/* * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. * * @APPLE_LICENSE_HEADER_START@ * * The contents of this file constitute Original Code as defined in and * are subject to the Apple Public Source License Version 1.1 (the * "License"). You may not use this file except in compliance with the * License. Please obtain a copy of the License at * http://www.apple.com/publicsource and read it before using this file. * * This Original Code and all software distributed under the License are * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the * License for the specific language governing rights and limitations * under the License. * * @APPLE_LICENSE_HEADER_END@ */ #include #include #include #include #include #include #include /* PCI config cycle probing * * boolean_t ml_probe_read(vm_offset_t paddr, unsigned int *val) * * Read the memory location at physical address paddr. * This is a part of a device probe, so there is a good chance we will * have a machine check here. So we have to be able to handle that. * We assume that machine checks are enabled both in MSR and HIDs */ ; Force a line boundry here .align 5 .globl EXT(ml_probe_read) LEXT(ml_probe_read) mfsprg r9,2 ; Get feature flags mfmsr r0 ; Save the current MSR neg r10,r3 ; Number of bytes to end of page rlwinm r2,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions rlwinm. r10,r10,0,20,31 ; Clear excess junk and test for page bndry mr r12,r3 ; Save the load address mtcrf 0x04,r9 ; Set the features cmplwi cr1,r10,4 ; At least 4 bytes left in page? rlwinm r2,r2,0,MSR_DR_BIT+1,MSR_IR_BIT-1 ; Clear translation beq- mprdoit ; We are right on the boundary... li r3,0 bltlr- cr1 ; No, just return failure... mprdoit: bt pfNoMSRirb,mprNoMSR ; No MSR... mtmsr r2 ; Translation and all off isync ; Toss prefetch b mprNoMSRx mprNoMSR: mr r5,r0 li r0,loadMSR ; Get the MSR setter SC mr r3,r2 ; Get new MSR sc ; Set it mr r0,r5 li r3,0 mprNoMSRx: ; ; We need to insure that there is no more than 1 BAT register that ; can get a hit. There could be repercussions beyond the ken ; of mortal man. It is best not to tempt fate. ; li r10,0 ; Clear a register mfdbatu r5,0 ; Save DBAT 0 high mfdbatl r6,0 ; Save DBAT 0 low mfdbatu r7,1 ; Save DBAT 1 high mfdbatu r8,2 ; Save DBAT 2 high mfdbatu r9,3 ; Save DBAT 3 high sync ; Make sure all is well mtdbatu 1,r10 ; Invalidate DBAT 1 mtdbatu 2,r10 ; Invalidate DBAT 2 mtdbatu 3,r10 ; Invalidate DBAT 3 rlwinm r10,r12,0,0,14 ; Round down to a 128k boundary ori r11,r10,0x32 ; Set uncached, coherent, R/W ori r10,r10,2 ; Make the upper half (128k, valid supervisor) mtdbatl 0,r11 ; Set lower BAT first mtdbatu 0,r10 ; Now the upper sync ; Just make sure ori r11,r2,lo16(MASK(MSR_DR)) ; Turn on data translation mtmsr r11 ; Do it for real isync ; Make sure of it eieio ; Make sure of all previous accesses sync ; Make sure it is all caught up lwz r11,0(r12) ; Get it and maybe machine check here eieio ; Make sure of ordering again sync ; Get caught up yet again isync ; Do not go further till we are here mtdbatu 0,r5 ; Restore DBAT 0 high mtdbatl 0,r6 ; Restore DBAT 0 low mtdbatu 1,r7 ; Restore DBAT 1 high mtdbatu 2,r8 ; Restore DBAT 2 high mtdbatu 3,r9 ; Restore DBAT 3 high sync li r3,1 ; We made it mtmsr r0 ; Restore translation and exceptions isync ; Toss speculations stw r11,0(r4) ; Save the loaded value blr ; Return... ; Force a line boundry here. This means we will be able to check addresses better .align 5 .globl EXT(ml_probe_read_mck) LEXT(ml_probe_read_mck) /* Read physical address * * unsigned int ml_phys_read_byte(vm_offset_t paddr) * * Read the byte at physical address paddr. Memory should not be cache inhibited. */ ; Force a line boundry here .align 5 .globl EXT(ml_phys_read_byte) LEXT(ml_phys_read_byte) mfmsr r10 ; Save the current MSR rlwinm r4,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions rlwinm r4,r4,0,MSR_DR_BIT+1,MSR_DR_BIT-1 ; Clear translation mtmsr r4 ; Translation and all off isync ; Toss prefetch lbz r3,0(r3) ; Get the byte sync mtmsr r10 ; Restore translation and rupts isync blr /* Read physical address * * unsigned int ml_phys_read(vm_offset_t paddr) * * Read the word at physical address paddr. Memory should not be cache inhibited. */ ; Force a line boundry here .align 5 .globl EXT(ml_phys_read) LEXT(ml_phys_read) mfmsr r0 ; Save the current MSR rlwinm r4,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions rlwinm r4,r4,0,MSR_DR_BIT+1,MSR_DR_BIT-1 ; Clear translation mtmsr r4 ; Translation and all off isync ; Toss prefetch lwz r3,0(r3) ; Get the word sync mtmsr r0 ; Restore translation and rupts isync blr /* Write physical address byte * * void ml_phys_write_byte(vm_offset_t paddr, unsigned int data) * * Write the byte at physical address paddr. Memory should not be cache inhibited. */ ; Force a line boundry here .align 5 .globl EXT(ml_phys_write_byte) LEXT(ml_phys_write_byte) mfmsr r0 ; Save the current MSR rlwinm r5,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions rlwinm r5,r5,0,MSR_DR_BIT+1,MSR_DR_BIT-1 ; Clear translation mtmsr r5 ; Translation and all off isync ; Toss prefetch stb r4,0(r3) ; Set the byte sync mtmsr r0 ; Restore translation and rupts isync blr /* Write physical address * * void ml_phys_write(vm_offset_t paddr, unsigned int data) * * Write the word at physical address paddr. Memory should not be cache inhibited. */ ; Force a line boundry here .align 5 .globl EXT(ml_phys_write) LEXT(ml_phys_write) mfmsr r0 ; Save the current MSR rlwinm r5,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear interruptions rlwinm r5,r5,0,MSR_DR_BIT+1,MSR_DR_BIT-1 ; Clear translation mtmsr r5 ; Translation and all off isync ; Toss prefetch stw r4,0(r3) ; Set the word sync mtmsr r0 ; Restore translation and rupts isync blr /* set interrupts enabled or disabled * * boolean_t set_interrupts_enabled(boolean_t enable) * * Set EE bit to "enable" and return old value as boolean */ ; Force a line boundry here .align 5 .globl EXT(set_interrupts_enabled) LEXT(set_interrupts_enabled) mfmsr r5 ; Get the current MSR mr r4,r3 ; Save the old value rlwinm r3,r5,17,31,31 ; Set return value rlwimi r5,r4,15,16,16 ; Insert new EE bit andi. r7,r5,lo16(MASK(MSR_EE)) ; Interruptions bne CheckPreemption NoPreemption: mtmsr r5 ; Slam enablement blr CheckPreemption: mfsprg r7,0 lwz r8,PP_NEED_AST(r7) lwz r7,PP_CPU_DATA(r7) li r6,AST_URGENT lwz r8,0(r8) lwz r7,CPU_PREEMPTION_LEVEL(r7) lis r0,HIGH_ADDR(DoPreemptCall) and. r8,r8,r6 ori r0,r0,LOW_ADDR(DoPreemptCall) beq+ NoPreemption cmpi cr0, r7, 0 bne+ NoPreemption sc mtmsr r5 blr /* Set machine into idle power-saving mode. * * void machine_idle_ppc(void) * * We will use the PPC NAP or DOZE for this. * This call always returns. Must be called with spllo (i.e., interruptions * enabled). * */ ; Force a line boundry here .align 5 .globl EXT(machine_idle_ppc) LEXT(machine_idle_ppc) mfmsr r3 ; Get the current MSR rlwinm r5,r3,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Turn off interruptions mtmsr r5 ; Hold up interruptions for now mfsprg r12,0 ; Get the per_proc_info mfspr r6,hid0 ; Get the current power-saving mode mfsprg r11,2 ; Get CPU specific features rlwinm r6,r6,0,sleep+1,doze-1 ; Clear all possible power-saving modes (not DPM though) mtcrf 0xC7,r11 ; Get the facility flags lis r4,hi16(napm) ; Assume we can nap bt pfWillNapb,yesnap ; Yeah, nap is ok... lis r4,hi16(dozem) ; Assume we can doze bt pfCanDozeb,yesnap ; We can sleep or doze one this machine... ori r3,r3,lo16(MASK(MSR_EE)) ; Flip on EE mtmsr r3 ; Turn interruptions back on blr ; Leave... yesnap: mftbu r9 ; Get the upper timebase mftb r7 ; Get the lower timebase mftbu r8 ; Get the upper one again cmplw r9,r8 ; Did the top tick? bne- yesnap ; Yeah, need to get it again... stw r8,napStamp(r12) ; Set high order time stamp stw r7,napStamp+4(r12) ; Set low order nap stamp bf pfL1nncb,minoflushl1 ; The L1 is coherent in nap/doze... ; ; 7450 does not keep L1 cache coherent across nap/sleep it must alwasy flush. ; It does not have a L1 flush assist, so we do not test for it here. ; ; Note that the time stamp take above is not completely accurate for 7450 ; because we are about to flush the L1 cache and that takes a bit of time. ; cror cr0_eq,pfL1ib,pfL1db ; Check for either I- or D-cache bf- cr0_eq,minoflushl1 ; No level 1 to flush... rlwinm. r0,r4,0,ice,dce ; Were either of the level 1s on? beq- minoflushl1 ; No, no need to flush... miswdl1: lwz r0,pfl1dSize(r12) ; Get the level 1 cache size rlwinm r2,r0,0,1,30 ; Double it add r0,r0,r2 ; Get 3 times cache size rlwinm r2,r5,0,MSR_DR_BIT+1,MSR_DR_BIT-1 ; Turn off data translation rlwinm r0,r0,26,6,31 ; Get 3/2 number of cache lines lis r3,0xFFF0 ; Dead recon ROM address for now mtctr r0 ; Number of lines to flush mtmsr r2 ; Do it isync miswfldl1a: lwz r2,0(r3) ; Flush anything else addi r3,r3,32 ; Next line bdnz miswfldl1a ; Flush the lot... miinvdl1: sync ; Make sure all flushes have been committed mtmsr r5 ; Put back data translation isync mfspr r8,hid0 ; Get the HID0 bits li r7,lo16(icem|dcem) ; Get the cache enable bits andc r8,r8,r7 ; Clear cache enables mtspr hid0,r8 ; and turn off L1 cache sync ; Make sure all is done ori r8,r8,lo16(icfim|dcfim) ; Set the HID0 bits for invalidate sync isync mtspr hid0,r8 ; Start the invalidate sync minoflushl1: ; ; We have to open up interruptions here because book 4 says that we should ; turn on only the POW bit and that we should have interrupts enabled ; The interrupt handler will detect that nap or doze is set if an interrupt ; is taken and set everything up to return directly to machine_idle_ret. ; So, make sure everything we need there is already set up... ; ori r7,r5,lo16(MASK(MSR_EE)) ; Flip on EE or r6,r6,r4 ; Set nap or doze oris r5,r7,hi16(MASK(MSR_POW)) ; Turn on power management in next MSR mtspr hid0,r6 ; Set up the HID for nap/doze isync ; Make sure it is set mtmsr r7 ; Enable for interrupts rlwinm. r11,r11,0,pfAltivecb,pfAltivecb ; Do we have altivec? beq- minovec ; No... dssall ; Stop the streams before we nap/doze minovec: sync ; Make sure queues are clear mtmsr r5 ; Nap or doze isync ; Make sure this takes before we proceed b minovec ; loop if POW does not take ; ; Note that the interrupt handler will turn off the nap/doze bits in the hid. ; Also remember that the interrupt handler will force return to here whenever ; the nap/doze bits are set. ; .globl EXT(machine_idle_ret) LEXT(machine_idle_ret) mtmsr r7 ; Make sure the MSR is what we want isync ; In case we turn on translation blr ; Return... /* Put machine to sleep. * This call never returns. We always exit sleep via a soft reset. * All external interruptions must be drained at this point and disabled. * * void ml_ppc_sleep(void) * * We will use the PPC SLEEP for this. * * There is one bit of hackery in here: we need to enable for * interruptions when we go to sleep and there may be a pending * decrimenter rupt. So we make the decrimenter 0x7FFFFFFF and enable for * interruptions. The decrimenter rupt vector recognizes this and returns * directly back here. * */ ; Force a line boundry here .align 5 .globl EXT(ml_ppc_sleep) LEXT(ml_ppc_sleep) #if 0 mfmsr r5 ; Hack to spin instead of sleep rlwinm r5,r5,0,MSR_DR_BIT+1,MSR_IR_BIT-1 ; Turn off translation rlwinm r5,r5,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Turn off interruptions mtmsr r5 ; No talking isync ; No interrupts allowed after we get the savearea mfsprg r6,0 ; Get the per_proc mfsprg r7,1 ; Get the pending savearea stw r7,savedSave(r6) ; Save the savearea for when we wake up deadsleep: addi r3,r3,1 ; Make analyzer happy addi r3,r3,1 addi r3,r3,1 b deadsleep ; Die the death of 1000 joys... #endif mfsprg r12,0 ; Get the per_proc_info mfspr r4,hid0 ; Get the current power-saving mode eqv r10,r10,r10 ; Get all foxes mfsprg r11,2 ; Get CPU specific features mfmsr r5 ; Get the current MSR rlwinm r10,r10,0,1,31 ; Make 0x7FFFFFFF rlwinm r4,r4,0,sleep+1,doze-1 ; Clear all possible power-saving modes (not DPM though) mtdec r10 ; Load decrimenter with 0x7FFFFFFF isync ; and make sure, mfdec r9 ; really sure, it gets there mtcrf 0x07,r11 ; Get the cache flags, etc oris r4,r4,hi16(sleepm) ; Set sleep rlwinm r5,r5,0,MSR_DR_BIT+1,MSR_IR_BIT-1 ; Turn off translation ; ; Note that we need translation off before we set the HID to sleep. Otherwise ; we will ignore any PTE misses that occur and cause an infinite loop. ; bt pfNoMSRirb,mpsNoMSR ; No MSR... mtmsr r5 ; Translation off isync ; Toss prefetch b mpsNoMSRx mpsNoMSR: li r0,loadMSR ; Get the MSR setter SC mr r3,r5 ; Get new MSR sc ; Set it mpsNoMSRx: ori r3,r5,lo16(MASK(MSR_EE)) ; Flip on EE sync mtspr hid0,r4 ; Set up the HID to sleep mtmsr r3 ; Enable for interrupts to drain decrimenter add r6,r4,r5 ; Just waste time add r6,r6,r4 ; A bit more add r6,r6,r5 ; A bit more mtmsr r5 ; Interruptions back off isync ; Toss prefetch mfsprg r7,1 ; Get the pending savearea stw r7,savedSave(r12) ; Save the savearea for when we wake up ; ; We are here with translation off, interrupts off, all possible ; interruptions drained off, and a decrimenter that will not pop. ; bl EXT(cacheInit) ; Clear out the caches. This will leave them on bl EXT(cacheDisable) ; Turn off all caches mfmsr r5 ; Get the current MSR oris r5,r5,hi16(MASK(MSR_POW)) ; Turn on power management in next MSR ; Leave EE off because power goes off shortly slSleepNow: sync ; Sync it all up mtmsr r5 ; Do sleep with interruptions enabled isync ; Take a pill b slSleepNow ; Go back to sleep if we wake up... /* Initialize all caches including the TLBs * * void cacheInit(void) * * This is used to force the caches to an initial clean state. First, we * check if the cache is on, if so, we need to flush the contents to memory. * Then we invalidate the L1. Next, we configure and invalidate the L2 etc. * Finally we turn on all of the caches * * Note that if translation is not disabled when this is called, the TLB will not * be completely clear after return. * */ ; Force a line boundry here .align 5 .globl EXT(cacheInit) LEXT(cacheInit) mfsprg r12,0 ; Get the per_proc_info mfspr r9,hid0 ; Get the current power-saving mode mfsprg r11,2 ; Get CPU specific features mfmsr r7 ; Get the current MSR rlwinm r4,r9,0,dpm+1,doze-1 ; Clear all possible power-saving modes (also disable DPM) rlwimi r11,r11,pfL23lckb+1,31,31 ; Move pfL23lck to another position (to keep from using non-volatile CRs) rlwinm r5,r7,0,MSR_DR_BIT+1,MSR_IR_BIT-1 ; Turn off translation rlwinm r5,r5,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Turn off interruptions mtcrf 0x87,r11 ; Get the feature flags mtspr hid0,r4 ; Set up the HID bt pfNoMSRirb,ciNoMSR ; No MSR... mtmsr r5 ; Translation and all off isync ; Toss prefetch b ciNoMSRx ciNoMSR: li r0,loadMSR ; Get the MSR setter SC mr r3,r5 ; Get new MSR sc ; Set it ciNoMSRx: bf pfAltivecb,cinoDSS ; No Altivec here... dssall ; Stop streams sync cinoDSS: lis r5,hi16(EXT(tlb_system_lock)) ; Get the TLBIE lock li r0,128 ; Get number of TLB entries ori r5,r5,lo16(EXT(tlb_system_lock)) ; Grab up the bottom part li r6,0 ; Start at 0 lwarx r2,0,r5 ; ? citlbhang: lwarx r2,0,r5 ; Get the TLBIE lock mr. r2,r2 ; Is it locked? bne- citlbhang ; It is locked, go wait... stwcx. r0,0,r5 ; Try to get it bne- citlbhang ; We was beat... mtctr r0 ; Set the CTR cipurgeTLB: tlbie r6 ; Purge this entry addi r6,r6,4096 ; Next page bdnz cipurgeTLB ; Do them all... mtcrf 0x80,r11 ; Set SMP capability sync ; Make sure all TLB purges are done eieio ; Order, order in the court bf pfSMPcapb,cinoSMP ; SMP incapable... tlbsync ; Sync all TLBs sync cinoSMP: stw r2,0(r5) ; Unlock TLBIE lock cror cr0_eq,pfL1ib,pfL1db ; Check for either I- or D-cache bf- cr0_eq,cinoL1 ; No level 1 to flush... rlwinm. r0,r9,0,ice,dce ; Were either of the level 1s on? beq- cinoL1 ; No, no need to flush... bf pfL1fab,ciswdl1 ; If no hw flush assist, go do by software... mfspr r8,msscr0 ; Get the memory system control register oris r8,r8,hi16(dl1hwfm) ; Turn on the hardware flush request mtspr msscr0,r8 ; Start the flush operation ciwdl1f: mfspr r8,msscr0 ; Get the control register again rlwinm. r8,r8,0,dl1hwf,dl1hwf ; Has the flush request been reset yet? bne ciwdl1f ; No, flush is still in progress... b ciinvdl1 ; Go invalidate l1... ; ; We need to either make this very complicated or to use ROM for ; the flush. The problem is that if during the following sequence a ; snoop occurs that invalidates one of the lines in the cache, the ; PLRU sequence will be altered making it possible to miss lines ; during the flush. So, we either need to dedicate an area of RAM ; to each processor, lock use of a RAM area, or use ROM. ROM is ; by far the easiest. Note that this is not an issue for machines ; that have harware flush assists. ; ciswdl1: lwz r0,pfl1dSize(r12) ; Get the level 1 cache size rlwinm r2,r0,0,1,30 ; Double it add r0,r0,r2 ; Get 3 times cache size rlwinm r0,r0,26,6,31 ; Get 3/2 number of cache lines lis r3,0xFFF0 ; Dead recon ROM address for now mtctr r0 ; Number of lines to flush ciswfldl1a: lwz r2,0(r3) ; Flush anything else addi r3,r3,32 ; Next line bdnz ciswfldl1a ; Flush the lot... ciinvdl1: sync ; Make sure all flushes have been committed mfspr r8,hid0 ; Get the HID0 bits rlwinm r8,r8,0,dce+1,ice-1 ; Clear cache enables mtspr hid0,r8 ; and turn off L1 cache sync ; Make sure all is done ori r8,r8,lo16(icem|dcem|icfim|dcfim) ; Set the HID0 bits for enable, and invalidate sync isync mtspr hid0,r8 ; Start the invalidate and turn on cache rlwinm r8,r8,0,dcfi+1,icfi-1 ; Turn off the invalidate bits mtspr hid0,r8 ; Turn off the invalidate (needed for some older machines) sync cinoL1: ; ; Flush and disable the level 2 ; bf pfL2b,cinol2 ; No level 2 cache to flush mfspr r8,l2cr ; Get the L2CR lwz r3,pfl2cr(r12) ; Get the L2CR value lis r0,hi16(l2sizm|l2clkm|l2ramm|l2ohm) ; Get confiuration bits xor r2,r8,r3 ; Get changing bits? ori r0,r0,lo16(l2slm|l2dfm|l2bypm) ; More config bits and. r0,r0,r2 ; Did any change? bne- ciinvdl2 ; Yes, just invalidate and get PLL synced... bf pfL2fab,ciswfl2 ; Flush not in hardware... mr r10,r3 ; Take a copy now bf 31,cinol2lck ; Skip if pfL23lck not set... oris r10,r10,hi16(l2ionlym|l2donlym) ; Set both instruction- and data-only sync mtspr l2cr,r10 ; Lock out the cache sync isync cinol2lck: ori r10,r10,lo16(l2hwfm) ; Request flush sync ; Make sure everything is done mtspr l2cr,r10 ; Request flush cihwfl2: mfspr r10,l2cr ; Get back the L2CR rlwinm. r10,r10,0,l2hwf,l2hwf ; Is the flush over? bne+ cihwfl2 ; Nope, keep going... b ciinvdl2 ; Flush done, go invalidate L2... ciswfl2: lwz r0,pfl2Size(r12) ; Get the L2 size oris r2,r3,hi16(l2dom) ; Set L2 to data only mode mtspr l2cr,r2 ; Go into data only mode sync ; Clean it up rlwinm r0,r0,27,5,31 ; Get the number of lines lis r10,0xFFF0 ; Dead recon ROM for now mtctr r0 ; Set the number of lines ciswfldl2a: lwz r0,0(r10) ; Load something to flush something addi r10,r10,32 ; Next line bdnz ciswfldl2a ; Do the lot... ciinvdl2: rlwinm r3,r3,0,l2e+1,31 ; Clear the enable bit b cinla ; Branch to next line... .align 5 cinlc: mtspr l2cr,r3 ; Disable L2 sync isync b ciinvl2 ; It is off, go invalidate it... cinla: b cinlb ; Branch to next... cinlb: sync ; Finish memory stuff isync ; Stop speculation b cinlc ; Jump back up and turn off cache... ciinvl2: sync isync oris r2,r3,hi16(l2im) ; Get the invalidate flag set mtspr l2cr,r2 ; Start the invalidate sync isync ciinvdl2a: mfspr r2,l2cr ; Get the L2CR bf pfL2ib,ciinvdl2b ; Flush not in hardware... rlwinm. r2,r2,0,l2i,l2i ; Is the invalidate still going? bne+ ciinvdl2a ; Assume so, this will take a looong time... sync b cinol2 ; No level 2 cache to flush ciinvdl2b: rlwinm. r2,r2,0,l2ip,l2ip ; Is the invalidate still going? bne+ ciinvdl2a ; Assume so, this will take a looong time... sync mtspr l2cr,r3 ; Turn off the invalidate request cinol2: ; ; Flush and enable the level 3 ; bf pfL3b,cinol3 ; No level 3 cache to flush mfspr r8,l3cr ; Get the L3CR lwz r3,pfl3cr(r12) ; Get the L3CR value lis r0,hi16(l3pem|l3sizm|l3dxm|l3clkm|l3spom|l3ckspm) ; Get configuration bits xor r2,r8,r3 ; Get changing bits? ori r0,r0,lo16(l3pspm|l3repm|l3rtm|l3cyam|l3dmemm|l3dmsizm) ; More config bits and. r0,r0,r2 ; Did any change? bne- ciinvdl3 ; Yes, just invalidate and get PLL synced... sync ; 7450 book says do this even though not needed mr r10,r3 ; Take a copy now bf 31,cinol3lck ; Skip if pfL23lck not set... oris r10,r10,hi16(l3iom) ; Set instruction-only ori r10,r10,lo16(l3donlym) ; Set data-only sync mtspr l3cr,r10 ; Lock out the cache sync isync cinol3lck: ori r10,r10,lo16(l3hwfm) ; Request flush sync ; Make sure everything is done mtspr l3cr,r10 ; Request flush cihwfl3: mfspr r10,l3cr ; Get back the L3CR rlwinm. r10,r10,0,l3hwf,l3hwf ; Is the flush over? bne+ cihwfl3 ; Nope, keep going... ciinvdl3: rlwinm r3,r3,0,l3e+1,31 ; Clear the enable bit sync ; Make sure of life, liberty, and justice mtspr l3cr,r3 ; Disable L3 sync ori r3,r3,lo16(l3im) ; Get the invalidate flag set mtspr l3cr,r3 ; Start the invalidate ciinvdl3b: mfspr r3,l3cr ; Get the L3CR rlwinm. r3,r3,0,l3i,l3i ; Is the invalidate still going? bne+ ciinvdl3b ; Assume so... sync mfspr r3,l3pdet ; ? rlwimi r3,r3,28,0,23 ; ? oris r3,r3,0xF000 ; ? ori r3,r3,0x0080 ; ? mtspr l3pdet,r3 ; ? isync mfspr r3,l3cr ; Get the L3CR rlwinm r3,r3,0,l3clken+1,l3clken-1 ; Clear the clock enable bit mtspr l3cr,r3 ; Disable the clock li r2,128 ; ? ciinvdl3c: addi r2,r2,-1 ; ? cmplwi r2,0 ; ? bne+ ciinvdl3c mfspr r10,msssr0 ; ? rlwinm r10,r10,0,vgL3TAG+1,vgL3TAG-1 ; ? mtspr msssr0,r10 ; ? sync oris r3,r3,hi16(l3em|l3clkenm) ; Turn on enable bit mtspr l3cr,r3 ; Enable it sync cinol3: bf pfL2b,cinol2a ; No level 2 cache to enable lwz r3,pfl2cr(r12) ; Get the L2CR value oris r3,r3,hi16(l2em) ; Turn on enable bit mtspr l2cr,r3 ; Enable it sync ; ; Invalidate and turn on L1s ; cinol2a: rlwinm r8,r9,0,dce+1,ice-1 ; Clear the I- and D- cache enables mtspr hid0,r8 ; Turn off dem caches sync ori r8,r9,lo16(icem|dcem|icfim|dcfim) ; Set the HID0 bits for enable, and invalidate rlwinm r9,r8,0,dcfi+1,icfi-1 ; Turn off the invalidate bits sync isync mtspr hid0,r8 ; Start the invalidate and turn on L1 cache mtspr hid0,r9 ; Turn off the invalidate (needed for some older machines) sync mtmsr r7 ; Restore MSR to entry isync blr ; Return... /* Disables all caches * * void cacheDisable(void) * * Turns off all caches on the processor. They are not flushed. * */ ; Force a line boundry here .align 5 .globl EXT(cacheDisable) LEXT(cacheDisable) mfsprg r11,2 ; Get CPU specific features mtcrf 0x83,r11 ; Set feature flags bf pfAltivecb,cdNoAlt ; No vectors... dssall ; Stop streams cdNoAlt: sync mfspr r5,hid0 ; Get the hid rlwinm r5,r5,0,dce+1,ice-1 ; Clear the I- and D- cache enables mtspr hid0,r5 ; Turn off dem caches sync bf pfL2b,cdNoL2 ; Skip if no L2... mfspr r5,l2cr ; Get the L2 rlwinm r5,r5,0,l2e+1,31 ; Turn off enable bit b cinlaa ; Branch to next line... .align 5 cinlcc: mtspr l2cr,r5 ; Disable L2 sync isync b cdNoL2 ; It is off, we are done... cinlaa: b cinlbb ; Branch to next... cinlbb: sync ; Finish memory stuff isync ; Stop speculation b cinlcc ; Jump back up and turn off cache... cdNoL2: bf pfL3b,cdNoL3 ; Skip down if no L3... mfspr r5,l3cr ; Get the L3 rlwinm r5,r5,0,l3e+1,31 ; Turn off enable bit rlwinm r5,r5,0,l3clken+1,l3clken-1 ; Turn off cache enable bit mtspr l3cr,r5 ; Disable the caches sync cdNoL3: blr ; Leave... /* Initialize processor thermal monitoring * void ml_thrm_init(void) * * Build initial TAU registers and start them all going. * We ca not do this at initial start up because we need to have the processor frequency first. * And just why is this in assembler when it does not have to be?? Cause I am just too * lazy to open up a "C" file, thats why. */ ; Force a line boundry here .align 5 .globl EXT(ml_thrm_init) LEXT(ml_thrm_init) mfsprg r12,0 ; Get the per_proc blok lis r11,hi16(EXT(gPEClockFrequencyInfo)) ; Get top of processor information mfsprg r10,2 ; Get CPU specific features ori r11,r11,lo16(EXT(gPEClockFrequencyInfo)) ; Get bottom of processor information mtcrf 0x40,r10 ; Get the installed features li r3,lo16(thrmtidm|thrmvm) ; Set for lower-than thermal event at 0 degrees bflr pfThermalb ; No thermal monitoring on this cpu mtspr thrm1,r3 ; Do it lwz r3,thrmthrottleTemp(r12) ; Get our throttle temprature rlwinm r3,r3,31-thrmthre,thrmthrs,thrmthre ; Position it ori r3,r3,lo16(thrmvm) ; Set for higher-than event mtspr thrm2,r3 ; Set it lis r4,hi16(1000000) ; Top of million ; ; Note: some CPU manuals say this is processor clocks, some say bus rate. The latter ; makes more sense because otherwise we can not get over about 400MHz. #if 0 lwz r3,PECFIcpurate(r11) ; Get the processor speed #else lwz r3,PECFIbusrate(r11) ; Get the bus speed #endif ori r4,r4,lo16(1000000) ; Bottom of million lis r7,hi16(thrmsitvm>>1) ; Get top of highest possible value divwu r3,r3,r4 ; Get number of cycles per microseconds ori r7,r7,lo16(thrmsitvm>>1) ; Get the bottom of the highest possible value addi r3,r3,1 ; Insure we have enough mulli r3,r3,20 ; Get 20 microseconds worth of cycles cmplw r3,r7 ; Check against max ble+ smallenuf ; It is ok... mr r3,r7 ; Saturate smallenuf: rlwinm r3,r3,31-thrmsitve,thrmsitvs,thrmsitve ; Position ori r3,r3,lo16(thrmem) ; Enable with at least 20micro sec sample stw r3,thrm3val(r12) ; Save this in case we need it later mtspr thrm3,r3 ; Do it blr /* Set thermal monitor bounds * void ml_thrm_set(unsigned int low, unsigned int high) * * Set TAU to interrupt below low and above high. A value of * zero disables interruptions in that direction. */ ; Force a line boundry here .align 5 .globl EXT(ml_thrm_set) LEXT(ml_thrm_set) mfmsr r0 ; Get the MSR rlwinm r6,r0,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Clear EE bit mtmsr r6 mfsprg r12,0 ; Get the per_proc blok rlwinm. r6,r3,31-thrmthre,thrmthrs,thrmthre ; Position it and see if enabled mfsprg r9,2 ; Get CPU specific features stw r3,thrmlowTemp(r12) ; Set the low temprature mtcrf 0x40,r9 ; See if we can thermal this machine rlwinm r9,r9,(((31-thrmtie)+(pfThermIntb+1))&31),thrmtie,thrmtie ; Set interrupt enable if this machine can handle it bf pfThermalb,tsetcant ; No can do... beq tsetlowo ; We are setting the low off... ori r6,r6,lo16(thrmtidm|thrmvm) ; Set the lower-than and valid bit or r6,r6,r9 ; Set interruption request if supported tsetlowo: mtspr thrm1,r6 ; Cram the register rlwinm. r6,r4,31-thrmthre,thrmthrs,thrmthre ; Position it and see if enabled stw r4,thrmhighTemp(r12) ; Set the high temprature beq tsethigho ; We are setting the high off... ori r6,r6,lo16(thrmvm) ; Set valid bit or r6,r6,r9 ; Set interruption request if supported tsethigho: mtspr thrm2,r6 ; Cram the register tsetcant: mtmsr r0 ; Reenable interruptions blr ; Leave... /* Read processor temprature * unsigned int ml_read_temp(void) * */ ; Force a line boundry here .align 5 .globl EXT(ml_read_temp) LEXT(ml_read_temp) mfmsr r9 ; Save the MSR rlwinm r8,r9,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Turn off interruptions li r5,15 ; Starting point for ranging (start at 15 so we do not overflow) mfsprg r7,2 ; Get CPU specific features mtmsr r8 ; Do not allow interruptions mtcrf 0x40,r7 ; See if we can thermal this machine bf pfThermalb,thrmcant ; No can do... mfspr r11,thrm1 ; Save thrm1 thrmrange: rlwinm r4,r5,31-thrmthre,thrmthrs,thrmthre ; Position it ori r4,r4,lo16(thrmtidm|thrmvm) ; Flip on the valid bit and make comparision for less than mtspr thrm1,r4 ; Set the test value thrmreada: mfspr r3,thrm1 ; Get the thermal register back rlwinm. r0,r3,0,thrmtiv,thrmtiv ; Has it settled yet? beq+ thrmreada ; Nope... rlwinm. r0,r3,0,thrmtin,thrmtin ; Are we still under the threshold? bne thrmsearch ; No, we went over... addi r5,r5,16 ; Start by trying every 16 degrees cmplwi r5,127 ; Have we hit the max? blt- thrmrange ; Got some more to do... thrmsearch: rlwinm r4,r5,31-thrmthre,thrmthrs,thrmthre ; Position it ori r4,r4,lo16(thrmtidm|thrmvm) ; Flip on the valid bit and make comparision for less than mtspr thrm1,r4 ; Set the test value thrmread: mfspr r3,thrm1 ; Get the thermal register back rlwinm. r0,r3,0,thrmtiv,thrmtiv ; Has it settled yet? beq+ thrmread ; Nope... rlwinm. r0,r3,0,thrmtin,thrmtin ; Are we still under the threshold? beq thrmdone ; No, we hit it... addic. r5,r5,-1 ; Go down a degree bge+ thrmsearch ; Try again (until we are below freezing)... thrmdone: addi r3,r5,1 ; Return the temprature (bump it up to make it correct) mtspr thrm1,r11 ; Restore the thermal register mtmsr r9 ; Re-enable interruptions blr ; Leave... thrmcant: eqv r3,r3,r3 ; Return bogus temprature because we can not read it mtmsr r9 ; Re-enable interruptions blr ; Leave... /* Throttle processor speed up or down * unsigned int ml_throttle(unsigned int step) * * Returns old speed and sets new. Both step and return are values from 0 to * 255 that define number of throttle steps, 0 being off and "ictcfim" is max * 2. * */ ; Force a line boundry here .align 5 .globl EXT(ml_throttle) LEXT(ml_throttle) mfmsr r9 ; Save the MSR rlwinm r8,r9,0,MSR_EE_BIT+1,MSR_EE_BIT-1 ; Turn off interruptions cmplwi r3,lo16(ictcfim>>1) ; See if we are going too far mtmsr r8 ; Do not allow interruptions ble+ throtok ; Throttle value is ok... li r3,lo16(ictcfim>>1) ; Set max throtok: rlwinm. r4,r3,1,ictcfib,ictcfie ; Set the throttle beq throtoff ; Skip if we are turning it off... ori r4,r4,lo16(thrmvm) ; Turn on the valid bit throtoff: mfspr r3,ictc ; Get the old throttle mtspr ictc,r4 ; Set the new rlwinm r3,r3,31,1,31 ; Shift throttle value over mtmsr r9 ; Restore interruptions blr ; Return... /* ** ml_get_timebase() ** ** Entry - R3 contains pointer to 64 bit structure. ** ** Exit - 64 bit structure filled in. ** */ ; Force a line boundry here .align 5 .globl EXT(ml_get_timebase) LEXT(ml_get_timebase) loop: mftbu r4 mftb r5 mftbu r6 cmpw r6, r4 bne- loop stw r4, 0(r3) stw r5, 4(r3) blr /* ** ml_sense_nmi() ** */ ; Force a line boundry here .align 5 .globl EXT(ml_sense_nmi) LEXT(ml_sense_nmi) blr ; Leave...