X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;ds=sidebyside;f=riscv%2Fprocessor.cc;h=6eb5ddb9fc8c975e1efadca13f66f2e96a1d4d23;hb=980a0121e0b7244020bba7a8f422d97e7886fb08;hp=c11657556f6e2b3539c3319500dde2ad4aff4978;hpb=07d126d57ede2d3c7e28fddd08338b2171f884bb;p=riscv-isa-sim.git

diff --git a/riscv/processor.cc b/riscv/processor.cc
index c116575..6eb5ddb 100644
--- a/riscv/processor.cc
+++ b/riscv/processor.cc
@@ -1,320 +1,549 @@
+// See LICENSE for license details.
+
 #include "processor.h"
+#include "extension.h"
 #include "common.h"
 #include "config.h"
 #include "sim.h"
+#include "mmu.h"
+#include "htif.h"
 #include "disasm.h"
-#include <inttypes.h>
+#include <cinttypes>
 #include <cmath>
 #include <cstdlib>
 #include <iostream>
 #include <assert.h>
+#include <limits.h>
+#include <stdexcept>
+#include <algorithm>
+
+#undef STATE
+#define STATE state
 
-processor_t::processor_t(sim_t* _sim, mmu_t* _mmu, uint32_t _id)
-  : sim(*_sim), mmu(*_mmu), id(_id), utidx(0)
+processor_t::processor_t(const char* isa, sim_t* sim, uint32_t id)
+  : sim(sim), ext(NULL), disassembler(new disassembler_t),
+    id(id), run(false), debug(false)
 {
+  parse_isa_string(isa);
+
+  mmu = new mmu_t(sim, this);
+
   reset(true);
 
-  // create microthreads
-  for (int i=0; i<MAX_UTS; i++)
-    uts[i] = new processor_t(&sim, &mmu, id, i);
+  register_base_instructions();
 }
 
-processor_t::processor_t(sim_t* _sim, mmu_t* _mmu, uint32_t _id,
-                         uint32_t _utidx)
-  : sim(*_sim), mmu(*_mmu), id(_id)
+processor_t::~processor_t()
 {
-  reset(true);
-  set_pcr(PCR_SR, sr | SR_EF | SR_EV);
-  utidx = _utidx;
+#ifdef RISCV_ENABLE_HISTOGRAM
+  if (histogram_enabled)
+  {
+    fprintf(stderr, "PC Histogram size:%zu\n", pc_histogram.size());
+    for (auto it : pc_histogram)
+      fprintf(stderr, "%0" PRIx64 " %" PRIu64 "\n", it.first, it.second);
+  }
+#endif
 
-  // microthreads don't possess their own microthreads
-  for (int i=0; i<MAX_UTS; i++)
-    uts[i] = NULL;
+  delete mmu;
+  delete disassembler;
 }
 
-processor_t::~processor_t()
+static void bad_isa_string(const char* isa)
 {
+  fprintf(stderr, "error: bad --isa option %s\n", isa);
+  abort();
 }
 
-void processor_t::reset(bool value)
+void processor_t::parse_isa_string(const char* str)
 {
-  if (run == !value)
-    return;
-  run = !value;
+  std::string lowercase, tmp;
+  for (const char *r = str; *r; r++)
+    lowercase += std::tolower(*r);
+
+  const char* p = lowercase.c_str();
+  const char* all_subsets = "imafdc";
 
-  // the ISA guarantees on boot that the PC is 0x2000 and the the processor
-  // is in supervisor mode, and in 64-bit mode, if supported, with traps
-  // and virtual memory disabled.
-  set_pcr(PCR_SR, SR_S | SR_S64 | SR_IM);
-  pc = 0x2000;
-
-  // the following state is undefined upon boot-up,
-  // but we zero it for determinism
-  XPR.reset();
-  FPR.reset();
-
-  evec = 0;
-  epc = 0;
-  badvaddr = 0;
-  cause = 0;
-  pcr_k0 = 0;
-  pcr_k1 = 0;
-  count = 0;
-  compare = 0;
-  cycle = 0;
-  set_fsr(0);
-
-  // vector stuff
-  vecbanks = 0xff;
-  vecbanks_count = 8;
-  utidx = -1;
-  vlmax = 32;
-  vl = 0;
-  nxfpr_bank = 256;
-  nxpr_use = 32;
-  nfpr_use = 32;
+  max_xlen = 64;
+  isa = reg_t(2) << 62;
+
+  if (strncmp(p, "rv32", 4) == 0)
+    max_xlen = 32, isa = reg_t(1) << 30, p += 4;
+  else if (strncmp(p, "rv64", 4) == 0)
+    p += 4;
+  else if (strncmp(p, "rv", 2) == 0)
+    p += 2;
+
+  if (!*p) {
+    p = all_subsets;
+  } else if (*p == 'g') { // treat "G" as "IMAFD"
+    tmp = std::string("imafd") + (p+1);
+    p = &tmp[0];
+  } else if (*p != 'i') {
+    bad_isa_string(str);
+  }
+
+  isa_string = "rv" + std::to_string(max_xlen) + p;
+  isa |= 1L << ('s' - 'a'); // advertise support for supervisor mode
+
+  while (*p) {
+    isa |= 1L << (*p - 'a');
+
+    if (auto next = strchr(all_subsets, *p)) {
+      all_subsets = next + 1;
+      p++;
+    } else if (*p == 'x') {
+      const char* ext = p+1, *end = ext;
+      while (islower(*end))
+        end++;
+      register_extension(find_extension(std::string(ext, end - ext).c_str())());
+      p = end;
+    } else {
+      bad_isa_string(str);
+    }
+  }
+
+  if (supports_extension('D') && !supports_extension('F'))
+    bad_isa_string(str);
+
+  // advertise support for supervisor and user modes
+  isa |= 1L << ('s' - 'a');
+  isa |= 1L << ('u' - 'a');
 }
 
-void processor_t::set_fsr(uint32_t val)
+void state_t::reset()
 {
-  fsr = val & ~FSR_ZERO; // clear FSR bits that read as zero
+  memset(this, 0, sizeof(*this));
+  prv = PRV_M;
+  pc = DEFAULT_RSTVEC;
+  mtvec = DEFAULT_MTVEC;
+  load_reservation = -1;
 }
 
-void processor_t::vcfg()
+void processor_t::set_debug(bool value)
 {
-  if (nxpr_use + nfpr_use < 2)
-    vlmax = nxfpr_bank * vecbanks_count;
-  else
-    vlmax = (nxfpr_bank / (nxpr_use + nfpr_use - 1)) * vecbanks_count;
+  debug = value;
+  if (ext)
+    ext->set_debug(value);
+}
 
-  vlmax = std::min(vlmax, MAX_UTS);
+void processor_t::set_histogram(bool value)
+{
+  histogram_enabled = value;
+#ifndef RISCV_ENABLE_HISTOGRAM
+  if (value) {
+    fprintf(stderr, "PC Histogram support has not been properly enabled;");
+    fprintf(stderr, " please re-build the riscv-isa-run project using \"configure --enable-histogram\".\n");
+  }
+#endif
 }
 
-void processor_t::setvl(int vlapp)
+void processor_t::reset(bool value)
 {
-  vl = std::min(vlmax, vlapp);
+  if (run == !value)
+    return;
+  run = !value;
+
+  state.reset();
+  set_csr(CSR_MSTATUS, state.mstatus);
+
+  if (ext)
+    ext->reset(); // reset the extension
 }
 
-void processor_t::take_interrupt()
+void processor_t::raise_interrupt(reg_t which)
 {
-  uint32_t interrupts = interrupts_pending;
-  interrupts &= (sr & SR_IM) >> SR_IM_SHIFT;
+  throw trap_t(((reg_t)1 << (max_xlen-1)) | which);
+}
 
-  if(interrupts && (sr & SR_ET))
-    for(int i = 0; ; i++, interrupts >>= 1)
-      if(interrupts & 1)
-        throw interrupt_t(i);
+static int ctz(reg_t val)
+{
+  int res = 0;
+  if (val)
+    while ((val & 1) == 0)
+      val >>= 1, res++;
+  return res;
 }
 
-void processor_t::step(size_t n, bool noisy)
+void processor_t::take_interrupt()
 {
-  if(!run)
-    return;
+  reg_t pending_interrupts = state.mip & state.mie;
 
-  size_t i = 0;
-  while(1) try
-  {
-    take_interrupt();
-
-    mmu_t& _mmu = mmu;
-    insn_t insn;
-    insn_func_t func;
-    reg_t npc = pc;
-
-    // execute_insn fetches and executes one instruction
-    #define execute_insn(noisy) \
-      do { \
-        insn = _mmu.load_insn(npc, sr & SR_EC, &func); \
-        if(noisy) disasm(insn,pc); \
-        npc = func(this, insn, npc); \
-        pc = npc; \
-      } while(0)
-
-    if(noisy) for( ; i < n; i++) // print out instructions as we go
-      execute_insn(true);
-    else 
-    {
-      // unrolled for speed
-      for( ; n > 3 && i < n-3; i+=4)
-      {
-        execute_insn(false);
-        execute_insn(false);
-        execute_insn(false);
-        execute_insn(false);
-      }
-      for( ; i < n; i++)
-        execute_insn(false);
-    }
+  reg_t mie = get_field(state.mstatus, MSTATUS_MIE);
+  reg_t m_enabled = state.prv < PRV_M || (state.prv == PRV_M && mie);
+  reg_t enabled_interrupts = pending_interrupts & ~state.mideleg & -m_enabled;
 
-    break;
-  }
-  catch(trap_t t)
-  {
-    // an exception occurred in the target processor
-    i++;
-    take_trap(t,noisy);
-  }
-  catch(interrupt_t t)
-  {
-    i++;
-    take_trap((1ULL << (8*sizeof(reg_t)-1)) + t.i, noisy);
-  }
-  catch(vt_command_t cmd)
-  {
-    // this microthread has finished
-    i++;
-    assert(cmd == vt_command_stop);
-    break;
-  }
+  reg_t sie = get_field(state.mstatus, MSTATUS_SIE);
+  reg_t s_enabled = state.prv < PRV_S || (state.prv == PRV_S && sie);
+  enabled_interrupts |= pending_interrupts & state.mideleg & -s_enabled;
+
+  if (enabled_interrupts)
+    raise_interrupt(ctz(enabled_interrupts));
+}
 
-  cycle += i;
+static bool validate_priv(reg_t priv)
+{
+  return priv == PRV_U || priv == PRV_S || priv == PRV_M;
+}
 
-  // update timer and possibly register a timer interrupt
-  uint32_t old_count = count;
-  count += i;
-  if(old_count < compare && uint64_t(old_count) + i >= compare)
-    interrupts_pending |= 1 << IRQ_TIMER;
+void processor_t::set_privilege(reg_t prv)
+{
+  assert(validate_priv(prv));
+  mmu->flush_tlb();
+  state.prv = prv;
 }
 
-void processor_t::take_trap(reg_t t, bool noisy)
+void processor_t::take_trap(trap_t& t, reg_t epc)
 {
-  if(noisy)
-  {
-    if ((sreg_t)t < 0)
-      printf("core %3d: interrupt %lld, pc 0x%016llx\n",
-             id, (long long)(t << 1 >> 1), (unsigned long long)pc);
-    else
-      printf("core %3d: trap %s, pc 0x%016llx\n",
-             id, trap_name(trap_t(t)), (unsigned long long)pc);
+  if (debug)
+    fprintf(stderr, "core %3d: exception %s, epc 0x%016" PRIx64 "\n",
+            id, t.name(), epc);
+
+  // by default, trap to M-mode, unless delegated to S-mode
+  reg_t bit = t.cause();
+  reg_t deleg = state.medeleg;
+  if (bit & ((reg_t)1 << (max_xlen-1)))
+    deleg = state.mideleg, bit &= ~((reg_t)1 << (max_xlen-1));
+  if (state.prv <= PRV_S && bit < max_xlen && ((deleg >> bit) & 1)) {
+    // handle the trap in S-mode
+    state.pc = state.stvec;
+    state.scause = t.cause();
+    state.sepc = epc;
+    if (t.has_badaddr())
+      state.sbadaddr = t.get_badaddr();
+
+    reg_t s = state.mstatus;
+    s = set_field(s, MSTATUS_SPIE, get_field(s, MSTATUS_UIE << state.prv));
+    s = set_field(s, MSTATUS_SPP, state.prv);
+    s = set_field(s, MSTATUS_SIE, 0);
+    set_csr(CSR_MSTATUS, s);
+    set_privilege(PRV_S);
+  } else {
+    state.pc = state.mtvec;
+    state.mcause = t.cause();
+    state.mepc = epc;
+    if (t.has_badaddr())
+      state.mbadaddr = t.get_badaddr();
+
+    reg_t s = state.mstatus;
+    s = set_field(s, MSTATUS_MPIE, get_field(s, MSTATUS_UIE << state.prv));
+    s = set_field(s, MSTATUS_MPP, state.prv);
+    s = set_field(s, MSTATUS_MIE, 0);
+    set_csr(CSR_MSTATUS, s);
+    set_privilege(PRV_M);
   }
 
-  // switch to supervisor, set previous supervisor bit, disable traps
-  set_pcr(PCR_SR, (((sr & ~SR_ET) | SR_S) & ~SR_PS) | ((sr & SR_S) ? SR_PS : 0));
-  cause = t;
-  epc = pc;
-  pc = evec;
-  badvaddr = mmu.get_badvaddr();
+  yield_load_reservation();
 }
 
-void processor_t::deliver_ipi()
+void processor_t::disasm(insn_t insn)
 {
-  if (run)
-    set_pcr(PCR_CLR_IPI, 1);
+  uint64_t bits = insn.bits() & ((1ULL << (8 * insn_length(insn.bits()))) - 1);
+  fprintf(stderr, "core %3d: 0x%016" PRIx64 " (0x%08" PRIx64 ") %s\n",
+          id, state.pc, bits, disassembler->disassemble(insn).c_str());
 }
 
-void processor_t::disasm(insn_t insn, reg_t pc)
+static bool validate_vm(int max_xlen, reg_t vm)
 {
-  // the disassembler is stateless, so we share it
-  static disassembler disasm;
-  printf("core %3d: 0x%016llx (0x%08x) %s\n", id, (unsigned long long)pc,
-         insn.bits, disasm.disassemble(insn).c_str());
+  if (max_xlen == 64 && (vm == VM_SV39 || vm == VM_SV48))
+    return true;
+  if (max_xlen == 32 && vm == VM_SV32)
+    return true;
+  return vm == VM_MBARE;
 }
 
-void processor_t::set_pcr(int which, reg_t val)
+void processor_t::set_csr(int which, reg_t val)
 {
+  val = zext_xlen(val);
+  reg_t delegable_ints = MIP_SSIP | MIP_STIP | (1 << IRQ_HOST) | (1 << IRQ_COP);
+  reg_t all_ints = delegable_ints | MIP_MSIP | MIP_MTIP;
   switch (which)
   {
-    case PCR_SR:
-      sr = val & ~SR_ZERO; // clear SR bits that read as zero
-#ifndef RISCV_ENABLE_64BIT
-      sr &= ~(SR_S64 | SR_U64);
-#endif
-#ifndef RISCV_ENABLE_FPU
-      sr &= ~SR_EF;
-#endif
-#ifndef RISCV_ENABLE_RVC
-      sr &= ~SR_EC;
-#endif
-#ifndef RISCV_ENABLE_VEC
-      sr &= ~SR_EV;
-#endif
-      // update MMU state and flush TLB
-      mmu.set_vm_enabled(sr & SR_VM);
-      mmu.set_supervisor(sr & SR_S);
-      mmu.flush_tlb();
-      // set the fixed-point register length
-      xprlen = ((sr & SR_S) ? (sr & SR_S64) : (sr & SR_U64)) ? 64 : 32;
-      break;
-    case PCR_EPC:
-      epc = val;
+    case CSR_FFLAGS:
+      dirty_fp_state;
+      state.fflags = val & (FSR_AEXC >> FSR_AEXC_SHIFT);
       break;
-    case PCR_EVEC: 
-      evec = val;
+    case CSR_FRM:
+      dirty_fp_state;
+      state.frm = val & (FSR_RD >> FSR_RD_SHIFT);
       break;
-    case PCR_COUNT:
-      count = val;
+    case CSR_FCSR:
+      dirty_fp_state;
+      state.fflags = (val & FSR_AEXC) >> FSR_AEXC_SHIFT;
+      state.frm = (val & FSR_RD) >> FSR_RD_SHIFT;
       break;
-    case PCR_COMPARE:
-      interrupts_pending &= ~(1 << IRQ_TIMER);
-      compare = val;
-      break;
-    case PCR_PTBR:
-      mmu.set_ptbr(val);
+    case CSR_MSTATUS: {
+      if ((val ^ state.mstatus) &
+          (MSTATUS_VM | MSTATUS_MPP | MSTATUS_MPRV | MSTATUS_PUM))
+        mmu->flush_tlb();
+
+      reg_t mask = MSTATUS_SIE | MSTATUS_SPIE | MSTATUS_MIE | MSTATUS_MPIE
+                 | MSTATUS_SPP | MSTATUS_FS | MSTATUS_MPRV | MSTATUS_PUM
+                 | (ext ? MSTATUS_XS : 0);
+
+      if (validate_vm(max_xlen, get_field(val, MSTATUS_VM)))
+        mask |= MSTATUS_VM;
+      if (validate_priv(get_field(val, MSTATUS_MPP)))
+        mask |= MSTATUS_MPP;
+
+      state.mstatus = (state.mstatus & ~mask) | (val & mask);
+
+      bool dirty = (state.mstatus & MSTATUS_FS) == MSTATUS_FS;
+      dirty |= (state.mstatus & MSTATUS_XS) == MSTATUS_XS;
+      if (max_xlen == 32)
+        state.mstatus = set_field(state.mstatus, MSTATUS32_SD, dirty);
+      else
+        state.mstatus = set_field(state.mstatus, MSTATUS64_SD, dirty);
+
+      // spike supports the notion of xlen < max_xlen, but current priv spec
+      // doesn't provide a mechanism to run RV32 software on an RV64 machine
+      xlen = max_xlen;
       break;
-    case PCR_SEND_IPI:
-      sim.send_ipi(val);
+    }
+    case CSR_MIP: {
+      reg_t mask = MIP_SSIP | MIP_STIP | MIP_MSIP;
+      state.mip = (state.mip & ~mask) | (val & mask);
       break;
-    case PCR_CLR_IPI:
-      if (val & 1)
-        interrupts_pending |= (1 << IRQ_IPI);
-      else
-        interrupts_pending &= ~(1 << IRQ_IPI);
+    }
+    case CSR_MIPI:
+      state.mip = set_field(state.mip, MIP_MSIP, val & 1);
       break;
-    case PCR_K0:
-      pcr_k0 = val;
+    case CSR_MIE:
+      state.mie = (state.mie & ~all_ints) | (val & all_ints);
       break;
-    case PCR_K1:
-      pcr_k1 = val;
+    case CSR_MIDELEG:
+      state.mideleg = (state.mideleg & ~delegable_ints) | (val & delegable_ints);
       break;
-    case PCR_VECBANK:
-      vecbanks = val & 0xff;
-      vecbanks_count = __builtin_popcountll(vecbanks);
+    case CSR_MEDELEG: {
+      reg_t mask = 0;
+#define DECLARE_CAUSE(name, value) mask |= 1ULL << (value);
+#include "encoding.h"
+#undef DECLARE_CAUSE
+      state.medeleg = (state.medeleg & ~mask) | (val & mask);
       break;
-    case PCR_TOHOST:
-      fromhost = 0;
-      tohost = val;
+    }
+    case CSR_MUCOUNTEREN:
+      state.mucounteren = val & 7;
       break;
-    case PCR_FROMHOST:
-      fromhost = val;
-      tohost = 0;
+    case CSR_MSCOUNTEREN:
+      state.mscounteren = val & 7;
       break;
+    case CSR_SSTATUS: {
+      reg_t mask = SSTATUS_SIE | SSTATUS_SPIE | SSTATUS_SPP | SSTATUS_FS
+                 | SSTATUS_XS | SSTATUS_PUM;
+      return set_csr(CSR_MSTATUS, (state.mstatus & ~mask) | (val & mask));
+    }
+    case CSR_SIP:
+      return set_csr(CSR_MIP,
+                     (state.mip & ~state.mideleg) | (val & state.mideleg));
+    case CSR_SIE:
+      return set_csr(CSR_MIE,
+                     (state.mie & ~state.mideleg) | (val & state.mideleg));
+    case CSR_SEPC: state.sepc = val; break;
+    case CSR_STVEC: state.stvec = val >> 2 << 2; break;
+    case CSR_SPTBR: state.sptbr = val; break;
+    case CSR_SSCRATCH: state.sscratch = val; break;
+    case CSR_SCAUSE: state.scause = val; break;
+    case CSR_SBADADDR: state.sbadaddr = val; break;
+    case CSR_MEPC: state.mepc = val; break;
+    case CSR_MTVEC: state.mtvec = val >> 2 << 2; break;
+    case CSR_MSCRATCH: state.mscratch = val; break;
+    case CSR_MCAUSE: state.mcause = val; break;
+    case CSR_MBADADDR: state.mbadaddr = val; break;
   }
 }
 
-reg_t processor_t::get_pcr(int which)
+reg_t processor_t::get_csr(int which)
 {
   switch (which)
   {
-    case PCR_SR:
-      return sr;
-    case PCR_EPC:
-      return epc;
-    case PCR_BADVADDR:
-      return badvaddr;
-    case PCR_EVEC:
-      return evec;
-    case PCR_COUNT:
-      return count;
-    case PCR_COMPARE:
-      return compare;
-    case PCR_CAUSE:
-      return cause;
-    case PCR_PTBR:
-      return mmu.get_ptbr();
-    case PCR_COREID:
-      return id;
-    case PCR_IMPL:
-      return 1;
-    case PCR_K0:
-      return pcr_k0;
-    case PCR_K1:
-      return pcr_k1;
-    case PCR_VECBANK:
-      return vecbanks;
-    case PCR_TOHOST:
-      return tohost;
-    case PCR_FROMHOST:
-      return fromhost;
+    case CSR_FFLAGS:
+      require_fp;
+      if (!supports_extension('F'))
+        break;
+      return state.fflags;
+    case CSR_FRM:
+      require_fp;
+      if (!supports_extension('F'))
+        break;
+      return state.frm;
+    case CSR_FCSR:
+      require_fp;
+      if (!supports_extension('F'))
+        break;
+      return (state.fflags << FSR_AEXC_SHIFT) | (state.frm << FSR_RD_SHIFT);
+    case CSR_TIME:
+    case CSR_INSTRET:
+    case CSR_CYCLE:
+      if ((state.mucounteren >> (which & (xlen-1))) & 1)
+        return get_csr(which + (CSR_MCYCLE - CSR_CYCLE));
+      break;
+    case CSR_STIME:
+    case CSR_SINSTRET:
+    case CSR_SCYCLE:
+      if ((state.mscounteren >> (which & (xlen-1))) & 1)
+        return get_csr(which + (CSR_MCYCLE - CSR_SCYCLE));
+      break;
+    case CSR_MUCOUNTEREN: return state.mucounteren;
+    case CSR_MSCOUNTEREN: return state.mscounteren;
+    case CSR_MUCYCLE_DELTA: return 0;
+    case CSR_MUTIME_DELTA: return 0;
+    case CSR_MUINSTRET_DELTA: return 0;
+    case CSR_MSCYCLE_DELTA: return 0;
+    case CSR_MSTIME_DELTA: return 0;
+    case CSR_MSINSTRET_DELTA: return 0;
+    case CSR_MUCYCLE_DELTAH: if (xlen > 32) break; else return 0;
+    case CSR_MUTIME_DELTAH: if (xlen > 32) break; else return 0;
+    case CSR_MUINSTRET_DELTAH: if (xlen > 32) break; else return 0;
+    case CSR_MSCYCLE_DELTAH: if (xlen > 32) break; else return 0;
+    case CSR_MSTIME_DELTAH: if (xlen > 32) break; else return 0;
+    case CSR_MSINSTRET_DELTAH: if (xlen > 32) break; else return 0;
+    case CSR_MCYCLE: return state.minstret;
+    case CSR_MINSTRET: return state.minstret;
+    case CSR_MCYCLEH: if (xlen > 32) break; else return state.minstret >> 32;
+    case CSR_MINSTRETH: if (xlen > 32) break; else return state.minstret >> 32;
+    case CSR_SSTATUS: {
+      reg_t mask = SSTATUS_SIE | SSTATUS_SPIE | SSTATUS_SPP | SSTATUS_FS
+                 | SSTATUS_XS | SSTATUS_PUM;
+      reg_t sstatus = state.mstatus & mask;
+      if ((sstatus & SSTATUS_FS) == SSTATUS_FS ||
+          (sstatus & SSTATUS_XS) == SSTATUS_XS)
+        sstatus |= (xlen == 32 ? SSTATUS32_SD : SSTATUS64_SD);
+      return sstatus;
+    }
+    case CSR_SIP: return state.mip & state.mideleg;
+    case CSR_SIE: return state.mie & state.mideleg;
+    case CSR_SEPC: return state.sepc;
+    case CSR_SBADADDR: return state.sbadaddr;
+    case CSR_STVEC: return state.stvec;
+    case CSR_SCAUSE:
+      if (max_xlen > xlen)
+        return state.scause | ((state.scause >> (max_xlen-1)) << (xlen-1));
+      return state.scause;
+    case CSR_SPTBR: return state.sptbr;
+    case CSR_SASID: return 0;
+    case CSR_SSCRATCH: return state.sscratch;
+    case CSR_MSTATUS: return state.mstatus;
+    case CSR_MIP: return state.mip;
+    case CSR_MIPI: return 0;
+    case CSR_MIE: return state.mie;
+    case CSR_MEPC: return state.mepc;
+    case CSR_MSCRATCH: return state.mscratch;
+    case CSR_MCAUSE: return state.mcause;
+    case CSR_MBADADDR: return state.mbadaddr;
+    case CSR_MISA: return isa;
+    case CSR_MARCHID: return 0;
+    case CSR_MIMPID: return 0;
+    case CSR_MVENDORID: return 0;
+    case CSR_MHARTID: return id;
+    case CSR_MTVEC: return state.mtvec;
+    case CSR_MEDELEG: return state.medeleg;
+    case CSR_MIDELEG: return state.mideleg;
+  }
+  throw trap_illegal_instruction();
+}
+
+reg_t illegal_instruction(processor_t* p, insn_t insn, reg_t pc)
+{
+  throw trap_illegal_instruction();
+}
+
+insn_func_t processor_t::decode_insn(insn_t insn)
+{
+  // look up opcode in hash table
+  size_t idx = insn.bits() % OPCODE_CACHE_SIZE;
+  insn_desc_t desc = opcode_cache[idx];
+
+  if (unlikely(insn.bits() != desc.match)) {
+    // fall back to linear search
+    insn_desc_t* p = &instructions[0];
+    while ((insn.bits() & p->mask) != p->match)
+      p++;
+    desc = *p;
+
+    if (p->mask != 0 && p > &instructions[0]) {
+      if (p->match != (p-1)->match && p->match != (p+1)->match) {
+        // move to front of opcode list to reduce miss penalty
+        while (--p >= &instructions[0])
+          *(p+1) = *p;
+        instructions[0] = desc;
+      }
+    }
+
+    opcode_cache[idx] = desc;
+    opcode_cache[idx].match = insn.bits();
+  }
+
+  return xlen == 64 ? desc.rv64 : desc.rv32;
+}
+
+void processor_t::register_insn(insn_desc_t desc)
+{
+  instructions.push_back(desc);
+}
+
+void processor_t::build_opcode_map()
+{
+  struct cmp {
+    bool operator()(const insn_desc_t& lhs, const insn_desc_t& rhs) {
+      if (lhs.match == rhs.match)
+        return lhs.mask > rhs.mask;
+      return lhs.match > rhs.match;
+    }
+  };
+  std::sort(instructions.begin(), instructions.end(), cmp());
+
+  for (size_t i = 0; i < OPCODE_CACHE_SIZE; i++)
+    opcode_cache[i] = {1, 0, &illegal_instruction, &illegal_instruction};
+}
+
+void processor_t::register_extension(extension_t* x)
+{
+  for (auto insn : x->get_instructions())
+    register_insn(insn);
+  build_opcode_map();
+  for (auto disasm_insn : x->get_disasms())
+    disassembler->add_insn(disasm_insn);
+  if (ext != NULL)
+    throw std::logic_error("only one extension may be registered");
+  ext = x;
+  x->set_processor(this);
+}
+
+void processor_t::register_base_instructions()
+{
+  #define DECLARE_INSN(name, match, mask) \
+    insn_bits_t name##_match = (match), name##_mask = (mask);
+  #include "encoding.h"
+  #undef DECLARE_INSN
+
+  #define DEFINE_INSN(name) \
+    REGISTER_INSN(this, name, name##_match, name##_mask)
+  #include "insn_list.h"
+  #undef DEFINE_INSN
+
+  register_insn({0, 0, &illegal_instruction, &illegal_instruction});
+  build_opcode_map();
+}
+
+bool processor_t::load(reg_t addr, size_t len, uint8_t* bytes)
+{
+  try {
+    auto res = get_csr(addr / (max_xlen / 8));
+    memcpy(bytes, &res, len);
+    return true;
+  } catch (trap_illegal_instruction& t) {
+    return false;
+  }
+}
+
+bool processor_t::store(reg_t addr, size_t len, const uint8_t* bytes)
+{
+  try {
+    reg_t value = 0;
+    memcpy(&value, bytes, len);
+    set_csr(addr / (max_xlen / 8), value);
+    return true;
+  } catch (trap_illegal_instruction& t) {
+    return false;
   }
-  return -1;
 }