* create boot_abs_to_virt() and boot_virt_to_abs() to use __boot_pa() and __boot_va() for early boot code. * change a lot of code to use __boot variants in early boot. nonlinear won't work when relocation is on anyway * switch rtas_call() and create_pte_mapping() to use __boot or nonlinear variants based on system_state variable * compute first page on the node in show_mem() with pfn_to_page(), not node_mem_map Signed-off-by: Dave Hansen --- memhotplug-dave/arch/ppc64/kernel/prom.c | 20 ++++++++-------- memhotplug-dave/arch/ppc64/kernel/rtas.c | 25 ++++++++++++--------- memhotplug-dave/arch/ppc64/kernel/smp.c | 5 ++-- memhotplug-dave/arch/ppc64/mm/hash_utils.c | 15 ++++++++---- memhotplug-dave/arch/ppc64/mm/init.c | 21 +++++++++++------ memhotplug-dave/include/asm-ppc64/abs_addr.h | 6 +++-- memhotplug-dave/include/asm-ppc64/memory_hotplug.h | 10 ++++++++ memhotplug-dave/include/asm-ppc64/nonlinear.h | 8 ++++++ memhotplug-dave/include/asm-ppc64/page.h | 19 +++++++++++---- mm/page_alloc.c | 0 10 files changed, 88 insertions(+), 41 deletions(-) diff -puN arch/ppc64/Kconfig~H-nonlinear-ppc64-daveh arch/ppc64/Kconfig diff -puN arch/ppc64/kernel/prom.c~H-nonlinear-ppc64-daveh arch/ppc64/kernel/prom.c --- memhotplug/arch/ppc64/kernel/prom.c~H-nonlinear-ppc64-daveh Thu Aug 5 09:41:13 2004 +++ memhotplug-dave/arch/ppc64/kernel/prom.c Thu Aug 5 09:41:13 2004 @@ -826,7 +826,7 @@ static void __init prom_initialize_tce_t prom_panic("ERROR, cannot find space for TCE table.\n"); } - vbase = (unsigned long)abs_to_virt(base); + vbase = (unsigned long)boot_abs_to_virt(base); /* Save away the TCE table attributes for later use. */ prom_tce_table[table].node = node; @@ -913,11 +913,11 @@ static void __init prom_hold_cpus(unsign extern unsigned long __secondary_hold_spinloop; extern unsigned long __secondary_hold_acknowledge; unsigned long *spinloop - = (void *)virt_to_abs(&__secondary_hold_spinloop); + = (void *)boot_virt_to_abs(&__secondary_hold_spinloop); unsigned long *acknowledge - = (void *)virt_to_abs(&__secondary_hold_acknowledge); + = (void *)boot_virt_to_abs(&__secondary_hold_acknowledge); unsigned long secondary_hold - = virt_to_abs(*PTRRELOC((unsigned long *)__secondary_hold)); + = boot_virt_to_abs(*PTRRELOC((unsigned long *)__secondary_hold)); struct systemcfg *_systemcfg = RELOC(systemcfg); struct paca_struct *lpaca = PTRRELOC(&paca[0]); struct prom_t *_prom = PTRRELOC(&prom); @@ -1699,7 +1699,7 @@ prom_init(unsigned long r3, unsigned lon */ #ifdef CONFIG_BLK_DEV_INITRD if ( _prom->bi_recs == NULL && r3 && r4 && r4 != 0xdeadbeef) { - RELOC(initrd_start) = (r3 >= KERNELBASE) ? __pa(r3) : r3; + RELOC(initrd_start) = (r3 >= KERNELBASE) ? __boot_pa(r3) : r3; RELOC(initrd_end) = RELOC(initrd_start) + r4; RELOC(initrd_below_start_ok) = 1; } @@ -1799,7 +1799,7 @@ prom_init(unsigned long r3, unsigned lon prom_debug("klimit=0x%x\n", RELOC(klimit)); prom_debug(" ->mem=0x%x\n", mem); - lmb_reserve(0, __pa(RELOC(klimit))); + lmb_reserve(0, __boot_pa(RELOC(klimit))); #ifdef CONFIG_BLK_DEV_INITRD if (RELOC(initrd_start)) { @@ -1807,7 +1807,7 @@ prom_init(unsigned long r3, unsigned lon initrd_len = RELOC(initrd_end) - RELOC(initrd_start); /* Move initrd if it's where we're going to copy kernel. */ - if (RELOC(initrd_start) < __pa(RELOC(klimit))) { + if (RELOC(initrd_start) < __boot_pa(RELOC(klimit))) { memmove((void *)mem, (void *)RELOC(initrd_start), initrd_len); RELOC(initrd_start) = mem; @@ -1840,8 +1840,8 @@ prom_init(unsigned long r3, unsigned lon #ifdef CONFIG_BLK_DEV_INITRD /* If we had an initrd, we convert its address to virtual */ if (RELOC(initrd_start)) { - RELOC(initrd_start) = (unsigned long)__va(RELOC(initrd_start)); - RELOC(initrd_end) = (unsigned long)__va(RELOC(initrd_end)); + RELOC(initrd_start) = (unsigned long)__boot_va(RELOC(initrd_start)); + RELOC(initrd_end) = (unsigned long)__boot_va(RELOC(initrd_end)); } #endif /* CONFIG_BLK_DEV_INITRD */ @@ -2309,7 +2309,7 @@ finish_device_tree(void) virt_irq_init(); dev_tree_size = finish_node(allnodes, 0, NULL, 0, 0, 1); - mem = (long)abs_to_virt(lmb_alloc(dev_tree_size, + mem = (long)boot_abs_to_virt(lmb_alloc(dev_tree_size, __alignof__(struct device_node))); if (finish_node(allnodes, mem, NULL, 0, 0, 0) != mem + dev_tree_size) BUG(); diff -puN arch/ppc64/kernel/rtas.c~H-nonlinear-ppc64-daveh arch/ppc64/kernel/rtas.c --- memhotplug/arch/ppc64/kernel/rtas.c~H-nonlinear-ppc64-daveh Thu Aug 5 09:41:13 2004 +++ memhotplug-dave/arch/ppc64/kernel/rtas.c Thu Aug 5 09:41:13 2004 @@ -56,7 +56,7 @@ call_rtas_display_status(char c) args->rets = (rtas_arg_t *)&(args->args[1]); args->args[0] = (int)c; - enter_rtas(__pa(args)); + enter_rtas(virt_to_phys(args)); spin_unlock_irqrestore(&rtas.lock, s); } @@ -90,7 +90,7 @@ __fetch_rtas_last_error(void) err_args.nret = 1; err_args.rets = (rtas_arg_t *)&(err_args.args[2]); - err_args.args[0] = (rtas_arg_t)__pa(rtas_err_buf); + err_args.args[0] = (rtas_arg_t)virt_to_phys(rtas_err_buf); err_args.args[1] = RTAS_ERROR_LOG_MAX; err_args.args[2] = 0; @@ -99,7 +99,7 @@ __fetch_rtas_last_error(void) PPCDBG(PPCDBG_RTAS, "\tentering rtas with 0x%lx\n", __pa(&err_args)); - enter_rtas(__pa(&rtas.args)); + enter_rtas(virt_to_phys(&rtas.args)); PPCDBG(PPCDBG_RTAS, "\treturned from rtas ...\n"); err_args = rtas.args; @@ -116,6 +116,7 @@ int rtas_call(int token, int nargs, int struct rtas_args *rtas_args; char * buff_copy = NULL; int ret; + unsigned long rtas_args_paddr; PPCDBG(PPCDBG_RTAS, "Entering rtas_call\n"); PPCDBG(PPCDBG_RTAS, "\ttoken = 0x%x\n", token); @@ -143,9 +144,13 @@ int rtas_call(int token, int nargs, int for (i = 0; i < nret; ++i) rtas_args->rets[i] = 0; - PPCDBG(PPCDBG_RTAS, "\tentering rtas with 0x%lx\n", - __pa(rtas_args)); - enter_rtas(__pa(rtas_args)); + if (likely(system_state >= SYSTEM_RUNNING)) + rtas_args_paddr = virt_to_phys(rtas_args); + else + rtas_args_paddr = __boot_pa(rtas_args); + + PPCDBG(PPCDBG_RTAS, "\tentering rtas with 0x%lx\n", rtas_args_paddr); + enter_rtas(rtas_args_paddr); PPCDBG(PPCDBG_RTAS, "\treturned from rtas ...\n"); /* A -1 return code indicates that the last command couldn't @@ -309,7 +314,7 @@ rtas_flash_firmware(void) */ rtas_firmware_flash_list.num_blocks = 0; flist = (struct flash_block_list *)&rtas_firmware_flash_list; - rtas_block_list = virt_to_abs(flist); + rtas_block_list = boot_virt_to_abs(flist); if (rtas_block_list >= 4UL*1024*1024*1024) { printk(KERN_ALERT "FLASH: kernel bug...flash list header addr above 4GB\n"); return; @@ -321,13 +326,13 @@ rtas_flash_firmware(void) for (f = flist; f; f = next) { /* Translate data addrs to absolute */ for (i = 0; i < f->num_blocks; i++) { - f->blocks[i].data = (char *)virt_to_abs(f->blocks[i].data); + f->blocks[i].data = (char *)boot_virt_to_abs(f->blocks[i].data); image_size += f->blocks[i].length; } next = f->next; /* Don't translate NULL pointer for last entry */ if (f->next) - f->next = (struct flash_block_list *)virt_to_abs(f->next); + f->next = (struct flash_block_list *)boot_virt_to_abs(f->next); else f->next = NULL; /* make num_blocks into the version/length field */ @@ -407,7 +412,7 @@ void rtas_os_term(char *str) do { status = rtas_call(rtas_token("ibm,os-term"), 1, 1, NULL, - __pa(rtas_os_term_buf)); + virt_to_phys(rtas_os_term_buf)); if (status == RTAS_BUSY) udelay(1); diff -puN arch/ppc64/kernel/smp.c~H-nonlinear-ppc64-daveh arch/ppc64/kernel/smp.c --- memhotplug/arch/ppc64/kernel/smp.c~H-nonlinear-ppc64-daveh Thu Aug 5 09:41:13 2004 +++ memhotplug-dave/arch/ppc64/kernel/smp.c Thu Aug 5 09:41:13 2004 @@ -504,7 +504,7 @@ void vpa_init(int cpu) /* Register the Virtual Processor Area (VPA) */ flags = 1UL << (63 - 18); - register_vpa(flags, cpu, __pa((unsigned long)&(paca[cpu].lppaca))); + register_vpa(flags, cpu, __boot_pa((unsigned long)&(paca[cpu].lppaca))); } static inline void smp_xics_do_message(int cpu, int msg) @@ -888,6 +888,7 @@ void __devinit smp_prepare_boot_cpu(void current_set[boot_cpuid] = current->thread_info; } +void dump_cpustuff(void); int __devinit __cpu_up(unsigned int cpu) { int c; @@ -910,7 +911,7 @@ int __devinit __cpu_up(unsigned int cpu) tmp = &stab_array[PAGE_SIZE * cpu]; memset(tmp, 0, PAGE_SIZE); paca[cpu].stab_addr = (unsigned long)tmp; - paca[cpu].stab_real = virt_to_abs(tmp); + paca[cpu].stab_real = boot_virt_to_abs(tmp); } /* The information for processor bringup must diff -puN arch/ppc64/kernel/vio.c~H-nonlinear-ppc64-daveh arch/ppc64/kernel/vio.c diff -puN arch/ppc64/mm/hash_utils.c~H-nonlinear-ppc64-daveh arch/ppc64/mm/hash_utils.c --- memhotplug/arch/ppc64/mm/hash_utils.c~H-nonlinear-ppc64-daveh Thu Aug 5 09:41:13 2004 +++ memhotplug-dave/arch/ppc64/mm/hash_utils.c Thu Aug 5 09:41:13 2004 @@ -96,8 +96,15 @@ static inline void create_pte_mapping(un unsigned long vpn, hash, hpteg; unsigned long vsid = get_kernel_vsid(addr); unsigned long va = (vsid << 28) | (addr & 0xfffffff); + unsigned long abs_page_nr, abs_page_addr; int ret; + if (system_state >= SYSTEM_RUNNING) + abs_page_addr = virt_to_abs((void *)addr); + else + abs_page_addr = boot_virt_to_abs(addr); + abs_page_nr = abs_page_addr >> PAGE_SHIFT; + if (large) vpn = va >> LARGE_PAGE_SHIFT; else @@ -109,12 +116,10 @@ static inline void create_pte_mapping(un if (systemcfg->platform == PLATFORM_PSERIES_LPAR) ret = pSeries_lpar_hpte_insert(hpteg, va, - virt_to_abs(addr) >> PAGE_SHIFT, - 0, mode, 1, large); + abs_page_nr, 0, mode, 1, large); else ret = pSeries_hpte_insert(hpteg, va, - virt_to_abs(addr) >> PAGE_SHIFT, - 0, mode, 1, large); + abs_page_nr, 0, mode, 1, large); if (ret == -1) { ppc64_terminate_msg(0x20, "create_pte_mapping"); @@ -156,7 +161,7 @@ void __init htab_initialize(void) ppc64_terminate_msg(0x20, "hpt space"); loop_forever(); } - htab_data.htab = abs_to_virt(table); + htab_data.htab = boot_abs_to_virt(table); /* htab absolute addr + encoded htabsize */ _SDR1 = table + __ilog2(pteg_count) - 11; diff -puN arch/ppc64/mm/init.c~H-nonlinear-ppc64-daveh arch/ppc64/mm/init.c --- memhotplug/arch/ppc64/mm/init.c~H-nonlinear-ppc64-daveh Thu Aug 5 09:41:13 2004 +++ memhotplug-dave/arch/ppc64/mm/init.c Thu Aug 5 09:41:13 2004 @@ -100,7 +100,7 @@ void show_mem(void) printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); for_each_pgdat(pgdat) { for (i = 0; i < pgdat->node_spanned_pages; i++) { - page = pgdat->node_mem_map + i; + page = pfn_to_page(i); total++; if (PageReserved(page)) reserved++; @@ -424,8 +424,8 @@ void free_initmem(void) addr = (unsigned long)__init_begin; for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) { - ClearPageReserved(virt_to_page(addr)); - set_page_count(virt_to_page(addr), 1); + ClearPageReserved(virt_to_page((void *)addr)); + set_page_count(virt_to_page((void *)addr), 1); free_page(addr); totalram_pages++; } @@ -533,6 +533,11 @@ void __init do_init_bootmem(void) unsigned long total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT; int boot_mapsize; +#ifdef CONFIG_NONLINEAR + setup_memsections(); + alloc_memsections(0, 0, total_pages); +#endif + /* * Find an area to use for the bootmem bitmap. Calculate the size of * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE. @@ -588,7 +593,7 @@ void __init paging_init(void) memset(zones_size, 0, sizeof(zones_size)); memset(zholes_size, 0, sizeof(zholes_size)); - zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT; + zones_size[ZONE_DMA] = top_pfn; zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT; free_area_init_node(0, &contig_page_data, zones_size, @@ -652,7 +657,8 @@ void __init mem_init(void) (unsigned long)nr_free_pages()<< (PAGE_SHIFT-10), codepages<< (PAGE_SHIFT-10), datapages<< (PAGE_SHIFT-10), initpages<< (PAGE_SHIFT-10), - PAGE_OFFSET, (unsigned long)__va(lmb_end_of_DRAM())); + PAGE_OFFSET, (unsigned long)__boot_va(lmb_end_of_DRAM())); + /* lmb_end_of_DRAM() is not a valid phsical address. hack around it for now */ } #else max_mapnr = num_physpages; @@ -661,7 +667,7 @@ void __init mem_init(void) for (addr = KERNELBASE; addr < (unsigned long)__va(lmb_end_of_DRAM()); addr += PAGE_SIZE) { - if (!PageReserved(virt_to_page(addr))) + if (!PageReserved(virt_to_page((void *)addr))) continue; if (addr < (unsigned long)_etext) codepages++; @@ -677,7 +683,7 @@ void __init mem_init(void) (unsigned long)nr_free_pages()<< (PAGE_SHIFT-10), codepages<< (PAGE_SHIFT-10), datapages<< (PAGE_SHIFT-10), initpages<< (PAGE_SHIFT-10), - PAGE_OFFSET, (unsigned long)__va(lmb_end_of_DRAM())); + PAGE_OFFSET, (unsigned long)__boot_va(lmb_end_of_DRAM())); #endif mem_init_done = 1; @@ -841,3 +847,4 @@ void pgtable_cache_init(void) if (!zero_cache) panic("pgtable_cache_init(): could not create zero_cache!\n"); } + diff -puN arch/ppc64/xmon/xmon.c~H-nonlinear-ppc64-daveh arch/ppc64/xmon/xmon.c diff -puN include/asm-ppc64/abs_addr.h~H-nonlinear-ppc64-daveh include/asm-ppc64/abs_addr.h --- memhotplug/include/asm-ppc64/abs_addr.h~H-nonlinear-ppc64-daveh Thu Aug 5 09:41:13 2004 +++ memhotplug-dave/include/asm-ppc64/abs_addr.h Thu Aug 5 09:41:13 2004 @@ -102,7 +102,9 @@ physRpn_to_absRpn(unsigned long rpn) #endif /* !CONFIG_MSCHUNKS */ /* Convenience macros */ -#define virt_to_abs(va) phys_to_abs(__pa(va)) -#define abs_to_virt(aa) __va(abs_to_phys(aa)) +#define virt_to_abs(va) phys_to_abs(virt_to_phys(va)) +#define abs_to_virt(aa) phys_to_virt(abs_to_phys(aa)) +#define boot_virt_to_abs(va) phys_to_abs(__boot_pa(va)) +#define boot_abs_to_virt(aa) __boot_va(abs_to_phys(aa)) #endif /* _ABS_ADDR_H */ diff -puN /dev/null include/asm-ppc64/memory_hotplug.h --- /dev/null Tue Mar 30 06:13:35 2004 +++ memhotplug-dave/include/asm-ppc64/memory_hotplug.h Thu Aug 5 09:41:13 2004 @@ -0,0 +1,10 @@ +#ifndef __ASM_MEMORY_HOTPLUG_H +#define __ASM_MEMORY_HOTPLUG_H + +/* copied from i386 version. Can they be shared? */ + +/* VM interface that may be used by firmware interface */ +extern int add_pages(u64 start, u64 size, unsigned long attr); +extern int remove_pages(u64 start, u64 size, unsigned long attr); + +#endif diff -puN /dev/null include/asm-ppc64/nonlinear.h --- /dev/null Tue Mar 30 06:13:35 2004 +++ memhotplug-dave/include/asm-ppc64/nonlinear.h Thu Aug 5 09:41:13 2004 @@ -0,0 +1,8 @@ +#ifndef __PPC64_NONLINEAR_H_ +#define __PPC64_NONLINEAR_H_ + +#define SECTION_SHIFT 28 /* Size of section - 256 Mbytes */ +#define MAX_MEM_SHIFT 40 +#define MAX_PHYS_SHIFT 40 + +#endif /* __PPC64_NONLINEAR_H_ */ diff -puN include/asm-ppc64/page.h~H-nonlinear-ppc64-daveh include/asm-ppc64/page.h --- memhotplug/include/asm-ppc64/page.h~H-nonlinear-ppc64-daveh Thu Aug 5 09:41:13 2004 +++ memhotplug-dave/include/asm-ppc64/page.h Thu Aug 5 09:41:13 2004 @@ -229,23 +229,32 @@ extern int page_is_ram(unsigned long phy #define IS_VALID_EA(x) \ ((!((x) & INVALID_EA_BITS)) && IS_VALID_REGION_ID(REGION_ID(x))) -#define __bpn_to_ba(x) ((((unsigned long)(x))<> PAGE_SHIFT) +#ifndef __ASSEMBLY__ #ifdef CONFIG_DISCONTIGMEM #define page_to_pfn(page) discontigmem_page_to_pfn(page) #define pfn_to_page(pfn) discontigmem_pfn_to_page(pfn) #define pfn_valid(pfn) discontigmem_pfn_valid(pfn) -#else +#elif defined CONFIG_NONLINEAR +extern unsigned long page_to_pfn(struct page *page); +extern struct page *pfn_to_page(unsigned long); +#else /* !CONFIG_NONLINEAR && !CONFIG_DISCONTIGMEM */ #define pfn_to_page(pfn) (mem_map + (pfn)) #define page_to_pfn(page) ((unsigned long)((page) - mem_map)) #define pfn_valid(pfn) ((pfn) < max_mapnr) #endif -#define virt_to_page(kaddr) pfn_to_page(__pa(kaddr) >> PAGE_SHIFT) -#define pfn_to_kaddr(pfn) __va((pfn) << PAGE_SHIFT) +#include +#include -#define virt_addr_valid(kaddr) pfn_valid(__pa(kaddr) >> PAGE_SHIFT) +#define pfn_to_kaddr(pfn) phys_to_virt((pfn) << PAGE_SHIFT) +#define virt_to_page(kaddr) pfn_to_page(virt_to_phys(kaddr) >> PAGE_SHIFT) + +#define virt_addr_valid(kaddr) pfn_valid(virt_to_phys(kaddr) >> PAGE_SHIFT) + +#endif /* __ASSEMBLY__ */ #define VM_DATA_DEFAULT_FLAGS (VM_READ | VM_WRITE | VM_EXEC | \ VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC) diff -puN include/linux/nonlinear.h~H-nonlinear-ppc64-daveh include/linux/nonlinear.h diff -puN mm/nonlinear.c~H-nonlinear-ppc64-daveh mm/nonlinear.c diff -puN mm/page_alloc.c~H-nonlinear-ppc64-daveh mm/page_alloc.c diff -L I-nonlinear-ppc64-dave-hmore -puN /dev/null /dev/null _