2 * Copyright (c) 1990 The Regents of the University of California.
5 * This code is derived from software contributed to Berkeley by
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)locore.s 7.3 (Berkeley) 5/13/91
35 * originally from: locore.s, by William F. Jolitz
37 * Substantially rewritten by David Greenman, Rod Grimes,
38 * Bruce Evans, Wolfgang Solfrank, Poul-Henning Kamp
42 #include "opt_bootp.h"
43 #include "opt_compat.h"
44 #include "opt_nfsroot.h"
47 #include <sys/syscall.h>
48 #include <sys/reboot.h>
50 #include <machine/asmacros.h>
51 #include <machine/cputypes.h>
52 #include <machine/psl.h>
53 #include <machine/pmap.h>
54 #include <machine/specialreg.h>
61 * Note: This version greatly munged to avoid various assembler errors
62 * that may be fixed in newer versions of gas. Perhaps newer versions
63 * will have more pleasant appearance.
67 * PTmap is recursive pagemap at top of virtual address space.
68 * Within PTmap, the page directory can be found (third indirection).
70 .globl PTmap,PTD,PTDpde
71 .set PTmap,(PTDPTDI << PDRSHIFT)
72 .set PTD,PTmap + (PTDPTDI * PAGE_SIZE)
73 .set PTDpde,PTD + (PTDPTDI * PDESIZE)
76 * Compiled KERNBASE location and the kernel load address
79 .set kernbase,KERNBASE
81 .set kernload,KERNLOAD
87 ALIGN_DATA /* just to be sure */
89 .space 0x2000 /* space for tmpstk - temporary stack */
93 bootinfo: .space BOOTINFO_SIZE /* bootinfo that we can handle */
96 KERNend: .long 0 /* phys addr end of kernel (just after bss) */
97 physfree: .long 0 /* phys addr of next free page */
100 IdlePTD: .long 0 /* phys addr of kernel PTD */
102 #if defined(PAE) || defined(PAE_TABLES)
104 IdlePDPT: .long 0 /* phys addr of kernel PDPT */
108 KPTmap: .long 0 /* address of kernel page tables */
111 KPTphys: .long 0 /* phys addr of kernel page tables */
114 proc0kstack: .long 0 /* address of proc 0 kstack space */
115 p0kpa: .long 0 /* phys addr of proc0's STACK */
117 vm86phystk: .long 0 /* PA of vm86/bios stack */
119 .globl vm86paddr, vm86pa
120 vm86paddr: .long 0 /* address of vm86 region */
121 vm86pa: .long 0 /* phys addr of vm86 region */
123 /**********************************************************************
129 #define R(foo) ((foo)-KERNBASE)
131 #define ALLOCPAGES(foo) \
132 movl R(physfree), %esi ; \
133 movl $((foo)*PAGE_SIZE), %eax ; \
135 movl %eax, R(physfree) ; \
137 movl $((foo)*PAGE_SIZE),%ecx ; \
145 * eax = page frame address
146 * ebx = index into page table
147 * ecx = how many pages to map
148 * base = base address of page dir/table
149 * prot = protection bits
151 #define fillkpt(base, prot) \
152 shll $PTESHIFT,%ebx ; \
156 1: movl %eax,(%ebx) ; \
157 addl $PAGE_SIZE,%eax ; /* increment physical address */ \
158 addl $PTESIZE,%ebx ; /* next pte */ \
163 * eax = physical address
164 * ecx = how many pages to map
165 * prot = protection bits
167 #define fillkptphys(prot) \
169 shrl $PAGE_SHIFT, %ebx ; \
170 fillkpt(R(KPTphys), prot)
173 /**********************************************************************
175 * This is where the bootblocks start us, set the ball rolling...
178 NON_GPROF_ENTRY(btext)
180 /* Tell the bios to warmboot next time */
183 /* Set up a real frame in case the double return in newboot is executed. */
187 /* Don't trust what the BIOS gives for eflags. */
192 * Don't trust what the BIOS gives for %fs and %gs. Trust the bootstrap
193 * to set %cs, %ds, %es and %ss.
200 * Clear the bss. Not all boot programs do it, and it is our job anyway.
202 * XXX we don't check that there is memory for our bss and page tables
205 * Note: we must be careful to not overwrite an active gdt or idt. They
206 * inactive from now until we switch to new ones, since we don't load any
207 * more segment registers or permit interrupts until after the switch.
217 call recover_bootinfo
219 /* Get onto a stack that we can trust. */
221 * XXX this step is delayed in case recover_bootinfo needs to return via
222 * the old stack, but it need not be, since recover_bootinfo actually
223 * returns via the old frame.
228 call create_pagetables
231 * If the CPU has support for VME, turn it on.
233 testl $CPUID_VME, R(cpu_feature)
240 /* Now enable paging */
241 #if defined(PAE) || defined(PAE_TABLES)
242 movl R(IdlePDPT), %eax
248 movl R(IdlePTD), %eax
249 movl %eax,%cr3 /* load ptd addr into mmu */
251 movl %cr0,%eax /* get control word */
252 orl $CR0_PE|CR0_PG,%eax /* enable paging */
253 movl %eax,%cr0 /* and let's page NOW! */
255 pushl $begin /* jump to high virtualized address */
258 /* now running relocated at KERNBASE where the system is linked to run */
260 /* set up bootstrap stack */
261 movl proc0kstack,%eax /* location of in-kernel stack */
264 * Only use bottom page for init386(). init386() calculates the
265 * PCB + FPU save area size and returns the true top of stack.
267 leal PAGE_SIZE(%eax),%esp
269 xorl %ebp,%ebp /* mark end of frames */
271 pushl physfree /* value of first for init386(first) */
272 call init386 /* wire 386 chip for unix operation */
275 * Clean up the stack in a way that db_numargs() understands, so
276 * that backtraces in ddb don't underrun the stack. Traps for
277 * inaccessible memory are more fatal than usual this early.
281 /* Switch to true top of stack. */
284 call mi_startup /* autoconfiguration, mountroot etc */
286 addl $0,%esp /* for db_numargs() again */
289 * Signal trampoline, copied to top of user stack
291 NON_GPROF_ENTRY(sigcode)
292 calll *SIGF_HANDLER(%esp)
293 leal SIGF_UC(%esp),%eax /* get ucontext */
295 testl $PSL_VM,UC_EFLAGS(%eax)
297 mov UC_GS(%eax),%gs /* restore %gs */
299 movl $SYS_sigreturn,%eax
300 pushl %eax /* junk to fake return addr. */
301 int $0x80 /* enter kernel with args */
306 #ifdef COMPAT_FREEBSD4
309 calll *SIGF_HANDLER(%esp)
310 leal SIGF_UC4(%esp),%eax /* get ucontext */
312 testl $PSL_VM,UC4_EFLAGS(%eax)
314 mov UC4_GS(%eax),%gs /* restore %gs */
316 movl $344,%eax /* 4.x SYS_sigreturn */
317 pushl %eax /* junk to fake return addr. */
318 int $0x80 /* enter kernel with args */
327 call *SIGF_HANDLER(%esp) /* call signal handler */
328 lea SIGF_SC(%esp),%eax /* get sigcontext */
330 testl $PSL_VM,SC_PS(%eax)
332 mov SC_GS(%eax),%gs /* restore %gs */
334 movl $103,%eax /* 3.x SYS_sigreturn */
335 pushl %eax /* junk to fake return addr. */
336 int $0x80 /* enter kernel with args */
338 #endif /* COMPAT_43 */
346 .long esigcode-sigcode
347 #ifdef COMPAT_FREEBSD4
348 .globl szfreebsd4_sigcode
350 .long esigcode-freebsd4_sigcode
355 .long esigcode-osigcode
359 /**********************************************************************
361 * Recover the bootinfo passed to us from the boot program
366 * This code is called in different ways depending on what loaded
367 * and started the kernel. This is used to detect how we get the
368 * arguments from the other code and what we do with them.
370 * Old disk boot blocks:
371 * (*btext)(howto, bootdev, cyloffset, esym);
372 * [return address == 0, and can NOT be returned to]
373 * [cyloffset was not supported by the FreeBSD boot code
374 * and always passed in as 0]
375 * [esym is also known as total in the boot code, and
376 * was never properly supported by the FreeBSD boot code]
378 * Old diskless netboot code:
379 * (*btext)(0,0,0,0,&nfsdiskless,0,0,0);
380 * [return address != 0, and can NOT be returned to]
381 * If we are being booted by this code it will NOT work,
382 * so we are just going to halt if we find this case.
384 * New uniform boot code:
385 * (*btext)(howto, bootdev, 0, 0, 0, &bootinfo)
386 * [return address != 0, and can be returned to]
388 * There may seem to be a lot of wasted arguments in here, but
389 * that is so the newer boot code can still load very old kernels
390 * and old boot code can load new kernels.
394 * The old style disk boot blocks fake a frame on the stack and
395 * did an lret to get here. The frame on the stack has a return
402 * We have some form of return address, so this is either the
403 * old diskless netboot code, or the new uniform code. That can
404 * be detected by looking at the 5th argument, if it is 0
405 * we are being booted by the new uniform boot code.
411 * Seems we have been loaded by the old diskless boot code, we
412 * don't stand a chance of running as the diskless structure
413 * changed considerably between the two, so just halt.
418 * We have been loaded by the new uniform boot code.
419 * Let's check the bootinfo version, and if we do not understand
420 * it we return to the loader with a status of 1 to indicate this error
423 movl 28(%ebp),%ebx /* &bootinfo.version */
424 movl BI_VERSION(%ebx),%eax
425 cmpl $1,%eax /* We only understand version 1 */
427 movl $1,%eax /* Return status */
430 * XXX this returns to our caller's caller (as is required) since
431 * we didn't set up a frame and our caller did.
437 * If we have a kernelname copy it in
439 movl BI_KERNELNAME(%ebx),%esi
441 je 2f /* No kernelname */
442 movl $MAXPATHLEN,%ecx /* Brute force!!! */
443 movl $R(kernelname),%edi
444 cmpb $'/',(%esi) /* Make sure it starts with a slash */
456 * Determine the size of the boot loader's copy of the bootinfo
457 * struct. This is impossible to do properly because old versions
458 * of the struct don't contain a size field and there are 2 old
459 * versions with the same version number.
461 movl $BI_ENDCOMMON,%ecx /* prepare for sizeless version */
462 testl $RB_BOOTINFO,8(%ebp) /* bi_size (and bootinfo) valid? */
463 je got_bi_size /* no, sizeless version */
464 movl BI_SIZE(%ebx),%ecx
468 * Copy the common part of the bootinfo struct
471 movl $R(bootinfo),%edi
472 cmpl $BOOTINFO_SIZE,%ecx
473 jbe got_common_bi_size
474 movl $BOOTINFO_SIZE,%ecx
483 * If we have a nfs_diskless structure copy it in
485 movl BI_NFS_DISKLESS(%ebx),%esi
488 movl $R(nfs_diskless),%edi
489 movl $NFSDISKLESS_SIZE,%ecx
493 movl $R(nfs_diskless_valid),%edi
499 * The old style disk boot.
500 * (*btext)(howto, bootdev, cyloffset, esym);
501 * Note that the newer boot code just falls into here to pick
502 * up howto and bootdev, cyloffset and esym are no longer used
506 movl %eax,R(boothowto)
513 /**********************************************************************
515 * Identify the CPU and initialize anything special about it
520 /* Try to toggle alignment check flag; does not exist on 386. */
537 /* NexGen CPU does not have aligment check flag. */
551 movl $CPU_NX586,R(cpu)
552 movl $0x4778654e,R(cpu_vendor) # store vendor string
553 movl $0x72446e65,R(cpu_vendor+4)
554 movl $0x6e657669,R(cpu_vendor+8)
555 movl $0,R(cpu_vendor+12)
558 try486: /* Try to toggle identification flag; does not exist on early 486s. */
578 * Cyrix CPUs do not change the undefined flags following
579 * execution of the divide instruction which divides 5 by 2.
581 * Note: CPUID is enabled on M2, so it passes another way.
591 jmp 3f /* You may use Intel CPU. */
596 * IBM Bluelighting CPU also doesn't change the undefined flags.
597 * Because IBM doesn't disclose the information for Bluelighting
598 * CPU, we couldn't distinguish it from Cyrix's (including IBM
599 * brand of Cyrix CPUs).
601 movl $0x69727943,R(cpu_vendor) # store vendor string
602 movl $0x736e4978,R(cpu_vendor+4)
603 movl $0x64616574,R(cpu_vendor+8)
606 trycpuid: /* Use the `cpuid' instruction. */
609 movl %eax,R(cpu_high) # highest capability
610 movl %ebx,R(cpu_vendor) # store vendor string
611 movl %edx,R(cpu_vendor+4)
612 movl %ecx,R(cpu_vendor+8)
613 movb $0,R(cpu_vendor+12)
617 movl %eax,R(cpu_id) # store cpu_id
618 movl %ebx,R(cpu_procinfo) # store cpu_procinfo
619 movl %edx,R(cpu_feature) # store cpu_feature
620 movl %ecx,R(cpu_feature2) # store cpu_feature2
621 rorl $8,%eax # extract family type
626 /* less than Pentium; must be 486 */
636 /* Greater than Pentium...call it a Pentium Pro */
642 /**********************************************************************
644 * Create the first page directory and its page tables.
650 /* Find end of kernel image (rounded up to a page boundary). */
653 /* Include symbols, if any. */
654 movl R(bootinfo+BI_ESYMTAB),%edi
659 addl %edi,R(bootinfo+BI_SYMTAB)
660 addl %edi,R(bootinfo+BI_ESYMTAB)
663 /* If we are told where the end of the kernel space is, believe it. */
664 movl R(bootinfo+BI_KERNEND),%edi
670 addl $PDRMASK,%esi /* Play conservative for now, and */
671 andl $~PDRMASK,%esi /* ... wrap to next 4M. */
672 movl %esi,R(KERNend) /* save end of kernel */
673 movl %esi,R(physfree) /* next free page is at end of kernel */
675 /* Allocate Kernel Page Tables */
678 addl $(KERNBASE-(KPTDI<<(PDRSHIFT-PAGE_SHIFT+PTESHIFT))),%esi
681 /* Allocate Page Table Directory */
682 #if defined(PAE) || defined(PAE_TABLES)
683 /* XXX only need 32 bytes (easier for now) */
685 movl %esi,R(IdlePDPT)
690 /* Allocate KSTACK */
691 ALLOCPAGES(TD0_KSTACK_PAGES)
694 movl %esi, R(proc0kstack)
696 ALLOCPAGES(1) /* vm86/bios stack */
697 movl %esi,R(vm86phystk)
699 ALLOCPAGES(3) /* pgtable + ext + IOPAGES */
702 movl %esi, R(vm86paddr)
705 * Enable PSE and PGE.
708 testl $CPUID_PSE, R(cpu_feature)
710 movl $PG_PS, R(pseflag)
717 testl $CPUID_PGE, R(cpu_feature)
719 movl $PG_G, R(pgeflag)
727 * Initialize page table pages mapping physical address zero through the
728 * end of the kernel. All of the page table entries allow read and write
729 * access. Write access to the first physical page is required by bios32
730 * calls, and write access to the first 1 MB of physical memory is required
731 * by ACPI for implementing suspend and resume. We do this even
732 * if we've enabled PSE above, we'll just switch the corresponding kernel
733 * PDEs before we turn on paging.
735 * XXX: We waste some pages here in the PSE case!
739 shrl $PAGE_SHIFT,%ecx
742 /* Map page table pages. */
747 /* Map page directory. */
748 #if defined(PAE) || defined(PAE_TABLES)
749 movl R(IdlePDPT), %eax
754 movl R(IdlePTD), %eax
758 /* Map proc0's KSTACK in the physical way ... */
760 movl $(TD0_KSTACK_PAGES), %ecx
764 movl $ISA_HOLE_START, %eax
765 movl $ISA_HOLE_LENGTH>>PAGE_SHIFT, %ecx
768 /* Map space for the vm86 region */
769 movl R(vm86phystk), %eax
773 /* Map page 0 into the vm86 page table */
777 fillkpt(R(vm86pa), $PG_RW|PG_U)
779 /* ...likewise for the ISA hole */
780 movl $ISA_HOLE_START, %eax
781 movl $ISA_HOLE_START>>PAGE_SHIFT, %ebx
782 movl $ISA_HOLE_LENGTH>>PAGE_SHIFT, %ecx
783 fillkpt(R(vm86pa), $PG_RW|PG_U)
786 * Create an identity mapping for low physical memory, including the kernel.
787 * The part of this mapping that covers the first 1 MB of physical memory
788 * becomes a permanent part of the kernel's address space. The rest of this
789 * mapping is destroyed in pmap_bootstrap(). Ordinarily, the same page table
790 * pages are shared by the identity mapping and the kernel's native mapping.
791 * However, the permanent identity mapping cannot contain PG_G mappings.
792 * Thus, if the kernel is loaded within the permanent identity mapping, that
793 * page table page must be duplicated and not shared.
795 * N.B. Due to errata concerning large pages and physical address zero,
796 * a PG_PS mapping is not used.
798 movl R(KPTphys), %eax
801 fillkpt(R(IdlePTD), $PG_RW)
802 #if KERNLOAD < (1 << PDRSHIFT)
803 testl $PG_G, R(pgeflag)
807 movl R(IdlePTD), %eax
810 movl $PAGE_SIZE, %ecx
818 * For the non-PSE case, install PDEs for PTs covering the KVA.
819 * For the PSE case, do the same, but clobber the ones corresponding
820 * to the kernel (from btext to KERNend) with 4M (2M for PAE) ('PS')
821 * PDEs immediately after.
823 movl R(KPTphys), %eax
826 fillkpt(R(IdlePTD), $PG_RW)
830 movl R(KERNend), %ecx
834 movl $(KPTDI+(KERNLOAD/(1 << PDRSHIFT))), %ebx
836 addl R(IdlePTD), %ebx
837 orl $(PG_V|PG_RW|PG_PS), %eax
839 addl $(1 << PDRSHIFT), %eax
844 /* install a pde recursively mapping page directory as a page table */
845 movl R(IdlePTD), %eax
848 fillkpt(R(IdlePTD), $PG_RW)
850 #if defined(PAE) || defined(PAE_TABLES)
851 movl R(IdlePTD), %eax
854 fillkpt(R(IdlePDPT), $0x0)
860 /* Xen Hypercall page */
862 .p2align PAGE_SHIFT, 0x90 /* Hypercall_page needs to be PAGE aligned */
864 NON_GPROF_ENTRY(hypercall_page)
865 .skip 0x1000, 0x90 /* Fill with "nop"s */