> > index 375e8104..de99795e 100644
> > --- a/x86_64/ldscript
> > +++ b/x86_64/ldscript
> > @@ -2,7 +2,7 @@
> > OUTPUT_FORMAT("elf64-x86-64", "elf64-x86-64",
> > "elf64-x86-64")
> > OUTPUT_ARCH(i386:x86-64)
> > -ENTRY(_start)
> > +ENTRY(boot_start)
> > SECTIONS
> > {
> > /*
> > @@ -11,22 +11,30 @@ SECTIONS
> > * be first in there. See also `i386/i386at/boothdr.S' and
> > * `gnumach_LINKFLAGS' in `i386/Makefrag.am'.
> > */
> > - . = _START;
> > - .text :
> > - AT (_START_MAP)
> > +
> > + . = _START_MAP;
> > + .boot :
> > + {
> > + *(.boot.text)
> > + *(.boot.data)
> > + } =0x90909090
> > +
> > + . += KERNEL_MAP_BASE;
> > + _start = .;
> > + .text : AT(((ADDR(.text)) - KERNEL_MAP_BASE))
> > {
> > - *(.text.start)
> > + *(.text*)
> > *(.text .stub .text.* .gnu.linkonce.t.*)
> > *(.text.unlikely .text.*_unlikely)
> > KEEP (*(.text.*personality*))
> > /* .gnu.warning sections are handled specially by elf32.em. */
> > *(.gnu.warning)
> > } =0x90909090
> > - .init :
> > + .init : AT(((ADDR(.init)) - KERNEL_MAP_BASE))
> > {
> > KEEP (*(.init))
> > } =0x90909090
> > - .fini :
> > + .fini : AT(((ADDR(.fini)) - KERNEL_MAP_BASE))
> > {
> > KEEP (*(.fini))
> > } =0x90909090
> > @@ -69,7 +77,7 @@ SECTIONS
> > PROVIDE_HIDDEN (__rela_iplt_end = .);
> > }
> > .plt : { *(.plt) *(.iplt) }
> > - .rodata : { *(.rodata .rodata.* .gnu.linkonce.r.*) }
> > + .rodata : AT(((ADDR(.rodata)) - KERNEL_MAP_BASE)) { *(.rodata
> > .rodata.* .gnu.linkonce.r.*) }
> > .rodata1 : { *(.rodata1) }
> > .eh_frame_hdr : { *(.eh_frame_hdr) }
> > .eh_frame : ONLY_IF_RO { KEEP (*(.eh_frame)) }
> > @@ -139,7 +147,7 @@ SECTIONS
> > .got : { *(.got) *(.igot) }
> > . = DATA_SEGMENT_RELRO_END (24, .);
> > .got.plt : { *(.got.plt) *(.igot.plt) }
> > - .data :
> > + .data : AT(((ADDR(.data)) - KERNEL_MAP_BASE))
> > {
> > *(.data .data.* .gnu.linkonce.d.*)
> > SORT(CONSTRUCTORS)
> > @@ -147,7 +155,7 @@ SECTIONS
> > .data1 : { *(.data1) }
> > _edata = .; PROVIDE (edata = .);
> > __bss_start = .;
> > - .bss :
> > + .bss : AT(((ADDR(.bss)) - KERNEL_MAP_BASE))
> > {
> > *(.dynbss)
> > *(.bss .bss.* .gnu.linkonce.b.*)
> > --
> > 2.30.2
> >
> >
>
> --
> Samuel
> ---
> Pour une évaluation indépendante, transparente et rigoureuse !
> Je soutiens la Commission d'Évaluation de l'Inria.
--
Samuel
---
Pour une évaluation indépendante, transparente et rigoureuse !
Je soutiens la Commission d'Évaluation de l'Inria.