Subversion Repositories or1k

/* dcache_model.c -- data cache simulation

   Copyright (C) 1999 Damjan Lampret, lampret@opencores.org

This file is part of OpenRISC 1000 Architectural Simulator.

This program is free software; you can redistribute it and/or modify

it under the terms of the GNU General Public License as published by

the Free Software Foundation; either version 2 of the License, or

(at your option) any later version.

This program is distributed in the hope that it will be useful,

but WITHOUT ANY WARRANTY; without even the implied warranty of

MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

GNU General Public License for more details.

You should have received a copy of the GNU General Public License

along with this program; if not, write to the Free Software

Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */

/* Cache functions.

   At the moment this functions only simulate functionality of data

   caches and do not influence on fetche/decode/execute stages and timings.

   They are here only to verify performance of various cache configurations.

 */

#include <stdio.h>

#include <string.h>

#include <errno.h>

#include <stdarg.h>

#include "config.h"

#ifdef HAVE_INTTYPES_H

#include <inttypes.h>

#endif

#include "port.h"

#include "arch.h"

#include "dcache_model.h"

#include "abstract.h"

#include "except.h"

#include "opcode/or32.h"

#include "spr_defs.h"

#include "execute.h"

#include "stats.h"

#include "sprs.h"

#include "sim-config.h"

#include "misc.h"

/* Data cache */

struct dc_set {

  struct {

    uint32_t line[MAX_DC_BLOCK_SIZE];

    oraddr_t tagaddr;  /* tag address */

    int lru;    /* least recently used */

  } way[MAX_DC_WAYS];

} dc[MAX_DC_SETS];

void dc_info(void)

{

  if (!(cpu_state.sprs[SPR_UPR] & SPR_UPR_DCP)) {

    PRINTF("DCache not implemented. Set UPR[DCP].\n");

    return;

  }

  PRINTF("Data cache %dKB: ", config.dc.nsets * config.dc.blocksize * config.dc.nways / 1024);

  PRINTF("%d ways, %d sets, block size %d bytes\n", config.dc.nways, config.dc.nsets, config.dc.blocksize);

}

/* First check if data is already in the cache and if it is:

    - increment DC read hit stats,

    - set 'lru' at this way to config.dc.ustates - 1 and

      decrement 'lru' of other ways unless they have reached 0,

   and if not:

    - increment DC read miss stats

    - find lru way and entry and replace old tag with tag of the 'dataaddr'

    - set 'lru' with config.dc.ustates - 1 and decrement 'lru' of other

      ways unless they have reached 0

    - refill cache line

*/

uint32_t dc_simulate_read(oraddr_t dataaddr, oraddr_t virt_addr, int width)

{

  int set, way = -1;

  int i;

  oraddr_t tagaddr;

  uint32_t tmp = 0;

  if (!(cpu_state.sprs[SPR_UPR] & SPR_UPR_DCP) ||

      !(cpu_state.sprs[SPR_SR] & SPR_SR_DCE)   ||

      data_ci) {

    if (width == 4)

      tmp = evalsim_mem32(dataaddr, virt_addr);

    else if (width == 2)

      tmp = evalsim_mem16(dataaddr, virt_addr);

    else if (width == 1)

      tmp = evalsim_mem8(dataaddr, virt_addr);

    if (cur_area && cur_area->log)

      fprintf (cur_area->log, "[%"PRIxADDR"] -> read %08"PRIx32"\n", dataaddr,

               tmp);

    return tmp;

  }

  /* Which set to check out? */

  set = (dataaddr / config.dc.blocksize) % config.dc.nsets;

  tagaddr = (dataaddr / config.dc.blocksize) / config.dc.nsets;

  /* Scan all ways and try to find a matching way. */

  for (i = 0; i < config.dc.nways; i++)

    if (dc[set].way[i].tagaddr == tagaddr)

      way = i;

  /* Did we find our cached data? */

  if (way >= 0) { /* Yes, we did. */

    dc_stats.readhit++;

    for (i = 0; i < config.dc.nways; i++)

      if (dc[set].way[i].lru > dc[set].way[way].lru)

        dc[set].way[i].lru--;

    dc[set].way[way].lru = config.dc.ustates - 1;

    runtime.sim.mem_cycles += config.dc.load_hitdelay;

    tmp = dc[set].way[way].line[(dataaddr & (config.dc.blocksize - 1)) >> 2];

    if (width == 4)

      return tmp;

    else if (width == 2) {

      tmp = ((tmp >> ((dataaddr & 2) ? 0 : 16)) & 0xffff);

      return tmp;

    }

    else if (width == 1) {

      tmp = ((tmp  >> (8 * (3 - (dataaddr & 3)))) & 0xff);

      return tmp;

    }

  } else {  /* No, we didn't. */

    int minlru = config.dc.ustates - 1;

    int minway = 0;

    dc_stats.readmiss++;

    for (i = 0; i < config.dc.nways; i++) {

      if (dc[set].way[i].lru < minlru) {

        minway = i;

        minlru = dc[set].way[i].lru;

      }

    }

    for (i = 0; i < (config.dc.blocksize); i += 4) {

      /* FIXME: What is the virtual address meant to be? (ie. What happens if

       * we read out of memory while refilling a cache line?) */

      tmp = evalsim_mem32((dataaddr & ~(config.dc.blocksize - 1)) + (((dataaddr & ~ADDR_C(3)) + i) & (config.dc.blocksize - 1)), 0);

      dc[set].way[minway].line[((dataaddr + i) & (config.dc.blocksize - 1)) >> 2] = tmp;

      if(!cur_area) {

        dc[set].way[minway].tagaddr = -1;

        dc[set].way[minway].lru = 0;

        return 0;

      } else if (cur_area->log)

        fprintf (cur_area->log, "[%"PRIxADDR"] -> read %08"PRIx32"\n", dataaddr,

                 tmp);

    }

    dc[set].way[minway].tagaddr = tagaddr;

    for (i = 0; i < config.dc.nways; i++)

      if (dc[set].way[i].lru)

        dc[set].way[i].lru--;

    dc[set].way[minway].lru = config.dc.ustates - 1;

    runtime.sim.mem_cycles += config.dc.load_missdelay;

    tmp = dc[set].way[minway].line[(dataaddr & (config.dc.blocksize - 1)) >> 2];

    if (width == 4)

      return tmp;

    else if (width == 2) {

      tmp = (tmp >> ((dataaddr & 2) ? 0 : 16)) & 0xffff;

      return tmp;

    }

    else if (width == 1) {

      tmp = (tmp  >> (8 * (3 - (dataaddr & 3)))) & 0xff;

      return tmp;

    }

  }

  return 0;

}

/* First check if data is already in the cache and if it is:

    - increment DC write hit stats,

    - set 'lru' at this way to config.dc.ustates - 1 and

      decrement 'lru' of other ways unless they have reached 0,

   and if not:

    - increment DC write miss stats

    - find lru way and entry and replace old tag with tag of the 'dataaddr'

    - set 'lru' with config.dc.ustates - 1 and decrement 'lru' of other

      ways unless they have reached 0

*/

void dc_simulate_write(oraddr_t dataaddr, oraddr_t virt_addr, uint32_t data,

                       int width)

{

  int set, way = -1;

  int i;

  oraddr_t tagaddr;

  uint32_t tmp;

  if (width == 4)

    setsim_mem32(dataaddr, virt_addr, data);

  else if (width == 2)

    setsim_mem16(dataaddr, virt_addr, data);

  else if (width == 1)

    setsim_mem8(dataaddr, virt_addr, data);

  if (!(cpu_state.sprs[SPR_UPR] & SPR_UPR_DCP) ||

      !(cpu_state.sprs[SPR_SR] & SPR_SR_DCE) ||

      data_ci || !cur_area)

    return;

  /* Which set to check out? */

  set = (dataaddr / config.dc.blocksize) % config.dc.nsets;

  tagaddr = (dataaddr / config.dc.blocksize) / config.dc.nsets;

  /* Scan all ways and try to find a matching way. */

  for (i = 0; i < config.dc.nways; i++)

    if (dc[set].way[i].tagaddr == tagaddr)

      way = i;

  /* Did we find our cached data? */

  if (way >= 0) { /* Yes, we did. */

    dc_stats.writehit++;

    for (i = 0; i < config.dc.nways; i++)

      if (dc[set].way[i].lru > dc[set].way[way].lru)

        dc[set].way[i].lru--;

    dc[set].way[way].lru = config.dc.ustates - 1;

    runtime.sim.mem_cycles += config.dc.store_hitdelay;

    tmp = dc[set].way[way].line[(dataaddr & (config.dc.blocksize - 1)) >> 2];

    if (width == 4)

      tmp = data;

    else if (width == 2) {

      tmp &= 0xffff << ((dataaddr & 2) ? 16 : 0);

      tmp |= (data & 0xffff) << ((dataaddr & 2) ? 0 : 16);

    }

    else if (width == 1) {

      tmp &= ~(0xff << (8 * (3 - (dataaddr & 3))));

      tmp |= (data & 0xff) << (8 * (3 - (dataaddr & 3)));

    }

    dc[set].way[way].line[(dataaddr & (config.dc.blocksize - 1)) >> 2] = tmp;

  }

  else {  /* No, we didn't. */

    int minlru = config.dc.ustates - 1;

    int minway = 0;

    dc_stats.writemiss++;

    for (i = 0; i < config.dc.nways; i++)

      if (dc[set].way[i].lru < minlru)

        minway = i;

    for (i = 0; i < (config.dc.blocksize); i += 4) {

      dc[set].way[minway].line[((dataaddr + i) & (config.dc.blocksize - 1)) >> 2] =

        /* FIXME: Same comment as in dc_simulate_read */

        evalsim_mem32((dataaddr & ~(config.dc.blocksize - 1)) + (((dataaddr & ~3ul)+ i) & (config.dc.blocksize - 1)), 0);

      if(!cur_area) {

        dc[set].way[minway].tagaddr = -1;

        dc[set].way[minway].lru = 0;

        return;

      }

    }

    dc[set].way[minway].tagaddr = tagaddr;

    for (i = 0; i < config.dc.nways; i++)

      if (dc[set].way[i].lru)

        dc[set].way[i].lru--;

    dc[set].way[minway].lru = config.dc.ustates - 1;

    runtime.sim.mem_cycles += config.dc.store_missdelay;

  }

}

/* First check if data is already in the cache and if it is:

    - invalidate block if way isn't locked

   otherwise don't do anything.

*/

void dc_inv(oraddr_t dataaddr)

{

  int set, way = -1;

  int i;

  oraddr_t tagaddr;

  if (!(cpu_state.sprs[SPR_UPR] & SPR_UPR_DCP))

    return;

  /* Which set to check out? */

  set = (dataaddr / config.dc.blocksize) % config.dc.nsets;

  tagaddr = (dataaddr / config.dc.blocksize) / config.dc.nsets;

  if (!(cpu_state.sprs[SPR_SR] & SPR_SR_DCE)) {

    for (i = 0; i < config.dc.nways; i++) {

      dc[set].way[i].tagaddr = -1;

      dc[set].way[i].lru = 0;

    }

    return;

  }

   /* Scan all ways and try to find a matching way. */

  for (i = 0; i < config.dc.nways; i++)

    if (dc[set].way[i].tagaddr == tagaddr)

      way = i;

  /* Did we find our cached data? */

  if (way >= 0) { /* Yes, we did. */

    dc[set].way[way].tagaddr = -1;

    dc[set].way[way].lru = 0;

  }

}

/*-----------------------------------------------------[ DC configuration ]---*/

void dc_enabled(union param_val val, void *dat)

{

  config.dc.enabled = val.int_val;

  if(val.int_val)

    cpu_state.sprs[SPR_UPR] |= SPR_UPR_DCP;

  else

    cpu_state.sprs[SPR_UPR] &= ~SPR_UPR_DCP;

}

void dc_nsets(union param_val val, void *dat)

{

  if (is_power2(val.int_val) && val.int_val <= MAX_DC_SETS){

    config.dc.nsets = val.int_val;

    cpu_state.sprs[SPR_DCCFGR] &= ~SPR_DCCFGR_NCS;

    cpu_state.sprs[SPR_DCCFGR] |= log2_int(val.int_val) << 3;

  }

  else {

    char tmp[200];

    sprintf (tmp, "value of power of two and lower or equal than %i expected.", MAX_DC_SETS);

    CONFIG_ERROR(tmp);

  }

}

void dc_nways(union param_val val, void *dat)

{

  if (is_power2(val.int_val) && val.int_val <= MAX_DC_WAYS){

    config.dc.nways = val.int_val;

    cpu_state.sprs[SPR_DCCFGR] &= ~SPR_DCCFGR_NCW;

    cpu_state.sprs[SPR_DCCFGR] |= log2_int(val.int_val);

  }

  else{

    char tmp[200];

    sprintf (tmp, "value of power of two and lower or equal than %i expected.",

    MAX_DC_WAYS);

    CONFIG_ERROR(tmp);

  }

}

void dc_blocksize(union param_val val, void *dat)

{

  if (is_power2(val.int_val)) {

    config.dc.blocksize = val.int_val;

    cpu_state.sprs[SPR_ICCFGR] &= ~SPR_ICCFGR_CBS;

    cpu_state.sprs[SPR_ICCFGR] |= log2_int(val.int_val) << 7;

  } else

    CONFIG_ERROR("value of power of two expected.");

}

void dc_ustates(union param_val val, void *dat)

{

  if (val.int_val >= 2 && val.int_val <= 4)

    config.dc.ustates = val.int_val;

  else

    CONFIG_ERROR("invalid USTATE.");

}

void dc_load_missdelay(union param_val val, void *dat)

{

  config.dc.load_missdelay = val.int_val;

}

void dc_load_hitdelay(union param_val val, void *dat)

{

  config.dc.load_hitdelay = val.int_val;

}

void dc_store_missdelay(union param_val val, void *dat)

{

  config.dc.store_missdelay = val.int_val;

}

void dc_store_hitdelay(union param_val val, void *dat)

{

  config.dc.store_hitdelay = val.int_val;

}

void reg_dc_sec(void)

{

  struct config_section *sec = reg_config_sec("dc", NULL, NULL);

  reg_config_param(sec, "enabled", paramt_int, dc_enabled);

  reg_config_param(sec, "nsets", paramt_int, dc_nsets);

  reg_config_param(sec, "nways", paramt_int, dc_nways);

  reg_config_param(sec, "blocksize", paramt_int, dc_blocksize);

  reg_config_param(sec, "ustates", paramt_int, dc_ustates);

  reg_config_param(sec, "load_missdelay", paramt_int, dc_load_missdelay);

  reg_config_param(sec, "load_hitdelay", paramt_int, dc_load_hitdelay);

  reg_config_param(sec, "store_missdelay", paramt_int, dc_store_missdelay);

  reg_config_param(sec, "store_hitdelay", paramt_int, dc_store_hitdelay);

}