arch/tile/include/asm/processor.h

   1 /*
   2  * Copyright 2010 Tilera Corporation. All Rights Reserved.
   3  *
   4  *   This program is free software; you can redistribute it and/or
   5  *   modify it under the terms of the GNU General Public License
   6  *   as published by the Free Software Foundation, version 2.
   7  *
   8  *   This program is distributed in the hope that it will be useful, but
   9  *   WITHOUT ANY WARRANTY; without even the implied warranty of
  10  *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  11  *   NON INFRINGEMENT.  See the GNU General Public License for
  12  *   more details.
  13  */
  14
  15 #ifndef _ASM_TILE_PROCESSOR_H
  16 #define _ASM_TILE_PROCESSOR_H
  17
  18 #include <arch/chip.h>
  19
  20 #ifndef __ASSEMBLY__
  21
  22 /*
  23  * NOTE: we don't include <linux/ptrace.h> or <linux/percpu.h> as one
  24  * normally would, due to #include dependencies.
  25  */
  26 #include <linux/types.h>
  27 #include <asm/ptrace.h>
  28 #include <asm/percpu.h>
  29
  30 #include <arch/spr_def.h>
  31
  32 struct task_struct;
  33 struct thread_struct;
  34
  35 typedef struct {
  36         unsigned long seg;
  37 } mm_segment_t;
  38
  39 /*
  40  * Default implementation of macro that returns current
  41  * instruction pointer ("program counter").
  42  */
  43 void *current_text_addr(void);
  44
  45 #if CHIP_HAS_TILE_DMA()
  46 /* Capture the state of a suspended DMA. */
  47 struct tile_dma_state {
  48         int enabled;
  49         unsigned long src;
  50         unsigned long dest;
  51         unsigned long strides;
  52         unsigned long chunk_size;
  53         unsigned long src_chunk;
  54         unsigned long dest_chunk;
  55         unsigned long byte;
  56         unsigned long status;
  57 };
  58
  59 /*
  60  * A mask of the DMA status register for selecting only the 'running'
  61  * and 'done' bits.
  62  */
  63 #define DMA_STATUS_MASK \
  64   (SPR_DMA_STATUS__RUNNING_MASK | SPR_DMA_STATUS__DONE_MASK)
  65 #endif
  66
  67 /*
  68  * Track asynchronous TLB events (faults and access violations)
  69  * that occur while we are in kernel mode from DMA or the SN processor.
  70  */
  71 struct async_tlb {
  72         short fault_num;         /* original fault number; 0 if none */
  73         char is_fault;           /* was it a fault (vs an access violation) */
  74         char is_write;           /* for fault: was it caused by a write? */
  75         unsigned long address;   /* what address faulted? */
  76 };
  77
  78 #ifdef CONFIG_HARDWALL
  79 struct hardwall_info;
  80 struct hardwall_task {
  81         /* Which hardwall is this task tied to? (or NULL if none) */
  82         struct hardwall_info *info;
  83         /* Chains this task into the list at info->task_head. */
  84         struct list_head list;
  85 };
  86 #ifdef __tilepro__
  87 #define HARDWALL_TYPES 1   /* udn */
  88 #else
  89 #define HARDWALL_TYPES 3   /* udn, idn, and ipi */
  90 #endif
  91 #endif
  92
  93 struct thread_struct {
  94         /* kernel stack pointer */
  95         unsigned long  ksp;
  96         /* kernel PC */
  97         unsigned long  pc;
  98         /* starting user stack pointer (for page migration) */
  99         unsigned long  usp0;
 100         /* pid of process that created this one */
 101         pid_t creator_pid;
 102 #if CHIP_HAS_TILE_DMA()
 103         /* DMA info for suspended threads (byte == 0 means no DMA state) */
 104         struct tile_dma_state tile_dma_state;
 105 #endif
 106         /* User EX_CONTEXT registers */
 107         unsigned long ex_context[2];
 108         /* User SYSTEM_SAVE registers */
 109         unsigned long system_save[4];
 110         /* User interrupt mask */
 111         unsigned long long interrupt_mask;
 112         /* User interrupt-control 0 state */
 113         unsigned long intctrl_0;
 114         /* Any other miscellaneous processor state bits */
 115         unsigned long proc_status;
 116 #if !CHIP_HAS_FIXED_INTVEC_BASE()
 117         /* Interrupt base for PL0 interrupts */
 118         unsigned long interrupt_vector_base;
 119 #endif
 120         /* Tile cache retry fifo high-water mark */
 121         unsigned long tile_rtf_hwm;
 122 #if CHIP_HAS_DSTREAM_PF()
 123         /* Data stream prefetch control */
 124         unsigned long dstream_pf;
 125 #endif
 126 #ifdef CONFIG_HARDWALL
 127         /* Hardwall information for various resources. */
 128         struct hardwall_task hardwall[HARDWALL_TYPES];
 129 #endif
 130 #if CHIP_HAS_TILE_DMA()
 131         /* Async DMA TLB fault information */
 132         struct async_tlb dma_async_tlb;
 133 #endif
 134 };
 135
 136 #endif /* !__ASSEMBLY__ */
 137
 138 /*
 139  * Start with "sp" this many bytes below the top of the kernel stack.
 140  * This allows us to be cache-aware when handling the initial save
 141  * of the pt_regs value to the stack.
 142  */
 143 #define STACK_TOP_DELTA 64
 144
 145 /*
 146  * When entering the kernel via a fault, start with the top of the
 147  * pt_regs structure this many bytes below the top of the page.
 148  * This aligns the pt_regs structure optimally for cache-line access.
 149  */
 150 #ifdef __tilegx__
 151 #define KSTK_PTREGS_GAP  48
 152 #else
 153 #define KSTK_PTREGS_GAP  56
 154 #endif
 155
 156 #ifndef __ASSEMBLY__
 157
 158 #ifdef __tilegx__
 159 #define TASK_SIZE_MAX           (_AC(1, UL) << (MAX_VA_WIDTH - 1))
 160 #else
 161 #define TASK_SIZE_MAX           PAGE_OFFSET
 162 #endif
 163
 164 /* TASK_SIZE and related variables are always checked in "current" context. */
 165 #ifdef CONFIG_COMPAT
 166 #define COMPAT_TASK_SIZE        (1UL << 31)
 167 #define TASK_SIZE               ((current_thread_info()->status & TS_COMPAT) ?\
 168                                  COMPAT_TASK_SIZE : TASK_SIZE_MAX)
 169 #else
 170 #define TASK_SIZE               TASK_SIZE_MAX
 171 #endif
 172
 173 #define VDSO_BASE       ((unsigned long)current->active_mm->context.vdso_base)
 174 #define VDSO_SYM(x)     (VDSO_BASE + (unsigned long)(x))
 175
 176 #define STACK_TOP               TASK_SIZE
 177
 178 /* STACK_TOP_MAX is used temporarily in execve and should not check COMPAT. */
 179 #define STACK_TOP_MAX           TASK_SIZE_MAX
 180
 181 /*
 182  * This decides where the kernel will search for a free chunk of vm
 183  * space during mmap's, if it is using bottom-up mapping.
 184  */
 185 #define TASK_UNMAPPED_BASE      (PAGE_ALIGN(TASK_SIZE / 3))
 186
 187 #define HAVE_ARCH_PICK_MMAP_LAYOUT
 188
 189 #define INIT_THREAD {                                                   \
 190         .ksp = (unsigned long)init_stack + THREAD_SIZE - STACK_TOP_DELTA, \
 191         .interrupt_mask = -1ULL                                         \
 192 }
 193
 194 /* Kernel stack top for the task that first boots on this cpu. */
 195 DECLARE_PER_CPU(unsigned long, boot_sp);
 196
 197 /* PC to boot from on this cpu. */
 198 DECLARE_PER_CPU(unsigned long, boot_pc);
 199
 200 /* Do necessary setup to start up a newly executed thread. */
 201 static inline void start_thread(struct pt_regs *regs,
 202                                 unsigned long pc, unsigned long usp)
 203 {
 204         regs->pc = pc;
 205         regs->sp = usp;
 206         single_step_execve();
 207 }
 208
 209 /* Free all resources held by a thread. */
 210 static inline void release_thread(struct task_struct *dead_task)
 211 {
 212         /* Nothing for now */
 213 }
 214
 215 extern void prepare_exit_to_usermode(struct pt_regs *regs, u32 flags);
 216
 217
 218 /*
 219  * Return saved (kernel) PC of a blocked thread.
 220  * Only used in a printk() in kernel/sched/core.c, so don't work too hard.
 221  */
 222 #define thread_saved_pc(t)   ((t)->thread.pc)
 223
 224 unsigned long get_wchan(struct task_struct *p);
 225
 226 /* Return initial ksp value for given task. */
 227 #define task_ksp0(task) \
 228         ((unsigned long)(task)->stack + THREAD_SIZE - STACK_TOP_DELTA)
 229
 230 /* Return some info about the user process TASK. */
 231 #define task_pt_regs(task) \
 232         ((struct pt_regs *)(task_ksp0(task) - KSTK_PTREGS_GAP) - 1)
 233 #define current_pt_regs()                                   \
 234         ((struct pt_regs *)((stack_pointer | (THREAD_SIZE - 1)) - \
 235                             STACK_TOP_DELTA - (KSTK_PTREGS_GAP - 1)) - 1)
 236 #define task_sp(task)   (task_pt_regs(task)->sp)
 237 #define task_pc(task)   (task_pt_regs(task)->pc)
 238 /* Aliases for pc and sp (used in fs/proc/array.c) */
 239 #define KSTK_EIP(task)  task_pc(task)
 240 #define KSTK_ESP(task)  task_sp(task)
 241
 242 /* Fine-grained unaligned JIT support */
 243 #define GET_UNALIGN_CTL(tsk, adr)       get_unalign_ctl((tsk), (adr))
 244 #define SET_UNALIGN_CTL(tsk, val)       set_unalign_ctl((tsk), (val))
 245
 246 extern int get_unalign_ctl(struct task_struct *tsk, unsigned long adr);
 247 extern int set_unalign_ctl(struct task_struct *tsk, unsigned int val);
 248
 249 /* Standard format for printing registers and other word-size data. */
 250 #ifdef __tilegx__
 251 # define REGFMT "0x%016lx"
 252 #else
 253 # define REGFMT "0x%08lx"
 254 #endif
 255
 256 /*
 257  * Do some slow action (e.g. read a slow SPR).
 258  * Note that this must also have compiler-barrier semantics since
 259  * it may be used in a busy loop reading memory.
 260  */
 261 static inline void cpu_relax(void)
 262 {
 263         __insn_mfspr(SPR_PASS);
 264         barrier();
 265 }
 266
 267 #define cpu_relax_lowlatency() cpu_relax()
 268
 269 /* Info on this processor (see fs/proc/cpuinfo.c) */
 270 struct seq_operations;
 271 extern const struct seq_operations cpuinfo_op;
 272
 273 /* Provide information about the chip model. */
 274 extern char chip_model[64];
 275
 276 /* Data on which physical memory controller corresponds to which NUMA node. */
 277 extern int node_controller[];
 278
 279 /* Does the heap allocator return hash-for-home pages by default? */
 280 extern int hash_default;
 281
 282 /* Should kernel stack pages be hash-for-home? */
 283 extern int kstack_hash;
 284
 285 /* Does MAP_ANONYMOUS return hash-for-home pages by default? */
 286 #define uheap_hash hash_default
 287
 288
 289 /* Are we using huge pages in the TLB for kernel data? */
 290 extern int kdata_huge;
 291
 292 /* Support standard Linux prefetching. */
 293 #define ARCH_HAS_PREFETCH
 294 #define prefetch(x) __builtin_prefetch(x)
 295 #define PREFETCH_STRIDE CHIP_L2_LINE_SIZE()
 296
 297 /* Bring a value into the L1D, faulting the TLB if necessary. */
 298 #ifdef __tilegx__
 299 #define prefetch_L1(x) __insn_prefetch_l1_fault((void *)(x))
 300 #else
 301 #define prefetch_L1(x) __insn_prefetch_L1((void *)(x))
 302 #endif
 303
 304 #else /* __ASSEMBLY__ */
 305
 306 /* Do some slow action (e.g. read a slow SPR). */
 307 #define CPU_RELAX       mfspr zero, SPR_PASS
 308
 309 #endif /* !__ASSEMBLY__ */
 310
 311 /* Assembly code assumes that the PL is in the low bits. */
 312 #if SPR_EX_CONTEXT_1_1__PL_SHIFT != 0
 313 # error Fix assembly assumptions about PL
 314 #endif
 315
 316 /* We sometimes use these macros for EX_CONTEXT_0_1 as well. */
 317 #if SPR_EX_CONTEXT_1_1__PL_SHIFT != SPR_EX_CONTEXT_0_1__PL_SHIFT || \
 318     SPR_EX_CONTEXT_1_1__PL_RMASK != SPR_EX_CONTEXT_0_1__PL_RMASK || \
 319     SPR_EX_CONTEXT_1_1__ICS_SHIFT != SPR_EX_CONTEXT_0_1__ICS_SHIFT || \
 320     SPR_EX_CONTEXT_1_1__ICS_RMASK != SPR_EX_CONTEXT_0_1__ICS_RMASK
 321 # error Fix assumptions that EX1 macros work for both PL0 and PL1
 322 #endif
 323
 324 /* Allow pulling apart and recombining the PL and ICS bits in EX_CONTEXT. */
 325 #define EX1_PL(ex1) \
 326   (((ex1) >> SPR_EX_CONTEXT_1_1__PL_SHIFT) & SPR_EX_CONTEXT_1_1__PL_RMASK)
 327 #define EX1_ICS(ex1) \
 328   (((ex1) >> SPR_EX_CONTEXT_1_1__ICS_SHIFT) & SPR_EX_CONTEXT_1_1__ICS_RMASK)
 329 #define PL_ICS_EX1(pl, ics) \
 330   (((pl) << SPR_EX_CONTEXT_1_1__PL_SHIFT) | \
 331    ((ics) << SPR_EX_CONTEXT_1_1__ICS_SHIFT))
 332
 333 /*
 334  * Provide symbolic constants for PLs.
 335  */
 336 #define USER_PL 0
 337 #if CONFIG_KERNEL_PL == 2
 338 #define GUEST_PL 1
 339 #endif
 340 #define KERNEL_PL CONFIG_KERNEL_PL
 341
 342 /* SYSTEM_SAVE_K_0 holds the current cpu number ORed with ksp0. */
 343 #ifdef __tilegx__
 344 #define CPU_SHIFT 48
 345 #if CHIP_VA_WIDTH() > CPU_SHIFT
 346 # error Too many VA bits!
 347 #endif
 348 #define MAX_CPU_ID ((1 << (64 - CPU_SHIFT)) - 1)
 349 #define raw_smp_processor_id() \
 350         ((int)(__insn_mfspr(SPR_SYSTEM_SAVE_K_0) >> CPU_SHIFT))
 351 #define get_current_ksp0() \
 352         ((unsigned long)(((long)__insn_mfspr(SPR_SYSTEM_SAVE_K_0) << \
 353                           (64 - CPU_SHIFT)) >> (64 - CPU_SHIFT)))
 354 #define next_current_ksp0(task) ({ \
 355         unsigned long __ksp0 = task_ksp0(task) & ((1UL << CPU_SHIFT) - 1); \
 356         unsigned long __cpu = (long)raw_smp_processor_id() << CPU_SHIFT; \
 357         __ksp0 | __cpu; \
 358 })
 359 #else
 360 #define LOG2_NR_CPU_IDS 6
 361 #define MAX_CPU_ID ((1 << LOG2_NR_CPU_IDS) - 1)
 362 #define raw_smp_processor_id() \
 363         ((int)__insn_mfspr(SPR_SYSTEM_SAVE_K_0) & MAX_CPU_ID)
 364 #define get_current_ksp0() \
 365         (__insn_mfspr(SPR_SYSTEM_SAVE_K_0) & ~MAX_CPU_ID)
 366 #define next_current_ksp0(task) ({ \
 367         unsigned long __ksp0 = task_ksp0(task); \
 368         int __cpu = raw_smp_processor_id(); \
 369         BUG_ON(__ksp0 & MAX_CPU_ID); \
 370         __ksp0 | __cpu; \
 371 })
 372 #endif
 373 #if CONFIG_NR_CPUS > (MAX_CPU_ID + 1)
 374 # error Too many cpus!
 375 #endif
 376
 377 #endif /* _ASM_TILE_PROCESSOR_H */