1 /* $NetBSD: arm32_machdep.c,v 1.44 2004/03/24 15:34:47 atatat Exp $ */
4 * Copyright (c) 2004 Olivier Houchard
5 * Copyright (c) 1994-1998 Mark Brinicombe.
6 * Copyright (c) 1994 Brini.
9 * This code is derived from software written for Brini by Mark Brinicombe
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by Mark Brinicombe
22 * for the NetBSD Project.
23 * 4. The name of the company nor the name of the author may be used to
24 * endorse or promote products derived from this software without specific
25 * prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
28 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
29 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
30 * IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
31 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
32 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
33 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * Machine dependant functions for kernel setup
42 * Updated : 18/04/01 updated for new wscons
45 #include "opt_compat.h"
47 #include "opt_platform.h"
48 #include "opt_sched.h"
49 #include "opt_timer.h"
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
54 #include <sys/param.h>
56 #include <sys/systm.h>
63 #include <sys/imgact.h>
65 #include <sys/kernel.h>
67 #include <sys/linker.h>
69 #include <sys/malloc.h>
70 #include <sys/msgbuf.h>
71 #include <sys/mutex.h>
73 #include <sys/ptrace.h>
74 #include <sys/rwlock.h>
75 #include <sys/sched.h>
76 #include <sys/signalvar.h>
77 #include <sys/syscallsubr.h>
78 #include <sys/sysctl.h>
79 #include <sys/sysent.h>
80 #include <sys/sysproto.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
90 #include <machine/armreg.h>
91 #include <machine/atags.h>
92 #include <machine/cpu.h>
93 #include <machine/devmap.h>
94 #include <machine/frame.h>
95 #include <machine/intr.h>
96 #include <machine/machdep.h>
97 #include <machine/md_var.h>
98 #include <machine/metadata.h>
99 #include <machine/pcb.h>
100 #include <machine/reg.h>
101 #include <machine/trap.h>
102 #include <machine/undefined.h>
103 #include <machine/vmparam.h>
104 #include <machine/sysarch.h>
107 #include <dev/fdt/fdt_common.h>
108 #include <dev/ofw/openfirm.h>
112 #define debugf(fmt, args...) printf(fmt, ##args)
114 #define debugf(fmt, args...)
117 struct pcpu __pcpu[MAXCPU];
118 struct pcpu *pcpup = &__pcpu[0];
120 static struct trapframe proc0_tf;
121 uint32_t cpu_reset_address = 0;
123 vm_offset_t vector_page;
127 int (*_arm_memcpy)(void *, void *, int, int) = NULL;
128 int (*_arm_bzero)(void *, int, int) = NULL;
129 int _min_memcpy_size = 0;
130 int _min_bzero_size = 0;
134 extern vm_offset_t ksym_start, ksym_end;
139 * This is the number of L2 page tables required for covering max
140 * (hypothetical) memsize of 4GB and all kernel mappings (vectors, msgbuf,
141 * stacks etc.), uprounded to be divisible by 4.
143 #define KERNEL_PT_MAX 78
145 static struct pv_addr kernel_pt_table[KERNEL_PT_MAX];
147 vm_paddr_t phys_avail[10];
148 vm_paddr_t dump_avail[4];
150 extern u_int data_abort_handler_address;
151 extern u_int prefetch_abort_handler_address;
152 extern u_int undefined_handler_address;
156 struct pv_addr systempage;
157 static struct pv_addr msgbufpv;
158 struct pv_addr irqstack;
159 struct pv_addr undstack;
160 struct pv_addr abtstack;
161 static struct pv_addr kernelstack;
165 #if defined(LINUX_BOOT_ABI)
166 #define LBABI_MAX_BANKS 10
169 struct arm_lbabi_tag *atag_list;
170 char linux_command_line[LBABI_MAX_COMMAND_LINE + 1];
171 char atags[LBABI_MAX_COMMAND_LINE * 2];
172 uint32_t memstart[LBABI_MAX_BANKS];
173 uint32_t memsize[LBABI_MAX_BANKS];
177 static uint32_t board_revision;
178 /* hex representation of uint64_t */
179 static char board_serial[32];
181 SYSCTL_NODE(_hw, OID_AUTO, board, CTLFLAG_RD, 0, "Board attributes");
182 SYSCTL_UINT(_hw_board, OID_AUTO, revision, CTLFLAG_RD,
183 &board_revision, 0, "Board revision");
184 SYSCTL_STRING(_hw_board, OID_AUTO, serial, CTLFLAG_RD,
185 board_serial, 0, "Board serial");
188 SYSCTL_INT(_hw, HW_FLOATINGPT, floatingpoint, CTLFLAG_RD,
189 &vfp_exists, 0, "Floating point support enabled");
192 board_set_serial(uint64_t serial)
195 snprintf(board_serial, sizeof(board_serial)-1,
200 board_set_revision(uint32_t revision)
203 board_revision = revision;
207 sendsig(catcher, ksi, mask)
214 struct trapframe *tf;
215 struct sigframe *fp, frame;
223 PROC_LOCK_ASSERT(p, MA_OWNED);
224 sig = ksi->ksi_signo;
225 code = ksi->ksi_code;
227 mtx_assert(&psp->ps_mtx, MA_OWNED);
229 onstack = sigonstack(tf->tf_usr_sp);
231 CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm,
234 /* Allocate and validate space for the signal handler context. */
235 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !(onstack) &&
236 SIGISMEMBER(psp->ps_sigonstack, sig)) {
237 fp = (struct sigframe *)(td->td_sigstk.ss_sp +
238 td->td_sigstk.ss_size);
239 #if defined(COMPAT_43)
240 td->td_sigstk.ss_flags |= SS_ONSTACK;
243 fp = (struct sigframe *)td->td_frame->tf_usr_sp;
245 /* make room on the stack */
248 /* make the stack aligned */
249 fp = (struct sigframe *)STACKALIGN(fp);
250 /* Populate the siginfo frame. */
251 get_mcontext(td, &frame.sf_uc.uc_mcontext, 0);
252 frame.sf_si = ksi->ksi_info;
253 frame.sf_uc.uc_sigmask = *mask;
254 frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK )
255 ? ((onstack) ? SS_ONSTACK : 0) : SS_DISABLE;
256 frame.sf_uc.uc_stack = td->td_sigstk;
257 mtx_unlock(&psp->ps_mtx);
258 PROC_UNLOCK(td->td_proc);
260 /* Copy the sigframe out to the user's stack. */
261 if (copyout(&frame, fp, sizeof(*fp)) != 0) {
262 /* Process has trashed its stack. Kill it. */
263 CTR2(KTR_SIG, "sendsig: sigexit td=%p fp=%p", td, fp);
268 /* Translate the signal if appropriate. */
269 if (p->p_sysent->sv_sigtbl && sig <= p->p_sysent->sv_sigsize)
270 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
273 * Build context to run handler in. We invoke the handler
274 * directly, only returning via the trampoline. Note the
275 * trampoline version numbers are coordinated with machine-
276 * dependent code in libc.
280 tf->tf_r1 = (register_t)&fp->sf_si;
281 tf->tf_r2 = (register_t)&fp->sf_uc;
283 /* the trampoline uses r5 as the uc address */
284 tf->tf_r5 = (register_t)&fp->sf_uc;
285 tf->tf_pc = (register_t)catcher;
286 tf->tf_usr_sp = (register_t)fp;
287 tf->tf_usr_lr = (register_t)(PS_STRINGS - *(p->p_sysent->sv_szsigcode));
289 CTR3(KTR_SIG, "sendsig: return td=%p pc=%#x sp=%#x", td, tf->tf_usr_lr,
293 mtx_lock(&psp->ps_mtx);
296 struct kva_md_info kmi;
301 * Initialize the vector page, and select whether or not to
302 * relocate the vectors.
304 * NOTE: We expect the vector page to be mapped at its expected
308 extern unsigned int page0[], page0_data[];
310 arm_vector_init(vm_offset_t va, int which)
312 unsigned int *vectors = (int *) va;
313 unsigned int *vectors_data = vectors + (page0_data - page0);
317 * Loop through the vectors we're taking over, and copy the
318 * vector's insn and data word.
320 for (vec = 0; vec < ARM_NVEC; vec++) {
321 if ((which & (1 << vec)) == 0) {
322 /* Don't want to take over this vector. */
325 vectors[vec] = page0[vec];
326 vectors_data[vec] = page0_data[vec];
329 /* Now sync the vectors. */
330 cpu_icache_sync_range(va, (ARM_NVEC * 2) * sizeof(u_int));
334 if (va == ARM_VECTORS_HIGH) {
336 * Assume the MD caller knows what it's doing here, and
337 * really does want the vector page relocated.
339 * Note: This has to be done here (and not just in
340 * cpu_setup()) because the vector page needs to be
341 * accessible *before* cpu_startup() is called.
344 * NOTE: If the CPU control register is not readable,
345 * this will totally fail! We'll just assume that
346 * any system that has high vector support has a
347 * readable CPU control register, for now. If we
348 * ever encounter one that does not, we'll have to
351 cpu_control(CPU_CONTROL_VECRELOC, CPU_CONTROL_VECRELOC);
356 cpu_startup(void *dummy)
358 struct pcb *pcb = thread0.td_pcb;
359 #ifdef ARM_TP_ADDRESS
360 #ifndef ARM_CACHE_LOCK_ENABLE
367 printf("real memory = %ju (%ju MB)\n", (uintmax_t)ptoa(physmem),
368 (uintmax_t)ptoa(physmem) / 1048576);
372 * Display the RAM layout.
377 printf("Physical memory chunk(s):\n");
378 for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) {
381 size = phys_avail[indx + 1] - phys_avail[indx];
382 printf(" 0x%08jx - 0x%08jx, %ju KBytes (%ju pages)\n",
383 (uintmax_t)phys_avail[indx],
384 (uintmax_t)phys_avail[indx + 1] - 1,
385 (uintmax_t)size / 1024, (uintmax_t)size / PAGE_SIZE);
389 vm_ksubmap_init(&kmi);
391 printf("avail memory = %ju (%ju MB)\n",
392 (uintmax_t)ptoa(cnt.v_free_count),
393 (uintmax_t)ptoa(cnt.v_free_count) / 1048576);
396 arm_devmap_print_table();
399 vm_pager_bufferinit();
400 pcb->un_32.pcb32_und_sp = (u_int)thread0.td_kstack +
401 USPACE_UNDEF_STACK_TOP;
402 pcb->un_32.pcb32_sp = (u_int)thread0.td_kstack +
403 USPACE_SVC_STACK_TOP;
404 vector_page_setprot(VM_PROT_READ);
405 pmap_set_pcb_pagedir(pmap_kernel(), pcb);
407 #ifdef ARM_TP_ADDRESS
408 #ifdef ARM_CACHE_LOCK_ENABLE
409 pmap_kenter_user(ARM_TP_ADDRESS, ARM_TP_ADDRESS);
410 arm_lock_cache_line(ARM_TP_ADDRESS);
412 m = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ | VM_ALLOC_ZERO);
413 pmap_kenter_user(ARM_TP_ADDRESS, VM_PAGE_TO_PHYS(m));
415 *(uint32_t *)ARM_RAS_START = 0;
416 *(uint32_t *)ARM_RAS_END = 0xffffffff;
420 SYSINIT(cpu, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
423 * Flush the D-cache for non-DMA I/O so that the I-cache can
424 * be made coherent later.
427 cpu_flush_dcache(void *ptr, size_t len)
430 cpu_dcache_wb_range((uintptr_t)ptr, len);
431 cpu_l2cache_wb_range((uintptr_t)ptr, len);
434 /* Get current clock frequency for the given cpu id. */
436 cpu_est_clockrate(int cpu_id, uint64_t *rate)
446 CTR2(KTR_SPARE2, "cpu_idle(%d) at %d",
448 #ifndef NO_EVENTTIMERS
454 if (!sched_runnable())
456 #ifndef NO_EVENTTIMERS
462 CTR2(KTR_SPARE2, "cpu_idle(%d) at %d done",
467 cpu_idle_wakeup(int cpu)
474 fill_regs(struct thread *td, struct reg *regs)
476 struct trapframe *tf = td->td_frame;
477 bcopy(&tf->tf_r0, regs->r, sizeof(regs->r));
478 regs->r_sp = tf->tf_usr_sp;
479 regs->r_lr = tf->tf_usr_lr;
480 regs->r_pc = tf->tf_pc;
481 regs->r_cpsr = tf->tf_spsr;
485 fill_fpregs(struct thread *td, struct fpreg *regs)
487 bzero(regs, sizeof(*regs));
492 set_regs(struct thread *td, struct reg *regs)
494 struct trapframe *tf = td->td_frame;
496 bcopy(regs->r, &tf->tf_r0, sizeof(regs->r));
497 tf->tf_usr_sp = regs->r_sp;
498 tf->tf_usr_lr = regs->r_lr;
499 tf->tf_pc = regs->r_pc;
500 tf->tf_spsr &= ~PSR_FLAGS;
501 tf->tf_spsr |= regs->r_cpsr & PSR_FLAGS;
506 set_fpregs(struct thread *td, struct fpreg *regs)
512 fill_dbregs(struct thread *td, struct dbreg *regs)
517 set_dbregs(struct thread *td, struct dbreg *regs)
524 ptrace_read_int(struct thread *td, vm_offset_t addr, u_int32_t *v)
529 PROC_LOCK_ASSERT(td->td_proc, MA_NOTOWNED);
530 iov.iov_base = (caddr_t) v;
531 iov.iov_len = sizeof(u_int32_t);
534 uio.uio_offset = (off_t)addr;
535 uio.uio_resid = sizeof(u_int32_t);
536 uio.uio_segflg = UIO_SYSSPACE;
537 uio.uio_rw = UIO_READ;
539 return proc_rwmem(td->td_proc, &uio);
543 ptrace_write_int(struct thread *td, vm_offset_t addr, u_int32_t v)
548 PROC_LOCK_ASSERT(td->td_proc, MA_NOTOWNED);
549 iov.iov_base = (caddr_t) &v;
550 iov.iov_len = sizeof(u_int32_t);
553 uio.uio_offset = (off_t)addr;
554 uio.uio_resid = sizeof(u_int32_t);
555 uio.uio_segflg = UIO_SYSSPACE;
556 uio.uio_rw = UIO_WRITE;
558 return proc_rwmem(td->td_proc, &uio);
562 ptrace_single_step(struct thread *td)
567 KASSERT(td->td_md.md_ptrace_instr == 0,
568 ("Didn't clear single step"));
571 error = ptrace_read_int(td, td->td_frame->tf_pc + 4,
572 &td->td_md.md_ptrace_instr);
575 error = ptrace_write_int(td, td->td_frame->tf_pc + 4,
578 td->td_md.md_ptrace_instr = 0;
579 td->td_md.md_ptrace_addr = td->td_frame->tf_pc + 4;
586 ptrace_clear_single_step(struct thread *td)
590 if (td->td_md.md_ptrace_instr) {
593 ptrace_write_int(td, td->td_md.md_ptrace_addr,
594 td->td_md.md_ptrace_instr);
596 td->td_md.md_ptrace_instr = 0;
602 ptrace_set_pc(struct thread *td, unsigned long addr)
604 td->td_frame->tf_pc = addr;
609 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
620 if (td->td_md.md_spinlock_count == 0) {
621 cspr = disable_interrupts(I32_bit | F32_bit);
622 td->td_md.md_spinlock_count = 1;
623 td->td_md.md_saved_cspr = cspr;
625 td->td_md.md_spinlock_count++;
637 cspr = td->td_md.md_saved_cspr;
638 td->td_md.md_spinlock_count--;
639 if (td->td_md.md_spinlock_count == 0)
640 restore_interrupts(cspr);
644 * Clear registers on exec
647 exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
649 struct trapframe *tf = td->td_frame;
651 memset(tf, 0, sizeof(*tf));
652 tf->tf_usr_sp = stack;
653 tf->tf_usr_lr = imgp->entry_addr;
654 tf->tf_svc_lr = 0x77777777;
655 tf->tf_pc = imgp->entry_addr;
656 tf->tf_spsr = PSR_USR32_MODE;
660 * Get machine context.
663 get_mcontext(struct thread *td, mcontext_t *mcp, int clear_ret)
665 struct trapframe *tf = td->td_frame;
666 __greg_t *gr = mcp->__gregs;
668 if (clear_ret & GET_MC_CLEAR_RET)
671 gr[_REG_R0] = tf->tf_r0;
672 gr[_REG_R1] = tf->tf_r1;
673 gr[_REG_R2] = tf->tf_r2;
674 gr[_REG_R3] = tf->tf_r3;
675 gr[_REG_R4] = tf->tf_r4;
676 gr[_REG_R5] = tf->tf_r5;
677 gr[_REG_R6] = tf->tf_r6;
678 gr[_REG_R7] = tf->tf_r7;
679 gr[_REG_R8] = tf->tf_r8;
680 gr[_REG_R9] = tf->tf_r9;
681 gr[_REG_R10] = tf->tf_r10;
682 gr[_REG_R11] = tf->tf_r11;
683 gr[_REG_R12] = tf->tf_r12;
684 gr[_REG_SP] = tf->tf_usr_sp;
685 gr[_REG_LR] = tf->tf_usr_lr;
686 gr[_REG_PC] = tf->tf_pc;
687 gr[_REG_CPSR] = tf->tf_spsr;
693 * Set machine context.
695 * However, we don't set any but the user modifiable flags, and we won't
696 * touch the cs selector.
699 set_mcontext(struct thread *td, const mcontext_t *mcp)
701 struct trapframe *tf = td->td_frame;
702 const __greg_t *gr = mcp->__gregs;
704 tf->tf_r0 = gr[_REG_R0];
705 tf->tf_r1 = gr[_REG_R1];
706 tf->tf_r2 = gr[_REG_R2];
707 tf->tf_r3 = gr[_REG_R3];
708 tf->tf_r4 = gr[_REG_R4];
709 tf->tf_r5 = gr[_REG_R5];
710 tf->tf_r6 = gr[_REG_R6];
711 tf->tf_r7 = gr[_REG_R7];
712 tf->tf_r8 = gr[_REG_R8];
713 tf->tf_r9 = gr[_REG_R9];
714 tf->tf_r10 = gr[_REG_R10];
715 tf->tf_r11 = gr[_REG_R11];
716 tf->tf_r12 = gr[_REG_R12];
717 tf->tf_usr_sp = gr[_REG_SP];
718 tf->tf_usr_lr = gr[_REG_LR];
719 tf->tf_pc = gr[_REG_PC];
720 tf->tf_spsr = gr[_REG_CPSR];
729 sys_sigreturn(td, uap)
731 struct sigreturn_args /* {
732 const struct __ucontext *sigcntxp;
736 struct trapframe *tf;
741 if (copyin(uap->sigcntxp, &sf, sizeof(sf)))
744 * Make sure the processor mode has not been tampered with and
745 * interrupts have not been disabled.
747 spsr = sf.sf_uc.uc_mcontext.__gregs[_REG_CPSR];
748 if ((spsr & PSR_MODE) != PSR_USR32_MODE ||
749 (spsr & (I32_bit | F32_bit)) != 0)
751 /* Restore register context. */
753 set_mcontext(td, &sf.sf_uc.uc_mcontext);
755 /* Restore signal mask. */
756 kern_sigprocmask(td, SIG_SETMASK, &sf.sf_uc.uc_sigmask, NULL, 0);
758 return (EJUSTRETURN);
763 * Construct a PCB from a trapframe. This is called from kdb_trap() where
764 * we want to start a backtrace from the function that caused us to enter
765 * the debugger. We have the context in the trapframe, but base the trace
766 * on the PCB. The PCB doesn't have to be perfect, as long as it contains
767 * enough for a backtrace.
770 makectx(struct trapframe *tf, struct pcb *pcb)
772 pcb->un_32.pcb32_r8 = tf->tf_r8;
773 pcb->un_32.pcb32_r9 = tf->tf_r9;
774 pcb->un_32.pcb32_r10 = tf->tf_r10;
775 pcb->un_32.pcb32_r11 = tf->tf_r11;
776 pcb->un_32.pcb32_r12 = tf->tf_r12;
777 pcb->un_32.pcb32_pc = tf->tf_pc;
778 pcb->un_32.pcb32_lr = tf->tf_usr_lr;
779 pcb->un_32.pcb32_sp = tf->tf_usr_sp;
783 * Make a standard dump_avail array. Can't make the phys_avail
784 * since we need to do that after we call pmap_bootstrap, but this
785 * is needed before pmap_boostrap.
787 * ARM_USE_SMALL_ALLOC uses dump_avail, so it must be filled before
788 * calling pmap_bootstrap.
791 arm_dump_avail_init(vm_offset_t ramsize, size_t max)
793 #ifdef LINUX_BOOT_ABI
795 * Linux boot loader passes us the actual banks of memory, so use them
796 * to construct the dump_avail array.
802 if (max < (membanks + 1) * 2)
803 panic("dump_avail[%d] too small for %d banks\n",
805 for (j = 0, i = 0; i < membanks; i++) {
806 dump_avail[j++] = round_page(memstart[i]);
807 dump_avail[j++] = trunc_page(memstart[i] + memsize[i]);
815 panic("dump_avail too small\n");
817 dump_avail[0] = round_page(PHYSADDR);
818 dump_avail[1] = trunc_page(PHYSADDR + ramsize);
824 * Fake up a boot descriptor table
827 fake_preload_metadata(struct arm_boot_params *abp __unused)
830 vm_offset_t zstart = 0, zend = 0;
832 vm_offset_t lastaddr;
834 static uint32_t fake_preload[35];
836 fake_preload[i++] = MODINFO_NAME;
837 fake_preload[i++] = strlen("kernel") + 1;
838 strcpy((char*)&fake_preload[i++], "kernel");
840 fake_preload[i++] = MODINFO_TYPE;
841 fake_preload[i++] = strlen("elf kernel") + 1;
842 strcpy((char*)&fake_preload[i++], "elf kernel");
844 fake_preload[i++] = MODINFO_ADDR;
845 fake_preload[i++] = sizeof(vm_offset_t);
846 fake_preload[i++] = KERNVIRTADDR;
847 fake_preload[i++] = MODINFO_SIZE;
848 fake_preload[i++] = sizeof(uint32_t);
849 fake_preload[i++] = (uint32_t)&end - KERNVIRTADDR;
851 if (*(uint32_t *)KERNVIRTADDR == MAGIC_TRAMP_NUMBER) {
852 fake_preload[i++] = MODINFO_METADATA|MODINFOMD_SSYM;
853 fake_preload[i++] = sizeof(vm_offset_t);
854 fake_preload[i++] = *(uint32_t *)(KERNVIRTADDR + 4);
855 fake_preload[i++] = MODINFO_METADATA|MODINFOMD_ESYM;
856 fake_preload[i++] = sizeof(vm_offset_t);
857 fake_preload[i++] = *(uint32_t *)(KERNVIRTADDR + 8);
858 lastaddr = *(uint32_t *)(KERNVIRTADDR + 8);
860 zstart = *(uint32_t *)(KERNVIRTADDR + 4);
865 lastaddr = (vm_offset_t)&end;
866 fake_preload[i++] = 0;
868 preload_metadata = (void *)fake_preload;
876 #if ARM_ARCH_6 || ARM_ARCH_7A || defined(CPU_MV_PJ4B)
879 pcpu_init(pcpup, 0, sizeof(struct pcpu));
880 PCPU_SET(curthread, &thread0);
886 #if defined(LINUX_BOOT_ABI)
888 linux_parse_boot_param(struct arm_boot_params *abp)
890 struct arm_lbabi_tag *walker;
895 * Linux boot ABI: r0 = 0, r1 is the board type (!= 0) and r2
896 * is atags or dtb pointer. If all of these aren't satisfied,
899 if (!(abp->abp_r0 == 0 && abp->abp_r1 != 0 && abp->abp_r2 != 0))
902 board_id = abp->abp_r1;
903 walker = (struct arm_lbabi_tag *)
904 (abp->abp_r2 + KERNVIRTADDR - KERNPHYSADDR);
906 /* xxx - Need to also look for binary device tree */
907 if (ATAG_TAG(walker) != ATAG_CORE)
911 while (ATAG_TAG(walker) != ATAG_NONE) {
912 switch (ATAG_TAG(walker)) {
916 if (membanks < LBABI_MAX_BANKS) {
917 memstart[membanks] = walker->u.tag_mem.start;
918 memsize[membanks] = walker->u.tag_mem.size;
925 serial = walker->u.tag_sn.low |
926 ((uint64_t)walker->u.tag_sn.high << 32);
927 board_set_serial(serial);
930 revision = walker->u.tag_rev.rev;
931 board_set_revision(revision);
934 /* XXX open question: Parse this for boothowto? */
935 bcopy(walker->u.tag_cmd.command, linux_command_line,
941 walker = ATAG_NEXT(walker);
944 /* Save a copy for later */
945 bcopy(atag_list, atags,
946 (char *)walker - (char *)atag_list + ATAG_SIZE(walker));
948 return fake_preload_metadata(abp);
952 #if defined(FREEBSD_BOOT_LOADER)
954 freebsd_parse_boot_param(struct arm_boot_params *abp)
956 vm_offset_t lastaddr = 0;
961 * Mask metadata pointer: it is supposed to be on page boundary. If
962 * the first argument (mdp) doesn't point to a valid address the
963 * bootloader must have passed us something else than the metadata
964 * ptr, so we give up. Also give up if we cannot find metadta section
965 * the loader creates that we get all this data out of.
968 if ((mdp = (void *)(abp->abp_r0 & ~PAGE_MASK)) == NULL)
970 preload_metadata = mdp;
971 kmdp = preload_search_by_type("elf kernel");
975 boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int);
976 kern_envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *);
977 lastaddr = MD_FETCH(kmdp, MODINFOMD_KERNEND, vm_offset_t);
979 ksym_start = MD_FETCH(kmdp, MODINFOMD_SSYM, uintptr_t);
980 ksym_end = MD_FETCH(kmdp, MODINFOMD_ESYM, uintptr_t);
982 preload_addr_relocate = KERNVIRTADDR - KERNPHYSADDR;
988 default_parse_boot_param(struct arm_boot_params *abp)
990 vm_offset_t lastaddr;
992 #if defined(LINUX_BOOT_ABI)
993 if ((lastaddr = linux_parse_boot_param(abp)) != 0)
996 #if defined(FREEBSD_BOOT_LOADER)
997 if ((lastaddr = freebsd_parse_boot_param(abp)) != 0)
1000 /* Fall back to hardcoded metadata. */
1001 lastaddr = fake_preload_metadata(abp);
1007 * Stub version of the boot parameter parsing routine. We are
1008 * called early in initarm, before even VM has been initialized.
1009 * This routine needs to preserve any data that the boot loader
1010 * has passed in before the kernel starts to grow past the end
1011 * of the BSS, traditionally the place boot-loaders put this data.
1013 * Since this is called so early, things that depend on the vm system
1014 * being setup (including access to some SoC's serial ports), about
1015 * all that can be done in this routine is to copy the arguments.
1017 * This is the default boot parameter parsing routine. Individual
1018 * kernels/boards can override this weak function with one of their
1019 * own. We just fake metadata...
1021 __weak_reference(default_parse_boot_param, parse_boot_param);
1027 init_proc0(vm_offset_t kstack)
1029 proc_linkup0(&proc0, &thread0);
1030 thread0.td_kstack = kstack;
1031 thread0.td_pcb = (struct pcb *)
1032 (thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
1033 thread0.td_pcb->pcb_flags = 0;
1034 thread0.td_frame = &proc0_tf;
1035 pcpup->pc_curpcb = thread0.td_pcb;
1039 set_stackptrs(int cpu)
1042 set_stackptr(PSR_IRQ32_MODE,
1043 irqstack.pv_va + ((IRQ_STACK_SIZE * PAGE_SIZE) * (cpu + 1)));
1044 set_stackptr(PSR_ABT32_MODE,
1045 abtstack.pv_va + ((ABT_STACK_SIZE * PAGE_SIZE) * (cpu + 1)));
1046 set_stackptr(PSR_UND32_MODE,
1047 undstack.pv_va + ((UND_STACK_SIZE * PAGE_SIZE) * (cpu + 1)));
1071 debugf("loader passed (static) kenv:\n");
1072 if (kern_envp == NULL) {
1073 debugf(" no env, null ptr\n");
1076 debugf(" kern_envp = 0x%08x\n", (uint32_t)kern_envp);
1079 for (cp = kern_envp; cp != NULL; cp = kenv_next(cp))
1080 debugf(" %x %s\n", (uint32_t)cp, cp);
1084 physmap_init(struct mem_region *availmem_regions, int availmem_regions_sz)
1087 vm_offset_t phys_kernelend, kernload;
1089 struct mem_region *mp, *mp1;
1091 phys_kernelend = KERNPHYSADDR + (virtual_avail - KERNVIRTADDR);
1092 kernload = KERNPHYSADDR;
1095 * Remove kernel physical address range from avail
1096 * regions list. Page align all regions.
1097 * Non-page aligned memory isn't very interesting to us.
1098 * Also, sort the entries for ascending addresses.
1101 cnt = availmem_regions_sz;
1102 debugf("processing avail regions:\n");
1103 for (mp = availmem_regions; mp->mr_size; mp++) {
1105 e = mp->mr_start + mp->mr_size;
1106 debugf(" %08x-%08x -> ", s, e);
1107 /* Check whether this region holds all of the kernel. */
1108 if (s < kernload && e > phys_kernelend) {
1109 availmem_regions[cnt].mr_start = phys_kernelend;
1110 availmem_regions[cnt++].mr_size = e - phys_kernelend;
1113 /* Look whether this regions starts within the kernel. */
1114 if (s >= kernload && s < phys_kernelend) {
1115 if (e <= phys_kernelend)
1119 /* Now look whether this region ends within the kernel. */
1120 if (e > kernload && e <= phys_kernelend) {
1121 if (s >= kernload) {
1126 /* Now page align the start and size of the region. */
1132 debugf("%08x-%08x = %x\n", s, e, sz);
1134 /* Check whether some memory is left here. */
1137 printf("skipping\n");
1139 (cnt - (mp - availmem_regions)) * sizeof(*mp));
1145 /* Do an insertion sort. */
1146 for (mp1 = availmem_regions; mp1 < mp; mp1++)
1147 if (s < mp1->mr_start)
1150 bcopy(mp1, mp1 + 1, (char *)mp - (char *)mp1);
1158 availmem_regions_sz = cnt;
1160 /* Fill in phys_avail table, based on availmem_regions */
1161 debugf("fill in phys_avail:\n");
1162 for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1164 debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1165 availmem_regions[i].mr_start,
1166 availmem_regions[i].mr_start + availmem_regions[i].mr_size,
1167 availmem_regions[i].mr_size);
1170 * We should not map the page at PA 0x0000000, the VM can't
1171 * handle it, as pmap_extract() == 0 means failure.
1173 if (availmem_regions[i].mr_start > 0 ||
1174 availmem_regions[i].mr_size > PAGE_SIZE) {
1175 phys_avail[j] = availmem_regions[i].mr_start;
1176 if (phys_avail[j] == 0)
1177 phys_avail[j] += PAGE_SIZE;
1178 phys_avail[j + 1] = availmem_regions[i].mr_start +
1179 availmem_regions[i].mr_size;
1184 phys_avail[j + 1] = 0;
1188 initarm(struct arm_boot_params *abp)
1190 struct mem_region memory_regions[FDT_MEM_REGIONS];
1191 struct mem_region availmem_regions[FDT_MEM_REGIONS];
1192 struct mem_region reserved_regions[FDT_MEM_REGIONS];
1193 struct pv_addr kernel_l1pt;
1194 struct pv_addr dpcpu;
1195 vm_offset_t dtbp, freemempos, l2_start, lastaddr;
1196 uint32_t memsize, l2size;
1200 int i = 0, j = 0, err_devmap = 0;
1201 int memory_regions_sz;
1202 int availmem_regions_sz;
1203 int reserved_regions_sz;
1204 vm_offset_t start, end;
1205 vm_offset_t rstart, rend;
1208 lastaddr = parse_boot_param(abp);
1213 * Find the dtb passed in by the boot loader.
1215 kmdp = preload_search_by_type("elf kernel");
1217 dtbp = MD_FETCH(kmdp, MODINFOMD_DTBP, vm_offset_t);
1219 dtbp = (vm_offset_t)NULL;
1221 #if defined(FDT_DTB_STATIC)
1223 * In case the device tree blob was not retrieved (from metadata) try
1224 * to use the statically embedded one.
1226 if (dtbp == (vm_offset_t)NULL)
1227 dtbp = (vm_offset_t)&fdt_static_dtb;
1230 if (OF_install(OFW_FDT, 0) == FALSE)
1233 if (OF_init((void *)dtbp) != 0)
1236 /* Grab physical memory regions information from device tree. */
1237 if (fdt_get_mem_regions(memory_regions, &memory_regions_sz,
1241 /* Grab physical memory regions information from device tree. */
1242 if (fdt_get_reserved_regions(reserved_regions, &reserved_regions_sz) != 0)
1243 reserved_regions_sz = 0;
1246 * Now exclude all the reserved regions
1249 for (i = 0; i < memory_regions_sz; i++) {
1250 start = memory_regions[i].mr_start;
1251 end = start + memory_regions[i].mr_size;
1252 for (j = 0; j < reserved_regions_sz; j++) {
1253 rstart = reserved_regions[j].mr_start;
1254 rend = rstart + reserved_regions[j].mr_size;
1256 * Restricted region is before available
1257 * Skip restricted region
1262 * Restricted region is behind available
1263 * No further processing required
1268 * Restricted region includes memory region
1269 * skip available region
1271 if ((start >= rstart) && (rend >= end)) {
1277 * Memory region includes restricted region
1279 if ((rstart > start) && (end > rend)) {
1280 availmem_regions[curr].mr_start = start;
1281 availmem_regions[curr++].mr_size = rstart - start;
1286 * Memory region partially overlaps with restricted
1288 if ((rstart >= start) && (rstart <= end)) {
1291 else if ((rend >= start) && (rend <= end)) {
1297 availmem_regions[curr].mr_start = start;
1298 availmem_regions[curr++].mr_size = end - start;
1302 availmem_regions_sz = curr;
1304 /* Platform-specific initialisation */
1305 initarm_early_init();
1309 /* Do basic tuning, hz etc */
1312 /* Calculate number of L2 tables needed for mapping vm_page_array */
1313 l2size = (memsize / PAGE_SIZE) * sizeof(struct vm_page);
1314 l2size = (l2size >> L1_S_SHIFT) + 1;
1317 * Add one table for end of kernel map, one for stacks, msgbuf and
1318 * L1 and L2 tables map and one for vectors map.
1322 /* Make it divisible by 4 */
1323 l2size = (l2size + 3) & ~3;
1325 freemempos = (lastaddr + PAGE_MASK) & ~PAGE_MASK;
1327 /* Define a macro to simplify memory allocation */
1328 #define valloc_pages(var, np) \
1329 alloc_pages((var).pv_va, (np)); \
1330 (var).pv_pa = (var).pv_va + (KERNPHYSADDR - KERNVIRTADDR);
1332 #define alloc_pages(var, np) \
1333 (var) = freemempos; \
1334 freemempos += (np * PAGE_SIZE); \
1335 memset((char *)(var), 0, ((np) * PAGE_SIZE));
1337 while (((freemempos - L1_TABLE_SIZE) & (L1_TABLE_SIZE - 1)) != 0)
1338 freemempos += PAGE_SIZE;
1339 valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE);
1341 for (i = 0; i < l2size; ++i) {
1342 if (!(i % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) {
1343 valloc_pages(kernel_pt_table[i],
1344 L2_TABLE_SIZE / PAGE_SIZE);
1347 kernel_pt_table[i].pv_va = kernel_pt_table[j].pv_va +
1348 L2_TABLE_SIZE_REAL * (i - j);
1349 kernel_pt_table[i].pv_pa =
1350 kernel_pt_table[i].pv_va - KERNVIRTADDR +
1356 * Allocate a page for the system page mapped to 0x00000000
1357 * or 0xffff0000. This page will just contain the system vectors
1358 * and can be shared by all processes.
1360 valloc_pages(systempage, 1);
1362 /* Allocate dynamic per-cpu area. */
1363 valloc_pages(dpcpu, DPCPU_SIZE / PAGE_SIZE);
1364 dpcpu_init((void *)dpcpu.pv_va, 0);
1366 /* Allocate stacks for all modes */
1367 valloc_pages(irqstack, IRQ_STACK_SIZE * MAXCPU);
1368 valloc_pages(abtstack, ABT_STACK_SIZE * MAXCPU);
1369 valloc_pages(undstack, UND_STACK_SIZE * MAXCPU);
1370 valloc_pages(kernelstack, KSTACK_PAGES * MAXCPU);
1371 valloc_pages(msgbufpv, round_page(msgbufsize) / PAGE_SIZE);
1374 * Now we start construction of the L1 page table
1375 * We start by mapping the L2 page tables into the L1.
1376 * This means that we can replace L1 mappings later on if necessary
1378 l1pagetable = kernel_l1pt.pv_va;
1381 * Try to map as much as possible of kernel text and data using
1382 * 1MB section mapping and for the rest of initial kernel address
1383 * space use L2 coarse tables.
1385 * Link L2 tables for mapping remainder of kernel (modulo 1MB)
1386 * and kernel structures
1388 l2_start = lastaddr & ~(L1_S_OFFSET);
1389 for (i = 0 ; i < l2size - 1; i++)
1390 pmap_link_l2pt(l1pagetable, l2_start + i * L1_S_SIZE,
1391 &kernel_pt_table[i]);
1393 pmap_curmaxkvaddr = l2_start + (l2size - 1) * L1_S_SIZE;
1395 /* Map kernel code and data */
1396 pmap_map_chunk(l1pagetable, KERNVIRTADDR, KERNPHYSADDR,
1397 (((uint32_t)(lastaddr) - KERNVIRTADDR) + PAGE_MASK) & ~PAGE_MASK,
1398 VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
1401 /* Map L1 directory and allocated L2 page tables */
1402 pmap_map_chunk(l1pagetable, kernel_l1pt.pv_va, kernel_l1pt.pv_pa,
1403 L1_TABLE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
1405 pmap_map_chunk(l1pagetable, kernel_pt_table[0].pv_va,
1406 kernel_pt_table[0].pv_pa,
1407 L2_TABLE_SIZE_REAL * l2size,
1408 VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
1410 /* Map allocated DPCPU, stacks and msgbuf */
1411 pmap_map_chunk(l1pagetable, dpcpu.pv_va, dpcpu.pv_pa,
1412 freemempos - dpcpu.pv_va,
1413 VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
1415 /* Link and map the vector page */
1416 pmap_link_l2pt(l1pagetable, ARM_VECTORS_HIGH,
1417 &kernel_pt_table[l2size - 1]);
1418 pmap_map_entry(l1pagetable, ARM_VECTORS_HIGH, systempage.pv_pa,
1419 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE, PTE_CACHE);
1421 /* Establish static device mappings. */
1422 err_devmap = initarm_devmap_init();
1423 arm_devmap_bootstrap(l1pagetable, NULL);
1424 vm_max_kernel_address = initarm_lastaddr();
1426 cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL * 2)) | DOMAIN_CLIENT);
1427 pmap_pa = kernel_l1pt.pv_pa;
1428 setttb(kernel_l1pt.pv_pa);
1430 cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL * 2));
1433 * Now that proper page tables are installed, call cpu_setup() to enable
1434 * instruction and data caches and other chip-specific features.
1439 * Only after the SOC registers block is mapped we can perform device
1440 * tree fixups, as they may attempt to read parameters from hardware.
1442 OF_interpret("perform-fixup", 0);
1444 initarm_gpio_init();
1448 physmem = memsize / PAGE_SIZE;
1450 debugf("initarm: console initialized\n");
1451 debugf(" arg1 kmdp = 0x%08x\n", (uint32_t)kmdp);
1452 debugf(" boothowto = 0x%08x\n", boothowto);
1453 debugf(" dtbp = 0x%08x\n", (uint32_t)dtbp);
1456 env = getenv("kernelname");
1458 strlcpy(kernelname, env, sizeof(kernelname));
1460 if (err_devmap != 0)
1461 printf("WARNING: could not fully configure devmap, error=%d\n",
1464 initarm_late_init();
1467 * Pages were allocated during the secondary bootstrap for the
1468 * stacks for different CPU modes.
1469 * We must now set the r13 registers in the different CPU modes to
1470 * point to these stacks.
1471 * Since the ARM stacks use STMFD etc. we must set r13 to the top end
1472 * of the stack memory.
1474 cpu_control(CPU_CONTROL_MMU_ENABLE, CPU_CONTROL_MMU_ENABLE);
1479 * We must now clean the cache again....
1480 * Cleaning may be done by reading new data to displace any
1481 * dirty data in the cache. This will have happened in setttb()
1482 * but since we are boot strapping the addresses used for the read
1483 * may have just been remapped and thus the cache could be out
1484 * of sync. A re-clean after the switch will cure this.
1485 * After booting there are no gross relocations of the kernel thus
1486 * this problem will not occur after initarm().
1488 cpu_idcache_wbinv_all();
1490 /* Set stack for exception handlers */
1491 data_abort_handler_address = (u_int)data_abort_handler;
1492 prefetch_abort_handler_address = (u_int)prefetch_abort_handler;
1493 undefined_handler_address = (u_int)undefinedinstruction_bounce;
1496 init_proc0(kernelstack.pv_va);
1498 arm_intrnames_init();
1499 arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL);
1500 arm_dump_avail_init(memsize, sizeof(dump_avail) / sizeof(dump_avail[0]));
1501 pmap_bootstrap(freemempos, &kernel_l1pt);
1502 msgbufp = (void *)msgbufpv.pv_va;
1503 msgbufinit(msgbufp, msgbufsize);
1507 * Prepare map of physical memory regions available to vm subsystem.
1509 physmap_init(availmem_regions, availmem_regions_sz);
1511 init_param2(physmem);
1514 return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP -
1515 sizeof(struct pcb)));