DBT-RISE-TGC/incl/iss/arch/riscv_hart_msu_vp.h

1200 lines
49 KiB
C++

/*******************************************************************************
* Copyright (C) 2017, 2018, 2021 MINRES Technologies GmbH
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
* Contributors:
* eyck@minres.com - initial implementation
******************************************************************************/
#ifndef _RISCV_HART_MSU_VP_H
#define _RISCV_HART_MSU_VP_H
#include "riscv_hart_common.h"
#include "iss/arch/traits.h"
#include "iss/instrumentation_if.h"
#include "iss/log_categories.h"
#include "iss/vm_if.h"
#ifndef FMT_HEADER_ONLY
#define FMT_HEADER_ONLY
#endif
#include <fmt/format.h>
#include <array>
#include <elfio/elfio.hpp>
#include <iomanip>
#include <sstream>
#include <type_traits>
#include <unordered_map>
#include <util/bit_field.h>
#include <util/ities.h>
#include <util/sparse_array.h>
#if defined(__GNUC__)
#define likely(x) __builtin_expect(!!(x), 1)
#define unlikely(x) __builtin_expect(!!(x), 0)
#else
#define likely(x) x
#define unlikely(x) x
#endif
namespace iss {
namespace arch {
template <typename BASE> class riscv_hart_msu_vp : public BASE {
protected:
const std::array<const char, 4> lvl = {{'U', 'S', 'H', 'M'}};
const std::array<const char *, 16> trap_str = {{""
"Instruction address misaligned", // 0
"Instruction access fault", // 1
"Illegal instruction", // 2
"Breakpoint", // 3
"Load address misaligned", // 4
"Load access fault", // 5
"Store/AMO address misaligned", // 6
"Store/AMO access fault", // 7
"Environment call from U-mode", // 8
"Environment call from S-mode", // 9
"Reserved", // a
"Environment call from M-mode", // b
"Instruction page fault", // c
"Load page fault", // d
"Reserved", // e
"Store/AMO page fault"}};
const std::array<const char *, 12> irq_str = {
{"User software interrupt", "Supervisor software interrupt", "Reserved", "Machine software interrupt",
"User timer interrupt", "Supervisor timer interrupt", "Reserved", "Machine timer interrupt",
"User external interrupt", "Supervisor external interrupt", "Reserved", "Machine external interrupt"}};
public:
using super = BASE;
using this_class = riscv_hart_msu_vp<BASE>;
using virt_addr_t = typename super::virt_addr_t;
using phys_addr_t = typename super::phys_addr_t;
using reg_t = typename super::reg_t;
using addr_t = typename super::addr_t;
using rd_csr_f = iss::status (this_class::*)(unsigned addr, reg_t &);
using wr_csr_f = iss::status (this_class::*)(unsigned addr, reg_t);
// primary template
template <class T, class Enable = void> struct hart_state {};
// specialization 32bit
template <typename T> class hart_state<T, typename std::enable_if<std::is_same<T, uint32_t>::value>::type> {
public:
BEGIN_BF_DECL(mstatus_t, T);
// SD bit is read-only and is set when either the FS or XS bits encode a Dirty state (i.e., SD=((FS==11) OR XS==11)))
BF_FIELD(SD, 31, 1);
// Trap SRET
BF_FIELD(TSR, 22, 1);
// Timeout Wait
BF_FIELD(TW, 21, 1);
// Trap Virtual Memory
BF_FIELD(TVM, 20, 1);
// Make eXecutable Readable
BF_FIELD(MXR, 19, 1);
// permit Supervisor User Memory access
BF_FIELD(SUM, 18, 1);
// Modify PRiVilege
BF_FIELD(MPRV, 17, 1);
// status of additional user-mode extensions and associated state, All off/None dirty or clean, some on/None dirty, some clean/Some dirty
BF_FIELD(XS, 15, 2);
// floating-point unit status Off/Initial/Clean/Dirty
BF_FIELD(FS, 13, 2);
// machine previous privilege
BF_FIELD(MPP, 11, 2);
// supervisor previous privilege
BF_FIELD(SPP, 8, 1);
// previous machine interrupt-enable
BF_FIELD(MPIE, 7, 1);
// previous supervisor interrupt-enable
BF_FIELD(SPIE, 5, 1);
// previous user interrupt-enable
BF_FIELD(UPIE, 4, 1);
// machine interrupt-enable
BF_FIELD(MIE, 3, 1);
// supervisor interrupt-enable
BF_FIELD(SIE, 1, 1);
// user interrupt-enable
BF_FIELD(UIE, 0, 1);
END_BF_DECL();
mstatus_t mstatus;
static const reg_t mstatus_reset_val = 0;
void write_mstatus(T val, unsigned priv_lvl) {
auto mask = get_mask(priv_lvl);
auto new_val = (mstatus.st.value & ~mask) | (val & mask);
mstatus = new_val;
}
T satp;
static constexpr T get_misa() { return (1UL << 30) | ISA_I | ISA_M | ISA_A | ISA_U | ISA_S | ISA_M; }
static constexpr uint32_t get_mask(unsigned priv_lvl) {
#if __cplusplus < 201402L
return priv_lvl == PRIV_U ? 0x80000011UL : priv_lvl == PRIV_S ? 0x800de133UL : 0x807ff9ddUL;
#else
switch (priv_lvl) {
case PRIV_U: return 0x80000011UL; // 0b1000 0000 0000 0000 0000 0000 0001 0001
case PRIV_S: return 0x800de133UL; // 0b1000 0000 0000 1101 1110 0001 0011 0011
default: return 0x807ff9ddUL; // 0b1000 0000 0111 1111 1111 1001 1011 1011
}
#endif
}
static inline vm_info decode_vm_info(uint32_t state, T sptbr) {
if (state == PRIV_M) return {0, 0, 0, 0};
if (state <= PRIV_S)
switch (bit_sub<31, 1>(sptbr)) {
case 0: return {0, 0, 0, 0}; // off
case 1: return {2, 10, 4, bit_sub<0, 22>(sptbr) << PGSHIFT}; // SV32
default: abort();
}
abort();
return {0, 0, 0, 0}; // dummy
}
};
// specialization 64bit
template <typename T> class hart_state<T, typename std::enable_if<std::is_same<T, uint64_t>::value>::type> {
public:
BEGIN_BF_DECL(mstatus_t, T);
// SD bit is read-only and is set when either the FS or XS bits encode a Dirty state (i.e., SD=((FS==11) OR XS==11)))
BF_FIELD(SD, 63, 1);
// value of XLEN for S-mode
BF_FIELD(SXL, 34, 2);
// value of XLEN for U-mode
BF_FIELD(UXL, 32, 2);
// Trap SRET
BF_FIELD(TSR, 22, 1);
// Timeout Wait
BF_FIELD(TW, 21, 1);
// Trap Virtual Memory
BF_FIELD(TVM, 20, 1);
// Make eXecutable Readable
BF_FIELD(MXR, 19, 1);
// permit Supervisor User Memory access
BF_FIELD(SUM, 18, 1);
// Modify PRiVilege
BF_FIELD(MPRV, 17, 1);
// status of additional user-mode extensions and associated state, All off/None dirty or clean, some on/None dirty, some clean/Some dirty
BF_FIELD(XS, 15, 2);
// floating-point unit status Off/Initial/Clean/Dirty
BF_FIELD(FS, 13, 2);
// machine previous privilege
BF_FIELD(MPP, 11, 2);
// supervisor previous privilege
BF_FIELD(SPP, 8, 1);
// previous machine interrupt-enable
BF_FIELD(MPIE, 7, 1);
// previous supervisor interrupt-enable
BF_FIELD(SPIE, 5, 1);
// previous user interrupt-enable
BF_FIELD(UPIE, 4, 1);
// machine interrupt-enable
BF_FIELD(MIE, 3, 1);
// supervisor interrupt-enable
BF_FIELD(SIE, 1, 1);
// user interrupt-enable
BF_FIELD(UIE, 0, 1);
END_BF_DECL();
mstatus_t mstatus;
static const reg_t mstatus_reset_val = 0xa00000000;
void write_mstatus(T val, unsigned priv_lvl) {
T old_val = mstatus;
auto mask = get_mask(priv_lvl);
auto new_val = (old_val & ~mask) | (val & mask);
if ((new_val & mstatus.SXL.Mask) == 0) {
new_val |= old_val & mstatus.SXL.Mask;
}
if ((new_val & mstatus.UXL.Mask) == 0) {
new_val |= old_val & mstatus.UXL.Mask;
}
mstatus = new_val;
}
T satp;
static constexpr T get_misa() { return (2ULL << 62) | ISA_I | ISA_M | ISA_A | ISA_U | ISA_S | ISA_M; }
static constexpr T get_mask(unsigned priv_lvl) {
uint64_t ret;
switch (priv_lvl) {
case PRIV_U: ret = 0x8000000f00000011ULL;break; // 0b1...0 1111 0000 0000 0111 1111 1111 1001 1011 1011
case PRIV_S: ret = 0x8000000f000de133ULL;break; // 0b1...0 0011 0000 0000 0000 1101 1110 0001 0011 0011
default: ret = 0x8000000f007ff9ddULL;break; // 0b1...0 1111 0000 0000 0111 1111 1111 1001 1011 1011
}
return ret;
}
static inline vm_info decode_vm_info(uint32_t state, T sptbr) {
if (state == PRIV_M) return {0, 0, 0, 0};
if (state <= PRIV_S)
switch (bit_sub<60, 4>(sptbr)) {
case 0: return {0, 0, 0, 0}; // off
case 8: return {3, 9, 8, bit_sub<0, 44>(sptbr) << PGSHIFT};// SV39
case 9: return {4, 9, 8, bit_sub<0, 44>(sptbr) << PGSHIFT};// SV48
case 10: return {5, 9, 8, bit_sub<0, 44>(sptbr) << PGSHIFT};// SV57
case 11: return {6, 9, 8, bit_sub<0, 44>(sptbr) << PGSHIFT};// SV64
default: abort();
}
abort();
return {0, 0, 0, 0}; // dummy
}
};
using hart_state_type = hart_state<reg_t>;
const typename super::reg_t PGSIZE = 1 << PGSHIFT;
const typename super::reg_t PGMASK = PGSIZE - 1;
constexpr reg_t get_irq_mask(size_t mode) {
std::array<const reg_t, 4> m = {{
0b000100010001, // U mode
0b001100110011, // S mode
0,
0b101110111011 // M mode
}};
return m[mode];
}
riscv_hart_msu_vp();
virtual ~riscv_hart_msu_vp() = default;
void reset(uint64_t address) override;
std::pair<uint64_t, bool> load_file(std::string name, int type = -1) override;
virtual phys_addr_t virt2phys(const iss::addr_t &addr) override;
iss::status read(const address_type type, const access_type access, const uint32_t space,
const uint64_t addr, const unsigned length, uint8_t *const data) override;
iss::status write(const address_type type, const access_type access, const uint32_t space,
const uint64_t addr, const unsigned length, const uint8_t *const data) override;
virtual uint64_t enter_trap(uint64_t flags) override { return riscv_hart_msu_vp::enter_trap(flags, fault_data); }
virtual uint64_t enter_trap(uint64_t flags, uint64_t addr) override;
virtual uint64_t leave_trap(uint64_t flags) override;
void wait_until(uint64_t flags) override;
void disass_output(uint64_t pc, const std::string instr) override {
CLOG(INFO, disass) << fmt::format("0x{:016x} {:40} [p:{};s:0x{:x};c:{}]",
pc, instr, lvl[this->reg.PRIV], (reg_t)state.mstatus, this->reg.icount);
};
iss::instrumentation_if *get_instrumentation_if() override { return &instr_if; }
protected:
struct riscv_instrumentation_if : public iss::instrumentation_if {
riscv_instrumentation_if(riscv_hart_msu_vp<BASE> &arch)
: arch(arch) {}
/**
* get the name of this architecture
*
* @return the name of this architecture
*/
const std::string core_type_name() const override { return traits<BASE>::core_type; }
virtual uint64_t get_pc() { return arch.get_pc(); };
virtual uint64_t get_next_pc() { return arch.get_next_pc(); };
virtual void set_curr_instr_cycles(unsigned cycles) { arch.cycle_offset += cycles - 1; };
riscv_hart_msu_vp<BASE> &arch;
};
friend struct riscv_instrumentation_if;
addr_t get_pc() { return this->reg.PC; }
addr_t get_next_pc() { return this->reg.NEXT_PC; }
virtual iss::status read_mem(phys_addr_t addr, unsigned length, uint8_t *const data);
virtual iss::status write_mem(phys_addr_t addr, unsigned length, const uint8_t *const data);
virtual iss::status read_csr(unsigned addr, reg_t &val);
virtual iss::status write_csr(unsigned addr, reg_t val);
hart_state_type state;
uint64_t cycle_offset;
reg_t fault_data;
std::array<vm_info, 2> vm;
uint64_t tohost = tohost_dflt;
uint64_t fromhost = fromhost_dflt;
unsigned to_host_wr_cnt = 0;
riscv_instrumentation_if instr_if;
using mem_type = util::sparse_array<uint8_t, 1ULL << 32>;
using csr_type = util::sparse_array<typename traits<BASE>::reg_t, 1ULL << 12, 12>;
using csr_page_type = typename csr_type::page_type;
mem_type mem;
csr_type csr;
void update_vm_info();
std::stringstream uart_buf;
std::unordered_map<reg_t, uint64_t> ptw;
std::unordered_map<uint64_t, uint8_t> atomic_reservation;
std::unordered_map<unsigned, rd_csr_f> csr_rd_cb;
std::unordered_map<unsigned, wr_csr_f> csr_wr_cb;
private:
iss::status read_cycle(unsigned addr, reg_t &val);
iss::status read_time(unsigned addr, reg_t &val);
iss::status read_status(unsigned addr, reg_t &val);
iss::status write_status(unsigned addr, reg_t val);
iss::status read_ie(unsigned addr, reg_t &val);
iss::status write_ie(unsigned addr, reg_t val);
iss::status read_ip(unsigned addr, reg_t &val);
iss::status write_ip(unsigned addr, reg_t val);
iss::status read_satp(unsigned addr, reg_t &val);
iss::status write_satp(unsigned addr, reg_t val);
iss::status read_fcsr(unsigned addr, reg_t &val);
iss::status write_fcsr(unsigned addr, reg_t val);
protected:
void check_interrupt();
};
template <typename BASE>
riscv_hart_msu_vp<BASE>::riscv_hart_msu_vp()
: state()
, cycle_offset(0)
, instr_if(*this) {
csr[misa] = hart_state_type::get_misa();
uart_buf.str("");
// read-only registers
csr_wr_cb[misa] = nullptr;
for (unsigned addr = mcycle; addr <= hpmcounter31; ++addr) csr_wr_cb[addr] = nullptr;
for (unsigned addr = mcycleh; addr <= hpmcounter31h; ++addr) csr_wr_cb[addr] = nullptr;
// special handling
csr_rd_cb[time] = &this_class::read_time;
csr_wr_cb[time] = nullptr;
csr_rd_cb[timeh] = &this_class::read_time;
csr_wr_cb[timeh] = nullptr;
csr_rd_cb[mcycle] = &this_class::read_cycle;
csr_rd_cb[mcycleh] = &this_class::read_cycle;
csr_rd_cb[minstret] = &this_class::read_cycle;
csr_rd_cb[minstreth] = &this_class::read_cycle;
csr_rd_cb[mstatus] = &this_class::read_status;
csr_wr_cb[mstatus] = &this_class::write_status;
csr_rd_cb[sstatus] = &this_class::read_status;
csr_wr_cb[sstatus] = &this_class::write_status;
csr_rd_cb[ustatus] = &this_class::read_status;
csr_wr_cb[ustatus] = &this_class::write_status;
csr_rd_cb[mip] = &this_class::read_ip;
csr_wr_cb[mip] = &this_class::write_ip;
csr_rd_cb[sip] = &this_class::read_ip;
csr_wr_cb[sip] = &this_class::write_ip;
csr_rd_cb[uip] = &this_class::read_ip;
csr_wr_cb[uip] = &this_class::write_ip;
csr_rd_cb[mie] = &this_class::read_ie;
csr_wr_cb[mie] = &this_class::write_ie;
csr_rd_cb[sie] = &this_class::read_ie;
csr_wr_cb[sie] = &this_class::write_ie;
csr_rd_cb[uie] = &this_class::read_ie;
csr_wr_cb[uie] = &this_class::write_ie;
csr_rd_cb[satp] = &this_class::read_satp;
csr_wr_cb[satp] = &this_class::write_satp;
csr_rd_cb[fcsr] = &this_class::read_fcsr;
csr_wr_cb[fcsr] = &this_class::write_fcsr;
csr_rd_cb[fflags] = &this_class::read_fcsr;
csr_wr_cb[fflags] = &this_class::write_fcsr;
csr_rd_cb[frm] = &this_class::read_fcsr;
csr_wr_cb[frm] = &this_class::write_fcsr;
}
template <typename BASE> std::pair<uint64_t, bool> riscv_hart_msu_vp<BASE>::load_file(std::string name, int type) {
FILE *fp = fopen(name.c_str(), "r");
if (fp) {
std::array<char, 5> buf;
auto n = fread(buf.data(), 1, 4, fp);
if (n != 4) throw std::runtime_error("input file has insufficient size");
buf[4] = 0;
if (strcmp(buf.data() + 1, "ELF") == 0) {
fclose(fp);
// Create elfio reader
ELFIO::elfio reader;
// Load ELF data
if (!reader.load(name)) throw std::runtime_error("could not process elf file");
// check elf properties
if (reader.get_class() != ELFCLASS32)
if (sizeof(reg_t) == 4) throw std::runtime_error("wrong elf class in file");
if (reader.get_type() != ET_EXEC) throw std::runtime_error("wrong elf type in file");
if (reader.get_machine() != EM_RISCV) throw std::runtime_error("wrong elf machine in file");
for (const auto pseg : reader.segments) {
const auto fsize = pseg->get_file_size(); // 0x42c/0x0
const auto seg_data = pseg->get_data();
if (fsize > 0) {
auto res = this->write(iss::address_type::PHYSICAL, iss::access_type::DEBUG_WRITE,
traits<BASE>::MEM, pseg->get_physical_address(),
fsize, reinterpret_cast<const uint8_t *const>(seg_data));
if (res != iss::Ok)
LOG(ERROR) << "problem writing " << fsize << "bytes to 0x" << std::hex
<< pseg->get_physical_address();
}
}
for (const auto sec : reader.sections) {
if (sec->get_name() == ".tohost") {
tohost = sec->get_address();
fromhost = tohost + 0x40;
}
}
return std::make_pair(reader.get_entry(), true);
}
throw std::runtime_error("memory load file is not a valid elf file");
}
throw std::runtime_error("memory load file not found");
}
template <typename BASE>
iss::status riscv_hart_msu_vp<BASE>::read(const address_type type, const access_type access, const uint32_t space,
const uint64_t addr, const unsigned length, uint8_t *const data) {
#ifndef NDEBUG
if (access && iss::access_type::DEBUG) {
LOG(TRACEALL) << "debug read of " << length << " bytes @addr 0x" << std::hex << addr;
} else if(access && iss::access_type::FETCH){
LOG(TRACEALL) << "fetch of " << length << " bytes @addr 0x" << std::hex << addr;
} else {
LOG(TRACE) << "read of " << length << " bytes @addr 0x" << std::hex << addr;
}
#endif
try {
switch (space) {
case traits<BASE>::MEM: {
if (unlikely((access == iss::access_type::FETCH || access == iss::access_type::DEBUG_FETCH) && (addr & 0x1) == 1)) {
fault_data = addr;
if (access && iss::access_type::DEBUG) throw trap_access(0, addr);
this->reg.trap_state = (1 << 31); // issue trap 0
return iss::Err;
}
try {
if (unlikely((addr & ~PGMASK) != ((addr + length - 1) & ~PGMASK))) { // we may cross a page boundary
vm_info vm = hart_state_type::decode_vm_info(this->reg.PRIV, state.satp);
if (vm.levels != 0) { // VM is active
auto split_addr = (addr + length) & ~PGMASK;
auto len1 = split_addr - addr;
auto res = read(type, access, space, addr, len1, data);
if (res == iss::Ok)
res = read(type, access, space, split_addr, length - len1, data + len1);
return res;
}
}
auto res = type==iss::address_type::PHYSICAL?
read_mem( BASE::v2p(phys_addr_t{access, space, addr}), length, data):
read_mem( BASE::v2p(iss::addr_t{access, type, space, addr}), length, data);
if (unlikely(res != iss::Ok)) this->reg.trap_state = (1 << 31) | (5 << 16); // issue trap 5 (load access fault
return res;
} catch (trap_access &ta) {
this->reg.trap_state = (1 << 31) | ta.id;
return iss::Err;
}
} break;
case traits<BASE>::CSR: {
if (length != sizeof(reg_t)) return iss::Err;
return read_csr(addr, *reinterpret_cast<reg_t *const>(data));
} break;
case traits<BASE>::FENCE: {
if ((addr + length) > mem.size()) return iss::Err;
switch (addr) {
case 2: // SFENCE:VMA lower
case 3: { // SFENCE:VMA upper
auto tvm = state.mstatus.TVM;
if (this->reg.PRIV == PRIV_S & tvm != 0) {
this->reg.trap_state = (1 << 31) | (2 << 16);
this->fault_data = this->reg.PC;
return iss::Err;
}
return iss::Ok;
}
}
} break;
case traits<BASE>::RES: {
auto it = atomic_reservation.find(addr);
if (it != atomic_reservation.end() && it->second != 0) {
memset(data, 0xff, length);
atomic_reservation.erase(addr);
} else
memset(data, 0, length);
} break;
default:
return iss::Err; // assert("Not supported");
}
return iss::Ok;
} catch (trap_access &ta) {
this->reg.trap_state = (1 << 31) | ta.id;
return iss::Err;
}
}
template <typename BASE>
iss::status riscv_hart_msu_vp<BASE>::write(const address_type type, const access_type access, const uint32_t space,
const uint64_t addr, const unsigned length, const uint8_t *const data) {
#ifndef NDEBUG
const char *prefix = (access && iss::access_type::DEBUG) ? "debug " : "";
switch (length) {
case 8:
LOG(TRACE) << prefix << "write of " << length << " bytes (0x" << std::hex << *(uint64_t *)&data[0] << std::dec
<< ") @addr 0x" << std::hex << addr;
break;
case 4:
LOG(TRACE) << prefix << "write of " << length << " bytes (0x" << std::hex << *(uint32_t *)&data[0] << std::dec
<< ") @addr 0x" << std::hex << addr;
break;
case 2:
LOG(TRACE) << prefix << "write of " << length << " bytes (0x" << std::hex << *(uint16_t *)&data[0] << std::dec
<< ") @addr 0x" << std::hex << addr;
break;
case 1:
LOG(TRACE) << prefix << "write of " << length << " bytes (0x" << std::hex << (uint16_t)data[0] << std::dec
<< ") @addr 0x" << std::hex << addr;
break;
default:
LOG(TRACE) << prefix << "write of " << length << " bytes @addr " << addr;
}
#endif
try {
switch (space) {
case traits<BASE>::MEM: {
if (unlikely((access && iss::access_type::FETCH) && (addr & 0x1) == 1)) {
fault_data = addr;
if (access && iss::access_type::DEBUG) throw trap_access(0, addr);
this->reg.trap_state = (1 << 31); // issue trap 0
return iss::Err;
}
try {
if (unlikely((addr & ~PGMASK) != ((addr + length - 1) & ~PGMASK))) { // we may cross a page boundary
vm_info vm = hart_state_type::decode_vm_info(this->reg.PRIV, state.satp);
if (vm.levels != 0) { // VM is active
auto split_addr = (addr + length) & ~PGMASK;
auto len1 = split_addr - addr;
auto res = write(type, access, space, addr, len1, data);
if (res == iss::Ok)
res = write(type, access, space, split_addr, length - len1, data + len1);
return res;
}
}
auto res = type==iss::address_type::PHYSICAL?
write_mem(phys_addr_t{access, space, addr}, length, data):
write_mem(BASE::v2p(iss::addr_t{access, type, space, addr}), length, data);
if (unlikely(res != iss::Ok))
this->reg.trap_state = (1 << 31) | (5 << 16); // issue trap 7 (Store/AMO access fault)
return res;
} catch (trap_access &ta) {
this->reg.trap_state = (1 << 31) | ta.id;
return iss::Err;
}
phys_addr_t paddr = BASE::v2p(iss::addr_t{access, type, space, addr});
if ((paddr.val + length) > mem.size()) return iss::Err;
switch (paddr.val) {
case 0x10013000: // UART0 base, TXFIFO reg
case 0x10023000: // UART1 base, TXFIFO reg
uart_buf << (char)data[0];
if (((char)data[0]) == '\n' || data[0] == 0) {
// LOG(INFO)<<"UART"<<((paddr.val>>16)&0x3)<<" send
// '"<<uart_buf.str()<<"'";
std::cout << uart_buf.str();
uart_buf.str("");
}
return iss::Ok;
case 0x10008000: { // HFROSC base, hfrosccfg reg
auto &p = mem(paddr.val / mem.page_size);
auto offs = paddr.val & mem.page_addr_mask;
std::copy(data, data + length, p.data() + offs);
auto &x = *(p.data() + offs + 3);
if (x & 0x40) x |= 0x80; // hfroscrdy = 1 if hfroscen==1
return iss::Ok;
}
case 0x10008008: { // HFROSC base, pllcfg reg
auto &p = mem(paddr.val / mem.page_size);
auto offs = paddr.val & mem.page_addr_mask;
std::copy(data, data + length, p.data() + offs);
auto &x = *(p.data() + offs + 3);
x |= 0x80; // set pll lock upon writing
return iss::Ok;
} break;
default: {}
}
} break;
case traits<BASE>::CSR: {
if (length != sizeof(reg_t)) return iss::Err;
return write_csr(addr, *reinterpret_cast<const reg_t *>(data));
} break;
case traits<BASE>::FENCE: {
if ((addr + length) > mem.size()) return iss::Err;
switch (addr) {
case 2:
case 3: {
ptw.clear();
auto tvm = state.mstatus.TVM;
if (this->reg.PRIV == PRIV_S & tvm != 0) {
this->reg.trap_state = (1 << 31) | (2 << 16);
this->fault_data = this->reg.PC;
return iss::Err;
}
return iss::Ok;
}
}
} break;
case traits<BASE>::RES: {
atomic_reservation[addr] = data[0];
} break;
default:
return iss::Err;
}
return iss::Ok;
} catch (trap_access &ta) {
this->reg.trap_state = (1 << 31) | ta.id;
return iss::Err;
}
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_csr(unsigned addr, reg_t &val) {
if (addr >= csr.size()) return iss::Err;
auto req_priv_lvl = (addr >> 8) & 0x3;
if (this->reg.PRIV < req_priv_lvl) throw illegal_instruction_fault(this->fault_data);
auto it = csr_rd_cb.find(addr);
if (it == csr_rd_cb.end()) {
val = csr[addr & csr.page_addr_mask];
return iss::Ok;
}
rd_csr_f f = it->second;
if (f == nullptr) throw illegal_instruction_fault(this->fault_data);
return (this->*f)(addr, val);
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::write_csr(unsigned addr, reg_t val) {
if (addr >= csr.size()) return iss::Err;
auto req_priv_lvl = (addr >> 8) & 0x3;
if (this->reg.PRIV < req_priv_lvl)
throw illegal_instruction_fault(this->fault_data);
if((addr&0xc00)==0xc00)
throw illegal_instruction_fault(this->fault_data);
auto it = csr_wr_cb.find(addr);
if (it == csr_wr_cb.end()) {
csr[addr & csr.page_addr_mask] = val;
return iss::Ok;
}
wr_csr_f f = it->second;
if (f == nullptr) throw illegal_instruction_fault(this->fault_data);
return (this->*f)(addr, val);
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_cycle(unsigned addr, reg_t &val) {
auto cycle_val = this->reg.icount + cycle_offset;
if (addr == mcycle) {
val = static_cast<reg_t>(cycle_val);
} else if (addr == mcycleh) {
if (sizeof(typename traits<BASE>::reg_t) != 4) return iss::Err;
val = static_cast<reg_t>(cycle_val >> 32);
}
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_time(unsigned addr, reg_t &val) {
uint64_t time_val = (this->reg.icount + cycle_offset) / (100000000 / 32768 - 1); //-> ~3052;
if (addr == time) {
val = static_cast<reg_t>(time_val);
} else if (addr == timeh) {
if (sizeof(typename traits<BASE>::reg_t) != 4) return iss::Err;
val = static_cast<reg_t>(time_val >> 32);
}
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_status(unsigned addr, reg_t &val) {
auto req_priv_lvl = (addr >> 8) & 0x3;
val = state.mstatus & hart_state_type::get_mask(req_priv_lvl);
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::write_status(unsigned addr, reg_t val) {
auto req_priv_lvl = (addr >> 8) & 0x3;
state.write_mstatus(val, req_priv_lvl);
check_interrupt();
update_vm_info();
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_ie(unsigned addr, reg_t &val) {
val = csr[mie];
if (addr < mie) val &= csr[mideleg];
if (addr < sie) val &= csr[sideleg];
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::write_ie(unsigned addr, reg_t val) {
auto req_priv_lvl = (addr >> 8) & 0x3;
auto mask = get_irq_mask(req_priv_lvl);
csr[mie] = (csr[mie] & ~mask) | (val & mask);
check_interrupt();
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_ip(unsigned addr, reg_t &val) {
val = csr[mip];
if (addr < mip) val &= csr[mideleg];
if (addr < sip) val &= csr[sideleg];
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::write_ip(unsigned addr, reg_t val) {
auto req_priv_lvl = (addr >> 8) & 0x3;
auto mask = get_irq_mask(req_priv_lvl);
mask &= ~(1 << 7); // MTIP is read only
csr[mip] = (csr[mip] & ~mask) | (val & mask);
check_interrupt();
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_satp(unsigned addr, reg_t &val) {
reg_t tvm = state.mstatus.TVM;
if (this->reg.PRIV == PRIV_S & tvm != 0) {
this->reg.trap_state = (1 << 31) | (2 << 16);
this->fault_data = this->reg.PC;
return iss::Err;
}
val = state.satp;
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::write_satp(unsigned addr, reg_t val) {
reg_t tvm = state.mstatus.TVM;
if (this->reg.PRIV == PRIV_S & tvm != 0) {
this->reg.trap_state = (1 << 31) | (2 << 16);
this->fault_data = this->reg.PC;
return iss::Err;
}
state.satp = val;
update_vm_info();
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_fcsr(unsigned addr, reg_t &val) {
switch (addr) {
case 1: // fflags, 4:0
val = bit_sub<0, 5>(this->get_fcsr());
break;
case 2: // frm, 7:5
val = bit_sub<5, 3>(this->get_fcsr());
break;
case 3: // fcsr
val = this->get_fcsr();
break;
default:
return iss::Err;
}
return iss::Ok;
}
template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::write_fcsr(unsigned addr, reg_t val) {
switch (addr) {
case 1: // fflags, 4:0
this->set_fcsr((this->get_fcsr() & 0xffffffe0) | (val & 0x1f));
break;
case 2: // frm, 7:5
this->set_fcsr((this->get_fcsr() & 0xffffff1f) | ((val & 0x7) << 5));
break;
case 3: // fcsr
this->set_fcsr(val & 0xff);
break;
default:
return iss::Err;
}
return iss::Ok;
}
template <typename BASE>
iss::status riscv_hart_msu_vp<BASE>::read_mem(phys_addr_t paddr, unsigned length, uint8_t *const data) {
if ((paddr.val + length) > mem.size()) return iss::Err;
switch (paddr.val) {
case 0x0200BFF8: { // CLINT base, mtime reg
if (sizeof(reg_t) < length) return iss::Err;
reg_t time_val;
this->read_csr(time, time_val);
std::copy((uint8_t *)&time_val, ((uint8_t *)&time_val) + length, data);
} break;
case 0x10008000: {
const mem_type::page_type &p = mem(paddr.val / mem.page_size);
uint64_t offs = paddr.val & mem.page_addr_mask;
std::copy(p.data() + offs, p.data() + offs + length, data);
if (this->reg.icount > 30000) data[3] |= 0x80;
} break;
default: {
const auto &p = mem(paddr.val / mem.page_size);
auto offs = paddr.val & mem.page_addr_mask;
std::copy(p.data() + offs, p.data() + offs + length, data);
}
}
return iss::Ok;
}
template <typename BASE>
iss::status riscv_hart_msu_vp<BASE>::write_mem(phys_addr_t paddr, unsigned length, const uint8_t *const data) {
if ((paddr.val + length) > mem.size()) return iss::Err;
switch (paddr.val) {
case 0x10013000: // UART0 base, TXFIFO reg
case 0x10023000: // UART1 base, TXFIFO reg
uart_buf << (char)data[0];
if (((char)data[0]) == '\n' || data[0] == 0) {
// LOG(INFO)<<"UART"<<((paddr.val>>16)&0x3)<<" send
// '"<<uart_buf.str()<<"'";
std::cout << uart_buf.str();
uart_buf.str("");
}
break;
case 0x10008000: { // HFROSC base, hfrosccfg reg
mem_type::page_type &p = mem(paddr.val / mem.page_size);
size_t offs = paddr.val & mem.page_addr_mask;
std::copy(data, data + length, p.data() + offs);
uint8_t &x = *(p.data() + offs + 3);
if (x & 0x40) x |= 0x80; // hfroscrdy = 1 if hfroscen==1
} break;
case 0x10008008: { // HFROSC base, pllcfg reg
mem_type::page_type &p = mem(paddr.val / mem.page_size);
size_t offs = paddr.val & mem.page_addr_mask;
std::copy(data, data + length, p.data() + offs);
uint8_t &x = *(p.data() + offs + 3);
x |= 0x80; // set pll lock upon writing
} break;
default: {
mem_type::page_type &p = mem(paddr.val / mem.page_size);
std::copy(data, data + length, p.data() + (paddr.val & mem.page_addr_mask));
// tohost handling in case of riscv-test
if (paddr.access && iss::access_type::FUNC) {
auto tohost_upper = (traits<BASE>::XLEN == 32 && paddr.val == (tohost + 4)) ||
(traits<BASE>::XLEN == 64 && paddr.val == tohost);
auto tohost_lower =
(traits<BASE>::XLEN == 32 && paddr.val == tohost) || (traits<BASE>::XLEN == 64 && paddr.val == tohost);
if (tohost_lower || tohost_upper) {
uint64_t hostvar = *reinterpret_cast<uint64_t *>(p.data() + (tohost & mem.page_addr_mask));
if (tohost_upper || (tohost_lower && to_host_wr_cnt > 0)) {
switch (hostvar >> 48) {
case 0:
if (hostvar != 0x1) {
LOG(FATAL) << "tohost value is 0x" << std::hex << hostvar << std::dec << " (" << hostvar
<< "), stopping simulation";
} else {
LOG(INFO) << "tohost value is 0x" << std::hex << hostvar << std::dec << " (" << hostvar
<< "), stopping simulation";
}
this->reg.trap_state=std::numeric_limits<uint32_t>::max();
this->interrupt_sim=hostvar;
break;
//throw(iss::simulation_stopped(hostvar));
case 0x0101: {
char c = static_cast<char>(hostvar & 0xff);
if (c == '\n' || c == 0) {
LOG(INFO) << "tohost send '" << uart_buf.str() << "'";
uart_buf.str("");
} else
uart_buf << c;
to_host_wr_cnt = 0;
} break;
default:
break;
}
} else if (tohost_lower)
to_host_wr_cnt++;
} else if ((traits<BASE>::XLEN == 32 && paddr.val == fromhost + 4) ||
(traits<BASE>::XLEN == 64 && paddr.val == fromhost)) {
uint64_t fhostvar = *reinterpret_cast<uint64_t *>(p.data() + (fromhost & mem.page_addr_mask));
*reinterpret_cast<uint64_t *>(p.data() + (tohost & mem.page_addr_mask)) = fhostvar;
}
}
}
}
return iss::Ok;
}
template <typename BASE> inline void riscv_hart_msu_vp<BASE>::reset(uint64_t address) {
BASE::reset(address);
state.mstatus = hart_state_type::mstatus_reset_val;
update_vm_info();
}
template <typename BASE> inline void riscv_hart_msu_vp<BASE>::update_vm_info() {
vm[1] = hart_state_type::decode_vm_info(this->reg.PRIV, state.satp);
BASE::addr_mode[3]=BASE::addr_mode[2] = vm[1].is_active()? iss::address_type::VIRTUAL : iss::address_type::PHYSICAL;
if (state.mstatus.MPRV)
vm[0] = hart_state_type::decode_vm_info(state.mstatus.MPP, state.satp);
else
vm[0] = vm[1];
BASE::addr_mode[1] = BASE::addr_mode[0]=vm[0].is_active() ? iss::address_type::VIRTUAL : iss::address_type::PHYSICAL;
ptw.clear();
}
template <typename BASE> void riscv_hart_msu_vp<BASE>::check_interrupt() {
auto status = state.mstatus;
auto ip = csr[mip];
auto ie = csr[mie];
auto ideleg = csr[mideleg];
// Multiple simultaneous interrupts and traps at the same privilege level are
// handled in the following decreasing priority order:
// external interrupts, software interrupts, timer interrupts, then finally
// any synchronous traps.
auto ena_irq = ip & ie;
bool mie = state.mstatus.MIE;
auto m_enabled = this->reg.PRIV < PRIV_M || (this->reg.PRIV == PRIV_M && mie);
auto enabled_interrupts = m_enabled ? ena_irq & ~ideleg : 0;
if (enabled_interrupts == 0) {
auto sie = state.mstatus.SIE;
auto s_enabled = this->reg.PRIV < PRIV_S || (this->reg.PRIV == PRIV_S && sie);
enabled_interrupts = s_enabled ? ena_irq & ideleg : 0;
}
if (enabled_interrupts != 0) {
int res = 0;
while ((enabled_interrupts & 1) == 0) enabled_interrupts >>= 1, res++;
this->reg.pending_trap = res << 16 | 1; // 0x80 << 24 | (cause << 16) | trap_id
}
}
template <typename BASE>
typename riscv_hart_msu_vp<BASE>::phys_addr_t riscv_hart_msu_vp<BASE>::virt2phys(const iss::addr_t &addr) {
const auto type = addr.access & iss::access_type::FUNC;
auto it = ptw.find(addr.val >> PGSHIFT);
if (it != ptw.end()) {
const reg_t pte = it->second;
const reg_t ad = PTE_A | (type == iss::access_type::WRITE) * PTE_D;
#ifdef RISCV_ENABLE_DIRTY
// set accessed and possibly dirty bits.
*(uint32_t *)ppte |= ad;
return {addr.getAccessType(), addr.space, (pte & (~PGMASK)) | (addr.val & PGMASK)};
#else
// take exception if access or possibly dirty bit is not set.
if ((pte & ad) == ad)
return {addr.access, addr.space, (pte & (~PGMASK)) | (addr.val & PGMASK)};
else
ptw.erase(it); // throw an exception
#endif
} else {
uint32_t mode = type != iss::access_type::FETCH && state.mstatus.MPRV ? // MPRV
state.mstatus.MPP :
this->reg.PRIV;
const vm_info &vm = this->vm[static_cast<uint16_t>(type) / 2];
const bool s_mode = mode == PRIV_S;
const bool sum = state.mstatus.SUM;
const bool mxr = state.mstatus.MXR;
// verify bits xlen-1:va_bits-1 are all equal
const int va_bits = PGSHIFT + vm.levels * vm.idxbits;
const reg_t mask = (reg_t(1) << (traits<BASE>::XLEN > -(va_bits - 1))) - 1;
const reg_t masked_msbs = (addr.val >> (va_bits - 1)) & mask;
const int levels = (masked_msbs != 0 && masked_msbs != mask) ? 0 : vm.levels;
reg_t base = vm.ptbase;
for (int i = levels - 1; i >= 0; i--) {
const int ptshift = i * vm.idxbits;
const reg_t idx = (addr.val >> (PGSHIFT + ptshift)) & ((1 << vm.idxbits) - 1);
// check that physical address of PTE is legal
reg_t pte = 0;
const uint8_t res = this->read(iss::address_type::PHYSICAL, addr.access,
traits<BASE>::MEM, base + idx * vm.ptesize, vm.ptesize, (uint8_t *)&pte);
if (res != 0) throw trap_load_access_fault(addr.val);
const reg_t ppn = pte >> PTE_PPN_SHIFT;
if (PTE_TABLE(pte)) { // next level of page table
base = ppn << PGSHIFT;
} else if ((pte & PTE_U) ? s_mode && (type == iss::access_type::FETCH || !sum) : !s_mode) {
break;
} else if (!(pte & PTE_V) || (!(pte & PTE_R) && (pte & PTE_W))) {
break;
} else if (type == iss::access_type::FETCH
? !(pte & PTE_X)
: type == iss::access_type::READ ? !(pte & PTE_R) && !(mxr && (pte & PTE_X))
: !((pte & PTE_R) && (pte & PTE_W))) {
break;
} else if ((ppn & ((reg_t(1) << ptshift) - 1)) != 0) {
break;
} else {
const reg_t ad = PTE_A | ((type == iss::access_type::WRITE) * PTE_D);
#ifdef RISCV_ENABLE_DIRTY
// set accessed and possibly dirty bits.
*(uint32_t *)ppte |= ad;
#else
// take exception if access or possibly dirty bit is not set.
if ((pte & ad) != ad) break;
#endif
// for superpage mappings, make a fake leaf PTE for the TLB's benefit.
const reg_t vpn = addr.val >> PGSHIFT;
const reg_t value = (ppn | (vpn & ((reg_t(1) << ptshift) - 1))) << PGSHIFT;
const reg_t offset = addr.val & PGMASK;
ptw[vpn] = value | (pte & 0xff);
return {addr.access, addr.space, value | offset};
}
}
}
switch (type) {
case access_type::FETCH:
this->fault_data = addr.val;
throw trap_instruction_page_fault(addr.val);
case access_type::READ:
this->fault_data = addr.val;
throw trap_load_page_fault(addr.val);
case access_type::WRITE:
this->fault_data = addr.val;
throw trap_store_page_fault(addr.val);
default:
abort();
}
}
template <typename BASE> uint64_t riscv_hart_msu_vp<BASE>::enter_trap(uint64_t flags, uint64_t addr) {
auto cur_priv = this->reg.PRIV;
// flags are ACTIVE[31:31], CAUSE[30:16], TRAPID[15:0]
// calculate and write mcause val
auto trap_id = bit_sub<0, 16>(flags);
auto cause = bit_sub<16, 15>(flags);
if (trap_id == 0 && cause == 11) cause = 0x8 + cur_priv; // adjust environment call cause
// calculate effective privilege level
auto new_priv = PRIV_M;
if (trap_id == 0) { // exception
if (cur_priv != PRIV_M && ((csr[medeleg] >> cause) & 0x1) != 0)
new_priv = (csr[sedeleg] >> cause) & 0x1 ? PRIV_U : PRIV_S;
// store ret addr in xepc register
csr[uepc | (new_priv << 8)] = static_cast<reg_t>(addr); // store actual address instruction of exception
/*
* write mtval if new_priv=M_MODE, spec says:
* When a hardware breakpoint is triggered, or an instruction-fetch, load,
* or store address-misaligned,
* access, or page-fault exception occurs, mtval is written with the
* faulting effective address.
*/
csr[utval | (new_priv << 8)] = fault_data;
fault_data = 0;
} else {
if (cur_priv != PRIV_M && ((csr[mideleg] >> cause) & 0x1) != 0)
new_priv = (csr[sideleg] >> cause) & 0x1 ? PRIV_U : PRIV_S;
csr[uepc | (new_priv << 8)] = this->reg.NEXT_PC; // store next address if interrupt
this->reg.pending_trap = 0;
}
size_t adr = ucause | (new_priv << 8);
csr[adr] = (trap_id << 31) + cause;
// update mstatus
// xPP field of mstatus is written with the active privilege mode at the time
// of the trap; the x PIE field of mstatus
// is written with the value of the active interrupt-enable bit at the time of
// the trap; and the x IE field of mstatus
// is cleared
// store the actual privilege level in yPP and store interrupt enable flags
switch (new_priv) {
case PRIV_M:
state.mstatus.MPP = cur_priv;
state.mstatus.MPIE = state.mstatus.MIE;
state.mstatus.MIE = false;
break;
case PRIV_S:
state.mstatus.SPP = cur_priv;
state.mstatus.SPIE = state.mstatus.SIE;
state.mstatus.SIE = false;
break;
case PRIV_U:
state.mstatus.UPIE = state.mstatus.UIE;
state.mstatus.UIE = false;
break;
default:
break;
}
// get trap vector
auto ivec = csr[utvec | (new_priv << 8)];
// calculate addr// set NEXT_PC to trap addressess to jump to based on MODE
// bits in mtvec
this->reg.NEXT_PC = ivec & ~0x1UL;
if ((ivec & 0x1) == 1 && trap_id != 0) this->reg.NEXT_PC += 4 * cause;
// reset trap state
this->reg.PRIV = new_priv;
this->reg.trap_state = 0;
std::array<char, 32> buffer;
sprintf(buffer.data(), "0x%016lx", addr);
if((flags&0xffffffff) != 0xffffffff)
CLOG(INFO, disass) << (trap_id ? "Interrupt" : "Trap") << " with cause '"
<< (trap_id ? irq_str[cause] : trap_str[cause]) << "' (" << cause << ")"
<< " at address " << buffer.data() << " occurred, changing privilege level from "
<< lvl[cur_priv] << " to " << lvl[new_priv];
update_vm_info();
return this->reg.NEXT_PC;
}
template <typename BASE> uint64_t riscv_hart_msu_vp<BASE>::leave_trap(uint64_t flags) {
auto cur_priv = this->reg.PRIV;
auto inst_priv = flags & 0x3;
auto status = state.mstatus;
auto tsr = state.mstatus.TSR;
if (cur_priv == PRIV_S && inst_priv == PRIV_S && tsr != 0) {
this->reg.trap_state = (1 << 31) | (2 << 16);
this->fault_data = this->reg.PC;
return this->reg.PC;
}
// pop the relevant lower-privilege interrupt enable and privilege mode stack
// clear respective yIE
switch (inst_priv) {
case PRIV_M:
this->reg.PRIV = state.mstatus.MPP;
state.mstatus.MPP = 0; // clear mpp to U mode
state.mstatus.MIE = state.mstatus.MPIE;
break;
case PRIV_S:
this->reg.PRIV = state.mstatus.SPP;
state.mstatus.SPP = 0; // clear spp to U mode
state.mstatus.SIE = state.mstatus.SPIE;
break;
case PRIV_U:
this->reg.PRIV = 0;
state.mstatus.UIE = state.mstatus.UPIE;
break;
}
// sets the pc to the value stored in the x epc register.
this->reg.NEXT_PC = csr[uepc | inst_priv << 8];
CLOG(INFO, disass) << "Executing xRET , changing privilege level from " << lvl[cur_priv] << " to "
<< lvl[this->reg.PRIV];
update_vm_info();
return this->reg.NEXT_PC;
}
template <typename BASE> void riscv_hart_msu_vp<BASE>::wait_until(uint64_t flags) {
auto status = state.mstatus;
auto tw = status.TW;
if (this->reg.PRIV == PRIV_S && tw != 0) {
this->reg.trap_state = (1 << 31) | (2 << 16);
this->fault_data = this->reg.PC;
}
}
}
}
#endif /* _RISCV_HART_MSU_VP_H */