LXR linux/arch/powerpc/mm/mem.c

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *  PowerPC version
   4 *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   5 *
   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)
  10 *
  11 *  Derived from "arch/i386/mm/init.c"
  12 *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  13 */
  14
  15#include <linux/memblock.h>
  16#include <linux/highmem.h>
  17#include <linux/suspend.h>
  18#include <linux/dma-direct.h>
  19
  20#include <asm/machdep.h>
  21#include <asm/rtas.h>
  22#include <asm/kasan.h>
  23#include <asm/sparsemem.h>
  24#include <asm/svm.h>
  25
  26#include <mm/mmu_decl.h>
  27
  28unsigned long long memory_limit;
  29bool init_mem_is_free;
  30
  31unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
  32EXPORT_SYMBOL(empty_zero_page);
  33
  34pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  35                              unsigned long size, pgprot_t vma_prot)
  36{
  37        if (ppc_md.phys_mem_access_prot)
  38                return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
  39
  40        if (!page_is_ram(pfn))
  41                vma_prot = pgprot_noncached(vma_prot);
  42
  43        return vma_prot;
  44}
  45EXPORT_SYMBOL(phys_mem_access_prot);
  46
  47#ifdef CONFIG_MEMORY_HOTPLUG
  48static DEFINE_MUTEX(linear_mapping_mutex);
  49
  50#ifdef CONFIG_NUMA
  51int memory_add_physaddr_to_nid(u64 start)
  52{
  53        return hot_add_scn_to_nid(start);
  54}
  55#endif
  56
  57int __weak create_section_mapping(unsigned long start, unsigned long end,
  58                                  int nid, pgprot_t prot)
  59{
  60        return -ENODEV;
  61}
  62
  63int __weak remove_section_mapping(unsigned long start, unsigned long end)
  64{
  65        return -ENODEV;
  66}
  67
  68int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
  69                                     struct mhp_params *params)
  70{
  71        int rc;
  72
  73        start = (unsigned long)__va(start);
  74        mutex_lock(&linear_mapping_mutex);
  75        rc = create_section_mapping(start, start + size, nid,
  76                                    params->pgprot);
  77        mutex_unlock(&linear_mapping_mutex);
  78        if (rc) {
  79                pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
  80                        start, start + size, rc);
  81                return -EFAULT;
  82        }
  83        return 0;
  84}
  85
  86void __ref arch_remove_linear_mapping(u64 start, u64 size)
  87{
  88        int ret;
  89
  90        /* Remove htab bolted mappings for this section of memory */
  91        start = (unsigned long)__va(start);
  92
  93        mutex_lock(&linear_mapping_mutex);
  94        ret = remove_section_mapping(start, start + size);
  95        mutex_unlock(&linear_mapping_mutex);
  96        if (ret)
  97                pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
  98                        start, start + size, ret);
  99
 100        /* Ensure all vmalloc mappings are flushed in case they also
 101         * hit that section of memory
 102         */
 103        vm_unmap_aliases();
 104}
 105
 106int __ref arch_add_memory(int nid, u64 start, u64 size,
 107                          struct mhp_params *params)
 108{
 109        unsigned long start_pfn = start >> PAGE_SHIFT;
 110        unsigned long nr_pages = size >> PAGE_SHIFT;
 111        int rc;
 112
 113        rc = arch_create_linear_mapping(nid, start, size, params);
 114        if (rc)
 115                return rc;
 116        rc = __add_pages(nid, start_pfn, nr_pages, params);
 117        if (rc)
 118                arch_remove_linear_mapping(start, size);
 119        return rc;
 120}
 121
 122void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
 123{
 124        unsigned long start_pfn = start >> PAGE_SHIFT;
 125        unsigned long nr_pages = size >> PAGE_SHIFT;
 126
 127        __remove_pages(start_pfn, nr_pages, altmap);
 128        arch_remove_linear_mapping(start, size);
 129}
 130#endif
 131
 132#ifndef CONFIG_NUMA
 133void __init mem_topology_setup(void)
 134{
 135        max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
 136        min_low_pfn = MEMORY_START >> PAGE_SHIFT;
 137#ifdef CONFIG_HIGHMEM
 138        max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
 139#endif
 140
 141        /* Place all memblock_regions in the same node and merge contiguous
 142         * memblock_regions
 143         */
 144        memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
 145}
 146
 147void __init initmem_init(void)
 148{
 149        sparse_init();
 150}
 151
 152/* mark pages that don't exist as nosave */
 153static int __init mark_nonram_nosave(void)
 154{
 155        unsigned long spfn, epfn, prev = 0;
 156        int i;
 157
 158        for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
 159                if (prev && prev < spfn)
 160                        register_nosave_region(prev, spfn);
 161
 162                prev = epfn;
 163        }
 164
 165        return 0;
 166}
 167#else /* CONFIG_NUMA */
 168static int __init mark_nonram_nosave(void)
 169{
 170        return 0;
 171}
 172#endif
 173
 174/*
 175 * Zones usage:
 176 *
 177 * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
 178 * everything else. GFP_DMA32 page allocations automatically fall back to
 179 * ZONE_DMA.
 180 *
 181 * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
 182 * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
 183 * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
 184 * ZONE_DMA.
 185 */
 186static unsigned long max_zone_pfns[MAX_NR_ZONES];
 187
 188/*
 189 * paging_init() sets up the page tables - in fact we've already done this.
 190 */
 191void __init paging_init(void)
 192{
 193        unsigned long long total_ram = memblock_phys_mem_size();
 194        phys_addr_t top_of_ram = memblock_end_of_DRAM();
 195
 196#ifdef CONFIG_HIGHMEM
 197        unsigned long v = __fix_to_virt(FIX_KMAP_END);
 198        unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
 199
 200        for (; v < end; v += PAGE_SIZE)
 201                map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
 202
 203        map_kernel_page(PKMAP_BASE, 0, __pgprot(0));    /* XXX gross */
 204        pkmap_page_table = virt_to_kpte(PKMAP_BASE);
 205#endif /* CONFIG_HIGHMEM */
 206
 207        printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
 208               (unsigned long long)top_of_ram, total_ram);
 209        printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 210               (long int)((top_of_ram - total_ram) >> 20));
 211
 212        /*
 213         * Allow 30-bit DMA for very limited Broadcom wifi chips on many
 214         * powerbooks.
 215         */
 216        if (IS_ENABLED(CONFIG_PPC32))
 217                zone_dma_bits = 30;
 218        else
 219                zone_dma_bits = 31;
 220
 221#ifdef CONFIG_ZONE_DMA
 222        max_zone_pfns[ZONE_DMA] = min(max_low_pfn,
 223                                      1UL << (zone_dma_bits - PAGE_SHIFT));
 224#endif
 225        max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 226#ifdef CONFIG_HIGHMEM
 227        max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
 228#endif
 229
 230        free_area_init(max_zone_pfns);
 231
 232        mark_nonram_nosave();
 233}
 234
 235void __init mem_init(void)
 236{
 237        /*
 238         * book3s is limited to 16 page sizes due to encoding this in
 239         * a 4-bit field for slices.
 240         */
 241        BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
 242
 243#ifdef CONFIG_SWIOTLB
 244        /*
 245         * Some platforms (e.g. 85xx) limit DMA-able memory way below
 246         * 4G. We force memblock to bottom-up mode to ensure that the
 247         * memory allocated in swiotlb_init() is DMA-able.
 248         * As it's the last memblock allocation, no need to reset it
 249         * back to to-down.
 250         */
 251        memblock_set_bottom_up(true);
 252        if (is_secure_guest())
 253                svm_swiotlb_init();
 254        else
 255                swiotlb_init(0);
 256#endif
 257
 258        high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 259        set_max_mapnr(max_pfn);
 260
 261        kasan_late_init();
 262
 263        memblock_free_all();
 264
 265#ifdef CONFIG_HIGHMEM
 266        {
 267                unsigned long pfn, highmem_mapnr;
 268
 269                highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 270                for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 271                        phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
 272                        struct page *page = pfn_to_page(pfn);
 273                        if (!memblock_is_reserved(paddr))
 274                                free_highmem_page(page);
 275                }
 276        }
 277#endif /* CONFIG_HIGHMEM */
 278
 279#if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
 280        /*
 281         * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
 282         * functions.... do it here for the non-smp case.
 283         */
 284        per_cpu(next_tlbcam_idx, smp_processor_id()) =
 285                (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
 286#endif
 287
 288#ifdef CONFIG_PPC32
 289        pr_info("Kernel virtual memory layout:\n");
 290#ifdef CONFIG_KASAN
 291        pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
 292                KASAN_SHADOW_START, KASAN_SHADOW_END);
 293#endif
 294        pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
 295#ifdef CONFIG_HIGHMEM
 296        pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
 297                PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
 298#endif /* CONFIG_HIGHMEM */
 299        if (ioremap_bot != IOREMAP_TOP)
 300                pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
 301                        ioremap_bot, IOREMAP_TOP);
 302        pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 303                VMALLOC_START, VMALLOC_END);
 304#ifdef MODULES_VADDR
 305        pr_info("  * 0x%08lx..0x%08lx  : modules\n",
 306                MODULES_VADDR, MODULES_END);
 307#endif
 308#endif /* CONFIG_PPC32 */
 309}
 310
 311void free_initmem(void)
 312{
 313        ppc_md.progress = ppc_printk_progress;
 314        mark_initmem_nx();
 315        init_mem_is_free = true;
 316        free_initmem_default(POISON_FREE_INITMEM);
 317}
 318
 319/*
 320 * System memory should not be in /proc/iomem but various tools expect it
 321 * (eg kdump).
 322 */
 323static int __init add_system_ram_resources(void)
 324{
 325        phys_addr_t start, end;
 326        u64 i;
 327
 328        for_each_mem_range(i, &start, &end) {
 329                struct resource *res;
 330
 331                res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 332                WARN_ON(!res);
 333
 334                if (res) {
 335                        res->name = "System RAM";
 336                        res->start = start;
 337                        /*
 338                         * In memblock, end points to the first byte after
 339                         * the range while in resourses, end points to the
 340                         * last byte in the range.
 341                         */
 342                        res->end = end - 1;
 343                        res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 344                        WARN_ON(request_resource(&iomem_resource, res) < 0);
 345                }
 346        }
 347
 348        return 0;
 349}
 350subsys_initcall(add_system_ram_resources);
 351
 352#ifdef CONFIG_STRICT_DEVMEM
 353/*
 354 * devmem_is_allowed(): check to see if /dev/mem access to a certain address
 355 * is valid. The argument is a physical page number.
 356 *
 357 * Access has to be given to non-kernel-ram areas as well, these contain the
 358 * PCI mmio resources as well as potential bios/acpi data regions.
 359 */
 360int devmem_is_allowed(unsigned long pfn)
 361{
 362        if (page_is_rtas_user_buf(pfn))
 363                return 1;
 364        if (iomem_is_exclusive(PFN_PHYS(pfn)))
 365                return 0;
 366        if (!page_is_ram(pfn))
 367                return 1;
 368        return 0;
 369}
 370#endif /* CONFIG_STRICT_DEVMEM */
 371
 372/*
 373 * This is defined in kernel/resource.c but only powerpc needs to export it, for
 374 * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
 375 */
 376EXPORT_SYMBOL_GPL(walk_system_ram_range);
 377