Leaked source code of windows server 2003
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

511 lines
15 KiB

// TITLE("Hibernation wake dispatcher")
//++
//
// Copyright (c) 1999 Intel Corporation
//
// Module Name:
//
// wakes.s
//
// Abstract:
//
//
// Author:
//
// Allen Kay ([email protected]) 8 June, 1999
//
// Environment:
//
// Firmware, OS Loader. Position-independent.
//
// Revision History:
//
//--
#include "ksia64.h"
#include "paldef.h"
.global HiberMapPage
.global HiberRemapPage
.global HiberWakeState
.global HiberFirstRemap
.global HiberLastRemap
.global HiberImagePageSelf
.global HiberBreakOnWake
// VOID
// WakeDispatcher(
// VOID
// )
//
// Routine description:
//
// This code performs the final stages of restarting the hibernation
// image. Pages that were loaded in temporary buffer space because
// the memory they belong in was in use by the firmware are copied
// to their final destination; an IMB is issued after this to ensure
// that the I-cache is coherent with any code that got copied; and
// necessary context is loaded into registers and NT is reentered.
//
// Because this code is part of the loader image that may be overwritten
// by this copy process, it must itself have been copied to a free
// page before it is executed. Note that because there is presently
// no mechanism for allocating multiple contiguous pages, this code cannot
// exceed one page (8K).
//
// Arguments:
//
// None.
//
// Return Value:
//
// Never returns.
LEAF_ENTRY(WakeDispatcherStartLocal)
.prologue
.regstk 1, 30, 2, 0
alloc t4 = ar.pfs, 1, 30, 2, 0
ARGPTR(a0)
rMapPage = loc11
rRemapPage = loc12
rWakeState = loc13
rPageCount = loc14
rPageSelf = loc15
rBreakOnWake = loc16
src1 = loc17
src2 = loc18
tmp = loc19
rKSEG0 = loc20
rpT0 = loc21
rpT1 = loc22
rpT2 = loc23
rpT3 = loc24
//
// Get variables into registers before copying any pages, as they may be
// overwritten.
//
movl rpT0 = HiberMapPage // pointer to source pages
movl rpT1 = HiberRemapPage // pointer to target pages
movl rpT2 = HiberWakeState // pointer to KPROCESSOR_STATE for
// restarting the hibernation image
;;
ld8 rMapPage = [rpT0] // load them
ld8 rRemapPage = [rpT1]
ld8 rWakeState = [rpT2]
;;
movl rpT0 = HiberFirstRemap // first page index
movl rpT1 = HiberLastRemap // last page index
movl rpT2 = HiberImagePageSelf // PFN where MemImage ends up
movl rpT3 = HiberBreakOnWake // "break on wake" flag
;;
ld4 t0 = [rpT0] // load them
ld4 t1 = [rpT1]
ld8 rPageSelf = [rpT2]
ld1 rBreakOnWake = [rpT3]
;;
sub rPageCount = t1, t0 // number of pages to copy
add rMapPage = t0, rMapPage // first source page
add rRemapPage = t0, rRemapPage // first target page
;;
cmp.eq pt1, pt0 = rPageCount, zero // nothing to copy
;;
(pt1) br.cond.spnt CopyDone
//
// Copy pages.
//
NextPage:
add rPageCount = -1, rPageCount // count page copied
movl rKSEG0 = KSEG0_BASE // physical -> KSEG0
ld8.fill t0 = [rMapPage], 8 // source page number
;;
shl t0 = t0, PAGE_SHIFT // page -> physical address
;;
add t0 = rKSEG0, t0 // physical -> KSEG0
ld8.fill t1 = [rRemapPage], 8 // destination address
;;
shl t1 = t1, PAGE_SHIFT // page -> physical address
;;
add t1 = rKSEG0, t1 // physical -> KSEG0
movl t2 = 1024 // 8KB = 1024 quadwords
;;
NextQuadWord:
add t2 = -1, t2 // count quadword
ld8.fill t3 = [t0], 8 // load a quadword
;;
st8.spill [t1] = t3, 8 // store it
;;
cmp.eq pt0, pt1 = t2, zero
;;
(pt1) br.cond.spnt NextQuadWord
cmp.eq pt0, pt1 = rPageCount, zero
;;
(pt1) br.cond.spnt NextPage
//
// All necessary pages have been copied. Check the break-on-wake flag,
// and change the signature NT will see when it wakes up if it was set.
//
CopyDone:
cmp.eq pt1, pt0 = rBreakOnWake, zero // no flag set, do nothing
;;
(pt1) br.cond.spnt SkipSigChange
shl rPageSelf = rPageSelf, PAGE_SHIFT // convert to physical
;;
add rPageSelf = rKSEG0, rPageSelf // make superpage address
movl t0 = 0x706b7262 // 'brkp'
;;
st4 [rPageSelf] = t0 // signature is first longword of MemImage
//
// Synchronize the I-cache, load essential NT context, and transfer control
// to the restored system.
//
SkipSigChange:
#if 0
PublicFunction(PalProc)
mov out0 = PAL_CACHE_FLUSH // call PAL cache flush routine
mov out1 = 1 // flush I-cache only
movl rpT0 = PalProc
;;
ld8 t0 = [rpT0]
;;
mov bt0 = t0
;;
br.call.spnt brp = bt0
#endif
//
// Restore context. Only the integer registers are restored; this code runs
// in the firmware environment, so floating point can't be used, and NT
// PALcode abstractions such as the PSR don't exist. It is the responsibility
// of NT's code that saves the hibernation context to put enough information
// in the integer registers in the CONTEXT to be able to finish restoring
// context to restart NT.
//
mov a0 = rWakeState // CONTEXT is the first thing
// in the KPROCESSOR_STATE
;;
//
// Restore all the registers.
//
add src1 = CxIntNats, a0
add src2 = CxPreds, a0
add tmp = CxIntGp, a0
;;
ld8.nt1 t17 = [src1], CxBrRp - CxIntNats
ld8.nt1 t16 = [src2], CxBrS0 - CxPreds
shr tmp = tmp, 3
;;
ld8.nt1 t0 = [src1], CxBrS1 - CxBrRp
ld8.nt1 t1 = [src2], CxBrS2 - CxBrS0
and tmp = 0x3f, tmp
;;
ld8.nt1 t2 = [src1], CxBrS3 - CxBrS1
ld8.nt1 t3 = [src2], CxBrS4 - CxBrS2
cmp4.ge pt1, pt0 = 1, tmp
;;
ld8.nt1 t4 = [src1], CxBrT0 - CxBrS3
ld8.nt1 t5 = [src2], CxBrT1 - CxBrS4
(pt1) sub loc5 = 1, tmp
;;
ld8.nt1 t6 = [src1], CxApUNAT - CxBrT0
ld8.nt1 t7 = [src2], CxApLC - CxBrT1
(pt0) add loc5 = -1, tmp
;;
ld8.nt1 loc0 = [src1], CxApEC - CxApUNAT
ld8.nt1 t8 = [src2], CxApCCV - CxApLC
(pt0) sub loc6 = 65, tmp
;;
ld8.nt1 t9 = [src1], CxApDCR - CxApEC
ld8.nt1 t10 = [src2], CxRsPFS - CxApCCV
(pt1) shr.u t17 = t17, loc5
;;
ld8.nt1 loc1 = [src1], CxRsBSP - CxApDCR
ld8.nt1 t11 = [src2], CxRsRSC - CxRsPFS
(pt0) shl loc7 = t17, loc5
;;
ld8.nt1 loc2 = [src1], CxStIIP - CxRsBSP
ld8.nt1 loc3 = [src2], CxStIFS - CxRsRSC
(pt0) shr.u loc8 = t17, loc6
;;
ld8.nt1 loc9 = [src1]
ld8.nt1 loc10 = [src2]
(pt0) or t17 = loc7, loc8
;;
mov ar.unat = t17
add src1 = CxFltS0, a0
shr t12 = loc2, 3
;;
add src2 = CxFltS1, a0
and t12 = 0x3f, t12 // current rnat save index
and t13 = 0x7f, loc10 // total frame size
;;
mov ar.ccv = t10
add t14 = t13, t12
mov ar.pfs = t11
;;
Rrc20:
cmp4.gt pt1, pt0 = 63, t14
;;
(pt0) add t14 = -63, t14
(pt0) add t13 = 1, t13
;;
nop.m 0
(pt1) shl t13 = t13, 3
(pt0) br.cond.spnt Rrc20
;;
add loc2 = loc2, t13
nop.f 0
mov pr = t16, -1
ldf.fill.nt1 fs0 = [src1], CxFltS2 - CxFltS0
ldf.fill.nt1 fs1 = [src2], CxFltS3 - CxFltS1
mov brp = t0
;;
ldf.fill.nt1 fs2 = [src1], CxFltT0 - CxFltS2
ldf.fill.nt1 fs3 = [src2], CxFltT1 - CxFltS3
mov bs0 = t1
;;
ldf.fill.nt1 ft0 = [src1], CxFltT2 - CxFltT0
ldf.fill.nt1 ft1 = [src2], CxFltT3 - CxFltT1
mov bs1 = t2
;;
ldf.fill.nt1 ft2 = [src1], CxFltT4 - CxFltT2
ldf.fill.nt1 ft3 = [src2], CxFltT5 - CxFltT3
mov bs2 = t3
;;
ldf.fill.nt1 ft4 = [src1], CxFltT6 - CxFltT4
ldf.fill.nt1 ft5 = [src2], CxFltT7 - CxFltT5
mov bs3 = t4
;;
ldf.fill.nt1 ft6 = [src1], CxFltT8 - CxFltT6
ldf.fill.nt1 ft7 = [src2], CxFltT9 - CxFltT7
mov bs4 = t5
;;
ldf.fill.nt1 ft8 = [src1], CxFltS4 - CxFltT8
ldf.fill.nt1 ft9 = [src2], CxFltS5 - CxFltT9
mov bt0 = t6
;;
ldf.fill.nt1 fs4 = [src1], CxFltS6 - CxFltS4
ldf.fill.nt1 fs5 = [src2], CxFltS7 - CxFltS5
mov bt1 = t7
;;
ldf.fill.nt1 fs6 = [src1], CxFltS8 - CxFltS6
ldf.fill.nt1 fs7 = [src2], CxFltS9 - CxFltS7
mov ar.lc = t8
;;
ldf.fill.nt1 fs8 = [src1], CxFltS10 - CxFltS8
ldf.fill.nt1 fs9 = [src2], CxFltS11 - CxFltS9
mov ar.ec = t9
;;
ldf.fill.nt1 fs10 = [src1], CxFltS12 - CxFltS10
ldf.fill.nt1 fs11 = [src2], CxFltS13 - CxFltS11
nop.i 0
;;
ldf.fill.nt1 fs12 = [src1], CxFltS14 - CxFltS12
ldf.fill.nt1 fs13 = [src2], CxFltS15 - CxFltS13
add loc6 = CxIntGp, a0
;;
ldf.fill.nt1 fs14 = [src1], CxFltS16 - CxFltS14
ldf.fill.nt1 fs15 = [src2], CxFltS17 - CxFltS15
add loc7 = CxIntT0, a0
;;
ldf.fill.nt1 fs16 = [src1], CxFltS18 - CxFltS16
ldf.fill.nt1 fs17 = [src2], CxFltS19 - CxFltS17
add t19 = CxRsRNAT, a0
;;
ldf.fill.nt1 fs18 = [src1]
ldf.fill.nt1 fs19 = [src2]
add t7 = CxStFPSR, a0
;;
ld8.nt1 loc8 = [t7] // load fpsr from context
ld8.nt1 loc5 = [t19] // load rnat from context
nop.i 0
ld8.fill.nt1 gp = [loc6], CxIntT1 - CxIntGp
ld8.fill.nt1 t0 = [loc7], CxIntS0 - CxIntT0
;;
ld8.fill.nt1 t1 = [loc6], CxIntS1 - CxIntT1
ld8.fill.nt1 s0 = [loc7], CxIntS2 - CxIntS0
;;
ld8.fill.nt1 s1 = [loc6], CxIntS3 - CxIntS1
ld8.fill.nt1 s2 = [loc7], CxIntV0 - CxIntS2
;;
ld8.fill.nt1 s3 = [loc6], CxIntTeb - CxIntS3
ld8.fill.nt1 v0 = [loc7], CxIntT2 - CxIntV0
;;
ld8.fill.nt1 teb = [loc6], CxIntT3 - CxIntTeb
ld8.fill.nt1 t2 = [loc7], CxIntSp - CxIntT2
;;
ld8.fill.nt1 t3 = [loc6], CxIntT4 - CxIntT3
ld8.fill.nt1 loc4 = [loc7], CxIntT5 - CxIntSp
;;
ld8.fill.nt1 t4 = [loc6], CxIntT6 - CxIntT4
ld8.fill.nt1 t5 = [loc7], CxIntT7 - CxIntT5
;;
ld8.fill.nt1 t6 = [loc6], CxIntT8 - CxIntT6
ld8.fill.nt1 t7 = [loc7], CxIntT9 - CxIntT7
;;
ld8.fill.nt1 t8 = [loc6], CxIntT10 - CxIntT8
ld8.fill.nt1 t9 = [loc7], CxIntT11 - CxIntT9
;;
ld8.fill.nt1 t10 = [loc6], CxIntT12 - CxIntT10
ld8.fill.nt1 t11 = [loc7], CxIntT13 - CxIntT11
;;
ld8.fill.nt1 t12 = [loc6], CxIntT14 - CxIntT12
ld8.fill.nt1 t13 = [loc7], CxIntT15 - CxIntT13
;;
ld8.fill.nt1 t14 = [loc6], CxIntT16 - CxIntT14
ld8.fill.nt1 t15 = [loc7], CxIntT17 - CxIntT15
;;
ld8.fill.nt1 t16 = [loc6], CxIntT18 - CxIntT16
ld8.fill.nt1 t17 = [loc7], CxIntT19 - CxIntT17
;;
ld8.fill.nt1 t18 = [loc6], CxIntT20 - CxIntT18
ld8.fill.nt1 t19 = [loc7], CxIntT21 - CxIntT19
;;
ld8.fill.nt1 t20 = [loc6], CxIntT22 - CxIntT20
ld8.fill.nt1 t21 = [loc7]
;;
rsm 1 << PSR_I
ld8.fill.nt1 t22 = [loc6]
;;
rsm 1 << PSR_IC
movl t0 = 1 << IFS_V
;;
mov ar.fpsr = loc8 // set fpsr
mov ar.unat = loc0
;;
srlz.d
or loc10 = t0, loc10 // set ifs valid bit
;;
mov cr.iip = loc9
mov cr.ifs = loc10
bsw.0
;;
mov cr.dcr = loc1
mov r17 = loc2 // put BSP in a shadow reg
or r16 = 0x3, loc3 // put RSE in eager mode
mov ar.rsc = r0 // put RSE in enforced lazy
nop.m 0
add r20 = CxStIPSR, a0
;;
ld8.nt1 r20 = [r20] // load IPSR
mov r18 = loc4 // put SP in a shadow reg
mov r19 = loc5 // put RNaTs in a shadow reg
;;
alloc t0 = 0, 0, 0, 0
mov cr.ipsr = r20
mov sp = r18
;;
loadrs
;;
mov ar.bspstore = r17
nop.i 0
;;
mov ar.rnat = r19 // set rnat register
mov ar.rsc = r16 // restore RSC
bsw.1
;;
invala
nop.i 0
rfi
;;
//
// This label is used to determine the size of the wake dispatcher code in the
// process of copying it to a free page.
//
WakeDispatcherEndLocal::
LEAF_EXIT(WakeDispatcherEndLocal)
.sdata
WakeDispatcherStart::
data4 @secrel(WakeDispatcherStartLocal)
WakeDispatcherEnd::
data4 @secrel(WakeDispatcherEndLocal)