linux/arch/ia64/kernel/mca_asm.S
Francois Wellenrieter 8a4b7b6f18 [IA64] Fix conversion of pal_min_state physical address
On return from INIT handler we must convert the address of the
minstate area from a kernel virtual uncached address (0xC...)
to physical uncached (0x8...).  A typo (or thinko?) in the code
converted to physical cached.

Signed-off-by: Tony Luck <tony.luck@intel.com>
2006-01-13 14:01:01 -08:00

1099 lines
27 KiB
ArmAsm

//
// assembly portion of the IA64 MCA handling
//
// Mods by cfleck to integrate into kernel build
// 00/03/15 davidm Added various stop bits to get a clean compile
//
// 00/03/29 cfleck Added code to save INIT handoff state in pt_regs format, switch to temp
// kstack, switch modes, jump to C INIT handler
//
// 02/01/04 J.Hall <jenna.s.hall@intel.com>
// Before entering virtual mode code:
// 1. Check for TLB CPU error
// 2. Restore current thread pointer to kr6
// 3. Move stack ptr 16 bytes to conform to C calling convention
//
// 04/11/12 Russ Anderson <rja@sgi.com>
// Added per cpu MCA/INIT stack save areas.
//
// 12/08/05 Keith Owens <kaos@sgi.com>
// Use per cpu MCA/INIT stacks for all data.
//
#include <linux/config.h>
#include <linux/threads.h>
#include <asm/asmmacro.h>
#include <asm/pgtable.h>
#include <asm/processor.h>
#include <asm/mca_asm.h>
#include <asm/mca.h>
#include "entry.h"
#define GET_IA64_MCA_DATA(reg) \
GET_THIS_PADDR(reg, ia64_mca_data) \
;; \
ld8 reg=[reg]
.global ia64_do_tlb_purge
.global ia64_os_mca_dispatch
.global ia64_os_init_dispatch_monarch
.global ia64_os_init_dispatch_slave
.text
.align 16
//StartMain////////////////////////////////////////////////////////////////////
/*
* Just the TLB purge part is moved to a separate function
* so we can re-use the code for cpu hotplug code as well
* Caller should now setup b1, so we can branch once the
* tlb flush is complete.
*/
ia64_do_tlb_purge:
#define O(member) IA64_CPUINFO_##member##_OFFSET
GET_THIS_PADDR(r2, cpu_info) // load phys addr of cpu_info into r2
;;
addl r17=O(PTCE_STRIDE),r2
addl r2=O(PTCE_BASE),r2
;;
ld8 r18=[r2],(O(PTCE_COUNT)-O(PTCE_BASE));; // r18=ptce_base
ld4 r19=[r2],4 // r19=ptce_count[0]
ld4 r21=[r17],4 // r21=ptce_stride[0]
;;
ld4 r20=[r2] // r20=ptce_count[1]
ld4 r22=[r17] // r22=ptce_stride[1]
mov r24=0
;;
adds r20=-1,r20
;;
#undef O
2:
cmp.ltu p6,p7=r24,r19
(p7) br.cond.dpnt.few 4f
mov ar.lc=r20
3:
ptc.e r18
;;
add r18=r22,r18
br.cloop.sptk.few 3b
;;
add r18=r21,r18
add r24=1,r24
;;
br.sptk.few 2b
4:
srlz.i // srlz.i implies srlz.d
;;
// Now purge addresses formerly mapped by TR registers
// 1. Purge ITR&DTR for kernel.
movl r16=KERNEL_START
mov r18=KERNEL_TR_PAGE_SHIFT<<2
;;
ptr.i r16, r18
ptr.d r16, r18
;;
srlz.i
;;
srlz.d
;;
// 2. Purge DTR for PERCPU data.
movl r16=PERCPU_ADDR
mov r18=PERCPU_PAGE_SHIFT<<2
;;
ptr.d r16,r18
;;
srlz.d
;;
// 3. Purge ITR for PAL code.
GET_THIS_PADDR(r2, ia64_mca_pal_base)
;;
ld8 r16=[r2]
mov r18=IA64_GRANULE_SHIFT<<2
;;
ptr.i r16,r18
;;
srlz.i
;;
// 4. Purge DTR for stack.
mov r16=IA64_KR(CURRENT_STACK)
;;
shl r16=r16,IA64_GRANULE_SHIFT
movl r19=PAGE_OFFSET
;;
add r16=r19,r16
mov r18=IA64_GRANULE_SHIFT<<2
;;
ptr.d r16,r18
;;
srlz.i
;;
// Now branch away to caller.
br.sptk.many b1
;;
//EndMain//////////////////////////////////////////////////////////////////////
//StartMain////////////////////////////////////////////////////////////////////
ia64_os_mca_dispatch:
// Serialize all MCA processing
mov r3=1;;
LOAD_PHYSICAL(p0,r2,ia64_mca_serialize);;
ia64_os_mca_spin:
xchg4 r4=[r2],r3;;
cmp.ne p6,p0=r4,r0
(p6) br ia64_os_mca_spin
mov r3=IA64_MCA_CPU_MCA_STACK_OFFSET // use the MCA stack
LOAD_PHYSICAL(p0,r2,1f) // return address
mov r19=1 // All MCA events are treated as monarch (for now)
br.sptk ia64_state_save // save the state that is not in minstate
1:
GET_IA64_MCA_DATA(r2)
// Using MCA stack, struct ia64_sal_os_state, variable proc_state_param
;;
add r3=IA64_MCA_CPU_MCA_STACK_OFFSET+MCA_SOS_OFFSET+IA64_SAL_OS_STATE_PROC_STATE_PARAM_OFFSET, r2
;;
ld8 r18=[r3] // Get processor state parameter on existing PALE_CHECK.
;;
tbit.nz p6,p7=r18,60
(p7) br.spnt done_tlb_purge_and_reload
// The following code purges TC and TR entries. Then reload all TC entries.
// Purge percpu data TC entries.
begin_tlb_purge_and_reload:
movl r18=ia64_reload_tr;;
LOAD_PHYSICAL(p0,r18,ia64_reload_tr);;
mov b1=r18;;
br.sptk.many ia64_do_tlb_purge;;
ia64_reload_tr:
// Finally reload the TR registers.
// 1. Reload DTR/ITR registers for kernel.
mov r18=KERNEL_TR_PAGE_SHIFT<<2
movl r17=KERNEL_START
;;
mov cr.itir=r18
mov cr.ifa=r17
mov r16=IA64_TR_KERNEL
mov r19=ip
movl r18=PAGE_KERNEL
;;
dep r17=0,r19,0, KERNEL_TR_PAGE_SHIFT
;;
or r18=r17,r18
;;
itr.i itr[r16]=r18
;;
itr.d dtr[r16]=r18
;;
srlz.i
srlz.d
;;
// 2. Reload DTR register for PERCPU data.
GET_THIS_PADDR(r2, ia64_mca_per_cpu_pte)
;;
movl r16=PERCPU_ADDR // vaddr
movl r18=PERCPU_PAGE_SHIFT<<2
;;
mov cr.itir=r18
mov cr.ifa=r16
;;
ld8 r18=[r2] // load per-CPU PTE
mov r16=IA64_TR_PERCPU_DATA;
;;
itr.d dtr[r16]=r18
;;
srlz.d
;;
// 3. Reload ITR for PAL code.
GET_THIS_PADDR(r2, ia64_mca_pal_pte)
;;
ld8 r18=[r2] // load PAL PTE
;;
GET_THIS_PADDR(r2, ia64_mca_pal_base)
;;
ld8 r16=[r2] // load PAL vaddr
mov r19=IA64_GRANULE_SHIFT<<2
;;
mov cr.itir=r19
mov cr.ifa=r16
mov r20=IA64_TR_PALCODE
;;
itr.i itr[r20]=r18
;;
srlz.i
;;
// 4. Reload DTR for stack.
mov r16=IA64_KR(CURRENT_STACK)
;;
shl r16=r16,IA64_GRANULE_SHIFT
movl r19=PAGE_OFFSET
;;
add r18=r19,r16
movl r20=PAGE_KERNEL
;;
add r16=r20,r16
mov r19=IA64_GRANULE_SHIFT<<2
;;
mov cr.itir=r19
mov cr.ifa=r18
mov r20=IA64_TR_CURRENT_STACK
;;
itr.d dtr[r20]=r16
;;
srlz.d
done_tlb_purge_and_reload:
// switch to per cpu MCA stack
mov r3=IA64_MCA_CPU_MCA_STACK_OFFSET // use the MCA stack
LOAD_PHYSICAL(p0,r2,1f) // return address
br.sptk ia64_new_stack
1:
// everything saved, now we can set the kernel registers
mov r3=IA64_MCA_CPU_MCA_STACK_OFFSET // use the MCA stack
LOAD_PHYSICAL(p0,r2,1f) // return address
br.sptk ia64_set_kernel_registers
1:
// This must be done in physical mode
GET_IA64_MCA_DATA(r2)
;;
mov r7=r2
// Enter virtual mode from physical mode
VIRTUAL_MODE_ENTER(r2, r3, ia64_os_mca_virtual_begin, r4)
// This code returns to SAL via SOS r2, in general SAL has no unwind
// data. To get a clean termination when backtracing the C MCA/INIT
// handler, set a dummy return address of 0 in this routine. That
// requires that ia64_os_mca_virtual_begin be a global function.
ENTRY(ia64_os_mca_virtual_begin)
.prologue
.save rp,r0
.body
mov ar.rsc=3 // set eager mode for C handler
mov r2=r7 // see GET_IA64_MCA_DATA above
;;
// Call virtual mode handler
alloc r14=ar.pfs,0,0,3,0
;;
DATA_PA_TO_VA(r2,r7)
;;
add out0=IA64_MCA_CPU_MCA_STACK_OFFSET+MCA_PT_REGS_OFFSET, r2
add out1=IA64_MCA_CPU_MCA_STACK_OFFSET+MCA_SWITCH_STACK_OFFSET, r2
add out2=IA64_MCA_CPU_MCA_STACK_OFFSET+MCA_SOS_OFFSET, r2
br.call.sptk.many b0=ia64_mca_handler
// Revert back to physical mode before going back to SAL
PHYSICAL_MODE_ENTER(r2, r3, ia64_os_mca_virtual_end, r4)
ia64_os_mca_virtual_end:
END(ia64_os_mca_virtual_begin)
// switch back to previous stack
alloc r14=ar.pfs,0,0,0,0 // remove the MCA handler frame
mov r3=IA64_MCA_CPU_MCA_STACK_OFFSET // use the MCA stack
LOAD_PHYSICAL(p0,r2,1f) // return address
br.sptk ia64_old_stack
1:
mov r3=IA64_MCA_CPU_MCA_STACK_OFFSET // use the MCA stack
LOAD_PHYSICAL(p0,r2,1f) // return address
br.sptk ia64_state_restore // restore the SAL state
1:
mov b0=r12 // SAL_CHECK return address
// release lock
LOAD_PHYSICAL(p0,r3,ia64_mca_serialize);;
st4.rel [r3]=r0
br b0
//EndMain//////////////////////////////////////////////////////////////////////
//StartMain////////////////////////////////////////////////////////////////////
//
// SAL to OS entry point for INIT on all processors. This has been defined for
// registration purposes with SAL as a part of ia64_mca_init. Monarch and
// slave INIT have identical processing, except for the value of the
// sos->monarch flag in r19.
//
ia64_os_init_dispatch_monarch:
mov r19=1 // Bow, bow, ye lower middle classes!
br.sptk ia64_os_init_dispatch
ia64_os_init_dispatch_slave:
mov r19=0 // <igor>yeth, mathter</igor>
ia64_os_init_dispatch:
mov r3=IA64_MCA_CPU_INIT_STACK_OFFSET // use the INIT stack
LOAD_PHYSICAL(p0,r2,1f) // return address
br.sptk ia64_state_save // save the state that is not in minstate
1:
// switch to per cpu INIT stack
mov r3=IA64_MCA_CPU_INIT_STACK_OFFSET // use the INIT stack
LOAD_PHYSICAL(p0,r2,1f) // return address
br.sptk ia64_new_stack
1:
// everything saved, now we can set the kernel registers
mov r3=IA64_MCA_CPU_INIT_STACK_OFFSET // use the INIT stack
LOAD_PHYSICAL(p0,r2,1f) // return address
br.sptk ia64_set_kernel_registers
1:
// This must be done in physical mode
GET_IA64_MCA_DATA(r2)
;;
mov r7=r2
// Enter virtual mode from physical mode
VIRTUAL_MODE_ENTER(r2, r3, ia64_os_init_virtual_begin, r4)
// This code returns to SAL via SOS r2, in general SAL has no unwind
// data. To get a clean termination when backtracing the C MCA/INIT
// handler, set a dummy return address of 0 in this routine. That
// requires that ia64_os_init_virtual_begin be a global function.
ENTRY(ia64_os_init_virtual_begin)
.prologue
.save rp,r0
.body
mov ar.rsc=3 // set eager mode for C handler
mov r2=r7 // see GET_IA64_MCA_DATA above
;;
// Call virtual mode handler
alloc r14=ar.pfs,0,0,3,0
;;
DATA_PA_TO_VA(r2,r7)
;;
add out0=IA64_MCA_CPU_INIT_STACK_OFFSET+MCA_PT_REGS_OFFSET, r2
add out1=IA64_MCA_CPU_INIT_STACK_OFFSET+MCA_SWITCH_STACK_OFFSET, r2
add out2=IA64_MCA_CPU_INIT_STACK_OFFSET+MCA_SOS_OFFSET, r2
br.call.sptk.many b0=ia64_init_handler
// Revert back to physical mode before going back to SAL
PHYSICAL_MODE_ENTER(r2, r3, ia64_os_init_virtual_end, r4)
ia64_os_init_virtual_end:
END(ia64_os_init_virtual_begin)
mov r3=IA64_MCA_CPU_INIT_STACK_OFFSET // use the INIT stack
LOAD_PHYSICAL(p0,r2,1f) // return address
br.sptk ia64_state_restore // restore the SAL state
1:
// switch back to previous stack
alloc r14=ar.pfs,0,0,0,0 // remove the INIT handler frame
mov r3=IA64_MCA_CPU_INIT_STACK_OFFSET // use the INIT stack
LOAD_PHYSICAL(p0,r2,1f) // return address
br.sptk ia64_old_stack
1:
mov b0=r12 // SAL_CHECK return address
br b0
//EndMain//////////////////////////////////////////////////////////////////////
// common defines for the stubs
#define ms r4
#define regs r5
#define temp1 r2 /* careful, it overlaps with input registers */
#define temp2 r3 /* careful, it overlaps with input registers */
#define temp3 r7
#define temp4 r14
//++
// Name:
// ia64_state_save()
//
// Stub Description:
//
// Save the state that is not in minstate. This is sensitive to the layout of
// struct ia64_sal_os_state in mca.h.
//
// r2 contains the return address, r3 contains either
// IA64_MCA_CPU_MCA_STACK_OFFSET or IA64_MCA_CPU_INIT_STACK_OFFSET.
//
// The OS to SAL section of struct ia64_sal_os_state is set to a default
// value of cold boot (MCA) or warm boot (INIT) and return to the same
// context. ia64_sal_os_state is also used to hold some registers that
// need to be saved and restored across the stack switches.
//
// Most input registers to this stub come from PAL/SAL
// r1 os gp, physical
// r8 pal_proc entry point
// r9 sal_proc entry point
// r10 sal gp
// r11 MCA - rendevzous state, INIT - reason code
// r12 sal return address
// r17 pal min_state
// r18 processor state parameter
// r19 monarch flag, set by the caller of this routine
//
// In addition to the SAL to OS state, this routine saves all the
// registers that appear in struct pt_regs and struct switch_stack,
// excluding those that are already in the PAL minstate area. This
// results in a partial pt_regs and switch_stack, the C code copies the
// remaining registers from PAL minstate to pt_regs and switch_stack. The
// resulting structures contain all the state of the original process when
// MCA/INIT occurred.
//
//--
ia64_state_save:
add regs=MCA_SOS_OFFSET, r3
add ms=MCA_SOS_OFFSET+8, r3
mov b0=r2 // save return address
cmp.eq p1,p2=IA64_MCA_CPU_MCA_STACK_OFFSET, r3
;;
GET_IA64_MCA_DATA(temp2)
;;
add temp1=temp2, regs // struct ia64_sal_os_state on MCA or INIT stack
add temp2=temp2, ms // struct ia64_sal_os_state+8 on MCA or INIT stack
;;
mov regs=temp1 // save the start of sos
st8 [temp1]=r1,16 // os_gp
st8 [temp2]=r8,16 // pal_proc
;;
st8 [temp1]=r9,16 // sal_proc
st8 [temp2]=r11,16 // rv_rc
mov r11=cr.iipa
;;
st8 [temp1]=r18,16 // proc_state_param
st8 [temp2]=r19,16 // monarch
mov r6=IA64_KR(CURRENT)
;;
st8 [temp1]=r12,16 // sal_ra
st8 [temp2]=r10,16 // sal_gp
mov r12=cr.isr
;;
st8 [temp1]=r17,16 // pal_min_state
st8 [temp2]=r6,16 // prev_IA64_KR_CURRENT
mov r6=IA64_KR(CURRENT_STACK)
;;
st8 [temp1]=r6,16 // prev_IA64_KR_CURRENT_STACK
st8 [temp2]=r0,16 // prev_task, starts off as NULL
mov r6=cr.ifa
;;
st8 [temp1]=r12,16 // cr.isr
st8 [temp2]=r6,16 // cr.ifa
mov r12=cr.itir
;;
st8 [temp1]=r12,16 // cr.itir
st8 [temp2]=r11,16 // cr.iipa
mov r12=cr.iim
;;
st8 [temp1]=r12,16 // cr.iim
(p1) mov r12=IA64_MCA_COLD_BOOT
(p2) mov r12=IA64_INIT_WARM_BOOT
mov r6=cr.iha
;;
st8 [temp2]=r6,16 // cr.iha
st8 [temp1]=r12 // os_status, default is cold boot
mov r6=IA64_MCA_SAME_CONTEXT
;;
st8 [temp1]=r6 // context, default is same context
// Save the pt_regs data that is not in minstate. The previous code
// left regs at sos.
add regs=MCA_PT_REGS_OFFSET-MCA_SOS_OFFSET, regs
;;
add temp1=PT(B6), regs
mov temp3=b6
mov temp4=b7
add temp2=PT(B7), regs
;;
st8 [temp1]=temp3,PT(AR_CSD)-PT(B6) // save b6
st8 [temp2]=temp4,PT(AR_SSD)-PT(B7) // save b7
mov temp3=ar.csd
mov temp4=ar.ssd
cover // must be last in group
;;
st8 [temp1]=temp3,PT(AR_UNAT)-PT(AR_CSD) // save ar.csd
st8 [temp2]=temp4,PT(AR_PFS)-PT(AR_SSD) // save ar.ssd
mov temp3=ar.unat
mov temp4=ar.pfs
;;
st8 [temp1]=temp3,PT(AR_RNAT)-PT(AR_UNAT) // save ar.unat
st8 [temp2]=temp4,PT(AR_BSPSTORE)-PT(AR_PFS) // save ar.pfs
mov temp3=ar.rnat
mov temp4=ar.bspstore
;;
st8 [temp1]=temp3,PT(LOADRS)-PT(AR_RNAT) // save ar.rnat
st8 [temp2]=temp4,PT(AR_FPSR)-PT(AR_BSPSTORE) // save ar.bspstore
mov temp3=ar.bsp
;;
sub temp3=temp3, temp4 // ar.bsp - ar.bspstore
mov temp4=ar.fpsr
;;
shl temp3=temp3,16 // compute ar.rsc to be used for "loadrs"
;;
st8 [temp1]=temp3,PT(AR_CCV)-PT(LOADRS) // save loadrs
st8 [temp2]=temp4,PT(F6)-PT(AR_FPSR) // save ar.fpsr
mov temp3=ar.ccv
;;
st8 [temp1]=temp3,PT(F7)-PT(AR_CCV) // save ar.ccv
stf.spill [temp2]=f6,PT(F8)-PT(F6)
;;
stf.spill [temp1]=f7,PT(F9)-PT(F7)
stf.spill [temp2]=f8,PT(F10)-PT(F8)
;;
stf.spill [temp1]=f9,PT(F11)-PT(F9)
stf.spill [temp2]=f10
;;
stf.spill [temp1]=f11
// Save the switch_stack data that is not in minstate nor pt_regs. The
// previous code left regs at pt_regs.
add regs=MCA_SWITCH_STACK_OFFSET-MCA_PT_REGS_OFFSET, regs
;;
add temp1=SW(F2), regs
add temp2=SW(F3), regs
;;
stf.spill [temp1]=f2,32
stf.spill [temp2]=f3,32
;;
stf.spill [temp1]=f4,32
stf.spill [temp2]=f5,32
;;
stf.spill [temp1]=f12,32
stf.spill [temp2]=f13,32
;;
stf.spill [temp1]=f14,32
stf.spill [temp2]=f15,32
;;
stf.spill [temp1]=f16,32
stf.spill [temp2]=f17,32
;;
stf.spill [temp1]=f18,32
stf.spill [temp2]=f19,32
;;
stf.spill [temp1]=f20,32
stf.spill [temp2]=f21,32
;;
stf.spill [temp1]=f22,32
stf.spill [temp2]=f23,32
;;
stf.spill [temp1]=f24,32
stf.spill [temp2]=f25,32
;;
stf.spill [temp1]=f26,32
stf.spill [temp2]=f27,32
;;
stf.spill [temp1]=f28,32
stf.spill [temp2]=f29,32
;;
stf.spill [temp1]=f30,SW(B2)-SW(F30)
stf.spill [temp2]=f31,SW(B3)-SW(F31)
mov temp3=b2
mov temp4=b3
;;
st8 [temp1]=temp3,16 // save b2
st8 [temp2]=temp4,16 // save b3
mov temp3=b4
mov temp4=b5
;;
st8 [temp1]=temp3,SW(AR_LC)-SW(B4) // save b4
st8 [temp2]=temp4 // save b5
mov temp3=ar.lc
;;
st8 [temp1]=temp3 // save ar.lc
// FIXME: Some proms are incorrectly accessing the minstate area as
// cached data. The C code uses region 6, uncached virtual. Ensure
// that there is no cache data lying around for the first 1K of the
// minstate area.
// Remove this code in September 2006, that gives platforms a year to
// fix their proms and get their customers updated.
add r1=32*1,r17
add r2=32*2,r17
add r3=32*3,r17
add r4=32*4,r17
add r5=32*5,r17
add r6=32*6,r17
add r7=32*7,r17
;;
fc r17
fc r1
fc r2
fc r3
fc r4
fc r5
fc r6
fc r7
add r17=32*8,r17
add r1=32*8,r1
add r2=32*8,r2
add r3=32*8,r3
add r4=32*8,r4
add r5=32*8,r5
add r6=32*8,r6
add r7=32*8,r7
;;
fc r17
fc r1
fc r2
fc r3
fc r4
fc r5
fc r6
fc r7
add r17=32*8,r17
add r1=32*8,r1
add r2=32*8,r2
add r3=32*8,r3
add r4=32*8,r4
add r5=32*8,r5
add r6=32*8,r6
add r7=32*8,r7
;;
fc r17
fc r1
fc r2
fc r3
fc r4
fc r5
fc r6
fc r7
add r17=32*8,r17
add r1=32*8,r1
add r2=32*8,r2
add r3=32*8,r3
add r4=32*8,r4
add r5=32*8,r5
add r6=32*8,r6
add r7=32*8,r7
;;
fc r17
fc r1
fc r2
fc r3
fc r4
fc r5
fc r6
fc r7
br.sptk b0
//EndStub//////////////////////////////////////////////////////////////////////
//++
// Name:
// ia64_state_restore()
//
// Stub Description:
//
// Restore the SAL/OS state. This is sensitive to the layout of struct
// ia64_sal_os_state in mca.h.
//
// r2 contains the return address, r3 contains either
// IA64_MCA_CPU_MCA_STACK_OFFSET or IA64_MCA_CPU_INIT_STACK_OFFSET.
//
// In addition to the SAL to OS state, this routine restores all the
// registers that appear in struct pt_regs and struct switch_stack,
// excluding those in the PAL minstate area.
//
//--
ia64_state_restore:
// Restore the switch_stack data that is not in minstate nor pt_regs.
add regs=MCA_SWITCH_STACK_OFFSET, r3
mov b0=r2 // save return address
;;
GET_IA64_MCA_DATA(temp2)
;;
add regs=temp2, regs
;;
add temp1=SW(F2), regs
add temp2=SW(F3), regs
;;
ldf.fill f2=[temp1],32
ldf.fill f3=[temp2],32
;;
ldf.fill f4=[temp1],32
ldf.fill f5=[temp2],32
;;
ldf.fill f12=[temp1],32
ldf.fill f13=[temp2],32
;;
ldf.fill f14=[temp1],32
ldf.fill f15=[temp2],32
;;
ldf.fill f16=[temp1],32
ldf.fill f17=[temp2],32
;;
ldf.fill f18=[temp1],32
ldf.fill f19=[temp2],32
;;
ldf.fill f20=[temp1],32
ldf.fill f21=[temp2],32
;;
ldf.fill f22=[temp1],32
ldf.fill f23=[temp2],32
;;
ldf.fill f24=[temp1],32
ldf.fill f25=[temp2],32
;;
ldf.fill f26=[temp1],32
ldf.fill f27=[temp2],32
;;
ldf.fill f28=[temp1],32
ldf.fill f29=[temp2],32
;;
ldf.fill f30=[temp1],SW(B2)-SW(F30)
ldf.fill f31=[temp2],SW(B3)-SW(F31)
;;
ld8 temp3=[temp1],16 // restore b2
ld8 temp4=[temp2],16 // restore b3
;;
mov b2=temp3
mov b3=temp4
ld8 temp3=[temp1],SW(AR_LC)-SW(B4) // restore b4
ld8 temp4=[temp2] // restore b5
;;
mov b4=temp3
mov b5=temp4
ld8 temp3=[temp1] // restore ar.lc
;;
mov ar.lc=temp3
// Restore the pt_regs data that is not in minstate. The previous code
// left regs at switch_stack.
add regs=MCA_PT_REGS_OFFSET-MCA_SWITCH_STACK_OFFSET, regs
;;
add temp1=PT(B6), regs
add temp2=PT(B7), regs
;;
ld8 temp3=[temp1],PT(AR_CSD)-PT(B6) // restore b6
ld8 temp4=[temp2],PT(AR_SSD)-PT(B7) // restore b7
;;
mov b6=temp3
mov b7=temp4
ld8 temp3=[temp1],PT(AR_UNAT)-PT(AR_CSD) // restore ar.csd
ld8 temp4=[temp2],PT(AR_PFS)-PT(AR_SSD) // restore ar.ssd
;;
mov ar.csd=temp3
mov ar.ssd=temp4
ld8 temp3=[temp1] // restore ar.unat
add temp1=PT(AR_CCV)-PT(AR_UNAT), temp1
ld8 temp4=[temp2],PT(AR_FPSR)-PT(AR_PFS) // restore ar.pfs
;;
mov ar.unat=temp3
mov ar.pfs=temp4
// ar.rnat, ar.bspstore, loadrs are restore in ia64_old_stack.
ld8 temp3=[temp1],PT(F6)-PT(AR_CCV) // restore ar.ccv
ld8 temp4=[temp2],PT(F7)-PT(AR_FPSR) // restore ar.fpsr
;;
mov ar.ccv=temp3
mov ar.fpsr=temp4
ldf.fill f6=[temp1],PT(F8)-PT(F6)
ldf.fill f7=[temp2],PT(F9)-PT(F7)
;;
ldf.fill f8=[temp1],PT(F10)-PT(F8)
ldf.fill f9=[temp2],PT(F11)-PT(F9)
;;
ldf.fill f10=[temp1]
ldf.fill f11=[temp2]
// Restore the SAL to OS state. The previous code left regs at pt_regs.
add regs=MCA_SOS_OFFSET-MCA_PT_REGS_OFFSET, regs
;;
add temp1=IA64_SAL_OS_STATE_COMMON_OFFSET, regs
add temp2=IA64_SAL_OS_STATE_COMMON_OFFSET+8, regs
;;
ld8 r12=[temp1],16 // sal_ra
ld8 r9=[temp2],16 // sal_gp
;;
ld8 r22=[temp1],16 // pal_min_state, virtual
ld8 r21=[temp2],16 // prev_IA64_KR_CURRENT
;;
ld8 r16=[temp1],16 // prev_IA64_KR_CURRENT_STACK
ld8 r20=[temp2],16 // prev_task
;;
ld8 temp3=[temp1],16 // cr.isr
ld8 temp4=[temp2],16 // cr.ifa
;;
mov cr.isr=temp3
mov cr.ifa=temp4
ld8 temp3=[temp1],16 // cr.itir
ld8 temp4=[temp2],16 // cr.iipa
;;
mov cr.itir=temp3
mov cr.iipa=temp4
ld8 temp3=[temp1],16 // cr.iim
ld8 temp4=[temp2],16 // cr.iha
;;
mov cr.iim=temp3
mov cr.iha=temp4
dep r22=0,r22,62,1 // pal_min_state, physical, uncached
mov IA64_KR(CURRENT)=r21
ld8 r8=[temp1] // os_status
ld8 r10=[temp2] // context
/* Wire IA64_TR_CURRENT_STACK to the stack that we are resuming to. To
* avoid any dependencies on the algorithm in ia64_switch_to(), just
* purge any existing CURRENT_STACK mapping and insert the new one.
*
* r16 contains prev_IA64_KR_CURRENT_STACK, r21 contains
* prev_IA64_KR_CURRENT, these values may have been changed by the C
* code. Do not use r8, r9, r10, r22, they contain values ready for
* the return to SAL.
*/
mov r15=IA64_KR(CURRENT_STACK) // physical granule mapped by IA64_TR_CURRENT_STACK
;;
shl r15=r15,IA64_GRANULE_SHIFT
;;
dep r15=-1,r15,61,3 // virtual granule
mov r18=IA64_GRANULE_SHIFT<<2 // for cr.itir.ps
;;
ptr.d r15,r18
;;
srlz.d
extr.u r19=r21,61,3 // r21 = prev_IA64_KR_CURRENT
shl r20=r16,IA64_GRANULE_SHIFT // r16 = prev_IA64_KR_CURRENT_STACK
movl r21=PAGE_KERNEL // page properties
;;
mov IA64_KR(CURRENT_STACK)=r16
cmp.ne p6,p0=RGN_KERNEL,r19 // new stack is in the kernel region?
or r21=r20,r21 // construct PA | page properties
(p6) br.spnt 1f // the dreaded cpu 0 idle task in region 5:(
;;
mov cr.itir=r18
mov cr.ifa=r21
mov r20=IA64_TR_CURRENT_STACK
;;
itr.d dtr[r20]=r21
;;
srlz.d
1:
br.sptk b0
//EndStub//////////////////////////////////////////////////////////////////////
//++
// Name:
// ia64_new_stack()
//
// Stub Description:
//
// Switch to the MCA/INIT stack.
//
// r2 contains the return address, r3 contains either
// IA64_MCA_CPU_MCA_STACK_OFFSET or IA64_MCA_CPU_INIT_STACK_OFFSET.
//
// On entry RBS is still on the original stack, this routine switches RBS
// to use the MCA/INIT stack.
//
// On entry, sos->pal_min_state is physical, on exit it is virtual.
//
//--
ia64_new_stack:
add regs=MCA_PT_REGS_OFFSET, r3
add temp2=MCA_SOS_OFFSET+IA64_SAL_OS_STATE_PAL_MIN_STATE_OFFSET, r3
mov b0=r2 // save return address
GET_IA64_MCA_DATA(temp1)
invala
;;
add temp2=temp2, temp1 // struct ia64_sal_os_state.pal_min_state on MCA or INIT stack
add regs=regs, temp1 // struct pt_regs on MCA or INIT stack
;;
// Address of minstate area provided by PAL is physical, uncacheable.
// Convert to Linux virtual address in region 6 for C code.
ld8 ms=[temp2] // pal_min_state, physical
;;
dep temp1=-1,ms,62,2 // set region 6
mov temp3=IA64_RBS_OFFSET-MCA_PT_REGS_OFFSET
;;
st8 [temp2]=temp1 // pal_min_state, virtual
add temp4=temp3, regs // start of bspstore on new stack
;;
mov ar.bspstore=temp4 // switch RBS to MCA/INIT stack
;;
flushrs // must be first in group
br.sptk b0
//EndStub//////////////////////////////////////////////////////////////////////
//++
// Name:
// ia64_old_stack()
//
// Stub Description:
//
// Switch to the old stack.
//
// r2 contains the return address, r3 contains either
// IA64_MCA_CPU_MCA_STACK_OFFSET or IA64_MCA_CPU_INIT_STACK_OFFSET.
//
// On entry, pal_min_state is virtual, on exit it is physical.
//
// On entry RBS is on the MCA/INIT stack, this routine switches RBS
// back to the previous stack.
//
// The psr is set to all zeroes. SAL return requires either all zeroes or
// just psr.mc set. Leaving psr.mc off allows INIT to be issued if this
// code does not perform correctly.
//
// The dirty registers at the time of the event were flushed to the
// MCA/INIT stack in ia64_pt_regs_save(). Restore the dirty registers
// before reverting to the previous bspstore.
//--
ia64_old_stack:
add regs=MCA_PT_REGS_OFFSET, r3
mov b0=r2 // save return address
GET_IA64_MCA_DATA(temp2)
LOAD_PHYSICAL(p0,temp1,1f)
;;
mov cr.ipsr=r0
mov cr.ifs=r0
mov cr.iip=temp1
;;
invala
rfi
1:
add regs=regs, temp2 // struct pt_regs on MCA or INIT stack
;;
add temp1=PT(LOADRS), regs
;;
ld8 temp2=[temp1],PT(AR_BSPSTORE)-PT(LOADRS) // restore loadrs
;;
ld8 temp3=[temp1],PT(AR_RNAT)-PT(AR_BSPSTORE) // restore ar.bspstore
mov ar.rsc=temp2
;;
loadrs
ld8 temp4=[temp1] // restore ar.rnat
;;
mov ar.bspstore=temp3 // back to old stack
;;
mov ar.rnat=temp4
;;
br.sptk b0
//EndStub//////////////////////////////////////////////////////////////////////
//++
// Name:
// ia64_set_kernel_registers()
//
// Stub Description:
//
// Set the registers that are required by the C code in order to run on an
// MCA/INIT stack.
//
// r2 contains the return address, r3 contains either
// IA64_MCA_CPU_MCA_STACK_OFFSET or IA64_MCA_CPU_INIT_STACK_OFFSET.
//
//--
ia64_set_kernel_registers:
add temp3=MCA_SP_OFFSET, r3
add temp4=MCA_SOS_OFFSET+IA64_SAL_OS_STATE_OS_GP_OFFSET, r3
mov b0=r2 // save return address
GET_IA64_MCA_DATA(temp1)
;;
add temp4=temp4, temp1 // &struct ia64_sal_os_state.os_gp
add r12=temp1, temp3 // kernel stack pointer on MCA/INIT stack
add r13=temp1, r3 // set current to start of MCA/INIT stack
add r20=temp1, r3 // physical start of MCA/INIT stack
;;
ld8 r1=[temp4] // OS GP from SAL OS state
;;
DATA_PA_TO_VA(r1,temp1)
DATA_PA_TO_VA(r12,temp2)
DATA_PA_TO_VA(r13,temp3)
;;
mov IA64_KR(CURRENT)=r13
/* Wire IA64_TR_CURRENT_STACK to the MCA/INIT handler stack. To avoid
* any dependencies on the algorithm in ia64_switch_to(), just purge
* any existing CURRENT_STACK mapping and insert the new one.
*/
mov r16=IA64_KR(CURRENT_STACK) // physical granule mapped by IA64_TR_CURRENT_STACK
;;
shl r16=r16,IA64_GRANULE_SHIFT
;;
dep r16=-1,r16,61,3 // virtual granule
mov r18=IA64_GRANULE_SHIFT<<2 // for cr.itir.ps
;;
ptr.d r16,r18
;;
srlz.d
shr.u r16=r20,IA64_GRANULE_SHIFT // r20 = physical start of MCA/INIT stack
movl r21=PAGE_KERNEL // page properties
;;
mov IA64_KR(CURRENT_STACK)=r16
or r21=r20,r21 // construct PA | page properties
;;
mov cr.itir=r18
mov cr.ifa=r13
mov r20=IA64_TR_CURRENT_STACK
;;
itr.d dtr[r20]=r21
;;
srlz.d
br.sptk b0
//EndStub//////////////////////////////////////////////////////////////////////
#undef ms
#undef regs
#undef temp1
#undef temp2
#undef temp3
#undef temp4
// Support function for mca.c, it is here to avoid using inline asm. Given the
// address of an rnat slot, if that address is below the current ar.bspstore
// then return the contents of that slot, otherwise return the contents of
// ar.rnat.
GLOBAL_ENTRY(ia64_get_rnat)
alloc r14=ar.pfs,1,0,0,0
mov ar.rsc=0
;;
mov r14=ar.bspstore
;;
cmp.lt p6,p7=in0,r14
;;
(p6) ld8 r8=[in0]
(p7) mov r8=ar.rnat
mov ar.rsc=3
br.ret.sptk.many rp
END(ia64_get_rnat)