1 /* head.S: Initial boot code for the Sparc64 port of Linux.
3 * Copyright (C) 1996, 1997, 2007 David S. Miller (davem@davemloft.net)
4 * Copyright (C) 1996 David Sitsky (David.Sitsky@anu.edu.au)
5 * Copyright (C) 1997, 1998 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
6 * Copyright (C) 1997 Miguel de Icaza (miguel@nuclecu.unam.mx)
9 #include <linux/version.h>
10 #include <linux/errno.h>
11 #include <linux/threads.h>
12 #include <linux/init.h>
13 #include <asm/thread_info.h>
15 #include <asm/pstate.h>
16 #include <asm/ptrace.h>
17 #include <asm/spitfire.h>
19 #include <asm/pgtable.h>
20 #include <asm/errno.h>
21 #include <asm/signal.h>
22 #include <asm/processor.h>
27 #include <asm/ttable.h>
29 #include <asm/cpudata.h>
31 /* This section from from _start to sparc64_boot_end should fit into
32 * 0x0000000000404000 to 0x0000000000408000.
35 .globl start, _start, stext, _stext
42 flushw /* Flush register file. */
44 /* This stuff has to be in sync with SILO and other potential boot loaders
45 * Fields should be kept upward compatible and whenever any change is made,
46 * HdrS version should be incremented.
48 .global root_flags, ram_flags, root_dev
49 .global sparc_ramdisk_image, sparc_ramdisk_size
50 .global sparc_ramdisk_image64
53 .word LINUX_VERSION_CODE
57 * 0x0300 : Supports being located at other than 0x4000
58 * 0x0202 : Supports kernel params string
59 * 0x0201 : Supports reboot_command
61 .half 0x0301 /* HdrS version */
75 sparc_ramdisk_image64:
79 /* PROM cif handler code address is in %o4. */
83 /* We need to remap the kernel. Use position independant
84 * code to remap us to KERNBASE.
86 * SILO can invoke us with 32-bit address masking enabled,
87 * so make sure that's clear.
90 andn %g1, PSTATE_AM, %g1
91 wrpr %g1, 0x0, %pstate
94 .globl prom_finddev_name, prom_chosen_path, prom_root_node
95 .globl prom_getprop_name, prom_mmu_name, prom_peer_name
96 .globl prom_callmethod_name, prom_translate_name, prom_root_compatible
97 .globl prom_map_name, prom_unmap_name, prom_mmu_ihandle_cache
98 .globl prom_boot_mapped_pc, prom_boot_mapping_mode
99 .globl prom_boot_mapping_phys_high, prom_boot_mapping_phys_low
100 .globl prom_compatible_name, prom_cpu_path, prom_cpu_compatible
101 .globl is_sun4v, sun4v_chip_type, prom_set_trap_table_name
104 prom_compatible_name:
116 prom_callmethod_name:
124 prom_set_trap_table_name:
125 .asciz "SUNW,set-trap-table"
129 .asciz "SUNW,UltraSPARC-T"
131 prom_root_compatible:
137 prom_mmu_ihandle_cache:
141 prom_boot_mapping_mode:
144 prom_boot_mapping_phys_high:
146 prom_boot_mapping_phys_low:
151 .word SUN4V_CHIP_INVALID
155 mov (1b - prom_peer_name), %l1
159 /* prom_root_node = prom_peer(0) */
160 stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "peer"
162 stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 1
163 stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
164 stx %l2, [%sp + 2047 + 128 + 0x18] ! arg1, 0
165 stx %g0, [%sp + 2047 + 128 + 0x20] ! ret1
167 add %sp, (2047 + 128), %o0 ! argument array
169 ldx [%sp + 2047 + 128 + 0x20], %l4 ! prom root node
170 mov (1b - prom_root_node), %l1
174 mov (1b - prom_getprop_name), %l1
175 mov (1b - prom_compatible_name), %l2
176 mov (1b - prom_root_compatible), %l5
181 /* prom_getproperty(prom_root_node, "compatible",
182 * &prom_root_compatible, 64)
184 stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "getprop"
186 stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 4
188 stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
189 stx %l4, [%sp + 2047 + 128 + 0x18] ! arg1, prom_root_node
190 stx %l2, [%sp + 2047 + 128 + 0x20] ! arg2, "compatible"
191 stx %l5, [%sp + 2047 + 128 + 0x28] ! arg3, &prom_root_compatible
193 stx %l3, [%sp + 2047 + 128 + 0x30] ! arg4, size
194 stx %g0, [%sp + 2047 + 128 + 0x38] ! ret1
196 add %sp, (2047 + 128), %o0 ! argument array
198 mov (1b - prom_finddev_name), %l1
199 mov (1b - prom_chosen_path), %l2
200 mov (1b - prom_boot_mapped_pc), %l3
205 sub %sp, (192 + 128), %sp
207 /* chosen_node = prom_finddevice("/chosen") */
208 stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "finddevice"
210 stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 1
211 stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
212 stx %l2, [%sp + 2047 + 128 + 0x18] ! arg1, "/chosen"
213 stx %g0, [%sp + 2047 + 128 + 0x20] ! ret1
215 add %sp, (2047 + 128), %o0 ! argument array
217 ldx [%sp + 2047 + 128 + 0x20], %l4 ! chosen device node
219 mov (1b - prom_getprop_name), %l1
220 mov (1b - prom_mmu_name), %l2
221 mov (1b - prom_mmu_ihandle_cache), %l5
226 /* prom_mmu_ihandle_cache = prom_getint(chosen_node, "mmu") */
227 stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "getprop"
229 stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 4
231 stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
232 stx %l4, [%sp + 2047 + 128 + 0x18] ! arg1, chosen_node
233 stx %l2, [%sp + 2047 + 128 + 0x20] ! arg2, "mmu"
234 stx %l5, [%sp + 2047 + 128 + 0x28] ! arg3, &prom_mmu_ihandle_cache
236 stx %l3, [%sp + 2047 + 128 + 0x30] ! arg4, sizeof(arg3)
237 stx %g0, [%sp + 2047 + 128 + 0x38] ! ret1
239 add %sp, (2047 + 128), %o0 ! argument array
241 mov (1b - prom_callmethod_name), %l1
242 mov (1b - prom_translate_name), %l2
245 lduw [%l5], %l5 ! prom_mmu_ihandle_cache
247 stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "call-method"
249 stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 3
251 stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 5
252 stx %l2, [%sp + 2047 + 128 + 0x18] ! arg1: "translate"
253 stx %l5, [%sp + 2047 + 128 + 0x20] ! arg2: prom_mmu_ihandle_cache
257 stx %l3, [%sp + 2047 + 128 + 0x28] ! arg3: vaddr, our PC
258 stx %g0, [%sp + 2047 + 128 + 0x30] ! res1
259 stx %g0, [%sp + 2047 + 128 + 0x38] ! res2
260 stx %g0, [%sp + 2047 + 128 + 0x40] ! res3
261 stx %g0, [%sp + 2047 + 128 + 0x48] ! res4
262 stx %g0, [%sp + 2047 + 128 + 0x50] ! res5
264 add %sp, (2047 + 128), %o0 ! argument array
266 ldx [%sp + 2047 + 128 + 0x40], %l1 ! translation mode
267 mov (1b - prom_boot_mapping_mode), %l4
270 mov (1b - prom_boot_mapping_phys_high), %l4
272 ldx [%sp + 2047 + 128 + 0x48], %l2 ! physaddr high
274 ldx [%sp + 2047 + 128 + 0x50], %l3 ! physaddr low
280 /* Leave service as-is, "call-method" */
282 stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 7
284 stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
285 mov (1b - prom_map_name), %l3
287 stx %l3, [%sp + 2047 + 128 + 0x18] ! arg1: "map"
288 /* Leave arg2 as-is, prom_mmu_ihandle_cache */
290 stx %l3, [%sp + 2047 + 128 + 0x28] ! arg3: mode (-1 default)
291 /* 4MB align the kernel image size. */
292 set (_end - KERNBASE), %l3
293 set ((4 * 1024 * 1024) - 1), %l4
296 stx %l3, [%sp + 2047 + 128 + 0x30] ! arg4: roundup(ksize, 4MB)
297 sethi %hi(KERNBASE), %l3
298 stx %l3, [%sp + 2047 + 128 + 0x38] ! arg5: vaddr (KERNBASE)
299 stx %g0, [%sp + 2047 + 128 + 0x40] ! arg6: empty
300 mov (1b - prom_boot_mapping_phys_low), %l3
303 stx %l3, [%sp + 2047 + 128 + 0x48] ! arg7: phys addr
305 add %sp, (2047 + 128), %o0 ! argument array
307 add %sp, (192 + 128), %sp
309 sethi %hi(prom_root_compatible), %g1
310 or %g1, %lo(prom_root_compatible), %g1
311 sethi %hi(prom_sun4v_name), %g7
312 or %g7, %lo(prom_sun4v_name), %g7
323 sethi %hi(is_sun4v), %g1
324 or %g1, %lo(is_sun4v), %g1
328 /* cpu_node = prom_finddevice("/cpu") */
329 mov (1b - prom_finddev_name), %l1
330 mov (1b - prom_cpu_path), %l2
333 sub %sp, (192 + 128), %sp
335 stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "finddevice"
337 stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 1
338 stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
339 stx %l2, [%sp + 2047 + 128 + 0x18] ! arg1, "/cpu"
340 stx %g0, [%sp + 2047 + 128 + 0x20] ! ret1
342 add %sp, (2047 + 128), %o0 ! argument array
344 ldx [%sp + 2047 + 128 + 0x20], %l4 ! cpu device node
346 mov (1b - prom_getprop_name), %l1
347 mov (1b - prom_compatible_name), %l2
348 mov (1b - prom_cpu_compatible), %l5
353 /* prom_getproperty(cpu_node, "compatible",
354 * &prom_cpu_compatible, 64)
356 stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "getprop"
358 stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 4
360 stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
361 stx %l4, [%sp + 2047 + 128 + 0x18] ! arg1, cpu_node
362 stx %l2, [%sp + 2047 + 128 + 0x20] ! arg2, "compatible"
363 stx %l5, [%sp + 2047 + 128 + 0x28] ! arg3, &prom_cpu_compatible
365 stx %l3, [%sp + 2047 + 128 + 0x30] ! arg4, size
366 stx %g0, [%sp + 2047 + 128 + 0x38] ! ret1
368 add %sp, (2047 + 128), %o0 ! argument array
370 add %sp, (192 + 128), %sp
372 sethi %hi(prom_cpu_compatible), %g1
373 or %g1, %lo(prom_cpu_compatible), %g1
374 sethi %hi(prom_niagara_prefix), %g7
375 or %g7, %lo(prom_niagara_prefix), %g7
386 sethi %hi(prom_cpu_compatible), %g1
387 or %g1, %lo(prom_cpu_compatible), %g1
391 mov SUN4V_CHIP_NIAGARA1, %g4
394 mov SUN4V_CHIP_NIAGARA2, %g4
396 mov SUN4V_CHIP_UNKNOWN, %g4
397 5: sethi %hi(sun4v_chip_type), %g2
398 or %g2, %lo(sun4v_chip_type), %g2
402 BRANCH_IF_SUN4V(g1, jump_to_sun4u_init)
403 BRANCH_IF_CHEETAH_BASE(g1,g7,cheetah_boot)
404 BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,cheetah_plus_boot)
405 ba,pt %xcc, spitfire_boot
409 /* Preserve OBP chosen DCU and DCR register settings. */
410 ba,pt %xcc, cheetah_generic_boot
414 mov DCR_BPE | DCR_RPE | DCR_SI | DCR_IFPOE | DCR_MS, %g1
417 sethi %uhi(DCU_ME|DCU_RE|DCU_HPE|DCU_SPE|DCU_SL|DCU_WE), %g7
418 or %g7, %ulo(DCU_ME|DCU_RE|DCU_HPE|DCU_SPE|DCU_SL|DCU_WE), %g7
420 or %g7, DCU_DM | DCU_IM | DCU_DC | DCU_IC, %g7
421 stxa %g7, [%g0] ASI_DCU_CONTROL_REG
424 cheetah_generic_boot:
425 mov TSB_EXTENSION_P, %g3
426 stxa %g0, [%g3] ASI_DMMU
427 stxa %g0, [%g3] ASI_IMMU
430 mov TSB_EXTENSION_S, %g3
431 stxa %g0, [%g3] ASI_DMMU
434 mov TSB_EXTENSION_N, %g3
435 stxa %g0, [%g3] ASI_DMMU
436 stxa %g0, [%g3] ASI_IMMU
439 ba,a,pt %xcc, jump_to_sun4u_init
442 /* Typically PROM has already enabled both MMU's and both on-chip
443 * caches, but we do it here anyway just to be paranoid.
445 mov (LSU_CONTROL_IC|LSU_CONTROL_DC|LSU_CONTROL_IM|LSU_CONTROL_DM), %g1
446 stxa %g1, [%g0] ASI_LSU_CONTROL
451 * Make sure we are in privileged mode, have address masking,
452 * using the ordinary globals and have enabled floating
455 * Again, typically PROM has left %pil at 13 or similar, and
456 * (PSTATE_PRIV | PSTATE_PEF | PSTATE_IE) in %pstate.
458 wrpr %g0, (PSTATE_PRIV|PSTATE_PEF|PSTATE_IE), %pstate
465 .section .text.init.refok
467 BRANCH_IF_SUN4V(g1, sun4v_init)
470 mov PRIMARY_CONTEXT, %g7
471 stxa %g0, [%g7] ASI_DMMU
474 mov SECONDARY_CONTEXT, %g7
475 stxa %g0, [%g7] ASI_DMMU
478 ba,pt %xcc, sun4u_continue
483 mov PRIMARY_CONTEXT, %g7
484 stxa %g0, [%g7] ASI_MMU
487 mov SECONDARY_CONTEXT, %g7
488 stxa %g0, [%g7] ASI_MMU
490 ba,pt %xcc, niagara_tlb_fixup
494 BRANCH_IF_ANY_CHEETAH(g1, g7, cheetah_tlb_fixup)
496 ba,pt %xcc, spitfire_tlb_fixup
500 mov 3, %g2 /* Set TLB type to hypervisor. */
501 sethi %hi(tlb_type), %g1
502 stw %g2, [%g1 + %lo(tlb_type)]
504 /* Patch copy/clear ops. */
505 sethi %hi(sun4v_chip_type), %g1
506 lduw [%g1 + %lo(sun4v_chip_type)], %g1
507 cmp %g1, SUN4V_CHIP_NIAGARA1
508 be,pt %xcc, niagara_patch
509 cmp %g1, SUN4V_CHIP_NIAGARA2
510 be,pt %xcc, niagara2_patch
513 call generic_patch_copyops
515 call generic_patch_bzero
517 call generic_patch_pageops
522 call niagara2_patch_copyops
524 call niagara_patch_bzero
526 call niagara2_patch_pageops
532 call niagara_patch_copyops
534 call niagara_patch_bzero
536 call niagara_patch_pageops
540 /* Patch TLB/cache ops. */
541 call hypervisor_patch_cachetlbops
544 ba,pt %xcc, tlb_fixup_done
548 mov 2, %g2 /* Set TLB type to cheetah+. */
549 BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,1f)
551 mov 1, %g2 /* Set TLB type to cheetah. */
553 1: sethi %hi(tlb_type), %g1
554 stw %g2, [%g1 + %lo(tlb_type)]
556 /* Patch copy/page operations to cheetah optimized versions. */
557 call cheetah_patch_copyops
559 call cheetah_patch_copy_page
561 call cheetah_patch_cachetlbops
564 ba,pt %xcc, tlb_fixup_done
568 /* Set TLB type to spitfire. */
570 sethi %hi(tlb_type), %g1
571 stw %g2, [%g1 + %lo(tlb_type)]
574 sethi %hi(init_thread_union), %g6
575 or %g6, %lo(init_thread_union), %g6
576 ldx [%g6 + TI_TASK], %g4
581 sllx %g1, THREAD_SHIFT, %g1
582 sub %g1, (STACKFRAME_SZ + STACK_BIAS), %g1
586 /* Set per-cpu pointer initially to zero, this makes
587 * the boot-cpu use the in-kernel-image per-cpu areas
588 * before setup_per_cpu_area() is invoked.
596 sethi %hi(__bss_start), %o0
597 or %o0, %lo(__bss_start), %o0
599 or %o1, %lo(_end), %o1
603 #ifdef CONFIG_LOCKDEP
604 /* We have this call this super early, as even prom_init can grab
605 * spinlocks and thus call into the lockdep code.
611 mov %l6, %o1 ! OpenPROM stack
613 mov %l7, %o0 ! OpenPROM cif handler
615 /* Initialize current_thread_info()->cpu as early as possible.
616 * In order to do that accurately we have to patch up the get_cpuid()
617 * assembler sequences. And that, in turn, requires that we know
618 * if we are on a Starfire box or not. While we're here, patch up
619 * the sun4v sequences as well.
621 call check_if_starfire
629 call hard_smp_processor_id
634 call boot_cpu_id_too_large
639 /* If we boot on a non-zero cpu, all of the per-cpu
640 * variable references we make before setting up the
641 * per-cpu areas will use a bogus offset. Put a
642 * compensating factor into __per_cpu_base to handle
645 * What the per-cpu code calculates is:
647 * __per_cpu_base + (cpu << __per_cpu_shift)
649 * These two variables are zero initially, so to
650 * make it all cancel out to zero we need to put
651 * "0 - (cpu << 0)" into __per_cpu_base so that the
652 * above formula evaluates to zero.
654 * We cannot even perform a printk() until this stuff
655 * is setup as that calls cpu_clock() which uses
659 sethi %hi(__per_cpu_base), %o2
660 stx %o1, [%o2 + %lo(__per_cpu_base)]
664 sth %o0, [%g6 + TI_CPU]
666 call prom_init_report
676 /* This is meant to allow the sharing of this code between
677 * boot processor invocation (via setup_tba() below) and
678 * secondary processor startup (via trampoline.S). The
679 * former does use this code, the latter does not yet due
680 * to some complexities. That should be fixed up at some
683 * There used to be enormous complexity wrt. transferring
684 * over from the firwmare's trap table to the Linux kernel's.
685 * For example, there was a chicken & egg problem wrt. building
686 * the OBP page tables, yet needing to be on the Linux kernel
687 * trap table (to translate PAGE_OFFSET addresses) in order to
690 * We now handle OBP tlb misses differently, via linear lookups
691 * into the prom_trans[] array. So that specific problem no
692 * longer exists. Yet, unfortunately there are still some issues
693 * preventing trampoline.S from using this code... ho hum.
695 .globl setup_trap_table
699 /* Force interrupts to be disabled. */
701 andn %l0, PSTATE_IE, %o1
702 wrpr %o1, 0x0, %pstate
706 /* Make the firmware call to jump over to the Linux trap table. */
707 sethi %hi(is_sun4v), %o0
708 lduw [%o0 + %lo(is_sun4v)], %o0
712 TRAP_LOAD_TRAP_BLOCK(%g2, %g3)
713 add %g2, TRAP_PER_CPU_FAULT_INFO, %g2
714 stxa %g2, [%g0] ASI_SCRATCHPAD
716 /* Compute physical address:
718 * paddr = kern_base + (mmfsa_vaddr - KERNBASE)
720 sethi %hi(KERNBASE), %g3
722 sethi %hi(kern_base), %g3
723 ldx [%g3 + %lo(kern_base)], %g3
725 sethi %hi(sparc64_ttable_tl0), %o0
727 set prom_set_trap_table_name, %g2
728 stx %g2, [%sp + 2047 + 128 + 0x00]
730 stx %g2, [%sp + 2047 + 128 + 0x08]
732 stx %g2, [%sp + 2047 + 128 + 0x10]
733 stx %o0, [%sp + 2047 + 128 + 0x18]
734 stx %o1, [%sp + 2047 + 128 + 0x20]
735 sethi %hi(p1275buf), %g2
736 or %g2, %lo(p1275buf), %g2
737 ldx [%g2 + 0x08], %o1
739 add %sp, (2047 + 128), %o0
744 1: sethi %hi(sparc64_ttable_tl0), %o0
745 set prom_set_trap_table_name, %g2
746 stx %g2, [%sp + 2047 + 128 + 0x00]
748 stx %g2, [%sp + 2047 + 128 + 0x08]
750 stx %g2, [%sp + 2047 + 128 + 0x10]
751 stx %o0, [%sp + 2047 + 128 + 0x18]
752 sethi %hi(p1275buf), %g2
753 or %g2, %lo(p1275buf), %g2
754 ldx [%g2 + 0x08], %o1
756 add %sp, (2047 + 128), %o0
758 /* Start using proper page size encodings in ctx register. */
759 2: sethi %hi(sparc64_kern_pri_context), %g3
760 ldx [%g3 + %lo(sparc64_kern_pri_context)], %g2
762 mov PRIMARY_CONTEXT, %g1
764 661: stxa %g2, [%g1] ASI_DMMU
765 .section .sun4v_1insn_patch, "ax"
767 stxa %g2, [%g1] ASI_MMU
772 BRANCH_IF_SUN4V(o2, 1f)
774 /* Kill PROM timer */
775 sethi %hi(0x80000000), %o2
777 wr %o2, 0, %tick_cmpr
779 BRANCH_IF_ANY_CHEETAH(o2, o3, 1f)
784 /* Disable STICK_INT interrupts. */
786 sethi %hi(0x80000000), %o2
791 wrpr %g0, %g0, %wstate
793 call init_irqwork_curcpu
796 /* Now we can restore interrupt state. */
807 /* The boot processor is the only cpu which invokes this
808 * routine, the other cpus set things up via trampoline.S.
809 * So save the OBP trap table address here.
812 sethi %hi(prom_tba), %o1
813 or %o1, %lo(prom_tba), %o1
816 call setup_trap_table
825 #include "winfixup.S"
827 #include "sun4v_tlb_miss.S"
828 #include "sun4v_ivec.S"
833 * The following skip makes sure the trap table in ttable.S is aligned
834 * on a 32K boundary as required by the v9 specs for TBA register.
836 * We align to a 32K boundary, then we have the 32K kernel TSB,
837 * the 64K kernel 4MB TSB, and then the 32K aligned trap table.
840 .skip 0x4000 + _start - 1b
848 .globl swapper_4m_tsb
854 /* Some care needs to be exercised if you try to move the
855 * location of the trap table relative to other things. For
856 * one thing there are br* instructions in some of the
857 * trap table entires which branch back to code in ktlb.S
858 * Those instructions can only handle a signed 16-bit
861 * There is a binutils bug (bugzilla #4558) which causes
862 * the relocation overflow checks for such instructions to
863 * not be done correctly. So bintuils will not notice the
864 * error and will instead write junk into the relocation and
865 * you'll have an unbootable kernel.
875 .globl prom_tba, tlb_type
877 tlb_type: .word 0 /* Must NOT end up in BSS */
878 .section ".fixup",#alloc,#execinstr
880 .globl __ret_efault, __retl_efault
883 restore %g0, -EFAULT, %o0