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_nfsroot.h"
46 #include <sys/reboot.h>
48 #include <machine/asmacros.h>
49 #include <machine/cputypes.h>
50 #include <machine/psl.h>
51 #include <machine/pmap.h>
52 #include <machine/specialreg.h>
59 * Note: This version greatly munged to avoid various assembler errors
60 * that may be fixed in newer versions of gas. Perhaps newer versions
61 * will have more pleasant appearance.
65 * PTmap is recursive pagemap at top of virtual address space.
66 * Within PTmap, the page directory can be found (third indirection).
68 .globl PTmap,PTD,PTDpde
69 .set PTmap,(PTDPTDI << PDRSHIFT)
70 .set PTD,PTmap + (PTDPTDI * PAGE_SIZE)
71 .set PTDpde,PTD + (PTDPTDI * PDESIZE)
74 * Compiled KERNBASE location and the kernel load address
77 .set kernbase,KERNBASE
79 .set kernload,KERNLOAD
85 ALIGN_DATA /* just to be sure */
87 .space 0x2000 /* space for tmpstk - temporary stack */
91 bootinfo: .space BOOTINFO_SIZE /* bootinfo that we can handle */
94 KERNend: .long 0 /* phys addr end of kernel (just after bss) */
95 physfree: .long 0 /* phys addr of next free page */
98 IdlePTD: .long 0 /* phys addr of kernel PTD */
100 #if defined(PAE) || defined(PAE_TABLES)
102 IdlePDPT: .long 0 /* phys addr of kernel PDPT */
106 KPTmap: .long 0 /* address of kernel page tables */
109 KPTphys: .long 0 /* phys addr of kernel page tables */
112 proc0kstack: .long 0 /* address of proc 0 kstack space */
113 p0kpa: .long 0 /* phys addr of proc0's STACK */
115 vm86phystk: .long 0 /* PA of vm86/bios stack */
117 .globl vm86paddr, vm86pa
118 vm86paddr: .long 0 /* address of vm86 region */
119 vm86pa: .long 0 /* phys addr of vm86 region */
121 /**********************************************************************
127 #define R(foo) ((foo)-KERNBASE)
129 #define ALLOCPAGES(foo) \
130 movl R(physfree), %esi ; \
131 movl $((foo)*PAGE_SIZE), %eax ; \
133 movl %eax, R(physfree) ; \
135 movl $((foo)*PAGE_SIZE),%ecx ; \
143 * eax = page frame address
144 * ebx = index into page table
145 * ecx = how many pages to map
146 * base = base address of page dir/table
147 * prot = protection bits
149 #define fillkpt(base, prot) \
150 shll $PTESHIFT,%ebx ; \
154 1: movl %eax,(%ebx) ; \
155 addl $PAGE_SIZE,%eax ; /* increment physical address */ \
156 addl $PTESIZE,%ebx ; /* next pte */ \
161 * eax = physical address
162 * ecx = how many pages to map
163 * prot = protection bits
165 #define fillkptphys(prot) \
167 shrl $PAGE_SHIFT, %ebx ; \
168 fillkpt(R(KPTphys), prot)
171 /**********************************************************************
173 * This is where the bootblocks start us, set the ball rolling...
176 NON_GPROF_ENTRY(btext)
178 /* Tell the bios to warmboot next time */
181 /* Set up a real frame in case the double return in newboot is executed. */
185 /* Don't trust what the BIOS gives for eflags. */
190 * Don't trust what the BIOS gives for %fs and %gs. Trust the bootstrap
191 * to set %cs, %ds, %es and %ss.
198 * Clear the bss. Not all boot programs do it, and it is our job anyway.
200 * XXX we don't check that there is memory for our bss and page tables
203 * Note: we must be careful to not overwrite an active gdt or idt. They
204 * inactive from now until we switch to new ones, since we don't load any
205 * more segment registers or permit interrupts until after the switch.
215 call recover_bootinfo
217 /* Get onto a stack that we can trust. */
219 * XXX this step is delayed in case recover_bootinfo needs to return via
220 * the old stack, but it need not be, since recover_bootinfo actually
221 * returns via the old frame.
226 call create_pagetables
229 * If the CPU has support for VME, turn it on.
231 testl $CPUID_VME, R(cpu_feature)
238 /* Now enable paging */
239 #if defined(PAE) || defined(PAE_TABLES)
240 movl R(IdlePDPT), %eax
246 movl R(IdlePTD), %eax
247 movl %eax,%cr3 /* load ptd addr into mmu */
249 movl %cr0,%edx /* get control word */
250 orl $CR0_PE|CR0_PG,%edx /* enable paging */
251 movl %edx,%cr0 /* and let's page NOW! */
253 pushl $begin /* jump to high virtualized address */
258 * Now running relocated at KERNBASE where the system is linked to run.
260 * Remove the lowest part of the double mapping of low memory to get
261 * some null pointer checks.
264 movl %eax,%cr3 /* invalidate TLB */
266 /* set up bootstrap stack */
267 movl proc0kstack,%eax /* location of in-kernel stack */
270 * Only use bottom page for init386(). init386() calculates the
271 * PCB + FPU save area size and returns the true top of stack.
273 leal PAGE_SIZE(%eax),%esp
275 xorl %ebp,%ebp /* mark end of frames */
277 pushl physfree /* value of first for init386(first) */
278 call init386 /* wire 386 chip for unix operation */
281 * Clean up the stack in a way that db_numargs() understands, so
282 * that backtraces in ddb don't underrun the stack. Traps for
283 * inaccessible memory are more fatal than usual this early.
287 /* Switch to true top of stack. */
290 call mi_startup /* autoconfiguration, mountroot etc */
292 addl $0,%esp /* for db_numargs() again */
294 /**********************************************************************
296 * Recover the bootinfo passed to us from the boot program
301 * This code is called in different ways depending on what loaded
302 * and started the kernel. This is used to detect how we get the
303 * arguments from the other code and what we do with them.
305 * Old disk boot blocks:
306 * (*btext)(howto, bootdev, cyloffset, esym);
307 * [return address == 0, and can NOT be returned to]
308 * [cyloffset was not supported by the FreeBSD boot code
309 * and always passed in as 0]
310 * [esym is also known as total in the boot code, and
311 * was never properly supported by the FreeBSD boot code]
313 * Old diskless netboot code:
314 * (*btext)(0,0,0,0,&nfsdiskless,0,0,0);
315 * [return address != 0, and can NOT be returned to]
316 * If we are being booted by this code it will NOT work,
317 * so we are just going to halt if we find this case.
319 * New uniform boot code:
320 * (*btext)(howto, bootdev, 0, 0, 0, &bootinfo)
321 * [return address != 0, and can be returned to]
323 * There may seem to be a lot of wasted arguments in here, but
324 * that is so the newer boot code can still load very old kernels
325 * and old boot code can load new kernels.
329 * The old style disk boot blocks fake a frame on the stack and
330 * did an lret to get here. The frame on the stack has a return
337 * We have some form of return address, so this is either the
338 * old diskless netboot code, or the new uniform code. That can
339 * be detected by looking at the 5th argument, if it is 0
340 * we are being booted by the new uniform boot code.
346 * Seems we have been loaded by the old diskless boot code, we
347 * don't stand a chance of running as the diskless structure
348 * changed considerably between the two, so just halt.
353 * We have been loaded by the new uniform boot code.
354 * Let's check the bootinfo version, and if we do not understand
355 * it we return to the loader with a status of 1 to indicate this error
358 movl 28(%ebp),%ebx /* &bootinfo.version */
359 movl BI_VERSION(%ebx),%eax
360 cmpl $1,%eax /* We only understand version 1 */
362 movl $1,%eax /* Return status */
365 * XXX this returns to our caller's caller (as is required) since
366 * we didn't set up a frame and our caller did.
372 * If we have a kernelname copy it in
374 movl BI_KERNELNAME(%ebx),%esi
376 je 2f /* No kernelname */
377 movl $MAXPATHLEN,%ecx /* Brute force!!! */
378 movl $R(kernelname),%edi
379 cmpb $'/',(%esi) /* Make sure it starts with a slash */
391 * Determine the size of the boot loader's copy of the bootinfo
392 * struct. This is impossible to do properly because old versions
393 * of the struct don't contain a size field and there are 2 old
394 * versions with the same version number.
396 movl $BI_ENDCOMMON,%ecx /* prepare for sizeless version */
397 testl $RB_BOOTINFO,8(%ebp) /* bi_size (and bootinfo) valid? */
398 je got_bi_size /* no, sizeless version */
399 movl BI_SIZE(%ebx),%ecx
403 * Copy the common part of the bootinfo struct
406 movl $R(bootinfo),%edi
407 cmpl $BOOTINFO_SIZE,%ecx
408 jbe got_common_bi_size
409 movl $BOOTINFO_SIZE,%ecx
418 * If we have a nfs_diskless structure copy it in
420 movl BI_NFS_DISKLESS(%ebx),%esi
423 movl $R(nfs_diskless),%edi
424 movl $NFSDISKLESS_SIZE,%ecx
428 movl $R(nfs_diskless_valid),%edi
434 * The old style disk boot.
435 * (*btext)(howto, bootdev, cyloffset, esym);
436 * Note that the newer boot code just falls into here to pick
437 * up howto and bootdev, cyloffset and esym are no longer used
441 movl %eax,R(boothowto)
448 /**********************************************************************
450 * Identify the CPU and initialize anything special about it
455 /* Try to toggle alignment check flag; does not exist on 386. */
472 /* NexGen CPU does not have aligment check flag. */
486 movl $CPU_NX586,R(cpu)
487 movl $0x4778654e,R(cpu_vendor) # store vendor string
488 movl $0x72446e65,R(cpu_vendor+4)
489 movl $0x6e657669,R(cpu_vendor+8)
490 movl $0,R(cpu_vendor+12)
493 try486: /* Try to toggle identification flag; does not exist on early 486s. */
513 * Cyrix CPUs do not change the undefined flags following
514 * execution of the divide instruction which divides 5 by 2.
516 * Note: CPUID is enabled on M2, so it passes another way.
526 jmp 3f /* You may use Intel CPU. */
531 * IBM Bluelighting CPU also doesn't change the undefined flags.
532 * Because IBM doesn't disclose the information for Bluelighting
533 * CPU, we couldn't distinguish it from Cyrix's (including IBM
534 * brand of Cyrix CPUs).
536 movl $0x69727943,R(cpu_vendor) # store vendor string
537 movl $0x736e4978,R(cpu_vendor+4)
538 movl $0x64616574,R(cpu_vendor+8)
541 trycpuid: /* Use the `cpuid' instruction. */
544 movl %eax,R(cpu_high) # highest capability
545 movl %ebx,R(cpu_vendor) # store vendor string
546 movl %edx,R(cpu_vendor+4)
547 movl %ecx,R(cpu_vendor+8)
548 movb $0,R(cpu_vendor+12)
552 movl %eax,R(cpu_id) # store cpu_id
553 movl %ebx,R(cpu_procinfo) # store cpu_procinfo
554 movl %edx,R(cpu_feature) # store cpu_feature
555 movl %ecx,R(cpu_feature2) # store cpu_feature2
556 rorl $8,%eax # extract family type
561 /* less than Pentium; must be 486 */
571 /* Greater than Pentium...call it a Pentium Pro */
577 /**********************************************************************
579 * Create the first page directory and its page tables.
585 /* Find end of kernel image (rounded up to a page boundary). */
588 /* Include symbols, if any. */
589 movl R(bootinfo+BI_ESYMTAB),%edi
594 addl %edi,R(bootinfo+BI_SYMTAB)
595 addl %edi,R(bootinfo+BI_ESYMTAB)
598 /* If we are told where the end of the kernel space is, believe it. */
599 movl R(bootinfo+BI_KERNEND),%edi
605 addl $PDRMASK,%esi /* Play conservative for now, and */
606 andl $~PDRMASK,%esi /* ... round up to PDR boundary */
607 movl %esi,R(KERNend) /* save end of kernel */
608 movl %esi,R(physfree) /* next free page is at end of kernel */
610 /* Allocate Kernel Page Tables */
613 addl $(KERNBASE-(KPTDI<<(PDRSHIFT-PAGE_SHIFT+PTESHIFT))),%esi
616 /* Allocate Page Table Directory */
617 #if defined(PAE) || defined(PAE_TABLES)
618 /* XXX only need 32 bytes (easier for now) */
620 movl %esi,R(IdlePDPT)
625 /* Allocate KSTACK */
626 ALLOCPAGES(TD0_KSTACK_PAGES)
629 movl %esi, R(proc0kstack)
631 ALLOCPAGES(1) /* vm86/bios stack */
632 movl %esi,R(vm86phystk)
634 ALLOCPAGES(3) /* pgtable + ext + IOPAGES */
637 movl %esi, R(vm86paddr)
640 * Enable PSE and PGE.
643 testl $CPUID_PSE, R(cpu_feature)
645 movl $PG_PS, R(pseflag)
652 testl $CPUID_PGE, R(cpu_feature)
654 movl $PG_G, R(pgeflag)
662 * Initialize page table pages mapping physical address zero through the
663 * (physical) end of the kernel. Many of these pages must be reserved,
664 * and we reserve them all and map them linearly for convenience. We do
665 * this even if we've enabled PSE above; we'll just switch the corresponding
666 * kernel PDEs before we turn on paging.
668 * XXX: We waste some pages here in the PSE case!
670 * This and all other page table entries allow read and write access for
671 * various reasons. Kernel mappings never have any access restrictions.
675 shrl $PAGE_SHIFT,%ecx
678 /* Map page table pages. */
683 /* Map page directory. */
684 #if defined(PAE) || defined(PAE_TABLES)
685 movl R(IdlePDPT), %eax
690 movl R(IdlePTD), %eax
694 /* Map proc0's KSTACK in the physical way ... */
696 movl $(TD0_KSTACK_PAGES), %ecx
700 movl $ISA_HOLE_START, %eax
701 movl $ISA_HOLE_LENGTH>>PAGE_SHIFT, %ecx
704 /* Map space for the vm86 region */
705 movl R(vm86phystk), %eax
709 /* Map page 0 into the vm86 page table */
713 fillkpt(R(vm86pa), $PG_RW|PG_U)
715 /* ...likewise for the ISA hole */
716 movl $ISA_HOLE_START, %eax
717 movl $ISA_HOLE_START>>PAGE_SHIFT, %ebx
718 movl $ISA_HOLE_LENGTH>>PAGE_SHIFT, %ecx
719 fillkpt(R(vm86pa), $PG_RW|PG_U)
722 * Create an identity mapping for low physical memory, including the kernel.
723 * This is only used to map the 2 instructions for jumping to 'begin' in
724 * locore (we map everything to avoid having to determine where these
725 * instructions are). ACPI resume will transiently restore the first PDE in
726 * this mapping (and depend on this PDE's page table created here not being
727 * destroyed). See pmap_bootstrap() for more details.
729 * Note: There are errata concerning large pages and physical address zero,
730 * so a PG_PS mapping should not be used for PDE 0. Our double mapping
731 * avoids this automatically by not using PG_PS for PDE #KPDI so that PAT
732 * bits can be set at the page level for i/o pages below 1 MB.
734 movl R(KPTphys), %eax
737 fillkpt(R(IdlePTD), $PG_RW)
740 * Install PDEs for PTs covering enough kva to bootstrap. Then for the PSE
741 * case, replace the PDEs whose coverage is strictly within the kernel
742 * (between KERNLOAD (rounded up) and KERNend) by large-page PDEs.
744 movl R(KPTphys), %eax
747 fillkpt(R(IdlePTD), $PG_RW)
751 movl R(KERNend), %ecx
752 movl $(KERNLOAD + PDRMASK) & ~PDRMASK, %eax
755 movl $KPTDI + ((KERNLOAD + PDRMASK) >> PDRSHIFT), %ebx
757 addl R(IdlePTD), %ebx
758 orl $(PG_V|PG_RW|PG_PS), %eax
760 addl $(1 << PDRSHIFT), %eax
765 /* install a pde recursively mapping page directory as a page table */
766 movl R(IdlePTD), %eax
769 fillkpt(R(IdlePTD), $PG_RW)
771 #if defined(PAE) || defined(PAE_TABLES)
772 movl R(IdlePTD), %eax
775 fillkpt(R(IdlePDPT), $0x0)
781 /* Xen Hypercall page */
783 .p2align PAGE_SHIFT, 0x90 /* Hypercall_page needs to be PAGE aligned */
785 NON_GPROF_ENTRY(hypercall_page)
786 .skip 0x1000, 0x90 /* Fill with "nop"s */