1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
6 * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
7 * and Cort Dougan (PReP) (cort@cs.nmt.edu)
8 * Copyright (C) 1996 Paul Mackerras
9 * PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
11 * Derived from "arch/i386/mm/init.c"
12 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
15 #include <linux/export.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/gfp.h>
21 #include <linux/types.h>
23 #include <linux/stddef.h>
24 #include <linux/init.h>
25 #include <linux/memblock.h>
26 #include <linux/highmem.h>
27 #include <linux/initrd.h>
28 #include <linux/pagemap.h>
29 #include <linux/suspend.h>
30 #include <linux/hugetlb.h>
31 #include <linux/slab.h>
32 #include <linux/vmalloc.h>
33 #include <linux/memremap.h>
34 #include <linux/dma-direct.h>
35 #include <linux/kprobes.h>
39 #include <asm/mmu_context.h>
42 #include <asm/machdep.h>
43 #include <asm/btext.h>
45 #include <asm/sections.h>
46 #include <asm/sparsemem.h>
48 #include <asm/fixmap.h>
49 #include <asm/swiotlb.h>
51 #include <asm/kasan.h>
53 #include <asm/mmzone.h>
54 #include <asm/ftrace.h>
56 #include <mm/mmu_decl.h>
58 #ifndef CPU_FTR_COHERENT_ICACHE
59 #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
60 #define CPU_FTR_NOEXECUTE 0
63 unsigned long long memory_limit;
64 bool init_mem_is_free;
68 EXPORT_SYMBOL(kmap_pte);
71 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
72 unsigned long size, pgprot_t vma_prot)
74 if (ppc_md.phys_mem_access_prot)
75 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
77 if (!page_is_ram(pfn))
78 vma_prot = pgprot_noncached(vma_prot);
82 EXPORT_SYMBOL(phys_mem_access_prot);
84 #ifdef CONFIG_MEMORY_HOTPLUG
87 int memory_add_physaddr_to_nid(u64 start)
89 return hot_add_scn_to_nid(start);
93 int __weak create_section_mapping(unsigned long start, unsigned long end,
94 int nid, pgprot_t prot)
99 int __weak remove_section_mapping(unsigned long start, unsigned long end)
104 #define FLUSH_CHUNK_SIZE SZ_1G
106 * flush_dcache_range_chunked(): Write any modified data cache blocks out to
107 * memory and invalidate them, in chunks of up to FLUSH_CHUNK_SIZE
108 * Does not invalidate the corresponding instruction cache blocks.
110 * @start: the start address
111 * @stop: the stop address (exclusive)
112 * @chunk: the max size of the chunks
114 static void flush_dcache_range_chunked(unsigned long start, unsigned long stop,
119 for (i = start; i < stop; i += chunk) {
120 flush_dcache_range(i, min(stop, i + chunk));
125 int __ref arch_add_memory(int nid, u64 start, u64 size,
126 struct mhp_params *params)
128 unsigned long start_pfn = start >> PAGE_SHIFT;
129 unsigned long nr_pages = size >> PAGE_SHIFT;
132 start = (unsigned long)__va(start);
133 rc = create_section_mapping(start, start + size, nid,
136 pr_warn("Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
137 start, start + size, rc);
141 return __add_pages(nid, start_pfn, nr_pages, params);
144 void __ref arch_remove_memory(int nid, u64 start, u64 size,
145 struct vmem_altmap *altmap)
147 unsigned long start_pfn = start >> PAGE_SHIFT;
148 unsigned long nr_pages = size >> PAGE_SHIFT;
151 __remove_pages(start_pfn, nr_pages, altmap);
153 /* Remove htab bolted mappings for this section of memory */
154 start = (unsigned long)__va(start);
155 flush_dcache_range_chunked(start, start + size, FLUSH_CHUNK_SIZE);
157 ret = remove_section_mapping(start, start + size);
160 /* Ensure all vmalloc mappings are flushed in case they also
161 * hit that section of memory
167 #ifndef CONFIG_NEED_MULTIPLE_NODES
168 void __init mem_topology_setup(void)
170 max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
171 min_low_pfn = MEMORY_START >> PAGE_SHIFT;
172 #ifdef CONFIG_HIGHMEM
173 max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
176 /* Place all memblock_regions in the same node and merge contiguous
179 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
182 void __init initmem_init(void)
187 /* mark pages that don't exist as nosave */
188 static int __init mark_nonram_nosave(void)
190 unsigned long spfn, epfn, prev = 0;
193 for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
194 if (prev && prev < spfn)
195 register_nosave_region(prev, spfn);
202 #else /* CONFIG_NEED_MULTIPLE_NODES */
203 static int __init mark_nonram_nosave(void)
212 * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
213 * everything else. GFP_DMA32 page allocations automatically fall back to
216 * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
217 * generic DMA mapping code. 32-bit only devices (if not handled by an IOMMU
218 * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
221 static unsigned long max_zone_pfns[MAX_NR_ZONES];
224 * paging_init() sets up the page tables - in fact we've already done this.
226 void __init paging_init(void)
228 unsigned long long total_ram = memblock_phys_mem_size();
229 phys_addr_t top_of_ram = memblock_end_of_DRAM();
231 #ifdef CONFIG_HIGHMEM
232 unsigned long v = __fix_to_virt(FIX_KMAP_END);
233 unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
235 for (; v < end; v += PAGE_SIZE)
236 map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
238 map_kernel_page(PKMAP_BASE, 0, __pgprot(0)); /* XXX gross */
239 pkmap_page_table = virt_to_kpte(PKMAP_BASE);
241 kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
242 #endif /* CONFIG_HIGHMEM */
244 printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
245 (unsigned long long)top_of_ram, total_ram);
246 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
247 (long int)((top_of_ram - total_ram) >> 20));
250 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
253 if (IS_ENABLED(CONFIG_PPC32))
258 #ifdef CONFIG_ZONE_DMA
259 max_zone_pfns[ZONE_DMA] = min(max_low_pfn,
260 1UL << (zone_dma_bits - PAGE_SHIFT));
262 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
263 #ifdef CONFIG_HIGHMEM
264 max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
267 free_area_init(max_zone_pfns);
269 mark_nonram_nosave();
272 void __init mem_init(void)
275 * book3s is limited to 16 page sizes due to encoding this in
276 * a 4-bit field for slices.
278 BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
280 #ifdef CONFIG_SWIOTLB
282 * Some platforms (e.g. 85xx) limit DMA-able memory way below
283 * 4G. We force memblock to bottom-up mode to ensure that the
284 * memory allocated in swiotlb_init() is DMA-able.
285 * As it's the last memblock allocation, no need to reset it
288 memblock_set_bottom_up(true);
289 if (is_secure_guest())
295 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
301 #ifdef CONFIG_HIGHMEM
303 unsigned long pfn, highmem_mapnr;
305 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
306 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
307 phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
308 struct page *page = pfn_to_page(pfn);
309 if (!memblock_is_reserved(paddr))
310 free_highmem_page(page);
313 #endif /* CONFIG_HIGHMEM */
315 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
317 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
318 * functions.... do it here for the non-smp case.
320 per_cpu(next_tlbcam_idx, smp_processor_id()) =
321 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
324 mem_init_print_info(NULL);
326 pr_info("Kernel virtual memory layout:\n");
328 pr_info(" * 0x%08lx..0x%08lx : kasan shadow mem\n",
329 KASAN_SHADOW_START, KASAN_SHADOW_END);
331 pr_info(" * 0x%08lx..0x%08lx : fixmap\n", FIXADDR_START, FIXADDR_TOP);
332 #ifdef CONFIG_HIGHMEM
333 pr_info(" * 0x%08lx..0x%08lx : highmem PTEs\n",
334 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
335 #endif /* CONFIG_HIGHMEM */
336 if (ioremap_bot != IOREMAP_TOP)
337 pr_info(" * 0x%08lx..0x%08lx : early ioremap\n",
338 ioremap_bot, IOREMAP_TOP);
339 pr_info(" * 0x%08lx..0x%08lx : vmalloc & ioremap\n",
340 VMALLOC_START, VMALLOC_END);
341 #endif /* CONFIG_PPC32 */
344 void free_initmem(void)
346 ppc_md.progress = ppc_printk_progress;
348 init_mem_is_free = true;
349 free_initmem_default(POISON_FREE_INITMEM);
350 ftrace_free_init_tramp();
354 * flush_coherent_icache() - if a CPU has a coherent icache, flush it
355 * @addr: The base address to use (can be any valid address, the whole cache will be flushed)
356 * Return true if the cache was flushed, false otherwise
358 static inline bool flush_coherent_icache(unsigned long addr)
361 * For a snooping icache, we still need a dummy icbi to purge all the
362 * prefetched instructions from the ifetch buffers. We also need a sync
363 * before the icbi to order the the actual stores to memory that might
364 * have modified instructions with the icbi.
366 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE)) {
368 allow_read_from_user((const void __user *)addr, L1_CACHE_BYTES);
370 prevent_read_from_user((const void __user *)addr, L1_CACHE_BYTES);
380 * invalidate_icache_range() - Flush the icache by issuing icbi across an address range
381 * @start: the start address
382 * @stop: the stop address (exclusive)
384 static void invalidate_icache_range(unsigned long start, unsigned long stop)
386 unsigned long shift = l1_icache_shift();
387 unsigned long bytes = l1_icache_bytes();
388 char *addr = (char *)(start & ~(bytes - 1));
389 unsigned long size = stop - (unsigned long)addr + (bytes - 1);
392 for (i = 0; i < size >> shift; i++, addr += bytes)
400 * flush_icache_range: Write any modified data cache blocks out to memory
401 * and invalidate the corresponding blocks in the instruction cache
403 * Generic code will call this after writing memory, before executing from it.
405 * @start: the start address
406 * @stop: the stop address (exclusive)
408 void flush_icache_range(unsigned long start, unsigned long stop)
410 if (flush_coherent_icache(start))
413 clean_dcache_range(start, stop);
415 if (IS_ENABLED(CONFIG_44x)) {
417 * Flash invalidate on 44x because we are passed kmapped
418 * addresses and this doesn't work for userspace pages due to
419 * the virtually tagged icache.
421 iccci((void *)start);
425 invalidate_icache_range(start, stop);
427 EXPORT_SYMBOL(flush_icache_range);
429 #if !defined(CONFIG_PPC_8xx) && !defined(CONFIG_PPC64)
431 * flush_dcache_icache_phys() - Flush a page by it's physical address
432 * @physaddr: the physical address of the page
434 static void flush_dcache_icache_phys(unsigned long physaddr)
436 unsigned long bytes = l1_dcache_bytes();
437 unsigned long nb = PAGE_SIZE / bytes;
438 unsigned long addr = physaddr & PAGE_MASK;
439 unsigned long msr, msr0;
440 unsigned long loop1 = addr, loop2 = addr;
443 msr = msr0 & ~MSR_DR;
445 * This must remain as ASM to prevent potential memory accesses
446 * while the data MMU is disabled
453 " addi %0, %0, %4;\n"
458 " addi %1, %1, %4;\n"
463 : "+&r" (loop1), "+&r" (loop2)
464 : "r" (nb), "r" (msr), "i" (bytes), "r" (msr0)
467 NOKPROBE_SYMBOL(flush_dcache_icache_phys)
468 #endif // !defined(CONFIG_PPC_8xx) && !defined(CONFIG_PPC64)
471 * This is called when a page has been modified by the kernel.
472 * It just marks the page as not i-cache clean. We do the i-cache
473 * flush later when the page is given to a user process, if necessary.
475 void flush_dcache_page(struct page *page)
477 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
479 /* avoid an atomic op if possible */
480 if (test_bit(PG_arch_1, &page->flags))
481 clear_bit(PG_arch_1, &page->flags);
483 EXPORT_SYMBOL(flush_dcache_page);
485 void flush_dcache_icache_page(struct page *page)
487 #ifdef CONFIG_HUGETLB_PAGE
488 if (PageCompound(page)) {
489 flush_dcache_icache_hugepage(page);
493 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
494 /* On 8xx there is no need to kmap since highmem is not supported */
495 __flush_dcache_icache(page_address(page));
497 if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
498 void *start = kmap_atomic(page);
499 __flush_dcache_icache(start);
500 kunmap_atomic(start);
502 unsigned long addr = page_to_pfn(page) << PAGE_SHIFT;
504 if (flush_coherent_icache(addr))
506 flush_dcache_icache_phys(addr);
510 EXPORT_SYMBOL(flush_dcache_icache_page);
513 * __flush_dcache_icache(): Flush a particular page from the data cache to RAM.
514 * Note: this is necessary because the instruction cache does *not*
515 * snoop from the data cache.
517 * @page: the address of the page to flush
519 void __flush_dcache_icache(void *p)
521 unsigned long addr = (unsigned long)p;
523 if (flush_coherent_icache(addr))
526 clean_dcache_range(addr, addr + PAGE_SIZE);
529 * We don't flush the icache on 44x. Those have a virtual icache and we
530 * don't have access to the virtual address here (it's not the page
531 * vaddr but where it's mapped in user space). The flushing of the
532 * icache on these is handled elsewhere, when a change in the address
533 * space occurs, before returning to user space.
536 if (mmu_has_feature(MMU_FTR_TYPE_44x))
539 invalidate_icache_range(addr, addr + PAGE_SIZE);
542 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
547 * We shouldn't have to do this, but some versions of glibc
548 * require it (ld.so assumes zero filled pages are icache clean)
551 flush_dcache_page(pg);
553 EXPORT_SYMBOL(clear_user_page);
555 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
558 copy_page(vto, vfrom);
561 * We should be able to use the following optimisation, however
562 * there are two problems.
563 * Firstly a bug in some versions of binutils meant PLT sections
564 * were not marked executable.
565 * Secondly the first word in the GOT section is blrl, used
566 * to establish the GOT address. Until recently the GOT was
567 * not marked executable.
571 if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
575 flush_dcache_page(pg);
578 void flush_icache_user_page(struct vm_area_struct *vma, struct page *page,
579 unsigned long addr, int len)
583 maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
584 flush_icache_range(maddr, maddr + len);
589 * System memory should not be in /proc/iomem but various tools expect it
592 static int __init add_system_ram_resources(void)
594 phys_addr_t start, end;
597 for_each_mem_range(i, &start, &end) {
598 struct resource *res;
600 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
604 res->name = "System RAM";
607 * In memblock, end points to the first byte after
608 * the range while in resourses, end points to the
609 * last byte in the range.
612 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
613 WARN_ON(request_resource(&iomem_resource, res) < 0);
619 subsys_initcall(add_system_ram_resources);
621 #ifdef CONFIG_STRICT_DEVMEM
623 * devmem_is_allowed(): check to see if /dev/mem access to a certain address
624 * is valid. The argument is a physical page number.
626 * Access has to be given to non-kernel-ram areas as well, these contain the
627 * PCI mmio resources as well as potential bios/acpi data regions.
629 int devmem_is_allowed(unsigned long pfn)
631 if (page_is_rtas_user_buf(pfn))
633 if (iomem_is_exclusive(PFN_PHYS(pfn)))
635 if (!page_is_ram(pfn))
639 #endif /* CONFIG_STRICT_DEVMEM */
642 * This is defined in kernel/resource.c but only powerpc needs to export it, for
643 * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
645 EXPORT_SYMBOL_GPL(walk_system_ram_range);