Linux Audio

Check our new training course

Loading...
v3.5.6
   1/*
   2 *  include/asm-s390/pgtable.h
   3 *
   4 *  S390 version
   5 *    Copyright (C) 1999,2000 IBM Deutschland Entwicklung GmbH, IBM Corporation
   6 *    Author(s): Hartmut Penner (hp@de.ibm.com)
   7 *               Ulrich Weigand (weigand@de.ibm.com)
   8 *               Martin Schwidefsky (schwidefsky@de.ibm.com)
   9 *
  10 *  Derived from "include/asm-i386/pgtable.h"
  11 */
  12
  13#ifndef _ASM_S390_PGTABLE_H
  14#define _ASM_S390_PGTABLE_H
  15
  16/*
  17 * The Linux memory management assumes a three-level page table setup. For
  18 * s390 31 bit we "fold" the mid level into the top-level page table, so
  19 * that we physically have the same two-level page table as the s390 mmu
  20 * expects in 31 bit mode. For s390 64 bit we use three of the five levels
  21 * the hardware provides (region first and region second tables are not
  22 * used).
  23 *
  24 * The "pgd_xxx()" functions are trivial for a folded two-level
  25 * setup: the pgd is never bad, and a pmd always exists (as it's folded
  26 * into the pgd entry)
  27 *
  28 * This file contains the functions and defines necessary to modify and use
  29 * the S390 page table tree.
  30 */
  31#ifndef __ASSEMBLY__
  32#include <linux/sched.h>
  33#include <linux/mm_types.h>
 
  34#include <asm/bug.h>
  35#include <asm/page.h>
  36
  37extern pgd_t swapper_pg_dir[] __attribute__ ((aligned (4096)));
  38extern void paging_init(void);
  39extern void vmem_map_init(void);
  40extern void fault_init(void);
  41
  42/*
  43 * The S390 doesn't have any external MMU info: the kernel page
  44 * tables contain all the necessary information.
  45 */
  46#define update_mmu_cache(vma, address, ptep)     do { } while (0)
 
  47
  48/*
  49 * ZERO_PAGE is a global shared page that is always zero; used
  50 * for zero-mapped memory areas etc..
  51 */
  52
  53extern unsigned long empty_zero_page;
  54extern unsigned long zero_page_mask;
  55
  56#define ZERO_PAGE(vaddr) \
  57	(virt_to_page((void *)(empty_zero_page + \
  58	 (((unsigned long)(vaddr)) &zero_page_mask))))
 
  59
  60#define is_zero_pfn is_zero_pfn
  61static inline int is_zero_pfn(unsigned long pfn)
  62{
  63	extern unsigned long zero_pfn;
  64	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
  65	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
  66}
  67
  68#define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
  69
  70#endif /* !__ASSEMBLY__ */
  71
  72/*
  73 * PMD_SHIFT determines the size of the area a second-level page
  74 * table can map
  75 * PGDIR_SHIFT determines what a third-level page table entry can map
  76 */
  77#ifndef CONFIG_64BIT
  78# define PMD_SHIFT	20
  79# define PUD_SHIFT	20
  80# define PGDIR_SHIFT	20
  81#else /* CONFIG_64BIT */
  82# define PMD_SHIFT	20
  83# define PUD_SHIFT	31
  84# define PGDIR_SHIFT	42
  85#endif /* CONFIG_64BIT */
  86
  87#define PMD_SIZE        (1UL << PMD_SHIFT)
  88#define PMD_MASK        (~(PMD_SIZE-1))
  89#define PUD_SIZE	(1UL << PUD_SHIFT)
  90#define PUD_MASK	(~(PUD_SIZE-1))
  91#define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
  92#define PGDIR_MASK	(~(PGDIR_SIZE-1))
  93
  94/*
  95 * entries per page directory level: the S390 is two-level, so
  96 * we don't really have any PMD directory physically.
  97 * for S390 segment-table entries are combined to one PGD
  98 * that leads to 1024 pte per pgd
  99 */
 100#define PTRS_PER_PTE	256
 101#ifndef CONFIG_64BIT
 102#define PTRS_PER_PMD	1
 103#define PTRS_PER_PUD	1
 104#else /* CONFIG_64BIT */
 105#define PTRS_PER_PMD	2048
 106#define PTRS_PER_PUD	2048
 107#endif /* CONFIG_64BIT */
 108#define PTRS_PER_PGD	2048
 109
 110#define FIRST_USER_ADDRESS  0
 111
 112#define pte_ERROR(e) \
 113	printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
 114#define pmd_ERROR(e) \
 115	printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
 116#define pud_ERROR(e) \
 117	printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
 118#define pgd_ERROR(e) \
 119	printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
 120
 121#ifndef __ASSEMBLY__
 122/*
 123 * The vmalloc area will always be on the topmost area of the kernel
 124 * mapping. We reserve 96MB (31bit) / 128GB (64bit) for vmalloc,
 125 * which should be enough for any sane case.
 126 * By putting vmalloc at the top, we maximise the gap between physical
 127 * memory and vmalloc to catch misplaced memory accesses. As a side
 128 * effect, this also makes sure that 64 bit module code cannot be used
 129 * as system call address.
 130 */
 131extern unsigned long VMALLOC_START;
 132extern unsigned long VMALLOC_END;
 133extern struct page *vmemmap;
 134
 135#define VMEM_MAX_PHYS ((unsigned long) vmemmap)
 136
 
 
 
 
 
 
 
 
 137/*
 138 * A 31 bit pagetable entry of S390 has following format:
 139 *  |   PFRA          |    |  OS  |
 140 * 0                   0IP0
 141 * 00000000001111111111222222222233
 142 * 01234567890123456789012345678901
 143 *
 144 * I Page-Invalid Bit:    Page is not available for address-translation
 145 * P Page-Protection Bit: Store access not possible for page
 146 *
 147 * A 31 bit segmenttable entry of S390 has following format:
 148 *  |   P-table origin      |  |PTL
 149 * 0                         IC
 150 * 00000000001111111111222222222233
 151 * 01234567890123456789012345678901
 152 *
 153 * I Segment-Invalid Bit:    Segment is not available for address-translation
 154 * C Common-Segment Bit:     Segment is not private (PoP 3-30)
 155 * PTL Page-Table-Length:    Page-table length (PTL+1*16 entries -> up to 256)
 156 *
 157 * The 31 bit segmenttable origin of S390 has following format:
 158 *
 159 *  |S-table origin   |     | STL |
 160 * X                   **GPS
 161 * 00000000001111111111222222222233
 162 * 01234567890123456789012345678901
 163 *
 164 * X Space-Switch event:
 165 * G Segment-Invalid Bit:     *
 166 * P Private-Space Bit:       Segment is not private (PoP 3-30)
 167 * S Storage-Alteration:
 168 * STL Segment-Table-Length:  Segment-table length (STL+1*16 entries -> up to 2048)
 169 *
 170 * A 64 bit pagetable entry of S390 has following format:
 171 * |			 PFRA			      |0IPC|  OS  |
 172 * 0000000000111111111122222222223333333333444444444455555555556666
 173 * 0123456789012345678901234567890123456789012345678901234567890123
 174 *
 175 * I Page-Invalid Bit:    Page is not available for address-translation
 176 * P Page-Protection Bit: Store access not possible for page
 177 * C Change-bit override: HW is not required to set change bit
 178 *
 179 * A 64 bit segmenttable entry of S390 has following format:
 180 * |        P-table origin                              |      TT
 181 * 0000000000111111111122222222223333333333444444444455555555556666
 182 * 0123456789012345678901234567890123456789012345678901234567890123
 183 *
 184 * I Segment-Invalid Bit:    Segment is not available for address-translation
 185 * C Common-Segment Bit:     Segment is not private (PoP 3-30)
 186 * P Page-Protection Bit: Store access not possible for page
 187 * TT Type 00
 188 *
 189 * A 64 bit region table entry of S390 has following format:
 190 * |        S-table origin                             |   TF  TTTL
 191 * 0000000000111111111122222222223333333333444444444455555555556666
 192 * 0123456789012345678901234567890123456789012345678901234567890123
 193 *
 194 * I Segment-Invalid Bit:    Segment is not available for address-translation
 195 * TT Type 01
 196 * TF
 197 * TL Table length
 198 *
 199 * The 64 bit regiontable origin of S390 has following format:
 200 * |      region table origon                          |       DTTL
 201 * 0000000000111111111122222222223333333333444444444455555555556666
 202 * 0123456789012345678901234567890123456789012345678901234567890123
 203 *
 204 * X Space-Switch event:
 205 * G Segment-Invalid Bit:  
 206 * P Private-Space Bit:    
 207 * S Storage-Alteration:
 208 * R Real space
 209 * TL Table-Length:
 210 *
 211 * A storage key has the following format:
 212 * | ACC |F|R|C|0|
 213 *  0   3 4 5 6 7
 214 * ACC: access key
 215 * F  : fetch protection bit
 216 * R  : referenced bit
 217 * C  : changed bit
 218 */
 219
 220/* Hardware bits in the page table entry */
 221#define _PAGE_CO	0x100		/* HW Change-bit override */
 222#define _PAGE_RO	0x200		/* HW read-only bit  */
 223#define _PAGE_INVALID	0x400		/* HW invalid bit    */
 
 224
 225/* Software bits in the page table entry */
 226#define _PAGE_SWT	0x001		/* SW pte type bit t */
 227#define _PAGE_SWX	0x002		/* SW pte type bit x */
 228#define _PAGE_SWC	0x004		/* SW pte changed bit (for KVM) */
 229#define _PAGE_SWR	0x008		/* SW pte referenced bit (for KVM) */
 230#define _PAGE_SPECIAL	0x010		/* SW associated with special page */
 
 
 
 231#define __HAVE_ARCH_PTE_SPECIAL
 232
 233/* Set of bits not changed in pte_modify */
 234#define _PAGE_CHG_MASK	(PAGE_MASK | _PAGE_SPECIAL | _PAGE_SWC | _PAGE_SWR)
 
 235
 236/* Six different types of pages. */
 237#define _PAGE_TYPE_EMPTY	0x400
 238#define _PAGE_TYPE_NONE		0x401
 239#define _PAGE_TYPE_SWAP		0x403
 240#define _PAGE_TYPE_FILE		0x601	/* bit 0x002 is used for offset !! */
 241#define _PAGE_TYPE_RO		0x200
 242#define _PAGE_TYPE_RW		0x000
 243
 244/*
 245 * Only four types for huge pages, using the invalid bit and protection bit
 246 * of a segment table entry.
 247 */
 248#define _HPAGE_TYPE_EMPTY	0x020	/* _SEGMENT_ENTRY_INV */
 249#define _HPAGE_TYPE_NONE	0x220
 250#define _HPAGE_TYPE_RO		0x200	/* _SEGMENT_ENTRY_RO  */
 251#define _HPAGE_TYPE_RW		0x000
 252
 253/*
 254 * PTE type bits are rather complicated. handle_pte_fault uses pte_present,
 255 * pte_none and pte_file to find out the pte type WITHOUT holding the page
 256 * table lock. ptep_clear_flush on the other hand uses ptep_clear_flush to
 257 * invalidate a given pte. ipte sets the hw invalid bit and clears all tlbs
 258 * for the page. The page table entry is set to _PAGE_TYPE_EMPTY afterwards.
 259 * This change is done while holding the lock, but the intermediate step
 260 * of a previously valid pte with the hw invalid bit set can be observed by
 261 * handle_pte_fault. That makes it necessary that all valid pte types with
 262 * the hw invalid bit set must be distinguishable from the four pte types
 263 * empty, none, swap and file.
 264 *
 265 *			irxt  ipte  irxt
 266 * _PAGE_TYPE_EMPTY	1000   ->   1000
 267 * _PAGE_TYPE_NONE	1001   ->   1001
 268 * _PAGE_TYPE_SWAP	1011   ->   1011
 269 * _PAGE_TYPE_FILE	11?1   ->   11?1
 270 * _PAGE_TYPE_RO	0100   ->   1100
 271 * _PAGE_TYPE_RW	0000   ->   1000
 272 *
 273 * pte_none is true for bits combinations 1000, 1010, 1100, 1110
 274 * pte_present is true for bits combinations 0000, 0010, 0100, 0110, 1001
 275 * pte_file is true for bits combinations 1101, 1111
 276 * swap pte is 1011 and 0001, 0011, 0101, 0111 are invalid.
 277 */
 278
 279#ifndef CONFIG_64BIT
 280
 281/* Bits in the segment table address-space-control-element */
 282#define _ASCE_SPACE_SWITCH	0x80000000UL	/* space switch event	    */
 283#define _ASCE_ORIGIN_MASK	0x7ffff000UL	/* segment table origin	    */
 284#define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
 285#define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
 286#define _ASCE_TABLE_LENGTH	0x7f	/* 128 x 64 entries = 8k	    */
 287
 288/* Bits in the segment table entry */
 
 289#define _SEGMENT_ENTRY_ORIGIN	0x7fffffc0UL	/* page table origin	    */
 290#define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
 291#define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
 292#define _SEGMENT_ENTRY_COMMON	0x10	/* common segment bit		    */
 293#define _SEGMENT_ENTRY_PTL	0x0f	/* page table length		    */
 
 294
 295#define _SEGMENT_ENTRY		(_SEGMENT_ENTRY_PTL)
 296#define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
 
 
 
 
 
 
 
 
 
 297
 298/* Page status table bits for virtualization */
 299#define RCP_ACC_BITS	0xf0000000UL
 300#define RCP_FP_BIT	0x08000000UL
 301#define RCP_PCL_BIT	0x00800000UL
 302#define RCP_HR_BIT	0x00400000UL
 303#define RCP_HC_BIT	0x00200000UL
 304#define RCP_GR_BIT	0x00040000UL
 305#define RCP_GC_BIT	0x00020000UL
 306
 307/* User dirty / referenced bit for KVM's migration feature */
 308#define KVM_UR_BIT	0x00008000UL
 309#define KVM_UC_BIT	0x00004000UL
 310
 311#else /* CONFIG_64BIT */
 312
 313/* Bits in the segment/region table address-space-control-element */
 314#define _ASCE_ORIGIN		~0xfffUL/* segment table origin		    */
 315#define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
 316#define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
 317#define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
 318#define _ASCE_REAL_SPACE	0x20	/* real space control		    */
 319#define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
 320#define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
 321#define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
 322#define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
 323#define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
 324#define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
 325
 326/* Bits in the region table entry */
 327#define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
 328#define _REGION_ENTRY_INV	0x20	/* invalid region table entry	    */
 
 329#define _REGION_ENTRY_TYPE_MASK	0x0c	/* region/segment table type mask   */
 330#define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
 331#define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
 332#define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
 333#define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
 334
 335#define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
 336#define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INV)
 337#define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
 338#define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INV)
 339#define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
 340#define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INV)
 
 
 
 
 341
 342/* Bits in the segment table entry */
 
 
 
 343#define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* segment table origin		    */
 344#define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
 345#define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
 346
 347#define _SEGMENT_ENTRY		(0)
 348#define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
 349
 350#define _SEGMENT_ENTRY_LARGE	0x400	/* STE-format control, large page   */
 351#define _SEGMENT_ENTRY_CO	0x100	/* change-recording override   */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 352
 353/* Page status table bits for virtualization */
 354#define RCP_ACC_BITS	0xf000000000000000UL
 355#define RCP_FP_BIT	0x0800000000000000UL
 356#define RCP_PCL_BIT	0x0080000000000000UL
 357#define RCP_HR_BIT	0x0040000000000000UL
 358#define RCP_HC_BIT	0x0020000000000000UL
 359#define RCP_GR_BIT	0x0004000000000000UL
 360#define RCP_GC_BIT	0x0002000000000000UL
 361
 362/* User dirty / referenced bit for KVM's migration feature */
 363#define KVM_UR_BIT	0x0000800000000000UL
 364#define KVM_UC_BIT	0x0000400000000000UL
 365
 366#endif /* CONFIG_64BIT */
 367
 
 
 
 
 
 
 368/*
 369 * A user page table pointer has the space-switch-event bit, the
 370 * private-space-control bit and the storage-alteration-event-control
 371 * bit set. A kernel page table pointer doesn't need them.
 372 */
 373#define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
 374				 _ASCE_ALT_EVENT)
 375
 376/*
 377 * Page protection definitions.
 378 */
 379#define PAGE_NONE	__pgprot(_PAGE_TYPE_NONE)
 380#define PAGE_RO		__pgprot(_PAGE_TYPE_RO)
 381#define PAGE_RW		__pgprot(_PAGE_TYPE_RW)
 382
 383#define PAGE_KERNEL	PAGE_RW
 384#define PAGE_COPY	PAGE_RO
 
 
 
 
 
 
 385
 386/*
 387 * On s390 the page table entry has an invalid bit and a read-only bit.
 388 * Read permission implies execute permission and write permission
 389 * implies read permission.
 390 */
 391         /*xwr*/
 392#define __P000	PAGE_NONE
 393#define __P001	PAGE_RO
 394#define __P010	PAGE_RO
 395#define __P011	PAGE_RO
 396#define __P100	PAGE_RO
 397#define __P101	PAGE_RO
 398#define __P110	PAGE_RO
 399#define __P111	PAGE_RO
 400
 401#define __S000	PAGE_NONE
 402#define __S001	PAGE_RO
 403#define __S010	PAGE_RW
 404#define __S011	PAGE_RW
 405#define __S100	PAGE_RO
 406#define __S101	PAGE_RO
 407#define __S110	PAGE_RW
 408#define __S111	PAGE_RW
 409
 410static inline int mm_exclusive(struct mm_struct *mm)
 411{
 412	return likely(mm == current->active_mm &&
 413		      atomic_read(&mm->context.attach_count) <= 1);
 414}
 
 
 
 415
 416static inline int mm_has_pgste(struct mm_struct *mm)
 417{
 418#ifdef CONFIG_PGSTE
 419	if (unlikely(mm->context.has_pgste))
 420		return 1;
 421#endif
 422	return 0;
 423}
 424/*
 425 * pgd/pmd/pte query functions
 426 */
 427#ifndef CONFIG_64BIT
 428
 429static inline int pgd_present(pgd_t pgd) { return 1; }
 430static inline int pgd_none(pgd_t pgd)    { return 0; }
 431static inline int pgd_bad(pgd_t pgd)     { return 0; }
 432
 433static inline int pud_present(pud_t pud) { return 1; }
 434static inline int pud_none(pud_t pud)	 { return 0; }
 
 435static inline int pud_bad(pud_t pud)	 { return 0; }
 436
 437#else /* CONFIG_64BIT */
 438
 439static inline int pgd_present(pgd_t pgd)
 440{
 441	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
 442		return 1;
 443	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
 444}
 445
 446static inline int pgd_none(pgd_t pgd)
 447{
 448	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
 449		return 0;
 450	return (pgd_val(pgd) & _REGION_ENTRY_INV) != 0UL;
 451}
 452
 453static inline int pgd_bad(pgd_t pgd)
 454{
 455	/*
 456	 * With dynamic page table levels the pgd can be a region table
 457	 * entry or a segment table entry. Check for the bit that are
 458	 * invalid for either table entry.
 459	 */
 460	unsigned long mask =
 461		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
 462		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
 463	return (pgd_val(pgd) & mask) != 0;
 464}
 465
 466static inline int pud_present(pud_t pud)
 467{
 468	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
 469		return 1;
 470	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
 471}
 472
 473static inline int pud_none(pud_t pud)
 474{
 475	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
 476		return 0;
 477	return (pud_val(pud) & _REGION_ENTRY_INV) != 0UL;
 
 
 
 
 
 
 
 478}
 479
 480static inline int pud_bad(pud_t pud)
 481{
 482	/*
 483	 * With dynamic page table levels the pud can be a region table
 484	 * entry or a segment table entry. Check for the bit that are
 485	 * invalid for either table entry.
 486	 */
 487	unsigned long mask =
 488		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
 489		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
 490	return (pud_val(pud) & mask) != 0;
 491}
 492
 493#endif /* CONFIG_64BIT */
 494
 495static inline int pmd_present(pmd_t pmd)
 496{
 497	return (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN) != 0UL;
 498}
 499
 500static inline int pmd_none(pmd_t pmd)
 501{
 502	return (pmd_val(pmd) & _SEGMENT_ENTRY_INV) != 0UL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 503}
 504
 505static inline int pmd_bad(pmd_t pmd)
 506{
 507	unsigned long mask = ~_SEGMENT_ENTRY_ORIGIN & ~_SEGMENT_ENTRY_INV;
 508	return (pmd_val(pmd) & mask) != _SEGMENT_ENTRY;
 
 
 
 509}
 510
 511static inline int pte_none(pte_t pte)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 512{
 513	return (pte_val(pte) & _PAGE_INVALID) && !(pte_val(pte) & _PAGE_SWT);
 
 
 
 
 
 
 
 514}
 515
 516static inline int pte_present(pte_t pte)
 517{
 518	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
 519	return (pte_val(pte) & mask) == _PAGE_TYPE_NONE ||
 520		(!(pte_val(pte) & _PAGE_INVALID) &&
 521		 !(pte_val(pte) & _PAGE_SWT));
 
 
 
 
 
 
 
 
 
 
 
 
 522}
 523
 524static inline int pte_file(pte_t pte)
 525{
 526	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT;
 527	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
 
 528}
 529
 530static inline int pte_special(pte_t pte)
 531{
 532	return (pte_val(pte) & _PAGE_SPECIAL);
 533}
 534
 535#define __HAVE_ARCH_PTE_SAME
 536static inline int pte_same(pte_t a, pte_t b)
 537{
 538	return pte_val(a) == pte_val(b);
 539}
 540
 541static inline pgste_t pgste_get_lock(pte_t *ptep)
 542{
 543	unsigned long new = 0;
 544#ifdef CONFIG_PGSTE
 545	unsigned long old;
 546
 547	preempt_disable();
 548	asm(
 549		"	lg	%0,%2\n"
 550		"0:	lgr	%1,%0\n"
 551		"	nihh	%0,0xff7f\n"	/* clear RCP_PCL_BIT in old */
 552		"	oihh	%1,0x0080\n"	/* set RCP_PCL_BIT in new */
 553		"	csg	%0,%1,%2\n"
 554		"	jl	0b\n"
 555		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
 556		: "Q" (ptep[PTRS_PER_PTE]) : "cc");
 557#endif
 558	return __pgste(new);
 559}
 560
 561static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
 562{
 563#ifdef CONFIG_PGSTE
 564	asm(
 565		"	nihh	%1,0xff7f\n"	/* clear RCP_PCL_BIT */
 566		"	stg	%1,%0\n"
 567		: "=Q" (ptep[PTRS_PER_PTE])
 568		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE]) : "cc");
 
 569	preempt_enable();
 570#endif
 571}
 572
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 573static inline pgste_t pgste_update_all(pte_t *ptep, pgste_t pgste)
 574{
 575#ifdef CONFIG_PGSTE
 576	unsigned long address, bits;
 577	unsigned char skey;
 578
 579	if (!pte_present(*ptep))
 580		return pgste;
 581	address = pte_val(*ptep) & PAGE_MASK;
 582	skey = page_get_storage_key(address);
 583	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
 584	/* Clear page changed & referenced bit in the storage key */
 585	if (bits & _PAGE_CHANGED)
 586		page_set_storage_key(address, skey ^ bits, 1);
 587	else if (bits)
 
 
 
 
 588		page_reset_referenced(address);
 
 589	/* Transfer page changed & referenced bit to guest bits in pgste */
 590	pgste_val(pgste) |= bits << 48;		/* RCP_GR_BIT & RCP_GC_BIT */
 591	/* Get host changed & referenced bits from pgste */
 592	bits |= (pgste_val(pgste) & (RCP_HR_BIT | RCP_HC_BIT)) >> 52;
 593	/* Clear host bits in pgste. */
 594	pgste_val(pgste) &= ~(RCP_HR_BIT | RCP_HC_BIT);
 595	pgste_val(pgste) &= ~(RCP_ACC_BITS | RCP_FP_BIT);
 596	/* Copy page access key and fetch protection bit to pgste */
 597	pgste_val(pgste) |=
 598		(unsigned long) (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
 599	/* Transfer changed and referenced to kvm user bits */
 600	pgste_val(pgste) |= bits << 45;		/* KVM_UR_BIT & KVM_UC_BIT */
 601	/* Transfer changed & referenced to pte sofware bits */
 602	pte_val(*ptep) |= bits << 1;		/* _PAGE_SWR & _PAGE_SWC */
 603#endif
 604	return pgste;
 605
 606}
 607
 608static inline pgste_t pgste_update_young(pte_t *ptep, pgste_t pgste)
 609{
 610#ifdef CONFIG_PGSTE
 611	int young;
 612
 613	if (!pte_present(*ptep))
 614		return pgste;
 615	young = page_reset_referenced(pte_val(*ptep) & PAGE_MASK);
 616	/* Transfer page referenced bit to pte software bit (host view) */
 617	if (young || (pgste_val(pgste) & RCP_HR_BIT))
 618		pte_val(*ptep) |= _PAGE_SWR;
 619	/* Clear host referenced bit in pgste. */
 620	pgste_val(pgste) &= ~RCP_HR_BIT;
 621	/* Transfer page referenced bit to guest bit in pgste */
 622	pgste_val(pgste) |= (unsigned long) young << 50; /* set RCP_GR_BIT */
 623#endif
 624	return pgste;
 625
 626}
 627
 628static inline void pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
 629{
 630#ifdef CONFIG_PGSTE
 631	unsigned long address;
 632	unsigned long okey, nkey;
 633
 634	if (!pte_present(entry))
 635		return;
 
 636	address = pte_val(entry) & PAGE_MASK;
 637	okey = nkey = page_get_storage_key(address);
 638	nkey &= ~(_PAGE_ACC_BITS | _PAGE_FP_BIT);
 639	/* Set page access key and fetch protection bit from pgste */
 640	nkey |= (pgste_val(pgste) & (RCP_ACC_BITS | RCP_FP_BIT)) >> 56;
 641	if (okey != nkey)
 642		page_set_storage_key(address, nkey, 1);
 
 643#endif
 644}
 645
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 646/**
 647 * struct gmap_struct - guest address space
 648 * @mm: pointer to the parent mm_struct
 649 * @table: pointer to the page directory
 650 * @asce: address space control element for gmap page table
 651 * @crst_list: list of all crst tables used in the guest address space
 
 652 */
 653struct gmap {
 654	struct list_head list;
 655	struct mm_struct *mm;
 656	unsigned long *table;
 657	unsigned long asce;
 
 658	struct list_head crst_list;
 
 659};
 660
 661/**
 662 * struct gmap_rmap - reverse mapping for segment table entries
 663 * @next: pointer to the next gmap_rmap structure in the list
 664 * @entry: pointer to a segment table entry
 
 665 */
 666struct gmap_rmap {
 667	struct list_head list;
 
 668	unsigned long *entry;
 
 669};
 670
 671/**
 672 * struct gmap_pgtable - gmap information attached to a page table
 673 * @vmaddr: address of the 1MB segment in the process virtual memory
 674 * @mapper: list of segment table entries maping a page table
 675 */
 676struct gmap_pgtable {
 677	unsigned long vmaddr;
 678	struct list_head mapper;
 679};
 680
 
 
 
 
 
 
 
 
 
 681struct gmap *gmap_alloc(struct mm_struct *mm);
 682void gmap_free(struct gmap *gmap);
 683void gmap_enable(struct gmap *gmap);
 684void gmap_disable(struct gmap *gmap);
 685int gmap_map_segment(struct gmap *gmap, unsigned long from,
 686		     unsigned long to, unsigned long length);
 687int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len);
 
 
 688unsigned long __gmap_fault(unsigned long address, struct gmap *);
 689unsigned long gmap_fault(unsigned long address, struct gmap *);
 690void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 691
 692/*
 693 * Certain architectures need to do special things when PTEs
 694 * within a page table are directly modified.  Thus, the following
 695 * hook is made available.
 696 */
 697static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
 698			      pte_t *ptep, pte_t entry)
 699{
 700	pgste_t pgste;
 701
 702	if (mm_has_pgste(mm)) {
 703		pgste = pgste_get_lock(ptep);
 704		pgste_set_pte(ptep, pgste, entry);
 705		*ptep = entry;
 
 706		pgste_set_unlock(ptep, pgste);
 707	} else
 
 
 708		*ptep = entry;
 
 709}
 710
 711/*
 712 * query functions pte_write/pte_dirty/pte_young only work if
 713 * pte_present() is true. Undefined behaviour if not..
 714 */
 715static inline int pte_write(pte_t pte)
 716{
 717	return (pte_val(pte) & _PAGE_RO) == 0;
 718}
 719
 720static inline int pte_dirty(pte_t pte)
 721{
 722#ifdef CONFIG_PGSTE
 723	if (pte_val(pte) & _PAGE_SWC)
 724		return 1;
 725#endif
 726	return 0;
 727}
 728
 729static inline int pte_young(pte_t pte)
 730{
 731#ifdef CONFIG_PGSTE
 732	if (pte_val(pte) & _PAGE_SWR)
 733		return 1;
 734#endif
 735	return 0;
 
 
 736}
 737
 738/*
 739 * pgd/pmd/pte modification functions
 740 */
 741
 742static inline void pgd_clear(pgd_t *pgd)
 743{
 744#ifdef CONFIG_64BIT
 745	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
 746		pgd_val(*pgd) = _REGION2_ENTRY_EMPTY;
 747#endif
 748}
 749
 750static inline void pud_clear(pud_t *pud)
 751{
 752#ifdef CONFIG_64BIT
 753	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
 754		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
 755#endif
 756}
 757
 758static inline void pmd_clear(pmd_t *pmdp)
 759{
 760	pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
 761}
 762
 763static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 764{
 765	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
 766}
 767
 768/*
 769 * The following pte modification functions only work if
 770 * pte_present() is true. Undefined behaviour if not..
 771 */
 772static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
 773{
 774	pte_val(pte) &= _PAGE_CHG_MASK;
 775	pte_val(pte) |= pgprot_val(newprot);
 
 
 
 
 
 
 
 
 
 
 
 
 776	return pte;
 777}
 778
 779static inline pte_t pte_wrprotect(pte_t pte)
 780{
 781	/* Do not clobber _PAGE_TYPE_NONE pages!  */
 782	if (!(pte_val(pte) & _PAGE_INVALID))
 783		pte_val(pte) |= _PAGE_RO;
 784	return pte;
 785}
 786
 787static inline pte_t pte_mkwrite(pte_t pte)
 788{
 789	pte_val(pte) &= ~_PAGE_RO;
 
 
 790	return pte;
 791}
 792
 793static inline pte_t pte_mkclean(pte_t pte)
 794{
 795#ifdef CONFIG_PGSTE
 796	pte_val(pte) &= ~_PAGE_SWC;
 797#endif
 798	return pte;
 799}
 800
 801static inline pte_t pte_mkdirty(pte_t pte)
 802{
 
 
 
 803	return pte;
 804}
 805
 806static inline pte_t pte_mkold(pte_t pte)
 807{
 808#ifdef CONFIG_PGSTE
 809	pte_val(pte) &= ~_PAGE_SWR;
 810#endif
 811	return pte;
 812}
 813
 814static inline pte_t pte_mkyoung(pte_t pte)
 815{
 
 
 
 816	return pte;
 817}
 818
 819static inline pte_t pte_mkspecial(pte_t pte)
 820{
 821	pte_val(pte) |= _PAGE_SPECIAL;
 822	return pte;
 823}
 824
 825#ifdef CONFIG_HUGETLB_PAGE
 826static inline pte_t pte_mkhuge(pte_t pte)
 827{
 828	/*
 829	 * PROT_NONE needs to be remapped from the pte type to the ste type.
 830	 * The HW invalid bit is also different for pte and ste. The pte
 831	 * invalid bit happens to be the same as the ste _SEGMENT_ENTRY_LARGE
 832	 * bit, so we don't have to clear it.
 833	 */
 834	if (pte_val(pte) & _PAGE_INVALID) {
 835		if (pte_val(pte) & _PAGE_SWT)
 836			pte_val(pte) |= _HPAGE_TYPE_NONE;
 837		pte_val(pte) |= _SEGMENT_ENTRY_INV;
 838	}
 839	/*
 840	 * Clear SW pte bits SWT and SWX, there are no SW bits in a segment
 841	 * table entry.
 842	 */
 843	pte_val(pte) &= ~(_PAGE_SWT | _PAGE_SWX);
 844	/*
 845	 * Also set the change-override bit because we don't need dirty bit
 846	 * tracking for hugetlbfs pages.
 847	 */
 848	pte_val(pte) |= (_SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_CO);
 849	return pte;
 850}
 851#endif
 852
 853/*
 854 * Get (and clear) the user dirty bit for a pte.
 855 */
 856static inline int ptep_test_and_clear_user_dirty(struct mm_struct *mm,
 857						 pte_t *ptep)
 858{
 859	pgste_t pgste;
 860	int dirty = 0;
 861
 862	if (mm_has_pgste(mm)) {
 863		pgste = pgste_get_lock(ptep);
 864		pgste = pgste_update_all(ptep, pgste);
 865		dirty = !!(pgste_val(pgste) & KVM_UC_BIT);
 866		pgste_val(pgste) &= ~KVM_UC_BIT;
 867		pgste_set_unlock(ptep, pgste);
 868		return dirty;
 869	}
 870	return dirty;
 871}
 872
 873/*
 874 * Get (and clear) the user referenced bit for a pte.
 875 */
 876static inline int ptep_test_and_clear_user_young(struct mm_struct *mm,
 877						 pte_t *ptep)
 878{
 879	pgste_t pgste;
 880	int young = 0;
 881
 882	if (mm_has_pgste(mm)) {
 883		pgste = pgste_get_lock(ptep);
 884		pgste = pgste_update_young(ptep, pgste);
 885		young = !!(pgste_val(pgste) & KVM_UR_BIT);
 886		pgste_val(pgste) &= ~KVM_UR_BIT;
 887		pgste_set_unlock(ptep, pgste);
 888	}
 889	return young;
 890}
 891
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 892#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
 893static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
 894					    unsigned long addr, pte_t *ptep)
 895{
 896	pgste_t pgste;
 897	pte_t pte;
 
 898
 899	if (mm_has_pgste(vma->vm_mm)) {
 900		pgste = pgste_get_lock(ptep);
 901		pgste = pgste_update_young(ptep, pgste);
 902		pte = *ptep;
 903		*ptep = pte_mkold(pte);
 904		pgste_set_unlock(ptep, pgste);
 905		return pte_young(pte);
 906	}
 907	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 908}
 909
 910#define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
 911static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
 912					 unsigned long address, pte_t *ptep)
 913{
 914	/* No need to flush TLB
 915	 * On s390 reference bits are in storage key and never in TLB
 916	 * With virtualization we handle the reference bit, without we
 917	 * we can simply return */
 918	return ptep_test_and_clear_young(vma, address, ptep);
 919}
 920
 921static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
 922{
 923	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
 924#ifndef CONFIG_64BIT
 925		/* pto must point to the start of the segment table */
 926		pte_t *pto = (pte_t *) (((unsigned long) ptep) & 0x7ffffc00);
 927#else
 928		/* ipte in zarch mode can do the math */
 929		pte_t *pto = ptep;
 930#endif
 931		asm volatile(
 932			"	ipte	%2,%3"
 933			: "=m" (*ptep) : "m" (*ptep),
 934			  "a" (pto), "a" (address));
 935	}
 936}
 937
 938/*
 939 * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
 940 * both clear the TLB for the unmapped pte. The reason is that
 941 * ptep_get_and_clear is used in common code (e.g. change_pte_range)
 942 * to modify an active pte. The sequence is
 943 *   1) ptep_get_and_clear
 944 *   2) set_pte_at
 945 *   3) flush_tlb_range
 946 * On s390 the tlb needs to get flushed with the modification of the pte
 947 * if the pte is active. The only way how this can be implemented is to
 948 * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
 949 * is a nop.
 950 */
 951#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
 952static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
 953				       unsigned long address, pte_t *ptep)
 954{
 955	pgste_t pgste;
 956	pte_t pte;
 957
 958	mm->context.flush_mm = 1;
 959	if (mm_has_pgste(mm))
 960		pgste = pgste_get_lock(ptep);
 
 
 961
 962	pte = *ptep;
 963	if (!mm_exclusive(mm))
 964		__ptep_ipte(address, ptep);
 965	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
 966
 967	if (mm_has_pgste(mm)) {
 968		pgste = pgste_update_all(&pte, pgste);
 969		pgste_set_unlock(ptep, pgste);
 970	}
 971	return pte;
 972}
 973
 974#define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 975static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 976					   unsigned long address,
 977					   pte_t *ptep)
 978{
 
 979	pte_t pte;
 980
 981	mm->context.flush_mm = 1;
 982	if (mm_has_pgste(mm))
 983		pgste_get_lock(ptep);
 
 984
 985	pte = *ptep;
 986	if (!mm_exclusive(mm))
 987		__ptep_ipte(address, ptep);
 
 
 
 
 988	return pte;
 989}
 990
 991static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 992					   unsigned long address,
 993					   pte_t *ptep, pte_t pte)
 994{
 995	*ptep = pte;
 996	if (mm_has_pgste(mm))
 997		pgste_set_unlock(ptep, *(pgste_t *)(ptep + PTRS_PER_PTE));
 
 
 
 
 
 
 998}
 999
1000#define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1001static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1002				     unsigned long address, pte_t *ptep)
1003{
1004	pgste_t pgste;
1005	pte_t pte;
1006
1007	if (mm_has_pgste(vma->vm_mm))
1008		pgste = pgste_get_lock(ptep);
 
 
1009
1010	pte = *ptep;
1011	__ptep_ipte(address, ptep);
1012	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1013
1014	if (mm_has_pgste(vma->vm_mm)) {
 
 
 
1015		pgste = pgste_update_all(&pte, pgste);
1016		pgste_set_unlock(ptep, pgste);
1017	}
1018	return pte;
1019}
1020
1021/*
1022 * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1023 * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1024 * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1025 * cannot be accessed while the batched unmap is running. In this case
1026 * full==1 and a simple pte_clear is enough. See tlb.h.
1027 */
1028#define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1029static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1030					    unsigned long address,
1031					    pte_t *ptep, int full)
1032{
1033	pgste_t pgste;
1034	pte_t pte;
1035
1036	if (mm_has_pgste(mm))
1037		pgste = pgste_get_lock(ptep);
 
 
1038
1039	pte = *ptep;
1040	if (!full)
1041		__ptep_ipte(address, ptep);
1042	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1043
1044	if (mm_has_pgste(mm)) {
1045		pgste = pgste_update_all(&pte, pgste);
1046		pgste_set_unlock(ptep, pgste);
1047	}
1048	return pte;
1049}
1050
1051#define __HAVE_ARCH_PTEP_SET_WRPROTECT
1052static inline pte_t ptep_set_wrprotect(struct mm_struct *mm,
1053				       unsigned long address, pte_t *ptep)
1054{
1055	pgste_t pgste;
1056	pte_t pte = *ptep;
1057
1058	if (pte_write(pte)) {
1059		mm->context.flush_mm = 1;
1060		if (mm_has_pgste(mm))
1061			pgste = pgste_get_lock(ptep);
 
 
1062
1063		if (!mm_exclusive(mm))
1064			__ptep_ipte(address, ptep);
1065		*ptep = pte_wrprotect(pte);
1066
1067		if (mm_has_pgste(mm))
 
1068			pgste_set_unlock(ptep, pgste);
 
 
1069	}
1070	return pte;
1071}
1072
1073#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1074static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1075					unsigned long address, pte_t *ptep,
1076					pte_t entry, int dirty)
1077{
1078	pgste_t pgste;
1079
1080	if (pte_same(*ptep, entry))
1081		return 0;
1082	if (mm_has_pgste(vma->vm_mm))
1083		pgste = pgste_get_lock(ptep);
 
 
1084
1085	__ptep_ipte(address, ptep);
1086	*ptep = entry;
1087
1088	if (mm_has_pgste(vma->vm_mm))
 
1089		pgste_set_unlock(ptep, pgste);
 
 
1090	return 1;
1091}
1092
1093/*
1094 * Conversion functions: convert a page and protection to a page entry,
1095 * and a page entry and page directory to the page they refer to.
1096 */
1097static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1098{
1099	pte_t __pte;
1100	pte_val(__pte) = physpage + pgprot_val(pgprot);
1101	return __pte;
1102}
1103
1104static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1105{
1106	unsigned long physpage = page_to_phys(page);
 
1107
1108	return mk_pte_phys(physpage, pgprot);
 
 
1109}
1110
1111#define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1112#define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1113#define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1114#define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
1115
1116#define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
1117#define pgd_offset_k(address) pgd_offset(&init_mm, address)
1118
1119#ifndef CONFIG_64BIT
1120
1121#define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1122#define pud_deref(pmd) ({ BUG(); 0UL; })
1123#define pgd_deref(pmd) ({ BUG(); 0UL; })
1124
1125#define pud_offset(pgd, address) ((pud_t *) pgd)
1126#define pmd_offset(pud, address) ((pmd_t *) pud + pmd_index(address))
1127
1128#else /* CONFIG_64BIT */
1129
1130#define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1131#define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
1132#define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)
1133
1134static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
1135{
1136	pud_t *pud = (pud_t *) pgd;
1137	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
1138		pud = (pud_t *) pgd_deref(*pgd);
1139	return pud  + pud_index(address);
1140}
1141
1142static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
1143{
1144	pmd_t *pmd = (pmd_t *) pud;
1145	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
1146		pmd = (pmd_t *) pud_deref(*pud);
1147	return pmd + pmd_index(address);
1148}
1149
1150#endif /* CONFIG_64BIT */
1151
1152#define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
1153#define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1154#define pte_page(x) pfn_to_page(pte_pfn(x))
1155
1156#define pmd_page(pmd) pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT)
1157
1158/* Find an entry in the lowest level page table.. */
1159#define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
1160#define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
1161#define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
1162#define pte_unmap(pte) do { } while (0)
1163
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1164/*
1165 * 31 bit swap entry format:
1166 * A page-table entry has some bits we have to treat in a special way.
1167 * Bits 0, 20 and bit 23 have to be zero, otherwise an specification
1168 * exception will occur instead of a page translation exception. The
1169 * specifiation exception has the bad habit not to store necessary
1170 * information in the lowcore.
1171 * Bit 21 and bit 22 are the page invalid bit and the page protection
1172 * bit. We set both to indicate a swapped page.
1173 * Bit 30 and 31 are used to distinguish the different page types. For
1174 * a swapped page these bits need to be zero.
1175 * This leaves the bits 1-19 and bits 24-29 to store type and offset.
1176 * We use the 5 bits from 25-29 for the type and the 20 bits from 1-19
1177 * plus 24 for the offset.
1178 * 0|     offset        |0110|o|type |00|
1179 * 0 0000000001111111111 2222 2 22222 33
1180 * 0 1234567890123456789 0123 4 56789 01
1181 *
1182 * 64 bit swap entry format:
1183 * A page-table entry has some bits we have to treat in a special way.
1184 * Bits 52 and bit 55 have to be zero, otherwise an specification
1185 * exception will occur instead of a page translation exception. The
1186 * specifiation exception has the bad habit not to store necessary
1187 * information in the lowcore.
1188 * Bit 53 and bit 54 are the page invalid bit and the page protection
1189 * bit. We set both to indicate a swapped page.
1190 * Bit 62 and 63 are used to distinguish the different page types. For
1191 * a swapped page these bits need to be zero.
1192 * This leaves the bits 0-51 and bits 56-61 to store type and offset.
1193 * We use the 5 bits from 57-61 for the type and the 53 bits from 0-51
1194 * plus 56 for the offset.
1195 * |                      offset                        |0110|o|type |00|
1196 *  0000000000111111111122222222223333333333444444444455 5555 5 55566 66
1197 *  0123456789012345678901234567890123456789012345678901 2345 6 78901 23
1198 */
1199#ifndef CONFIG_64BIT
1200#define __SWP_OFFSET_MASK (~0UL >> 12)
1201#else
1202#define __SWP_OFFSET_MASK (~0UL >> 11)
1203#endif
1204static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1205{
1206	pte_t pte;
1207	offset &= __SWP_OFFSET_MASK;
1208	pte_val(pte) = _PAGE_TYPE_SWAP | ((type & 0x1f) << 2) |
1209		((offset & 1UL) << 7) | ((offset & ~1UL) << 11);
1210	return pte;
1211}
1212
1213#define __swp_type(entry)	(((entry).val >> 2) & 0x1f)
1214#define __swp_offset(entry)	(((entry).val >> 11) | (((entry).val >> 7) & 1))
1215#define __swp_entry(type,offset) ((swp_entry_t) { pte_val(mk_swap_pte((type),(offset))) })
1216
1217#define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1218#define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1219
1220#ifndef CONFIG_64BIT
1221# define PTE_FILE_MAX_BITS	26
1222#else /* CONFIG_64BIT */
1223# define PTE_FILE_MAX_BITS	59
1224#endif /* CONFIG_64BIT */
1225
1226#define pte_to_pgoff(__pte) \
1227	((((__pte).pte >> 12) << 7) + (((__pte).pte >> 1) & 0x7f))
1228
1229#define pgoff_to_pte(__off) \
1230	((pte_t) { ((((__off) & 0x7f) << 1) + (((__off) >> 7) << 12)) \
1231		   | _PAGE_TYPE_FILE })
1232
1233#endif /* !__ASSEMBLY__ */
1234
1235#define kern_addr_valid(addr)   (1)
1236
1237extern int vmem_add_mapping(unsigned long start, unsigned long size);
1238extern int vmem_remove_mapping(unsigned long start, unsigned long size);
1239extern int s390_enable_sie(void);
1240
1241/*
1242 * No page table caches to initialise
1243 */
1244#define pgtable_cache_init()	do { } while (0)
 
1245
1246#include <asm-generic/pgtable.h>
1247
1248#endif /* _S390_PAGE_H */
v3.15
   1/*
 
 
   2 *  S390 version
   3 *    Copyright IBM Corp. 1999, 2000
   4 *    Author(s): Hartmut Penner (hp@de.ibm.com)
   5 *               Ulrich Weigand (weigand@de.ibm.com)
   6 *               Martin Schwidefsky (schwidefsky@de.ibm.com)
   7 *
   8 *  Derived from "include/asm-i386/pgtable.h"
   9 */
  10
  11#ifndef _ASM_S390_PGTABLE_H
  12#define _ASM_S390_PGTABLE_H
  13
  14/*
  15 * The Linux memory management assumes a three-level page table setup. For
  16 * s390 31 bit we "fold" the mid level into the top-level page table, so
  17 * that we physically have the same two-level page table as the s390 mmu
  18 * expects in 31 bit mode. For s390 64 bit we use three of the five levels
  19 * the hardware provides (region first and region second tables are not
  20 * used).
  21 *
  22 * The "pgd_xxx()" functions are trivial for a folded two-level
  23 * setup: the pgd is never bad, and a pmd always exists (as it's folded
  24 * into the pgd entry)
  25 *
  26 * This file contains the functions and defines necessary to modify and use
  27 * the S390 page table tree.
  28 */
  29#ifndef __ASSEMBLY__
  30#include <linux/sched.h>
  31#include <linux/mm_types.h>
  32#include <linux/page-flags.h>
  33#include <asm/bug.h>
  34#include <asm/page.h>
  35
  36extern pgd_t swapper_pg_dir[] __attribute__ ((aligned (4096)));
  37extern void paging_init(void);
  38extern void vmem_map_init(void);
 
  39
  40/*
  41 * The S390 doesn't have any external MMU info: the kernel page
  42 * tables contain all the necessary information.
  43 */
  44#define update_mmu_cache(vma, address, ptep)     do { } while (0)
  45#define update_mmu_cache_pmd(vma, address, ptep) do { } while (0)
  46
  47/*
  48 * ZERO_PAGE is a global shared page that is always zero; used
  49 * for zero-mapped memory areas etc..
  50 */
  51
  52extern unsigned long empty_zero_page;
  53extern unsigned long zero_page_mask;
  54
  55#define ZERO_PAGE(vaddr) \
  56	(virt_to_page((void *)(empty_zero_page + \
  57	 (((unsigned long)(vaddr)) &zero_page_mask))))
  58#define __HAVE_COLOR_ZERO_PAGE
  59
  60/* TODO: s390 cannot support io_remap_pfn_range... */
 
 
 
 
 
 
 
 
 
  61#endif /* !__ASSEMBLY__ */
  62
  63/*
  64 * PMD_SHIFT determines the size of the area a second-level page
  65 * table can map
  66 * PGDIR_SHIFT determines what a third-level page table entry can map
  67 */
  68#ifndef CONFIG_64BIT
  69# define PMD_SHIFT	20
  70# define PUD_SHIFT	20
  71# define PGDIR_SHIFT	20
  72#else /* CONFIG_64BIT */
  73# define PMD_SHIFT	20
  74# define PUD_SHIFT	31
  75# define PGDIR_SHIFT	42
  76#endif /* CONFIG_64BIT */
  77
  78#define PMD_SIZE        (1UL << PMD_SHIFT)
  79#define PMD_MASK        (~(PMD_SIZE-1))
  80#define PUD_SIZE	(1UL << PUD_SHIFT)
  81#define PUD_MASK	(~(PUD_SIZE-1))
  82#define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
  83#define PGDIR_MASK	(~(PGDIR_SIZE-1))
  84
  85/*
  86 * entries per page directory level: the S390 is two-level, so
  87 * we don't really have any PMD directory physically.
  88 * for S390 segment-table entries are combined to one PGD
  89 * that leads to 1024 pte per pgd
  90 */
  91#define PTRS_PER_PTE	256
  92#ifndef CONFIG_64BIT
  93#define PTRS_PER_PMD	1
  94#define PTRS_PER_PUD	1
  95#else /* CONFIG_64BIT */
  96#define PTRS_PER_PMD	2048
  97#define PTRS_PER_PUD	2048
  98#endif /* CONFIG_64BIT */
  99#define PTRS_PER_PGD	2048
 100
 101#define FIRST_USER_ADDRESS  0
 102
 103#define pte_ERROR(e) \
 104	printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
 105#define pmd_ERROR(e) \
 106	printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
 107#define pud_ERROR(e) \
 108	printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
 109#define pgd_ERROR(e) \
 110	printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
 111
 112#ifndef __ASSEMBLY__
 113/*
 114 * The vmalloc and module area will always be on the topmost area of the kernel
 115 * mapping. We reserve 96MB (31bit) / 128GB (64bit) for vmalloc and modules.
 116 * On 64 bit kernels we have a 2GB area at the top of the vmalloc area where
 117 * modules will reside. That makes sure that inter module branches always
 118 * happen without trampolines and in addition the placement within a 2GB frame
 119 * is branch prediction unit friendly.
 
 120 */
 121extern unsigned long VMALLOC_START;
 122extern unsigned long VMALLOC_END;
 123extern struct page *vmemmap;
 124
 125#define VMEM_MAX_PHYS ((unsigned long) vmemmap)
 126
 127#ifdef CONFIG_64BIT
 128extern unsigned long MODULES_VADDR;
 129extern unsigned long MODULES_END;
 130#define MODULES_VADDR	MODULES_VADDR
 131#define MODULES_END	MODULES_END
 132#define MODULES_LEN	(1UL << 31)
 133#endif
 134
 135/*
 136 * A 31 bit pagetable entry of S390 has following format:
 137 *  |   PFRA          |    |  OS  |
 138 * 0                   0IP0
 139 * 00000000001111111111222222222233
 140 * 01234567890123456789012345678901
 141 *
 142 * I Page-Invalid Bit:    Page is not available for address-translation
 143 * P Page-Protection Bit: Store access not possible for page
 144 *
 145 * A 31 bit segmenttable entry of S390 has following format:
 146 *  |   P-table origin      |  |PTL
 147 * 0                         IC
 148 * 00000000001111111111222222222233
 149 * 01234567890123456789012345678901
 150 *
 151 * I Segment-Invalid Bit:    Segment is not available for address-translation
 152 * C Common-Segment Bit:     Segment is not private (PoP 3-30)
 153 * PTL Page-Table-Length:    Page-table length (PTL+1*16 entries -> up to 256)
 154 *
 155 * The 31 bit segmenttable origin of S390 has following format:
 156 *
 157 *  |S-table origin   |     | STL |
 158 * X                   **GPS
 159 * 00000000001111111111222222222233
 160 * 01234567890123456789012345678901
 161 *
 162 * X Space-Switch event:
 163 * G Segment-Invalid Bit:     *
 164 * P Private-Space Bit:       Segment is not private (PoP 3-30)
 165 * S Storage-Alteration:
 166 * STL Segment-Table-Length:  Segment-table length (STL+1*16 entries -> up to 2048)
 167 *
 168 * A 64 bit pagetable entry of S390 has following format:
 169 * |			 PFRA			      |0IPC|  OS  |
 170 * 0000000000111111111122222222223333333333444444444455555555556666
 171 * 0123456789012345678901234567890123456789012345678901234567890123
 172 *
 173 * I Page-Invalid Bit:    Page is not available for address-translation
 174 * P Page-Protection Bit: Store access not possible for page
 175 * C Change-bit override: HW is not required to set change bit
 176 *
 177 * A 64 bit segmenttable entry of S390 has following format:
 178 * |        P-table origin                              |      TT
 179 * 0000000000111111111122222222223333333333444444444455555555556666
 180 * 0123456789012345678901234567890123456789012345678901234567890123
 181 *
 182 * I Segment-Invalid Bit:    Segment is not available for address-translation
 183 * C Common-Segment Bit:     Segment is not private (PoP 3-30)
 184 * P Page-Protection Bit: Store access not possible for page
 185 * TT Type 00
 186 *
 187 * A 64 bit region table entry of S390 has following format:
 188 * |        S-table origin                             |   TF  TTTL
 189 * 0000000000111111111122222222223333333333444444444455555555556666
 190 * 0123456789012345678901234567890123456789012345678901234567890123
 191 *
 192 * I Segment-Invalid Bit:    Segment is not available for address-translation
 193 * TT Type 01
 194 * TF
 195 * TL Table length
 196 *
 197 * The 64 bit regiontable origin of S390 has following format:
 198 * |      region table origon                          |       DTTL
 199 * 0000000000111111111122222222223333333333444444444455555555556666
 200 * 0123456789012345678901234567890123456789012345678901234567890123
 201 *
 202 * X Space-Switch event:
 203 * G Segment-Invalid Bit:  
 204 * P Private-Space Bit:    
 205 * S Storage-Alteration:
 206 * R Real space
 207 * TL Table-Length:
 208 *
 209 * A storage key has the following format:
 210 * | ACC |F|R|C|0|
 211 *  0   3 4 5 6 7
 212 * ACC: access key
 213 * F  : fetch protection bit
 214 * R  : referenced bit
 215 * C  : changed bit
 216 */
 217
 218/* Hardware bits in the page table entry */
 219#define _PAGE_CO	0x100		/* HW Change-bit override */
 220#define _PAGE_PROTECT	0x200		/* HW read-only bit  */
 221#define _PAGE_INVALID	0x400		/* HW invalid bit    */
 222#define _PAGE_LARGE	0x800		/* Bit to mark a large pte */
 223
 224/* Software bits in the page table entry */
 225#define _PAGE_PRESENT	0x001		/* SW pte present bit */
 226#define _PAGE_TYPE	0x002		/* SW pte type bit */
 227#define _PAGE_YOUNG	0x004		/* SW pte young bit */
 228#define _PAGE_DIRTY	0x008		/* SW pte dirty bit */
 229#define _PAGE_READ	0x010		/* SW pte read bit */
 230#define _PAGE_WRITE	0x020		/* SW pte write bit */
 231#define _PAGE_SPECIAL	0x040		/* SW associated with special page */
 232#define _PAGE_UNUSED	0x080		/* SW bit for pgste usage state */
 233#define __HAVE_ARCH_PTE_SPECIAL
 234
 235/* Set of bits not changed in pte_modify */
 236#define _PAGE_CHG_MASK		(PAGE_MASK | _PAGE_SPECIAL | _PAGE_CO | \
 237				 _PAGE_DIRTY | _PAGE_YOUNG)
 238
 239/*
 240 * handle_pte_fault uses pte_present, pte_none and pte_file to find out the
 241 * pte type WITHOUT holding the page table lock. The _PAGE_PRESENT bit
 242 * is used to distinguish present from not-present ptes. It is changed only
 243 * with the page table lock held.
 244 *
 245 * The following table gives the different possible bit combinations for
 246 * the pte hardware and software bits in the last 12 bits of a pte:
 247 *
 248 *				842100000000
 249 *				000084210000
 250 *				000000008421
 251 *				.IR...wrdytp
 252 * empty			.10...000000
 253 * swap				.10...xxxx10
 254 * file				.11...xxxxx0
 255 * prot-none, clean, old	.11...000001
 256 * prot-none, clean, young	.11...000101
 257 * prot-none, dirty, old	.10...001001
 258 * prot-none, dirty, young	.10...001101
 259 * read-only, clean, old	.11...010001
 260 * read-only, clean, young	.01...010101
 261 * read-only, dirty, old	.11...011001
 262 * read-only, dirty, young	.01...011101
 263 * read-write, clean, old	.11...110001
 264 * read-write, clean, young	.01...110101
 265 * read-write, dirty, old	.10...111001
 266 * read-write, dirty, young	.00...111101
 267 *
 268 * pte_present is true for the bit pattern .xx...xxxxx1, (pte & 0x001) == 0x001
 269 * pte_none    is true for the bit pattern .10...xxxx00, (pte & 0x603) == 0x400
 270 * pte_file    is true for the bit pattern .11...xxxxx0, (pte & 0x601) == 0x600
 271 * pte_swap    is true for the bit pattern .10...xxxx10, (pte & 0x603) == 0x402
 
 
 
 
 
 
 
 
 272 */
 273
 274#ifndef CONFIG_64BIT
 275
 276/* Bits in the segment table address-space-control-element */
 277#define _ASCE_SPACE_SWITCH	0x80000000UL	/* space switch event	    */
 278#define _ASCE_ORIGIN_MASK	0x7ffff000UL	/* segment table origin	    */
 279#define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
 280#define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
 281#define _ASCE_TABLE_LENGTH	0x7f	/* 128 x 64 entries = 8k	    */
 282
 283/* Bits in the segment table entry */
 284#define _SEGMENT_ENTRY_BITS	0x7fffffffUL	/* Valid segment table bits */
 285#define _SEGMENT_ENTRY_ORIGIN	0x7fffffc0UL	/* page table origin	    */
 286#define _SEGMENT_ENTRY_PROTECT	0x200	/* page protection bit		    */
 287#define _SEGMENT_ENTRY_INVALID	0x20	/* invalid segment table entry	    */
 288#define _SEGMENT_ENTRY_COMMON	0x10	/* common segment bit		    */
 289#define _SEGMENT_ENTRY_PTL	0x0f	/* page table length		    */
 290#define _SEGMENT_ENTRY_NONE	_SEGMENT_ENTRY_PROTECT
 291
 292#define _SEGMENT_ENTRY		(_SEGMENT_ENTRY_PTL)
 293#define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INVALID)
 294
 295/*
 296 * Segment table entry encoding (I = invalid, R = read-only bit):
 297 *		..R...I.....
 298 * prot-none	..1...1.....
 299 * read-only	..1...0.....
 300 * read-write	..0...0.....
 301 * empty	..0...1.....
 302 */
 303
 304/* Page status table bits for virtualization */
 305#define PGSTE_ACC_BITS	0xf0000000UL
 306#define PGSTE_FP_BIT	0x08000000UL
 307#define PGSTE_PCL_BIT	0x00800000UL
 308#define PGSTE_HR_BIT	0x00400000UL
 309#define PGSTE_HC_BIT	0x00200000UL
 310#define PGSTE_GR_BIT	0x00040000UL
 311#define PGSTE_GC_BIT	0x00020000UL
 312#define PGSTE_IN_BIT	0x00008000UL	/* IPTE notify bit */
 
 
 
 313
 314#else /* CONFIG_64BIT */
 315
 316/* Bits in the segment/region table address-space-control-element */
 317#define _ASCE_ORIGIN		~0xfffUL/* segment table origin		    */
 318#define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
 319#define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
 320#define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
 321#define _ASCE_REAL_SPACE	0x20	/* real space control		    */
 322#define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
 323#define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
 324#define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
 325#define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
 326#define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
 327#define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
 328
 329/* Bits in the region table entry */
 330#define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
 331#define _REGION_ENTRY_PROTECT	0x200	/* region protection bit	    */
 332#define _REGION_ENTRY_INVALID	0x20	/* invalid region table entry	    */
 333#define _REGION_ENTRY_TYPE_MASK	0x0c	/* region/segment table type mask   */
 334#define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
 335#define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
 336#define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
 337#define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
 338
 339#define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
 340#define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID)
 341#define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
 342#define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID)
 343#define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
 344#define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID)
 345
 346#define _REGION3_ENTRY_LARGE	0x400	/* RTTE-format control, large page  */
 347#define _REGION3_ENTRY_RO	0x200	/* page protection bit		    */
 348#define _REGION3_ENTRY_CO	0x100	/* change-recording override	    */
 349
 350/* Bits in the segment table entry */
 351#define _SEGMENT_ENTRY_BITS	0xfffffffffffffe33UL
 352#define _SEGMENT_ENTRY_BITS_LARGE 0xfffffffffff1ff33UL
 353#define _SEGMENT_ENTRY_ORIGIN_LARGE ~0xfffffUL /* large page address	    */
 354#define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* segment table origin		    */
 355#define _SEGMENT_ENTRY_PROTECT	0x200	/* page protection bit		    */
 356#define _SEGMENT_ENTRY_INVALID	0x20	/* invalid segment table entry	    */
 357
 358#define _SEGMENT_ENTRY		(0)
 359#define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INVALID)
 360
 361#define _SEGMENT_ENTRY_LARGE	0x400	/* STE-format control, large page   */
 362#define _SEGMENT_ENTRY_CO	0x100	/* change-recording override   */
 363#define _SEGMENT_ENTRY_SPLIT	0x001	/* THP splitting bit */
 364#define _SEGMENT_ENTRY_YOUNG	0x002	/* SW segment young bit */
 365#define _SEGMENT_ENTRY_NONE	_SEGMENT_ENTRY_YOUNG
 366
 367/*
 368 * Segment table entry encoding (R = read-only, I = invalid, y = young bit):
 369 *			..R...I...y.
 370 * prot-none, old	..0...1...1.
 371 * prot-none, young	..1...1...1.
 372 * read-only, old	..1...1...0.
 373 * read-only, young	..1...0...1.
 374 * read-write, old	..0...1...0.
 375 * read-write, young	..0...0...1.
 376 * The segment table origin is used to distinguish empty (origin==0) from
 377 * read-write, old segment table entries (origin!=0)
 378 */
 379
 380#define _SEGMENT_ENTRY_SPLIT_BIT 0	/* THP splitting bit number */
 381
 382/* Set of bits not changed in pmd_modify */
 383#define _SEGMENT_CHG_MASK	(_SEGMENT_ENTRY_ORIGIN | _SEGMENT_ENTRY_LARGE \
 384				 | _SEGMENT_ENTRY_SPLIT | _SEGMENT_ENTRY_CO)
 385
 386/* Page status table bits for virtualization */
 387#define PGSTE_ACC_BITS	0xf000000000000000UL
 388#define PGSTE_FP_BIT	0x0800000000000000UL
 389#define PGSTE_PCL_BIT	0x0080000000000000UL
 390#define PGSTE_HR_BIT	0x0040000000000000UL
 391#define PGSTE_HC_BIT	0x0020000000000000UL
 392#define PGSTE_GR_BIT	0x0004000000000000UL
 393#define PGSTE_GC_BIT	0x0002000000000000UL
 394#define PGSTE_IN_BIT	0x0000800000000000UL	/* IPTE notify bit */
 
 
 
 395
 396#endif /* CONFIG_64BIT */
 397
 398/* Guest Page State used for virtualization */
 399#define _PGSTE_GPS_ZERO		0x0000000080000000UL
 400#define _PGSTE_GPS_USAGE_MASK	0x0000000003000000UL
 401#define _PGSTE_GPS_USAGE_STABLE 0x0000000000000000UL
 402#define _PGSTE_GPS_USAGE_UNUSED 0x0000000001000000UL
 403
 404/*
 405 * A user page table pointer has the space-switch-event bit, the
 406 * private-space-control bit and the storage-alteration-event-control
 407 * bit set. A kernel page table pointer doesn't need them.
 408 */
 409#define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
 410				 _ASCE_ALT_EVENT)
 411
 412/*
 413 * Page protection definitions.
 414 */
 415#define PAGE_NONE	__pgprot(_PAGE_PRESENT | _PAGE_INVALID)
 416#define PAGE_READ	__pgprot(_PAGE_PRESENT | _PAGE_READ | \
 417				 _PAGE_INVALID | _PAGE_PROTECT)
 418#define PAGE_WRITE	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
 419				 _PAGE_INVALID | _PAGE_PROTECT)
 420
 421#define PAGE_SHARED	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
 422				 _PAGE_YOUNG | _PAGE_DIRTY)
 423#define PAGE_KERNEL	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
 424				 _PAGE_YOUNG | _PAGE_DIRTY)
 425#define PAGE_KERNEL_RO	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_YOUNG | \
 426				 _PAGE_PROTECT)
 427
 428/*
 429 * On s390 the page table entry has an invalid bit and a read-only bit.
 430 * Read permission implies execute permission and write permission
 431 * implies read permission.
 432 */
 433         /*xwr*/
 434#define __P000	PAGE_NONE
 435#define __P001	PAGE_READ
 436#define __P010	PAGE_READ
 437#define __P011	PAGE_READ
 438#define __P100	PAGE_READ
 439#define __P101	PAGE_READ
 440#define __P110	PAGE_READ
 441#define __P111	PAGE_READ
 442
 443#define __S000	PAGE_NONE
 444#define __S001	PAGE_READ
 445#define __S010	PAGE_WRITE
 446#define __S011	PAGE_WRITE
 447#define __S100	PAGE_READ
 448#define __S101	PAGE_READ
 449#define __S110	PAGE_WRITE
 450#define __S111	PAGE_WRITE
 451
 452/*
 453 * Segment entry (large page) protection definitions.
 454 */
 455#define SEGMENT_NONE	__pgprot(_SEGMENT_ENTRY_INVALID | \
 456				 _SEGMENT_ENTRY_NONE)
 457#define SEGMENT_READ	__pgprot(_SEGMENT_ENTRY_INVALID | \
 458				 _SEGMENT_ENTRY_PROTECT)
 459#define SEGMENT_WRITE	__pgprot(_SEGMENT_ENTRY_INVALID)
 460
 461static inline int mm_has_pgste(struct mm_struct *mm)
 462{
 463#ifdef CONFIG_PGSTE
 464	if (unlikely(mm->context.has_pgste))
 465		return 1;
 466#endif
 467	return 0;
 468}
 469/*
 470 * pgd/pmd/pte query functions
 471 */
 472#ifndef CONFIG_64BIT
 473
 474static inline int pgd_present(pgd_t pgd) { return 1; }
 475static inline int pgd_none(pgd_t pgd)    { return 0; }
 476static inline int pgd_bad(pgd_t pgd)     { return 0; }
 477
 478static inline int pud_present(pud_t pud) { return 1; }
 479static inline int pud_none(pud_t pud)	 { return 0; }
 480static inline int pud_large(pud_t pud)	 { return 0; }
 481static inline int pud_bad(pud_t pud)	 { return 0; }
 482
 483#else /* CONFIG_64BIT */
 484
 485static inline int pgd_present(pgd_t pgd)
 486{
 487	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
 488		return 1;
 489	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
 490}
 491
 492static inline int pgd_none(pgd_t pgd)
 493{
 494	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
 495		return 0;
 496	return (pgd_val(pgd) & _REGION_ENTRY_INVALID) != 0UL;
 497}
 498
 499static inline int pgd_bad(pgd_t pgd)
 500{
 501	/*
 502	 * With dynamic page table levels the pgd can be a region table
 503	 * entry or a segment table entry. Check for the bit that are
 504	 * invalid for either table entry.
 505	 */
 506	unsigned long mask =
 507		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INVALID &
 508		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
 509	return (pgd_val(pgd) & mask) != 0;
 510}
 511
 512static inline int pud_present(pud_t pud)
 513{
 514	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
 515		return 1;
 516	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
 517}
 518
 519static inline int pud_none(pud_t pud)
 520{
 521	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
 522		return 0;
 523	return (pud_val(pud) & _REGION_ENTRY_INVALID) != 0UL;
 524}
 525
 526static inline int pud_large(pud_t pud)
 527{
 528	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3)
 529		return 0;
 530	return !!(pud_val(pud) & _REGION3_ENTRY_LARGE);
 531}
 532
 533static inline int pud_bad(pud_t pud)
 534{
 535	/*
 536	 * With dynamic page table levels the pud can be a region table
 537	 * entry or a segment table entry. Check for the bit that are
 538	 * invalid for either table entry.
 539	 */
 540	unsigned long mask =
 541		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INVALID &
 542		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
 543	return (pud_val(pud) & mask) != 0;
 544}
 545
 546#endif /* CONFIG_64BIT */
 547
 548static inline int pmd_present(pmd_t pmd)
 549{
 550	return pmd_val(pmd) != _SEGMENT_ENTRY_INVALID;
 551}
 552
 553static inline int pmd_none(pmd_t pmd)
 554{
 555	return pmd_val(pmd) == _SEGMENT_ENTRY_INVALID;
 556}
 557
 558static inline int pmd_large(pmd_t pmd)
 559{
 560#ifdef CONFIG_64BIT
 561	return (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) != 0;
 562#else
 563	return 0;
 564#endif
 565}
 566
 567static inline int pmd_prot_none(pmd_t pmd)
 568{
 569	return (pmd_val(pmd) & _SEGMENT_ENTRY_INVALID) &&
 570		(pmd_val(pmd) & _SEGMENT_ENTRY_NONE);
 571}
 572
 573static inline int pmd_bad(pmd_t pmd)
 574{
 575#ifdef CONFIG_64BIT
 576	if (pmd_large(pmd))
 577		return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS_LARGE) != 0;
 578#endif
 579	return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS) != 0;
 580}
 581
 582#define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
 583extern void pmdp_splitting_flush(struct vm_area_struct *vma,
 584				 unsigned long addr, pmd_t *pmdp);
 585
 586#define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
 587extern int pmdp_set_access_flags(struct vm_area_struct *vma,
 588				 unsigned long address, pmd_t *pmdp,
 589				 pmd_t entry, int dirty);
 590
 591#define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
 592extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
 593				  unsigned long address, pmd_t *pmdp);
 594
 595#define __HAVE_ARCH_PMD_WRITE
 596static inline int pmd_write(pmd_t pmd)
 597{
 598	if (pmd_prot_none(pmd))
 599		return 0;
 600	return (pmd_val(pmd) & _SEGMENT_ENTRY_PROTECT) == 0;
 601}
 602
 603static inline int pmd_young(pmd_t pmd)
 604{
 605	int young = 0;
 606#ifdef CONFIG_64BIT
 607	if (pmd_prot_none(pmd))
 608		young = (pmd_val(pmd) & _SEGMENT_ENTRY_PROTECT) != 0;
 609	else
 610		young = (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG) != 0;
 611#endif
 612	return young;
 613}
 614
 615static inline int pte_present(pte_t pte)
 616{
 617	/* Bit pattern: (pte & 0x001) == 0x001 */
 618	return (pte_val(pte) & _PAGE_PRESENT) != 0;
 619}
 620
 621static inline int pte_none(pte_t pte)
 622{
 623	/* Bit pattern: pte == 0x400 */
 624	return pte_val(pte) == _PAGE_INVALID;
 625}
 626
 627static inline int pte_swap(pte_t pte)
 628{
 629	/* Bit pattern: (pte & 0x603) == 0x402 */
 630	return (pte_val(pte) & (_PAGE_INVALID | _PAGE_PROTECT |
 631				_PAGE_TYPE | _PAGE_PRESENT))
 632		== (_PAGE_INVALID | _PAGE_TYPE);
 633}
 634
 635static inline int pte_file(pte_t pte)
 636{
 637	/* Bit pattern: (pte & 0x601) == 0x600 */
 638	return (pte_val(pte) & (_PAGE_INVALID | _PAGE_PROTECT | _PAGE_PRESENT))
 639		== (_PAGE_INVALID | _PAGE_PROTECT);
 640}
 641
 642static inline int pte_special(pte_t pte)
 643{
 644	return (pte_val(pte) & _PAGE_SPECIAL);
 645}
 646
 647#define __HAVE_ARCH_PTE_SAME
 648static inline int pte_same(pte_t a, pte_t b)
 649{
 650	return pte_val(a) == pte_val(b);
 651}
 652
 653static inline pgste_t pgste_get_lock(pte_t *ptep)
 654{
 655	unsigned long new = 0;
 656#ifdef CONFIG_PGSTE
 657	unsigned long old;
 658
 659	preempt_disable();
 660	asm(
 661		"	lg	%0,%2\n"
 662		"0:	lgr	%1,%0\n"
 663		"	nihh	%0,0xff7f\n"	/* clear PCL bit in old */
 664		"	oihh	%1,0x0080\n"	/* set PCL bit in new */
 665		"	csg	%0,%1,%2\n"
 666		"	jl	0b\n"
 667		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
 668		: "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
 669#endif
 670	return __pgste(new);
 671}
 672
 673static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
 674{
 675#ifdef CONFIG_PGSTE
 676	asm(
 677		"	nihh	%1,0xff7f\n"	/* clear PCL bit */
 678		"	stg	%1,%0\n"
 679		: "=Q" (ptep[PTRS_PER_PTE])
 680		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
 681		: "cc", "memory");
 682	preempt_enable();
 683#endif
 684}
 685
 686static inline pgste_t pgste_get(pte_t *ptep)
 687{
 688	unsigned long pgste = 0;
 689#ifdef CONFIG_PGSTE
 690	pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
 691#endif
 692	return __pgste(pgste);
 693}
 694
 695static inline void pgste_set(pte_t *ptep, pgste_t pgste)
 696{
 697#ifdef CONFIG_PGSTE
 698	*(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
 699#endif
 700}
 701
 702static inline pgste_t pgste_update_all(pte_t *ptep, pgste_t pgste)
 703{
 704#ifdef CONFIG_PGSTE
 705	unsigned long address, bits, skey;
 
 706
 707	if (pte_val(*ptep) & _PAGE_INVALID)
 708		return pgste;
 709	address = pte_val(*ptep) & PAGE_MASK;
 710	skey = (unsigned long) page_get_storage_key(address);
 711	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
 712	if (!(pgste_val(pgste) & PGSTE_HC_BIT) && (bits & _PAGE_CHANGED)) {
 713		/* Transfer dirty + referenced bit to host bits in pgste */
 714		pgste_val(pgste) |= bits << 52;
 715		page_set_storage_key(address, skey ^ bits, 0);
 716	} else if (!(pgste_val(pgste) & PGSTE_HR_BIT) &&
 717		   (bits & _PAGE_REFERENCED)) {
 718		/* Transfer referenced bit to host bit in pgste */
 719		pgste_val(pgste) |= PGSTE_HR_BIT;
 720		page_reset_referenced(address);
 721	}
 722	/* Transfer page changed & referenced bit to guest bits in pgste */
 723	pgste_val(pgste) |= bits << 48;		/* GR bit & GC bit */
 
 
 
 
 
 724	/* Copy page access key and fetch protection bit to pgste */
 725	pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
 726	pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
 
 
 
 
 727#endif
 728	return pgste;
 729
 730}
 731
 732static inline pgste_t pgste_update_young(pte_t *ptep, pgste_t pgste)
 733{
 734#ifdef CONFIG_PGSTE
 735	if (pte_val(*ptep) & _PAGE_INVALID)
 
 
 736		return pgste;
 737	/* Get referenced bit from storage key */
 738	if (page_reset_referenced(pte_val(*ptep) & PAGE_MASK))
 739		pgste_val(pgste) |= PGSTE_HR_BIT | PGSTE_GR_BIT;
 
 
 
 
 
 740#endif
 741	return pgste;
 
 742}
 743
 744static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry)
 745{
 746#ifdef CONFIG_PGSTE
 747	unsigned long address;
 748	unsigned long nkey;
 749
 750	if (pte_val(entry) & _PAGE_INVALID)
 751		return;
 752	VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
 753	address = pte_val(entry) & PAGE_MASK;
 754	/*
 755	 * Set page access key and fetch protection bit from pgste.
 756	 * The guest C/R information is still in the PGSTE, set real
 757	 * key C/R to 0.
 758	 */
 759	nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
 760	page_set_storage_key(address, nkey, 0);
 761#endif
 762}
 763
 764static inline void pgste_set_pte(pte_t *ptep, pte_t entry)
 765{
 766	if (!MACHINE_HAS_ESOP &&
 767	    (pte_val(entry) & _PAGE_PRESENT) &&
 768	    (pte_val(entry) & _PAGE_WRITE)) {
 769		/*
 770		 * Without enhanced suppression-on-protection force
 771		 * the dirty bit on for all writable ptes.
 772		 */
 773		pte_val(entry) |= _PAGE_DIRTY;
 774		pte_val(entry) &= ~_PAGE_PROTECT;
 775	}
 776	*ptep = entry;
 777}
 778
 779/**
 780 * struct gmap_struct - guest address space
 781 * @mm: pointer to the parent mm_struct
 782 * @table: pointer to the page directory
 783 * @asce: address space control element for gmap page table
 784 * @crst_list: list of all crst tables used in the guest address space
 785 * @pfault_enabled: defines if pfaults are applicable for the guest
 786 */
 787struct gmap {
 788	struct list_head list;
 789	struct mm_struct *mm;
 790	unsigned long *table;
 791	unsigned long asce;
 792	void *private;
 793	struct list_head crst_list;
 794	bool pfault_enabled;
 795};
 796
 797/**
 798 * struct gmap_rmap - reverse mapping for segment table entries
 799 * @gmap: pointer to the gmap_struct
 800 * @entry: pointer to a segment table entry
 801 * @vmaddr: virtual address in the guest address space
 802 */
 803struct gmap_rmap {
 804	struct list_head list;
 805	struct gmap *gmap;
 806	unsigned long *entry;
 807	unsigned long vmaddr;
 808};
 809
 810/**
 811 * struct gmap_pgtable - gmap information attached to a page table
 812 * @vmaddr: address of the 1MB segment in the process virtual memory
 813 * @mapper: list of segment table entries mapping a page table
 814 */
 815struct gmap_pgtable {
 816	unsigned long vmaddr;
 817	struct list_head mapper;
 818};
 819
 820/**
 821 * struct gmap_notifier - notify function block for page invalidation
 822 * @notifier_call: address of callback function
 823 */
 824struct gmap_notifier {
 825	struct list_head list;
 826	void (*notifier_call)(struct gmap *gmap, unsigned long address);
 827};
 828
 829struct gmap *gmap_alloc(struct mm_struct *mm);
 830void gmap_free(struct gmap *gmap);
 831void gmap_enable(struct gmap *gmap);
 832void gmap_disable(struct gmap *gmap);
 833int gmap_map_segment(struct gmap *gmap, unsigned long from,
 834		     unsigned long to, unsigned long len);
 835int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len);
 836unsigned long __gmap_translate(unsigned long address, struct gmap *);
 837unsigned long gmap_translate(unsigned long address, struct gmap *);
 838unsigned long __gmap_fault(unsigned long address, struct gmap *);
 839unsigned long gmap_fault(unsigned long address, struct gmap *);
 840void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
 841void __gmap_zap(unsigned long address, struct gmap *);
 842
 843void gmap_register_ipte_notifier(struct gmap_notifier *);
 844void gmap_unregister_ipte_notifier(struct gmap_notifier *);
 845int gmap_ipte_notify(struct gmap *, unsigned long start, unsigned long len);
 846void gmap_do_ipte_notify(struct mm_struct *, pte_t *);
 847
 848static inline pgste_t pgste_ipte_notify(struct mm_struct *mm,
 849					pte_t *ptep, pgste_t pgste)
 850{
 851#ifdef CONFIG_PGSTE
 852	if (pgste_val(pgste) & PGSTE_IN_BIT) {
 853		pgste_val(pgste) &= ~PGSTE_IN_BIT;
 854		gmap_do_ipte_notify(mm, ptep);
 855	}
 856#endif
 857	return pgste;
 858}
 859
 860/*
 861 * Certain architectures need to do special things when PTEs
 862 * within a page table are directly modified.  Thus, the following
 863 * hook is made available.
 864 */
 865static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
 866			      pte_t *ptep, pte_t entry)
 867{
 868	pgste_t pgste;
 869
 870	if (mm_has_pgste(mm)) {
 871		pgste = pgste_get_lock(ptep);
 872		pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
 873		pgste_set_key(ptep, pgste, entry);
 874		pgste_set_pte(ptep, entry);
 875		pgste_set_unlock(ptep, pgste);
 876	} else {
 877		if (!(pte_val(entry) & _PAGE_INVALID) && MACHINE_HAS_EDAT1)
 878			pte_val(entry) |= _PAGE_CO;
 879		*ptep = entry;
 880	}
 881}
 882
 883/*
 884 * query functions pte_write/pte_dirty/pte_young only work if
 885 * pte_present() is true. Undefined behaviour if not..
 886 */
 887static inline int pte_write(pte_t pte)
 888{
 889	return (pte_val(pte) & _PAGE_WRITE) != 0;
 890}
 891
 892static inline int pte_dirty(pte_t pte)
 893{
 894	return (pte_val(pte) & _PAGE_DIRTY) != 0;
 
 
 
 
 895}
 896
 897static inline int pte_young(pte_t pte)
 898{
 899	return (pte_val(pte) & _PAGE_YOUNG) != 0;
 900}
 901
 902#define __HAVE_ARCH_PTE_UNUSED
 903static inline int pte_unused(pte_t pte)
 904{
 905	return pte_val(pte) & _PAGE_UNUSED;
 906}
 907
 908/*
 909 * pgd/pmd/pte modification functions
 910 */
 911
 912static inline void pgd_clear(pgd_t *pgd)
 913{
 914#ifdef CONFIG_64BIT
 915	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
 916		pgd_val(*pgd) = _REGION2_ENTRY_EMPTY;
 917#endif
 918}
 919
 920static inline void pud_clear(pud_t *pud)
 921{
 922#ifdef CONFIG_64BIT
 923	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
 924		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
 925#endif
 926}
 927
 928static inline void pmd_clear(pmd_t *pmdp)
 929{
 930	pmd_val(*pmdp) = _SEGMENT_ENTRY_INVALID;
 931}
 932
 933static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 934{
 935	pte_val(*ptep) = _PAGE_INVALID;
 936}
 937
 938/*
 939 * The following pte modification functions only work if
 940 * pte_present() is true. Undefined behaviour if not..
 941 */
 942static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
 943{
 944	pte_val(pte) &= _PAGE_CHG_MASK;
 945	pte_val(pte) |= pgprot_val(newprot);
 946	/*
 947	 * newprot for PAGE_NONE, PAGE_READ and PAGE_WRITE has the
 948	 * invalid bit set, clear it again for readable, young pages
 949	 */
 950	if ((pte_val(pte) & _PAGE_YOUNG) && (pte_val(pte) & _PAGE_READ))
 951		pte_val(pte) &= ~_PAGE_INVALID;
 952	/*
 953	 * newprot for PAGE_READ and PAGE_WRITE has the page protection
 954	 * bit set, clear it again for writable, dirty pages
 955	 */
 956	if ((pte_val(pte) & _PAGE_DIRTY) && (pte_val(pte) & _PAGE_WRITE))
 957		pte_val(pte) &= ~_PAGE_PROTECT;
 958	return pte;
 959}
 960
 961static inline pte_t pte_wrprotect(pte_t pte)
 962{
 963	pte_val(pte) &= ~_PAGE_WRITE;
 964	pte_val(pte) |= _PAGE_PROTECT;
 
 965	return pte;
 966}
 967
 968static inline pte_t pte_mkwrite(pte_t pte)
 969{
 970	pte_val(pte) |= _PAGE_WRITE;
 971	if (pte_val(pte) & _PAGE_DIRTY)
 972		pte_val(pte) &= ~_PAGE_PROTECT;
 973	return pte;
 974}
 975
 976static inline pte_t pte_mkclean(pte_t pte)
 977{
 978	pte_val(pte) &= ~_PAGE_DIRTY;
 979	pte_val(pte) |= _PAGE_PROTECT;
 
 980	return pte;
 981}
 982
 983static inline pte_t pte_mkdirty(pte_t pte)
 984{
 985	pte_val(pte) |= _PAGE_DIRTY;
 986	if (pte_val(pte) & _PAGE_WRITE)
 987		pte_val(pte) &= ~_PAGE_PROTECT;
 988	return pte;
 989}
 990
 991static inline pte_t pte_mkold(pte_t pte)
 992{
 993	pte_val(pte) &= ~_PAGE_YOUNG;
 994	pte_val(pte) |= _PAGE_INVALID;
 
 995	return pte;
 996}
 997
 998static inline pte_t pte_mkyoung(pte_t pte)
 999{
1000	pte_val(pte) |= _PAGE_YOUNG;
1001	if (pte_val(pte) & _PAGE_READ)
1002		pte_val(pte) &= ~_PAGE_INVALID;
1003	return pte;
1004}
1005
1006static inline pte_t pte_mkspecial(pte_t pte)
1007{
1008	pte_val(pte) |= _PAGE_SPECIAL;
1009	return pte;
1010}
1011
1012#ifdef CONFIG_HUGETLB_PAGE
1013static inline pte_t pte_mkhuge(pte_t pte)
1014{
1015	pte_val(pte) |= _PAGE_LARGE;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1016	return pte;
1017}
1018#endif
1019
1020/*
1021 * Get (and clear) the user dirty bit for a pte.
1022 */
1023static inline int ptep_test_and_clear_user_dirty(struct mm_struct *mm,
1024						 pte_t *ptep)
1025{
1026	pgste_t pgste;
1027	int dirty = 0;
1028
1029	if (mm_has_pgste(mm)) {
1030		pgste = pgste_get_lock(ptep);
1031		pgste = pgste_update_all(ptep, pgste);
1032		dirty = !!(pgste_val(pgste) & PGSTE_HC_BIT);
1033		pgste_val(pgste) &= ~PGSTE_HC_BIT;
1034		pgste_set_unlock(ptep, pgste);
1035		return dirty;
1036	}
1037	return dirty;
1038}
1039
1040/*
1041 * Get (and clear) the user referenced bit for a pte.
1042 */
1043static inline int ptep_test_and_clear_user_young(struct mm_struct *mm,
1044						 pte_t *ptep)
1045{
1046	pgste_t pgste;
1047	int young = 0;
1048
1049	if (mm_has_pgste(mm)) {
1050		pgste = pgste_get_lock(ptep);
1051		pgste = pgste_update_young(ptep, pgste);
1052		young = !!(pgste_val(pgste) & PGSTE_HR_BIT);
1053		pgste_val(pgste) &= ~PGSTE_HR_BIT;
1054		pgste_set_unlock(ptep, pgste);
1055	}
1056	return young;
1057}
1058
1059static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
1060{
1061	unsigned long pto = (unsigned long) ptep;
1062
1063#ifndef CONFIG_64BIT
1064	/* pto in ESA mode must point to the start of the segment table */
1065	pto &= 0x7ffffc00;
1066#endif
1067	/* Invalidation + global TLB flush for the pte */
1068	asm volatile(
1069		"	ipte	%2,%3"
1070		: "=m" (*ptep) : "m" (*ptep), "a" (pto), "a" (address));
1071}
1072
1073static inline void __ptep_ipte_local(unsigned long address, pte_t *ptep)
1074{
1075	unsigned long pto = (unsigned long) ptep;
1076
1077#ifndef CONFIG_64BIT
1078	/* pto in ESA mode must point to the start of the segment table */
1079	pto &= 0x7ffffc00;
1080#endif
1081	/* Invalidation + local TLB flush for the pte */
1082	asm volatile(
1083		"	.insn rrf,0xb2210000,%2,%3,0,1"
1084		: "=m" (*ptep) : "m" (*ptep), "a" (pto), "a" (address));
1085}
1086
1087static inline void ptep_flush_direct(struct mm_struct *mm,
1088				     unsigned long address, pte_t *ptep)
1089{
1090	int active, count;
1091
1092	if (pte_val(*ptep) & _PAGE_INVALID)
1093		return;
1094	active = (mm == current->active_mm) ? 1 : 0;
1095	count = atomic_add_return(0x10000, &mm->context.attach_count);
1096	if (MACHINE_HAS_TLB_LC && (count & 0xffff) <= active &&
1097	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
1098		__ptep_ipte_local(address, ptep);
1099	else
1100		__ptep_ipte(address, ptep);
1101	atomic_sub(0x10000, &mm->context.attach_count);
1102}
1103
1104static inline void ptep_flush_lazy(struct mm_struct *mm,
1105				   unsigned long address, pte_t *ptep)
1106{
1107	int active, count;
1108
1109	if (pte_val(*ptep) & _PAGE_INVALID)
1110		return;
1111	active = (mm == current->active_mm) ? 1 : 0;
1112	count = atomic_add_return(0x10000, &mm->context.attach_count);
1113	if ((count & 0xffff) <= active) {
1114		pte_val(*ptep) |= _PAGE_INVALID;
1115		mm->context.flush_mm = 1;
1116	} else
1117		__ptep_ipte(address, ptep);
1118	atomic_sub(0x10000, &mm->context.attach_count);
1119}
1120
1121#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1122static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1123					    unsigned long addr, pte_t *ptep)
1124{
1125	pgste_t pgste;
1126	pte_t pte;
1127	int young;
1128
1129	if (mm_has_pgste(vma->vm_mm)) {
1130		pgste = pgste_get_lock(ptep);
1131		pgste = pgste_ipte_notify(vma->vm_mm, ptep, pgste);
 
 
 
 
1132	}
1133
1134	pte = *ptep;
1135	ptep_flush_direct(vma->vm_mm, addr, ptep);
1136	young = pte_young(pte);
1137	pte = pte_mkold(pte);
1138
1139	if (mm_has_pgste(vma->vm_mm)) {
1140		pgste_set_pte(ptep, pte);
1141		pgste_set_unlock(ptep, pgste);
1142	} else
1143		*ptep = pte;
1144
1145	return young;
1146}
1147
1148#define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1149static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1150					 unsigned long address, pte_t *ptep)
1151{
 
 
 
 
1152	return ptep_test_and_clear_young(vma, address, ptep);
1153}
1154
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1155/*
1156 * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
1157 * both clear the TLB for the unmapped pte. The reason is that
1158 * ptep_get_and_clear is used in common code (e.g. change_pte_range)
1159 * to modify an active pte. The sequence is
1160 *   1) ptep_get_and_clear
1161 *   2) set_pte_at
1162 *   3) flush_tlb_range
1163 * On s390 the tlb needs to get flushed with the modification of the pte
1164 * if the pte is active. The only way how this can be implemented is to
1165 * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
1166 * is a nop.
1167 */
1168#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1169static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1170				       unsigned long address, pte_t *ptep)
1171{
1172	pgste_t pgste;
1173	pte_t pte;
1174
1175	if (mm_has_pgste(mm)) {
 
1176		pgste = pgste_get_lock(ptep);
1177		pgste = pgste_ipte_notify(mm, ptep, pgste);
1178	}
1179
1180	pte = *ptep;
1181	ptep_flush_lazy(mm, address, ptep);
1182	pte_val(*ptep) = _PAGE_INVALID;
 
1183
1184	if (mm_has_pgste(mm)) {
1185		pgste = pgste_update_all(&pte, pgste);
1186		pgste_set_unlock(ptep, pgste);
1187	}
1188	return pte;
1189}
1190
1191#define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1192static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
1193					   unsigned long address,
1194					   pte_t *ptep)
1195{
1196	pgste_t pgste;
1197	pte_t pte;
1198
1199	if (mm_has_pgste(mm)) {
1200		pgste = pgste_get_lock(ptep);
1201		pgste_ipte_notify(mm, ptep, pgste);
1202	}
1203
1204	pte = *ptep;
1205	ptep_flush_lazy(mm, address, ptep);
1206
1207	if (mm_has_pgste(mm)) {
1208		pgste = pgste_update_all(&pte, pgste);
1209		pgste_set(ptep, pgste);
1210	}
1211	return pte;
1212}
1213
1214static inline void ptep_modify_prot_commit(struct mm_struct *mm,
1215					   unsigned long address,
1216					   pte_t *ptep, pte_t pte)
1217{
1218	pgste_t pgste;
1219
1220	if (mm_has_pgste(mm)) {
1221		pgste = pgste_get(ptep);
1222		pgste_set_key(ptep, pgste, pte);
1223		pgste_set_pte(ptep, pte);
1224		pgste_set_unlock(ptep, pgste);
1225	} else
1226		*ptep = pte;
1227}
1228
1229#define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1230static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1231				     unsigned long address, pte_t *ptep)
1232{
1233	pgste_t pgste;
1234	pte_t pte;
1235
1236	if (mm_has_pgste(vma->vm_mm)) {
1237		pgste = pgste_get_lock(ptep);
1238		pgste = pgste_ipte_notify(vma->vm_mm, ptep, pgste);
1239	}
1240
1241	pte = *ptep;
1242	ptep_flush_direct(vma->vm_mm, address, ptep);
1243	pte_val(*ptep) = _PAGE_INVALID;
1244
1245	if (mm_has_pgste(vma->vm_mm)) {
1246		if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
1247		    _PGSTE_GPS_USAGE_UNUSED)
1248			pte_val(pte) |= _PAGE_UNUSED;
1249		pgste = pgste_update_all(&pte, pgste);
1250		pgste_set_unlock(ptep, pgste);
1251	}
1252	return pte;
1253}
1254
1255/*
1256 * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1257 * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1258 * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1259 * cannot be accessed while the batched unmap is running. In this case
1260 * full==1 and a simple pte_clear is enough. See tlb.h.
1261 */
1262#define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1263static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1264					    unsigned long address,
1265					    pte_t *ptep, int full)
1266{
1267	pgste_t pgste;
1268	pte_t pte;
1269
1270	if (!full && mm_has_pgste(mm)) {
1271		pgste = pgste_get_lock(ptep);
1272		pgste = pgste_ipte_notify(mm, ptep, pgste);
1273	}
1274
1275	pte = *ptep;
1276	if (!full)
1277		ptep_flush_lazy(mm, address, ptep);
1278	pte_val(*ptep) = _PAGE_INVALID;
1279
1280	if (!full && mm_has_pgste(mm)) {
1281		pgste = pgste_update_all(&pte, pgste);
1282		pgste_set_unlock(ptep, pgste);
1283	}
1284	return pte;
1285}
1286
1287#define __HAVE_ARCH_PTEP_SET_WRPROTECT
1288static inline pte_t ptep_set_wrprotect(struct mm_struct *mm,
1289				       unsigned long address, pte_t *ptep)
1290{
1291	pgste_t pgste;
1292	pte_t pte = *ptep;
1293
1294	if (pte_write(pte)) {
1295		if (mm_has_pgste(mm)) {
 
1296			pgste = pgste_get_lock(ptep);
1297			pgste = pgste_ipte_notify(mm, ptep, pgste);
1298		}
1299
1300		ptep_flush_lazy(mm, address, ptep);
1301		pte = pte_wrprotect(pte);
 
1302
1303		if (mm_has_pgste(mm)) {
1304			pgste_set_pte(ptep, pte);
1305			pgste_set_unlock(ptep, pgste);
1306		} else
1307			*ptep = pte;
1308	}
1309	return pte;
1310}
1311
1312#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1313static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1314					unsigned long address, pte_t *ptep,
1315					pte_t entry, int dirty)
1316{
1317	pgste_t pgste;
1318
1319	if (pte_same(*ptep, entry))
1320		return 0;
1321	if (mm_has_pgste(vma->vm_mm)) {
1322		pgste = pgste_get_lock(ptep);
1323		pgste = pgste_ipte_notify(vma->vm_mm, ptep, pgste);
1324	}
1325
1326	ptep_flush_direct(vma->vm_mm, address, ptep);
 
1327
1328	if (mm_has_pgste(vma->vm_mm)) {
1329		pgste_set_pte(ptep, entry);
1330		pgste_set_unlock(ptep, pgste);
1331	} else
1332		*ptep = entry;
1333	return 1;
1334}
1335
1336/*
1337 * Conversion functions: convert a page and protection to a page entry,
1338 * and a page entry and page directory to the page they refer to.
1339 */
1340static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1341{
1342	pte_t __pte;
1343	pte_val(__pte) = physpage + pgprot_val(pgprot);
1344	return pte_mkyoung(__pte);
1345}
1346
1347static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1348{
1349	unsigned long physpage = page_to_phys(page);
1350	pte_t __pte = mk_pte_phys(physpage, pgprot);
1351
1352	if (pte_write(__pte) && PageDirty(page))
1353		__pte = pte_mkdirty(__pte);
1354	return __pte;
1355}
1356
1357#define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1358#define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1359#define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1360#define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
1361
1362#define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
1363#define pgd_offset_k(address) pgd_offset(&init_mm, address)
1364
1365#ifndef CONFIG_64BIT
1366
1367#define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1368#define pud_deref(pmd) ({ BUG(); 0UL; })
1369#define pgd_deref(pmd) ({ BUG(); 0UL; })
1370
1371#define pud_offset(pgd, address) ((pud_t *) pgd)
1372#define pmd_offset(pud, address) ((pmd_t *) pud + pmd_index(address))
1373
1374#else /* CONFIG_64BIT */
1375
1376#define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1377#define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
1378#define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)
1379
1380static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
1381{
1382	pud_t *pud = (pud_t *) pgd;
1383	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
1384		pud = (pud_t *) pgd_deref(*pgd);
1385	return pud  + pud_index(address);
1386}
1387
1388static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
1389{
1390	pmd_t *pmd = (pmd_t *) pud;
1391	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
1392		pmd = (pmd_t *) pud_deref(*pud);
1393	return pmd + pmd_index(address);
1394}
1395
1396#endif /* CONFIG_64BIT */
1397
1398#define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
1399#define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1400#define pte_page(x) pfn_to_page(pte_pfn(x))
1401
1402#define pmd_page(pmd) pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT)
1403
1404/* Find an entry in the lowest level page table.. */
1405#define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
1406#define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
1407#define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
1408#define pte_unmap(pte) do { } while (0)
1409
1410#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1411static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot)
1412{
1413	/*
1414	 * pgprot is PAGE_NONE, PAGE_READ, or PAGE_WRITE (see __Pxxx / __Sxxx)
1415	 * Convert to segment table entry format.
1416	 */
1417	if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE))
1418		return pgprot_val(SEGMENT_NONE);
1419	if (pgprot_val(pgprot) == pgprot_val(PAGE_READ))
1420		return pgprot_val(SEGMENT_READ);
1421	return pgprot_val(SEGMENT_WRITE);
1422}
1423
1424static inline pmd_t pmd_mkyoung(pmd_t pmd)
1425{
1426#ifdef CONFIG_64BIT
1427	if (pmd_prot_none(pmd)) {
1428		pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1429	} else {
1430		pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
1431		pmd_val(pmd) &= ~_SEGMENT_ENTRY_INVALID;
1432	}
1433#endif
1434	return pmd;
1435}
1436
1437static inline pmd_t pmd_mkold(pmd_t pmd)
1438{
1439#ifdef CONFIG_64BIT
1440	if (pmd_prot_none(pmd)) {
1441		pmd_val(pmd) &= ~_SEGMENT_ENTRY_PROTECT;
1442	} else {
1443		pmd_val(pmd) &= ~_SEGMENT_ENTRY_YOUNG;
1444		pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
1445	}
1446#endif
1447	return pmd;
1448}
1449
1450static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1451{
1452	int young;
1453
1454	young = pmd_young(pmd);
1455	pmd_val(pmd) &= _SEGMENT_CHG_MASK;
1456	pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1457	if (young)
1458		pmd = pmd_mkyoung(pmd);
1459	return pmd;
1460}
1461
1462static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot)
1463{
1464	pmd_t __pmd;
1465	pmd_val(__pmd) = physpage + massage_pgprot_pmd(pgprot);
1466	return pmd_mkyoung(__pmd);
1467}
1468
1469static inline pmd_t pmd_mkwrite(pmd_t pmd)
1470{
1471	/* Do not clobber PROT_NONE segments! */
1472	if (!pmd_prot_none(pmd))
1473		pmd_val(pmd) &= ~_SEGMENT_ENTRY_PROTECT;
1474	return pmd;
1475}
1476#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLB_PAGE */
1477
1478static inline void __pmdp_csp(pmd_t *pmdp)
1479{
1480	register unsigned long reg2 asm("2") = pmd_val(*pmdp);
1481	register unsigned long reg3 asm("3") = pmd_val(*pmdp) |
1482					       _SEGMENT_ENTRY_INVALID;
1483	register unsigned long reg4 asm("4") = ((unsigned long) pmdp) + 5;
1484
1485	asm volatile(
1486		"	csp %1,%3"
1487		: "=m" (*pmdp)
1488		: "d" (reg2), "d" (reg3), "d" (reg4), "m" (*pmdp) : "cc");
1489}
1490
1491static inline void __pmdp_idte(unsigned long address, pmd_t *pmdp)
1492{
1493	unsigned long sto;
1494
1495	sto = (unsigned long) pmdp - pmd_index(address) * sizeof(pmd_t);
1496	asm volatile(
1497		"	.insn	rrf,0xb98e0000,%2,%3,0,0"
1498		: "=m" (*pmdp)
1499		: "m" (*pmdp), "a" (sto), "a" ((address & HPAGE_MASK))
1500		: "cc" );
1501}
1502
1503static inline void __pmdp_idte_local(unsigned long address, pmd_t *pmdp)
1504{
1505	unsigned long sto;
1506
1507	sto = (unsigned long) pmdp - pmd_index(address) * sizeof(pmd_t);
1508	asm volatile(
1509		"	.insn	rrf,0xb98e0000,%2,%3,0,1"
1510		: "=m" (*pmdp)
1511		: "m" (*pmdp), "a" (sto), "a" ((address & HPAGE_MASK))
1512		: "cc" );
1513}
1514
1515static inline void pmdp_flush_direct(struct mm_struct *mm,
1516				     unsigned long address, pmd_t *pmdp)
1517{
1518	int active, count;
1519
1520	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1521		return;
1522	if (!MACHINE_HAS_IDTE) {
1523		__pmdp_csp(pmdp);
1524		return;
1525	}
1526	active = (mm == current->active_mm) ? 1 : 0;
1527	count = atomic_add_return(0x10000, &mm->context.attach_count);
1528	if (MACHINE_HAS_TLB_LC && (count & 0xffff) <= active &&
1529	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
1530		__pmdp_idte_local(address, pmdp);
1531	else
1532		__pmdp_idte(address, pmdp);
1533	atomic_sub(0x10000, &mm->context.attach_count);
1534}
1535
1536static inline void pmdp_flush_lazy(struct mm_struct *mm,
1537				   unsigned long address, pmd_t *pmdp)
1538{
1539	int active, count;
1540
1541	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1542		return;
1543	active = (mm == current->active_mm) ? 1 : 0;
1544	count = atomic_add_return(0x10000, &mm->context.attach_count);
1545	if ((count & 0xffff) <= active) {
1546		pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
1547		mm->context.flush_mm = 1;
1548	} else if (MACHINE_HAS_IDTE)
1549		__pmdp_idte(address, pmdp);
1550	else
1551		__pmdp_csp(pmdp);
1552	atomic_sub(0x10000, &mm->context.attach_count);
1553}
1554
1555#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1556
1557#define __HAVE_ARCH_PGTABLE_DEPOSIT
1558extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
1559				       pgtable_t pgtable);
1560
1561#define __HAVE_ARCH_PGTABLE_WITHDRAW
1562extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
1563
1564static inline int pmd_trans_splitting(pmd_t pmd)
1565{
1566	return pmd_val(pmd) & _SEGMENT_ENTRY_SPLIT;
1567}
1568
1569static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1570			      pmd_t *pmdp, pmd_t entry)
1571{
1572	if (!(pmd_val(entry) & _SEGMENT_ENTRY_INVALID) && MACHINE_HAS_EDAT1)
1573		pmd_val(entry) |= _SEGMENT_ENTRY_CO;
1574	*pmdp = entry;
1575}
1576
1577static inline pmd_t pmd_mkhuge(pmd_t pmd)
1578{
1579	pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE;
1580	return pmd;
1581}
1582
1583static inline pmd_t pmd_wrprotect(pmd_t pmd)
1584{
1585	/* Do not clobber PROT_NONE segments! */
1586	if (!pmd_prot_none(pmd))
1587		pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1588	return pmd;
1589}
1590
1591static inline pmd_t pmd_mkdirty(pmd_t pmd)
1592{
1593	/* No dirty bit in the segment table entry. */
1594	return pmd;
1595}
1596
1597#define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1598static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1599					    unsigned long address, pmd_t *pmdp)
1600{
1601	pmd_t pmd;
1602
1603	pmd = *pmdp;
1604	pmdp_flush_direct(vma->vm_mm, address, pmdp);
1605	*pmdp = pmd_mkold(pmd);
1606	return pmd_young(pmd);
1607}
1608
1609#define __HAVE_ARCH_PMDP_GET_AND_CLEAR
1610static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
1611				       unsigned long address, pmd_t *pmdp)
1612{
1613	pmd_t pmd = *pmdp;
1614
1615	pmdp_flush_direct(mm, address, pmdp);
1616	pmd_clear(pmdp);
1617	return pmd;
1618}
1619
1620#define __HAVE_ARCH_PMDP_CLEAR_FLUSH
1621static inline pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
1622				     unsigned long address, pmd_t *pmdp)
1623{
1624	return pmdp_get_and_clear(vma->vm_mm, address, pmdp);
1625}
1626
1627#define __HAVE_ARCH_PMDP_INVALIDATE
1628static inline void pmdp_invalidate(struct vm_area_struct *vma,
1629				   unsigned long address, pmd_t *pmdp)
1630{
1631	pmdp_flush_direct(vma->vm_mm, address, pmdp);
1632}
1633
1634#define __HAVE_ARCH_PMDP_SET_WRPROTECT
1635static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1636				      unsigned long address, pmd_t *pmdp)
1637{
1638	pmd_t pmd = *pmdp;
1639
1640	if (pmd_write(pmd)) {
1641		pmdp_flush_direct(mm, address, pmdp);
1642		set_pmd_at(mm, address, pmdp, pmd_wrprotect(pmd));
1643	}
1644}
1645
1646#define pfn_pmd(pfn, pgprot)	mk_pmd_phys(__pa((pfn) << PAGE_SHIFT), (pgprot))
1647#define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))
1648
1649static inline int pmd_trans_huge(pmd_t pmd)
1650{
1651	return pmd_val(pmd) & _SEGMENT_ENTRY_LARGE;
1652}
1653
1654static inline int has_transparent_hugepage(void)
1655{
1656	return MACHINE_HAS_HPAGE ? 1 : 0;
1657}
1658
1659static inline unsigned long pmd_pfn(pmd_t pmd)
1660{
1661	return pmd_val(pmd) >> PAGE_SHIFT;
1662}
1663#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1664
1665/*
1666 * 31 bit swap entry format:
1667 * A page-table entry has some bits we have to treat in a special way.
1668 * Bits 0, 20 and bit 23 have to be zero, otherwise an specification
1669 * exception will occur instead of a page translation exception. The
1670 * specifiation exception has the bad habit not to store necessary
1671 * information in the lowcore.
1672 * Bits 21, 22, 30 and 31 are used to indicate the page type.
1673 * A swap pte is indicated by bit pattern (pte & 0x603) == 0x402
 
 
1674 * This leaves the bits 1-19 and bits 24-29 to store type and offset.
1675 * We use the 5 bits from 25-29 for the type and the 20 bits from 1-19
1676 * plus 24 for the offset.
1677 * 0|     offset        |0110|o|type |00|
1678 * 0 0000000001111111111 2222 2 22222 33
1679 * 0 1234567890123456789 0123 4 56789 01
1680 *
1681 * 64 bit swap entry format:
1682 * A page-table entry has some bits we have to treat in a special way.
1683 * Bits 52 and bit 55 have to be zero, otherwise an specification
1684 * exception will occur instead of a page translation exception. The
1685 * specifiation exception has the bad habit not to store necessary
1686 * information in the lowcore.
1687 * Bits 53, 54, 62 and 63 are used to indicate the page type.
1688 * A swap pte is indicated by bit pattern (pte & 0x603) == 0x402
 
 
1689 * This leaves the bits 0-51 and bits 56-61 to store type and offset.
1690 * We use the 5 bits from 57-61 for the type and the 53 bits from 0-51
1691 * plus 56 for the offset.
1692 * |                      offset                        |0110|o|type |00|
1693 *  0000000000111111111122222222223333333333444444444455 5555 5 55566 66
1694 *  0123456789012345678901234567890123456789012345678901 2345 6 78901 23
1695 */
1696#ifndef CONFIG_64BIT
1697#define __SWP_OFFSET_MASK (~0UL >> 12)
1698#else
1699#define __SWP_OFFSET_MASK (~0UL >> 11)
1700#endif
1701static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1702{
1703	pte_t pte;
1704	offset &= __SWP_OFFSET_MASK;
1705	pte_val(pte) = _PAGE_INVALID | _PAGE_TYPE | ((type & 0x1f) << 2) |
1706		((offset & 1UL) << 7) | ((offset & ~1UL) << 11);
1707	return pte;
1708}
1709
1710#define __swp_type(entry)	(((entry).val >> 2) & 0x1f)
1711#define __swp_offset(entry)	(((entry).val >> 11) | (((entry).val >> 7) & 1))
1712#define __swp_entry(type,offset) ((swp_entry_t) { pte_val(mk_swap_pte((type),(offset))) })
1713
1714#define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1715#define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1716
1717#ifndef CONFIG_64BIT
1718# define PTE_FILE_MAX_BITS	26
1719#else /* CONFIG_64BIT */
1720# define PTE_FILE_MAX_BITS	59
1721#endif /* CONFIG_64BIT */
1722
1723#define pte_to_pgoff(__pte) \
1724	((((__pte).pte >> 12) << 7) + (((__pte).pte >> 1) & 0x7f))
1725
1726#define pgoff_to_pte(__off) \
1727	((pte_t) { ((((__off) & 0x7f) << 1) + (((__off) >> 7) << 12)) \
1728		   | _PAGE_INVALID | _PAGE_PROTECT })
1729
1730#endif /* !__ASSEMBLY__ */
1731
1732#define kern_addr_valid(addr)   (1)
1733
1734extern int vmem_add_mapping(unsigned long start, unsigned long size);
1735extern int vmem_remove_mapping(unsigned long start, unsigned long size);
1736extern int s390_enable_sie(void);
1737
1738/*
1739 * No page table caches to initialise
1740 */
1741static inline void pgtable_cache_init(void) { }
1742static inline void check_pgt_cache(void) { }
1743
1744#include <asm-generic/pgtable.h>
1745
1746#endif /* _S390_PAGE_H */