2 * This file contains miscellaneous low-level functions.
3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
5 * Largely rewritten by Cort Dougan (cort@cs.nmt.edu)
9 * Copyright (C) 2002-2003 Eric Biederman <ebiederm@xmission.com>
10 * GameCube/ppc32 port Copyright (C) 2004 Albert Herranz
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
19 #include <linux/sys.h>
20 #include <asm/unistd.h>
21 #include <asm/errno.h>
24 #include <asm/cache.h>
25 #include <asm/cputable.h>
27 #include <asm/ppc_asm.h>
28 #include <asm/thread_info.h>
29 #include <asm/asm-offsets.h>
30 #include <asm/processor.h>
31 #include <asm/kexec.h>
36 * This returns the high 64 bits of the product of two 64-bit numbers.
48 1: beqlr cr1 /* all done if high part of A is 0 */
63 * sub_reloc_offset(x) returns x - reloc_offset().
65 _GLOBAL(sub_reloc_offset)
77 * reloc_got2 runs through the .got2 section adding an offset
82 lis r7,__got2_start@ha
83 addi r7,r7,__got2_start@l
85 addi r8,r8,__got2_end@l
105 * call_setup_cpu - call the setup_cpu function for this cpu
106 * r3 = data offset, r24 = cpu number
108 * Setup function is called with:
110 * r4 = ptr to CPU spec (relocated)
112 _GLOBAL(call_setup_cpu)
113 addis r4,r3,cur_cpu_spec@ha
114 addi r4,r4,cur_cpu_spec@l
117 lwz r5,CPU_SPEC_SETUP(r4)
124 #if defined(CONFIG_CPU_FREQ_PMAC) && defined(CONFIG_6xx)
126 /* This gets called by via-pmu.c to switch the PLL selection
127 * on 750fx CPU. This function should really be moved to some
128 * other place (as most of the cpufreq code in via-pmu
130 _GLOBAL(low_choose_750fx_pll)
136 /* If switching to PLL1, disable HID0:BTIC */
147 /* Calc new HID1 value */
148 mfspr r4,SPRN_HID1 /* Build a HID1:PS bit from parameter */
149 rlwinm r5,r3,16,15,15 /* Clear out HID1:PS from value read */
150 rlwinm r4,r4,0,16,14 /* Could have I used rlwimi here ? */
154 /* Store new HID1 image */
158 addis r6,r6,nap_save_hid1@ha
159 stw r4,nap_save_hid1@l(r6)
161 /* If switching to PLL0, enable HID0:BTIC */
176 _GLOBAL(low_choose_7447a_dfs)
182 /* Calc new HID1 value */
184 insrwi r4,r3,1,9 /* insert parameter into bit 9 */
194 #endif /* CONFIG_CPU_FREQ_PMAC && CONFIG_6xx */
197 * complement mask on the msr then "or" some values on.
198 * _nmask_and_or_msr(nmask, value_to_or)
200 _GLOBAL(_nmask_and_or_msr)
201 mfmsr r0 /* Get current msr */
202 andc r0,r0,r3 /* And off the bits set in r3 (first parm) */
203 or r0,r0,r4 /* Or on the bits in r4 (second parm) */
204 SYNC /* Some chip revs have problems here... */
205 mtmsr r0 /* Update machine state */
214 #if defined(CONFIG_40x)
215 sync /* Flush to memory before changing mapping */
217 isync /* Flush shadow TLB */
218 #elif defined(CONFIG_44x)
222 /* Load high watermark */
223 lis r4,tlb_44x_hwater@ha
224 lwz r5,tlb_44x_hwater@l(r4)
226 1: tlbwe r3,r3,PPC44x_TLB_PAGEID
232 #elif defined(CONFIG_FSL_BOOKE)
233 /* Invalidate all entries in TLB0 */
236 /* Invalidate all entries in TLB1 */
239 /* Invalidate all entries in TLB2 */
242 /* Invalidate all entries in TLB3 */
248 #endif /* CONFIG_SMP */
249 #else /* !(CONFIG_40x || CONFIG_44x || CONFIG_FSL_BOOKE) */
250 #if defined(CONFIG_SMP)
256 rlwinm r0,r10,0,17,15 /* clear bit 16 (MSR_EE) */
257 rlwinm r0,r0,0,28,26 /* clear DR */
261 lis r9,mmu_hash_lock@h
262 ori r9,r9,mmu_hash_lock@l
274 stw r0,0(r9) /* clear mmu_hash_lock */
278 #else /* CONFIG_SMP */
282 #endif /* CONFIG_SMP */
283 #endif /* ! defined(CONFIG_40x) */
287 * Flush MMU TLB for a particular address
290 #if defined(CONFIG_40x)
294 /* There are only 64 TLB entries, so r3 < 64, which means bit 25 is clear.
295 * Since 25 is the V bit in the TLB_TAG, loading this value will invalidate
297 tlbwe r3, r3, TLB_TAG
300 #elif defined(CONFIG_44x)
302 mfspr r5,SPRN_PID /* Get PID */
303 rlwimi r4,r5,0,24,31 /* Set TID */
305 /* We have to run the search with interrupts disabled, even critical
306 * and debug interrupts (in fact the only critical exceptions we have
307 * are debug and machine check). Otherwise an interrupt which causes
308 * a TLB miss can clobber the MMUCR between the mtspr and the tlbsx. */
310 lis r6,(MSR_EE|MSR_CE|MSR_ME|MSR_DE)@ha
311 addi r6,r6,(MSR_EE|MSR_CE|MSR_ME|MSR_DE)@l
319 /* There are only 64 TLB entries, so r3 < 64,
320 * which means bit 22, is clear. Since 22 is
321 * the V bit in the TLB_PAGEID, loading this
322 * value will invalidate the TLB entry.
324 tlbwe r3, r3, PPC44x_TLB_PAGEID
327 #elif defined(CONFIG_FSL_BOOKE)
328 rlwinm r4, r3, 0, 0, 19
329 ori r5, r4, 0x08 /* TLBSEL = 1 */
330 ori r6, r4, 0x10 /* TLBSEL = 2 */
331 ori r7, r4, 0x18 /* TLBSEL = 3 */
337 #if defined(CONFIG_SMP)
339 #endif /* CONFIG_SMP */
340 #else /* !(CONFIG_40x || CONFIG_44x || CONFIG_FSL_BOOKE) */
341 #if defined(CONFIG_SMP)
347 rlwinm r0,r10,0,17,15 /* clear bit 16 (MSR_EE) */
348 rlwinm r0,r0,0,28,26 /* clear DR */
352 lis r9,mmu_hash_lock@h
353 ori r9,r9,mmu_hash_lock@l
365 stw r0,0(r9) /* clear mmu_hash_lock */
369 #else /* CONFIG_SMP */
372 #endif /* CONFIG_SMP */
373 #endif /* ! CONFIG_40x */
377 * Flush instruction cache.
378 * This is a no-op on the 601.
380 _GLOBAL(flush_instruction_cache)
381 #if defined(CONFIG_8xx)
384 mtspr SPRN_IC_CST, r5
385 #elif defined(CONFIG_4xx)
397 #elif CONFIG_FSL_BOOKE
400 ori r3,r3,L1CSR0_CFI|L1CSR0_CLFC
401 /* msync; isync recommended here */
405 END_FTR_SECTION_IFSET(CPU_FTR_UNIFIED_ID_CACHE)
407 ori r3,r3,L1CSR1_ICFI|L1CSR1_ICLFR
411 rlwinm r3,r3,16,16,31
413 beqlr /* for 601, do nothing */
414 /* 603/604 processor - use invalidate-all bit in HID0 */
418 #endif /* CONFIG_8xx/4xx */
423 * Write any modified data cache blocks out to memory
424 * and invalidate the corresponding instruction cache blocks.
425 * This is a no-op on the 601.
427 * flush_icache_range(unsigned long start, unsigned long stop)
429 _GLOBAL(__flush_icache_range)
431 blr /* for 601, do nothing */
432 END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
433 li r5,L1_CACHE_BYTES-1
437 srwi. r4,r4,L1_CACHE_SHIFT
442 addi r3,r3,L1_CACHE_BYTES
444 sync /* wait for dcbst's to get to ram */
447 addi r6,r6,L1_CACHE_BYTES
449 sync /* additional sync needed on g4 */
453 * Write any modified data cache blocks out to memory.
454 * Does not invalidate the corresponding cache lines (especially for
455 * any corresponding instruction cache).
457 * clean_dcache_range(unsigned long start, unsigned long stop)
459 _GLOBAL(clean_dcache_range)
460 li r5,L1_CACHE_BYTES-1
464 srwi. r4,r4,L1_CACHE_SHIFT
469 addi r3,r3,L1_CACHE_BYTES
471 sync /* wait for dcbst's to get to ram */
475 * Write any modified data cache blocks out to memory and invalidate them.
476 * Does not invalidate the corresponding instruction cache blocks.
478 * flush_dcache_range(unsigned long start, unsigned long stop)
480 _GLOBAL(flush_dcache_range)
481 li r5,L1_CACHE_BYTES-1
485 srwi. r4,r4,L1_CACHE_SHIFT
490 addi r3,r3,L1_CACHE_BYTES
492 sync /* wait for dcbst's to get to ram */
496 * Like above, but invalidate the D-cache. This is used by the 8xx
497 * to invalidate the cache so the PPC core doesn't get stale data
498 * from the CPM (no cache snooping here :-).
500 * invalidate_dcache_range(unsigned long start, unsigned long stop)
502 _GLOBAL(invalidate_dcache_range)
503 li r5,L1_CACHE_BYTES-1
507 srwi. r4,r4,L1_CACHE_SHIFT
512 addi r3,r3,L1_CACHE_BYTES
514 sync /* wait for dcbi's to get to ram */
518 * Flush a particular page from the data cache to RAM.
519 * Note: this is necessary because the instruction cache does *not*
520 * snoop from the data cache.
521 * This is a no-op on the 601 which has a unified cache.
523 * void __flush_dcache_icache(void *page)
525 _GLOBAL(__flush_dcache_icache)
528 END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
529 rlwinm r3,r3,0,0,19 /* Get page base address */
530 li r4,4096/L1_CACHE_BYTES /* Number of lines in a page */
533 0: dcbst 0,r3 /* Write line to ram */
534 addi r3,r3,L1_CACHE_BYTES
539 addi r6,r6,L1_CACHE_BYTES
546 * Flush a particular page from the data cache to RAM, identified
547 * by its physical address. We turn off the MMU so we can just use
548 * the physical address (this may be a highmem page without a kernel
551 * void __flush_dcache_icache_phys(unsigned long physaddr)
553 _GLOBAL(__flush_dcache_icache_phys)
555 blr /* for 601, do nothing */
556 END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
558 rlwinm r0,r10,0,28,26 /* clear DR */
561 rlwinm r3,r3,0,0,19 /* Get page base address */
562 li r4,4096/L1_CACHE_BYTES /* Number of lines in a page */
565 0: dcbst 0,r3 /* Write line to ram */
566 addi r3,r3,L1_CACHE_BYTES
571 addi r6,r6,L1_CACHE_BYTES
574 mtmsr r10 /* restore DR */
579 * Clear pages using the dcbz instruction, which doesn't cause any
580 * memory traffic (except to write out any cache lines which get
581 * displaced). This only works on cacheable memory.
583 * void clear_pages(void *page, int order) ;
586 li r0,4096/L1_CACHE_BYTES
598 addi r3,r3,L1_CACHE_BYTES
603 * Copy a whole page. We use the dcbz instruction on the destination
604 * to reduce memory traffic (it eliminates the unnecessary reads of
605 * the destination into cache). This requires that the destination
608 #define COPY_16_BYTES \
623 /* don't use prefetch on 8xx */
624 li r0,4096/L1_CACHE_BYTES
630 #else /* not 8xx, we can prefetch */
633 #if MAX_COPY_PREFETCH > 1
634 li r0,MAX_COPY_PREFETCH
638 addi r11,r11,L1_CACHE_BYTES
640 #else /* MAX_COPY_PREFETCH == 1 */
642 li r11,L1_CACHE_BYTES+4
643 #endif /* MAX_COPY_PREFETCH */
644 li r0,4096/L1_CACHE_BYTES - MAX_COPY_PREFETCH
652 #if L1_CACHE_BYTES >= 32
654 #if L1_CACHE_BYTES >= 64
657 #if L1_CACHE_BYTES >= 128
667 crnot 4*cr0+eq,4*cr0+eq
668 li r0,MAX_COPY_PREFETCH
671 #endif /* CONFIG_8xx */
674 * void atomic_clear_mask(atomic_t mask, atomic_t *addr)
675 * void atomic_set_mask(atomic_t mask, atomic_t *addr);
677 _GLOBAL(atomic_clear_mask)
684 _GLOBAL(atomic_set_mask)
693 * Extended precision shifts.
695 * Updated to be valid for shift counts from 0 to 63 inclusive.
698 * R3/R4 has 64 bit value
702 * ashrdi3: arithmetic right shift (sign propagation)
703 * lshrdi3: logical right shift
704 * ashldi3: left shift
708 srw r4,r4,r5 # LSW = count > 31 ? 0 : LSW >> count
709 addi r7,r5,32 # could be xori, or addi with -32
710 slw r6,r3,r6 # t1 = count > 31 ? 0 : MSW << (32-count)
711 rlwinm r8,r7,0,32 # t3 = (count < 32) ? 32 : 0
712 sraw r7,r3,r7 # t2 = MSW >> (count-32)
713 or r4,r4,r6 # LSW |= t1
714 slw r7,r7,r8 # t2 = (count < 32) ? 0 : t2
715 sraw r3,r3,r5 # MSW = MSW >> count
716 or r4,r4,r7 # LSW |= t2
721 slw r3,r3,r5 # MSW = count > 31 ? 0 : MSW << count
722 addi r7,r5,32 # could be xori, or addi with -32
723 srw r6,r4,r6 # t1 = count > 31 ? 0 : LSW >> (32-count)
724 slw r7,r4,r7 # t2 = count < 32 ? 0 : LSW << (count-32)
725 or r3,r3,r6 # MSW |= t1
726 slw r4,r4,r5 # LSW = LSW << count
727 or r3,r3,r7 # MSW |= t2
732 srw r4,r4,r5 # LSW = count > 31 ? 0 : LSW >> count
733 addi r7,r5,32 # could be xori, or addi with -32
734 slw r6,r3,r6 # t1 = count > 31 ? 0 : MSW << (32-count)
735 srw r7,r3,r7 # t2 = count < 32 ? 0 : MSW >> (count-32)
736 or r4,r4,r6 # LSW |= t1
737 srw r3,r3,r5 # MSW = MSW >> count
738 or r4,r4,r7 # LSW |= t2
748 * Create a kernel thread
749 * kernel_thread(fn, arg, flags)
751 _GLOBAL(kernel_thread)
755 mr r30,r3 /* function */
756 mr r31,r4 /* argument */
757 ori r3,r5,CLONE_VM /* flags */
758 oris r3,r3,CLONE_UNTRACED>>16
759 li r4,0 /* new sp (unused) */
762 cmpwi 0,r3,0 /* parent or child? */
763 bne 1f /* return if parent */
764 li r0,0 /* make top-level stack frame */
766 mtlr r30 /* fn addr in lr */
767 mr r3,r31 /* load arg and call fn */
770 li r0,__NR_exit /* exit if function returns */
778 _GLOBAL(kernel_execve)
786 * This routine is just here to keep GCC happy - sigh...
793 * Must be relocatable PIC code callable as a C function.
795 .globl relocate_new_kernel
798 /* r4 = reboot_code_buffer */
799 /* r5 = start_address */
804 * Set Machine Status Register to a known status,
805 * switch the MMU off and jump to 1: in a single step.
809 ori r8, r8, MSR_RI|MSR_ME
811 addi r8, r4, 1f - relocate_new_kernel
817 /* from this point address translation is turned off */
818 /* and interrupts are disabled */
820 /* set a new stack at the bottom of our page... */
821 /* (not really needed now) */
822 addi r1, r4, KEXEC_CONTROL_CODE_SIZE - 8 /* for LR Save+Back Chain */
826 li r6, 0 /* checksum */
830 0: /* top, read another word for the indirection page */
834 /* is it a destination page? (r8) */
835 rlwinm. r7, r0, 0, 31, 31 /* IND_DESTINATION (1<<0) */
838 rlwinm r8, r0, 0, 0, 19 /* clear kexec flags, page align */
841 2: /* is it an indirection page? (r3) */
842 rlwinm. r7, r0, 0, 30, 30 /* IND_INDIRECTION (1<<1) */
845 rlwinm r3, r0, 0, 0, 19 /* clear kexec flags, page align */
849 2: /* are we done? */
850 rlwinm. r7, r0, 0, 29, 29 /* IND_DONE (1<<2) */
854 2: /* is it a source page? (r9) */
855 rlwinm. r7, r0, 0, 28, 28 /* IND_SOURCE (1<<3) */
858 rlwinm r9, r0, 0, 0, 19 /* clear kexec flags, page align */
865 lwzu r0, 4(r9) /* do the copy */
879 /* To be certain of avoiding problems with self-modifying code
880 * execute a serializing instruction here.
885 /* jump to the entry point, usually the setup routine */
891 relocate_new_kernel_end:
893 .globl relocate_new_kernel_size
894 relocate_new_kernel_size:
895 .long relocate_new_kernel_end - relocate_new_kernel