arch/x86/platform/efi/efi_64.c

   1 /*
   2  * x86_64 specific EFI support functions
   3  * Based on Extensible Firmware Interface Specification version 1.0
   4  *
   5  * Copyright (C) 2005-2008 Intel Co.
   6  *      Fenghua Yu <fenghua.yu@intel.com>
   7  *      Bibo Mao <bibo.mao@intel.com>
   8  *      Chandramouli Narayanan <mouli@linux.intel.com>
   9  *      Huang Ying <ying.huang@intel.com>
  10  *
  11  * Code to convert EFI to E820 map has been implemented in elilo bootloader
  12  * based on a EFI patch by Edgar Hucek. Based on the E820 map, the page table
  13  * is setup appropriately for EFI runtime code.
  14  * - mouli 06/14/2007.
  15  *
  16  */
  17
  18 #include <linux/kernel.h>
  19 #include <linux/init.h>
  20 #include <linux/mm.h>
  21 #include <linux/types.h>
  22 #include <linux/spinlock.h>
  23 #include <linux/bootmem.h>
  24 #include <linux/ioport.h>
  25 #include <linux/module.h>
  26 #include <linux/efi.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/io.h>
  29 #include <linux/reboot.h>
  30 #include <linux/slab.h>
  31
  32 #include <asm/setup.h>
  33 #include <asm/page.h>
  34 #include <asm/e820.h>
  35 #include <asm/pgtable.h>
  36 #include <asm/tlbflush.h>
  37 #include <asm/proto.h>
  38 #include <asm/efi.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/fixmap.h>
  41 #include <asm/realmode.h>
  42
  43 static pgd_t *save_pgd __initdata;
  44 static unsigned long efi_flags __initdata;
  45
  46 /*
  47  * We allocate runtime services regions bottom-up, starting from -4G, i.e.
  48  * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
  49  */
  50 static u64 efi_va       = -4 * (1UL << 30);
  51 #define EFI_VA_END      (-68 * (1UL << 30))
  52
  53 /*
  54  * Scratch space used for switching the pagetable in the EFI stub
  55  */
  56 struct efi_scratch {
  57         u64 r15;
  58         u64 prev_cr3;
  59         pgd_t *efi_pgt;
  60         bool use_pgd;
  61 };
  62
  63 static void __init early_code_mapping_set_exec(int executable)
  64 {
  65         efi_memory_desc_t *md;
  66         void *p;
  67
  68         if (!(__supported_pte_mask & _PAGE_NX))
  69                 return;
  70
  71         /* Make EFI service code area executable */
  72         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
  73                 md = p;
  74                 if (md->type == EFI_RUNTIME_SERVICES_CODE ||
  75                     md->type == EFI_BOOT_SERVICES_CODE)
  76                         efi_set_executable(md, executable);
  77         }
  78 }
  79
  80 void __init efi_call_phys_prelog(void)
  81 {
  82         unsigned long vaddress;
  83         int pgd;
  84         int n_pgds;
  85
  86         if (!efi_enabled(EFI_OLD_MEMMAP))
  87                 return;
  88
  89         early_code_mapping_set_exec(1);
  90         local_irq_save(efi_flags);
  91
  92         n_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT), PGDIR_SIZE);
  93         save_pgd = kmalloc(n_pgds * sizeof(pgd_t), GFP_KERNEL);
  94
  95         for (pgd = 0; pgd < n_pgds; pgd++) {
  96                 save_pgd[pgd] = *pgd_offset_k(pgd * PGDIR_SIZE);
  97                 vaddress = (unsigned long)__va(pgd * PGDIR_SIZE);
  98                 set_pgd(pgd_offset_k(pgd * PGDIR_SIZE), *pgd_offset_k(vaddress));
  99         }
 100         __flush_tlb_all();
 101 }
 102
 103 void __init efi_call_phys_epilog(void)
 104 {
 105         /*
 106          * After the lock is released, the original page table is restored.
 107          */
 108         int pgd;
 109         int n_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
 110
 111         if (!efi_enabled(EFI_OLD_MEMMAP))
 112                 return;
 113
 114         for (pgd = 0; pgd < n_pgds; pgd++)
 115                 set_pgd(pgd_offset_k(pgd * PGDIR_SIZE), save_pgd[pgd]);
 116         kfree(save_pgd);
 117         __flush_tlb_all();
 118         local_irq_restore(efi_flags);
 119         early_code_mapping_set_exec(0);
 120 }
 121
 122 /*
 123  * Add low kernel mappings for passing arguments to EFI functions.
 124  */
 125 void efi_sync_low_kernel_mappings(void)
 126 {
 127         unsigned num_pgds;
 128         pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
 129
 130         if (efi_enabled(EFI_OLD_MEMMAP))
 131                 return;
 132
 133         num_pgds = pgd_index(MODULES_END - 1) - pgd_index(PAGE_OFFSET);
 134
 135         memcpy(pgd + pgd_index(PAGE_OFFSET),
 136                 init_mm.pgd + pgd_index(PAGE_OFFSET),
 137                 sizeof(pgd_t) * num_pgds);
 138 }
 139
 140 void efi_setup_page_tables(void)
 141 {
 142         efi_scratch.efi_pgt = (pgd_t *)(unsigned long)real_mode_header->trampoline_pgd;
 143
 144         if (!efi_enabled(EFI_OLD_MEMMAP))
 145                 efi_scratch.use_pgd = true;
 146 }
 147
 148 static void __init __map_region(efi_memory_desc_t *md, u64 va)
 149 {
 150         pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
 151         unsigned long pf = 0;
 152
 153         if (!(md->attribute & EFI_MEMORY_WB))
 154                 pf |= _PAGE_PCD;
 155
 156         if (kernel_map_pages_in_pgd(pgd, md->phys_addr, va, md->num_pages, pf))
 157                 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
 158                            md->phys_addr, va);
 159 }
 160
 161 void __init efi_map_region(efi_memory_desc_t *md)
 162 {
 163         unsigned long size = md->num_pages << PAGE_SHIFT;
 164         u64 pa = md->phys_addr;
 165
 166         if (efi_enabled(EFI_OLD_MEMMAP))
 167                 return old_map_region(md);
 168
 169         /*
 170          * Make sure the 1:1 mappings are present as a catch-all for b0rked
 171          * firmware which doesn't update all internal pointers after switching
 172          * to virtual mode and would otherwise crap on us.
 173          */
 174         __map_region(md, md->phys_addr);
 175
 176         efi_va -= size;
 177
 178         /* Is PA 2M-aligned? */
 179         if (!(pa & (PMD_SIZE - 1))) {
 180                 efi_va &= PMD_MASK;
 181         } else {
 182                 u64 pa_offset = pa & (PMD_SIZE - 1);
 183                 u64 prev_va = efi_va;
 184
 185                 /* get us the same offset within this 2M page */
 186                 efi_va = (efi_va & PMD_MASK) + pa_offset;
 187
 188                 if (efi_va > prev_va)
 189                         efi_va -= PMD_SIZE;
 190         }
 191
 192         if (efi_va < EFI_VA_END) {
 193                 pr_warn(FW_WARN "VA address range overflow!\n");
 194                 return;
 195         }
 196
 197         /* Do the VA map */
 198         __map_region(md, efi_va);
 199         md->virt_addr = efi_va;
 200 }
 201
 202 /*
 203  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
 204  * md->virt_addr is the original virtual address which had been mapped in kexec
 205  * 1st kernel.
 206  */
 207 void __init efi_map_region_fixed(efi_memory_desc_t *md)
 208 {
 209         __map_region(md, md->virt_addr);
 210 }
 211
 212 void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
 213                                  u32 type, u64 attribute)
 214 {
 215         unsigned long last_map_pfn;
 216
 217         if (type == EFI_MEMORY_MAPPED_IO)
 218                 return ioremap(phys_addr, size);
 219
 220         last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
 221         if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
 222                 unsigned long top = last_map_pfn << PAGE_SHIFT;
 223                 efi_ioremap(top, size - (top - phys_addr), type, attribute);
 224         }
 225
 226         if (!(attribute & EFI_MEMORY_WB))
 227                 efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
 228
 229         return (void __iomem *)__va(phys_addr);
 230 }
 231
 232 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
 233 {
 234         efi_setup = phys_addr + sizeof(struct setup_data);
 235 }
 236
 237 void __init efi_runtime_mkexec(void)
 238 {
 239         if (!efi_enabled(EFI_OLD_MEMMAP))
 240                 return;
 241
 242         if (__supported_pte_mask & _PAGE_NX)
 243                 runtime_code_page_mkexec();
 244 }