kernel-ark/arch/sparc64/kernel/head.S
David S. Miller c9c1083074 [SPARC64]: Fix boot failures on SunBlade-150
The sequence to move over to the Linux trap tables from
the firmware ones needs to be more air tight.  It turns
out that to be %100 safe we do need to be able to translate
OBP mappings in our TLB miss handlers early.

In order not to eat up a lot of kernel image memory with
static page tables, just use the translations array in
the OBP TLB miss handlers.  That solves the bulk of the
problem.

Furthermore, to make sure the OBP TLB miss path will work
even before the fixed MMU globals are loaded, explicitly
load %g1 to TLB_SFSR at the beginning of the i-TLB and
d-TLB miss handlers.

To ease the OBP TLB miss walking of the prom_trans[] array,
we sort it then delete all of the non-OBP entries in there
(for example, there are entries for the kernel image itself
which we're not interested in at all).

We also save about 32K of kernel image size with this change.
Not a bad side effect :-)

There are still some reasons why trampoline.S can't use the
setup_trap_table() yet.  The most noteworthy are:

1) OBP boots secondary processors with non-bias'd stack for
   some reason.  This is easily fixed by using a small bootup
   stack in the kernel image explicitly for this purpose.

2) Doing a firmware call via the normal C call prom_set_trap_table()
   goes through the whole OBP enter/exit sequence that saves and
   restores OBP and Linux kernel state in the MMUs.  This path
   unfortunately does a "flush %g6" while loading up the OBP locked
   TLB entries for the firmware call.

   If we setup the %g6 in the trampoline.S code properly, that
   is in the PAGE_OFFSET linear mapping, but we're not on the
   kernel trap table yet so those addresses won't translate properly.

   One idea is to do a by-hand firmware call like we do in the
   early bootup code and elsewhere here in trampoline.S  But this
   fails as well, as aparently the secondary processors are not
   booted with OBP's special locked TLB entries loaded.  These
   are necessary for the firwmare to processes TLB misses correctly
   up until the point where we take over the trap table.

This does need to be resolved at some point.

Signed-off-by: David S. Miller <davem@davemloft.net>
2005-10-12 12:22:46 -07:00

578 lines
14 KiB
ArmAsm

/* $Id: head.S,v 1.87 2002/02/09 19:49:31 davem Exp $
* head.S: Initial boot code for the Sparc64 port of Linux.
*
* Copyright (C) 1996,1997 David S. Miller (davem@caip.rutgers.edu)
* Copyright (C) 1996 David Sitsky (David.Sitsky@anu.edu.au)
* Copyright (C) 1997,1998 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
* Copyright (C) 1997 Miguel de Icaza (miguel@nuclecu.unam.mx)
*/
#include <linux/config.h>
#include <linux/version.h>
#include <linux/errno.h>
#include <asm/thread_info.h>
#include <asm/asi.h>
#include <asm/pstate.h>
#include <asm/ptrace.h>
#include <asm/spitfire.h>
#include <asm/page.h>
#include <asm/pgtable.h>
#include <asm/errno.h>
#include <asm/signal.h>
#include <asm/processor.h>
#include <asm/lsu.h>
#include <asm/dcr.h>
#include <asm/dcu.h>
#include <asm/head.h>
#include <asm/ttable.h>
#include <asm/mmu.h>
/* This section from from _start to sparc64_boot_end should fit into
* 0x0000000000404000 to 0x0000000000408000.
*/
.text
.globl start, _start, stext, _stext
_start:
start:
_stext:
stext:
! 0x0000000000404000
b sparc64_boot
flushw /* Flush register file. */
/* This stuff has to be in sync with SILO and other potential boot loaders
* Fields should be kept upward compatible and whenever any change is made,
* HdrS version should be incremented.
*/
.global root_flags, ram_flags, root_dev
.global sparc_ramdisk_image, sparc_ramdisk_size
.global sparc_ramdisk_image64
.ascii "HdrS"
.word LINUX_VERSION_CODE
/* History:
*
* 0x0300 : Supports being located at other than 0x4000
* 0x0202 : Supports kernel params string
* 0x0201 : Supports reboot_command
*/
.half 0x0301 /* HdrS version */
root_flags:
.half 1
root_dev:
.half 0
ram_flags:
.half 0
sparc_ramdisk_image:
.word 0
sparc_ramdisk_size:
.word 0
.xword reboot_command
.xword bootstr_info
sparc_ramdisk_image64:
.xword 0
.word _end
/* PROM cif handler code address is in %o4. */
sparc64_boot:
1: rd %pc, %g7
set 1b, %g1
cmp %g1, %g7
be,pn %xcc, sparc64_boot_after_remap
mov %o4, %l7
/* We need to remap the kernel. Use position independant
* code to remap us to KERNBASE.
*
* SILO can invoke us with 32-bit address masking enabled,
* so make sure that's clear.
*/
rdpr %pstate, %g1
andn %g1, PSTATE_AM, %g1
wrpr %g1, 0x0, %pstate
ba,a,pt %xcc, 1f
.globl prom_finddev_name, prom_chosen_path
.globl prom_getprop_name, prom_mmu_name
.globl prom_callmethod_name, prom_translate_name
.globl prom_map_name, prom_unmap_name, prom_mmu_ihandle_cache
.globl prom_boot_mapped_pc, prom_boot_mapping_mode
.globl prom_boot_mapping_phys_high, prom_boot_mapping_phys_low
prom_finddev_name:
.asciz "finddevice"
prom_chosen_path:
.asciz "/chosen"
prom_getprop_name:
.asciz "getprop"
prom_mmu_name:
.asciz "mmu"
prom_callmethod_name:
.asciz "call-method"
prom_translate_name:
.asciz "translate"
prom_map_name:
.asciz "map"
prom_unmap_name:
.asciz "unmap"
.align 4
prom_mmu_ihandle_cache:
.word 0
prom_boot_mapped_pc:
.word 0
prom_boot_mapping_mode:
.word 0
.align 8
prom_boot_mapping_phys_high:
.xword 0
prom_boot_mapping_phys_low:
.xword 0
1:
rd %pc, %l0
mov (1b - prom_finddev_name), %l1
mov (1b - prom_chosen_path), %l2
mov (1b - prom_boot_mapped_pc), %l3
sub %l0, %l1, %l1
sub %l0, %l2, %l2
sub %l0, %l3, %l3
stw %l0, [%l3]
sub %sp, (192 + 128), %sp
/* chosen_node = prom_finddevice("/chosen") */
stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "finddevice"
mov 1, %l3
stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 1
stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
stx %l2, [%sp + 2047 + 128 + 0x18] ! arg1, "/chosen"
stx %g0, [%sp + 2047 + 128 + 0x20] ! ret1
call %l7
add %sp, (2047 + 128), %o0 ! argument array
ldx [%sp + 2047 + 128 + 0x20], %l4 ! chosen device node
mov (1b - prom_getprop_name), %l1
mov (1b - prom_mmu_name), %l2
mov (1b - prom_mmu_ihandle_cache), %l5
sub %l0, %l1, %l1
sub %l0, %l2, %l2
sub %l0, %l5, %l5
/* prom_mmu_ihandle_cache = prom_getint(chosen_node, "mmu") */
stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "getprop"
mov 4, %l3
stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 4
mov 1, %l3
stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
stx %l4, [%sp + 2047 + 128 + 0x18] ! arg1, chosen_node
stx %l2, [%sp + 2047 + 128 + 0x20] ! arg2, "mmu"
stx %l5, [%sp + 2047 + 128 + 0x28] ! arg3, &prom_mmu_ihandle_cache
mov 4, %l3
stx %l3, [%sp + 2047 + 128 + 0x30] ! arg4, sizeof(arg3)
stx %g0, [%sp + 2047 + 128 + 0x38] ! ret1
call %l7
add %sp, (2047 + 128), %o0 ! argument array
mov (1b - prom_callmethod_name), %l1
mov (1b - prom_translate_name), %l2
sub %l0, %l1, %l1
sub %l0, %l2, %l2
lduw [%l5], %l5 ! prom_mmu_ihandle_cache
stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "call-method"
mov 3, %l3
stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 3
mov 5, %l3
stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 5
stx %l2, [%sp + 2047 + 128 + 0x18] ! arg1: "translate"
stx %l5, [%sp + 2047 + 128 + 0x20] ! arg2: prom_mmu_ihandle_cache
/* PAGE align */
srlx %l0, 13, %l3
sllx %l3, 13, %l3
stx %l3, [%sp + 2047 + 128 + 0x28] ! arg3: vaddr, our PC
stx %g0, [%sp + 2047 + 128 + 0x30] ! res1
stx %g0, [%sp + 2047 + 128 + 0x38] ! res2
stx %g0, [%sp + 2047 + 128 + 0x40] ! res3
stx %g0, [%sp + 2047 + 128 + 0x48] ! res4
stx %g0, [%sp + 2047 + 128 + 0x50] ! res5
call %l7
add %sp, (2047 + 128), %o0 ! argument array
ldx [%sp + 2047 + 128 + 0x40], %l1 ! translation mode
mov (1b - prom_boot_mapping_mode), %l4
sub %l0, %l4, %l4
stw %l1, [%l4]
mov (1b - prom_boot_mapping_phys_high), %l4
sub %l0, %l4, %l4
ldx [%sp + 2047 + 128 + 0x48], %l2 ! physaddr high
stx %l2, [%l4 + 0x0]
ldx [%sp + 2047 + 128 + 0x50], %l3 ! physaddr low
/* 4MB align */
srlx %l3, 22, %l3
sllx %l3, 22, %l3
stx %l3, [%l4 + 0x8]
/* Leave service as-is, "call-method" */
mov 7, %l3
stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 7
mov 1, %l3
stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1
mov (1b - prom_map_name), %l3
sub %l0, %l3, %l3
stx %l3, [%sp + 2047 + 128 + 0x18] ! arg1: "map"
/* Leave arg2 as-is, prom_mmu_ihandle_cache */
mov -1, %l3
stx %l3, [%sp + 2047 + 128 + 0x28] ! arg3: mode (-1 default)
sethi %hi(8 * 1024 * 1024), %l3
stx %l3, [%sp + 2047 + 128 + 0x30] ! arg4: size (8MB)
sethi %hi(KERNBASE), %l3
stx %l3, [%sp + 2047 + 128 + 0x38] ! arg5: vaddr (KERNBASE)
stx %g0, [%sp + 2047 + 128 + 0x40] ! arg6: empty
mov (1b - prom_boot_mapping_phys_low), %l3
sub %l0, %l3, %l3
ldx [%l3], %l3
stx %l3, [%sp + 2047 + 128 + 0x48] ! arg7: phys addr
call %l7
add %sp, (2047 + 128), %o0 ! argument array
add %sp, (192 + 128), %sp
sparc64_boot_after_remap:
BRANCH_IF_CHEETAH_BASE(g1,g7,cheetah_boot)
BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,cheetah_plus_boot)
ba,pt %xcc, spitfire_boot
nop
cheetah_plus_boot:
/* Preserve OBP chosen DCU and DCR register settings. */
ba,pt %xcc, cheetah_generic_boot
nop
cheetah_boot:
mov DCR_BPE | DCR_RPE | DCR_SI | DCR_IFPOE | DCR_MS, %g1
wr %g1, %asr18
sethi %uhi(DCU_ME|DCU_RE|DCU_HPE|DCU_SPE|DCU_SL|DCU_WE), %g7
or %g7, %ulo(DCU_ME|DCU_RE|DCU_HPE|DCU_SPE|DCU_SL|DCU_WE), %g7
sllx %g7, 32, %g7
or %g7, DCU_DM | DCU_IM | DCU_DC | DCU_IC, %g7
stxa %g7, [%g0] ASI_DCU_CONTROL_REG
membar #Sync
cheetah_generic_boot:
mov TSB_EXTENSION_P, %g3
stxa %g0, [%g3] ASI_DMMU
stxa %g0, [%g3] ASI_IMMU
membar #Sync
mov TSB_EXTENSION_S, %g3
stxa %g0, [%g3] ASI_DMMU
membar #Sync
mov TSB_EXTENSION_N, %g3
stxa %g0, [%g3] ASI_DMMU
stxa %g0, [%g3] ASI_IMMU
membar #Sync
ba,a,pt %xcc, jump_to_sun4u_init
spitfire_boot:
/* Typically PROM has already enabled both MMU's and both on-chip
* caches, but we do it here anyway just to be paranoid.
*/
mov (LSU_CONTROL_IC|LSU_CONTROL_DC|LSU_CONTROL_IM|LSU_CONTROL_DM), %g1
stxa %g1, [%g0] ASI_LSU_CONTROL
membar #Sync
jump_to_sun4u_init:
/*
* Make sure we are in privileged mode, have address masking,
* using the ordinary globals and have enabled floating
* point.
*
* Again, typically PROM has left %pil at 13 or similar, and
* (PSTATE_PRIV | PSTATE_PEF | PSTATE_IE) in %pstate.
*/
wrpr %g0, (PSTATE_PRIV|PSTATE_PEF|PSTATE_IE), %pstate
wr %g0, 0, %fprs
set sun4u_init, %g2
jmpl %g2 + %g0, %g0
nop
sun4u_init:
/* Set ctx 0 */
mov PRIMARY_CONTEXT, %g7
stxa %g0, [%g7] ASI_DMMU
membar #Sync
mov SECONDARY_CONTEXT, %g7
stxa %g0, [%g7] ASI_DMMU
membar #Sync
BRANCH_IF_ANY_CHEETAH(g1,g7,cheetah_tlb_fixup)
ba,pt %xcc, spitfire_tlb_fixup
nop
cheetah_tlb_fixup:
mov 2, %g2 /* Set TLB type to cheetah+. */
BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,1f)
mov 1, %g2 /* Set TLB type to cheetah. */
1: sethi %hi(tlb_type), %g1
stw %g2, [%g1 + %lo(tlb_type)]
/* Patch copy/page operations to cheetah optimized versions. */
call cheetah_patch_copyops
nop
call cheetah_patch_copy_page
nop
call cheetah_patch_cachetlbops
nop
ba,pt %xcc, tlb_fixup_done
nop
spitfire_tlb_fixup:
/* Set TLB type to spitfire. */
mov 0, %g2
sethi %hi(tlb_type), %g1
stw %g2, [%g1 + %lo(tlb_type)]
tlb_fixup_done:
sethi %hi(init_thread_union), %g6
or %g6, %lo(init_thread_union), %g6
ldx [%g6 + TI_TASK], %g4
mov %sp, %l6
mov %o4, %l7
wr %g0, ASI_P, %asi
mov 1, %g1
sllx %g1, THREAD_SHIFT, %g1
sub %g1, (STACKFRAME_SZ + STACK_BIAS), %g1
add %g6, %g1, %sp
mov 0, %fp
/* Set per-cpu pointer initially to zero, this makes
* the boot-cpu use the in-kernel-image per-cpu areas
* before setup_per_cpu_area() is invoked.
*/
clr %g5
wrpr %g0, 0, %wstate
wrpr %g0, 0x0, %tl
/* Clear the bss */
sethi %hi(__bss_start), %o0
or %o0, %lo(__bss_start), %o0
sethi %hi(_end), %o1
or %o1, %lo(_end), %o1
call __bzero
sub %o1, %o0, %o1
mov %l6, %o1 ! OpenPROM stack
call prom_init
mov %l7, %o0 ! OpenPROM cif handler
/* Off we go.... */
call start_kernel
nop
/* Not reached... */
/* This is meant to allow the sharing of this code between
* boot processor invocation (via setup_tba() below) and
* secondary processor startup (via trampoline.S). The
* former does use this code, the latter does not yet due
* to some complexities. That should be fixed up at some
* point.
*
* There used to be enormous complexity wrt. transferring
* over from the firwmare's trap table to the Linux kernel's.
* For example, there was a chicken & egg problem wrt. building
* the OBP page tables, yet needing to be on the Linux kernel
* trap table (to translate PAGE_OFFSET addresses) in order to
* do that.
*
* We now handle OBP tlb misses differently, via linear lookups
* into the prom_trans[] array. So that specific problem no
* longer exists. Yet, unfortunately there are still some issues
* preventing trampoline.S from using this code... ho hum.
*/
.globl setup_trap_table
setup_trap_table:
save %sp, -192, %sp
/* Force interrupts to be disabled. */
rdpr %pstate, %o1
andn %o1, PSTATE_IE, %o1
wrpr %o1, 0x0, %pstate
wrpr %g0, 15, %pil
/* Make the firmware call to jump over to the Linux trap table. */
call prom_set_trap_table
sethi %hi(sparc64_ttable_tl0), %o0
/* Start using proper page size encodings in ctx register. */
sethi %hi(sparc64_kern_pri_context), %g3
ldx [%g3 + %lo(sparc64_kern_pri_context)], %g2
mov PRIMARY_CONTEXT, %g1
stxa %g2, [%g1] ASI_DMMU
membar #Sync
/* The Linux trap handlers expect various trap global registers
* to be setup with some fixed values. So here we set these
* up very carefully. These globals are:
*
* Alternate Globals (PSTATE_AG):
*
* %g6 --> current_thread_info()
*
* MMU Globals (PSTATE_MG):
*
* %g1 --> TLB_SFSR
* %g2 --> ((_PAGE_VALID | _PAGE_SZ4MB |
* _PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_W)
* ^ 0xfffff80000000000)
* (this %g2 value is used for computing the PAGE_OFFSET kernel
* TLB entries quickly, the virtual address of the fault XOR'd
* with this %g2 value is the PTE to load into the TLB)
* %g3 --> VPTE_BASE_CHEETAH or VPTE_BASE_SPITFIRE
*
* Interrupt Globals (PSTATE_IG, setup by init_irqwork_curcpu()):
*
* %g6 --> __irq_work[smp_processor_id()]
*/
rdpr %pstate, %o1
mov %g6, %o2
wrpr %o1, PSTATE_AG, %pstate
mov %o2, %g6
#define KERN_HIGHBITS ((_PAGE_VALID|_PAGE_SZ4MB)^0xfffff80000000000)
#define KERN_LOWBITS (_PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_W)
wrpr %o1, PSTATE_MG, %pstate
mov TSB_REG, %g1
stxa %g0, [%g1] ASI_DMMU
membar #Sync
stxa %g0, [%g1] ASI_IMMU
membar #Sync
mov TLB_SFSR, %g1
sethi %uhi(KERN_HIGHBITS), %g2
or %g2, %ulo(KERN_HIGHBITS), %g2
sllx %g2, 32, %g2
or %g2, KERN_LOWBITS, %g2
BRANCH_IF_ANY_CHEETAH(g3,g7,8f)
ba,pt %xcc, 9f
nop
8:
sethi %uhi(VPTE_BASE_CHEETAH), %g3
or %g3, %ulo(VPTE_BASE_CHEETAH), %g3
ba,pt %xcc, 2f
sllx %g3, 32, %g3
9:
sethi %uhi(VPTE_BASE_SPITFIRE), %g3
or %g3, %ulo(VPTE_BASE_SPITFIRE), %g3
sllx %g3, 32, %g3
2:
clr %g7
#undef KERN_HIGHBITS
#undef KERN_LOWBITS
/* Kill PROM timer */
sethi %hi(0x80000000), %o2
sllx %o2, 32, %o2
wr %o2, 0, %tick_cmpr
BRANCH_IF_ANY_CHEETAH(o2,o3,1f)
ba,pt %xcc, 2f
nop
/* Disable STICK_INT interrupts. */
1:
sethi %hi(0x80000000), %o2
sllx %o2, 32, %o2
wr %o2, %asr25
2:
wrpr %g0, %g0, %wstate
wrpr %o1, 0x0, %pstate
call init_irqwork_curcpu
nop
/* Now we can turn interrupts back on. */
rdpr %pstate, %o1
or %o1, PSTATE_IE, %o1
wrpr %o1, 0, %pstate
wrpr %g0, 0x0, %pil
ret
restore
.globl setup_tba
setup_tba: /* i0 = is_starfire */
save %sp, -192, %sp
/* The boot processor is the only cpu which invokes this
* routine, the other cpus set things up via trampoline.S.
* So save the OBP trap table address here.
*/
rdpr %tba, %g7
sethi %hi(prom_tba), %o1
or %o1, %lo(prom_tba), %o1
stx %g7, [%o1]
call setup_trap_table
nop
ret
restore
sparc64_boot_end:
#include "systbls.S"
#include "ktlb.S"
#include "etrap.S"
#include "rtrap.S"
#include "winfixup.S"
#include "entry.S"
/*
* The following skip makes sure the trap table in ttable.S is aligned
* on a 32K boundary as required by the v9 specs for TBA register.
*/
1:
.skip 0x4000 + _start - 1b
#ifdef CONFIG_SBUS
/* This is just a hack to fool make depend config.h discovering
strategy: As the .S files below need config.h, but
make depend does not find it for them, we include config.h
in head.S */
#endif
! 0x0000000000408000
#include "ttable.S"
.data
.align 8
.globl prom_tba, tlb_type
prom_tba: .xword 0
tlb_type: .word 0 /* Must NOT end up in BSS */
.section ".fixup",#alloc,#execinstr
.globl __ret_efault, __retl_efault
__ret_efault:
ret
restore %g0, -EFAULT, %o0
__retl_efault:
retl
mov -EFAULT, %o0