Files
sdk/runtime/vm/os_linux.cc
T
Ryan Macnak 04ba20aa98 [vm] Support RISC-V.
Implements a backend targeting RV32GC and RV64GC, based on Linux standardizing around GC. The assembler is written to make it easy to disable usage of C, but because the sizes of some instruction sequences are compile-time constants, an additional build configuration would need to be defined to make use of it.

The assembler and disassembler cover every RV32/64GC instruction. The simulator covers all instructions except accessing CSRs and the floating point state accessible through such, include accrued exceptions and dynamic rounding mode.

Quirks:
  - RISC-V is a compare-and-branch architecture, but some existing "architecture-independent" parts of the Dart compiler assume a condition code architecture. To avoid rewriting these parts, we use a peephole in the assembler to map to compare-and-branch. See Assembler::BranchIf. Luckily nothing depended on taking multiple branches on the same condition code set.
  - There are no hardware overflow checks, so we must use Hacker's Delight style software checks. Often these are very cheap: if the sign of one operand is known, a single branch is needed.
  - The ranges of RISC-V branches and jumps are such that we use 3 levels of generation for forward branches, instead of the 2 levels of near and far branches used on ARM[64]. Nearly all code is handled by the first two levels with 20-bits of range, with enormous regex matchers triggering the third level that uses aupic+jalr to get 32-bits of range.
  - For PC-relative calls in AOT, we always generate auipc+jalr pairs with 32-bits of range, so we never generate trampolines.
  - Only a subset of registers are available in some compressed instructions, so we assign the most popular uses to these registers. In particular, THR, TMP[2], CODE and PP. This has the effect of assigning CODE and PP to volatile registers in the C calling convention, whereas they are assigned preserved registers on the other architectures. As on ARM64, PP is untagged; this is so short indices can be accessed with a compressed instruction.
  - There are no push or pop instructions, so combining pushes and pops is preferred so we can update SP once.
  - The C calling convention has a strongly aligned stack, but unlike on ARM64 we don't need to use an alternate stack pointer. The author ensured language was added to the RISC-V psABI making the OS responsible for realigning the stack pointer for signal handlers, allowing Dart to leave the stack pointer misaligned from the C calling convention's point of view until a foreign call.
  - We don't bother with the link register tracking done on ARM[64]. Instead we make use of an alternate link register to avoid inline spilling in the write barrier.

Unimplemented:
 - non-trivial FFI cases
 - Compressed pointers - No intention to implement.
 - Unboxed SIMD - We might make use of the V extension registers when the V extension is ratified.
 - BigInt intrinsics

TEST=existing tests for IL level, new tests for assembler/disassembler/simulator
Bug: https://github.com/dart-lang/sdk/issues/38587
Bug: https://github.com/dart-lang/sdk/issues/48164
Change-Id: I991d1df4be5bf55efec5371b767b332d37dfa3e0
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/217289
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Daco Harkes <dacoharkes@google.com>
Reviewed-by: Slava Egorov <vegorov@google.com>
Commit-Queue: Ryan Macnak <rmacnak@google.com>
2022-01-20 00:57:57 +00:00

668 lines
18 KiB
C++

// Copyright (c) 2012, the Dart project authors. Please see the AUTHORS file
// for details. All rights reserved. Use of this source code is governed by a
// BSD-style license that can be found in the LICENSE file.
#include "vm/globals.h"
#if defined(DART_HOST_OS_LINUX)
#include "vm/os.h"
#include <errno.h> // NOLINT
#include <fcntl.h> // NOLINT
#include <limits.h> // NOLINT
#include <malloc.h> // NOLINT
#include <sys/mman.h> // NOLINT
#include <sys/resource.h> // NOLINT
#include <sys/stat.h> // NOLINT
#include <sys/syscall.h> // NOLINT
#include <sys/time.h> // NOLINT
#include <sys/types.h> // NOLINT
#include <time.h> // NOLINT
#include <unistd.h> // NOLINT
#include "platform/memory_sanitizer.h"
#include "platform/utils.h"
#include "vm/code_comments.h"
#include "vm/code_observers.h"
#include "vm/dart.h"
#include "vm/flags.h"
#include "vm/isolate.h"
#include "vm/lockers.h"
#include "vm/os_thread.h"
#include "vm/zone.h"
namespace dart {
#ifndef PRODUCT
DEFINE_FLAG(bool,
generate_perf_events_symbols,
false,
"Generate events symbols for profiling with perf (disables dual "
"code mapping)");
DEFINE_FLAG(bool,
generate_perf_jitdump,
false,
"Generate jitdump file to use with perf-inject (disables dual code "
"mapping)");
DECLARE_FLAG(bool, write_protect_code);
DECLARE_FLAG(bool, write_protect_vm_isolate);
#if !defined(DART_PRECOMPILED_RUNTIME)
DECLARE_FLAG(bool, code_comments);
#endif
// Linux CodeObservers.
// Simple perf support: generate /tmp/perf-<pid>.map file that maps
// memory ranges to symbol names for JIT generated code. This allows
// perf-report to resolve addresses falling into JIT generated code.
// However perf-annotate does not work in this mode because JIT code
// is transient and does not exist anymore at the moment when you
// invoke perf-report.
class PerfCodeObserver : public CodeObserver {
public:
PerfCodeObserver() : out_file_(NULL) {
Dart_FileOpenCallback file_open = Dart::file_open_callback();
if (file_open == NULL) {
return;
}
intptr_t pid = getpid();
char* filename = OS::SCreate(NULL, "/tmp/perf-%" Pd ".map", pid);
out_file_ = (*file_open)(filename, true);
free(filename);
}
~PerfCodeObserver() {
Dart_FileCloseCallback file_close = Dart::file_close_callback();
if ((file_close == NULL) || (out_file_ == NULL)) {
return;
}
(*file_close)(out_file_);
}
virtual bool IsActive() const {
return FLAG_generate_perf_events_symbols && (out_file_ != NULL);
}
virtual void Notify(const char* name,
uword base,
uword prologue_offset,
uword size,
bool optimized,
const CodeComments* comments) {
Dart_FileWriteCallback file_write = Dart::file_write_callback();
if ((file_write == NULL) || (out_file_ == NULL)) {
return;
}
const char* marker = optimized ? "*" : "";
char* buffer =
OS::SCreate(Thread::Current()->zone(), "%" Px " %" Px " %s%s\n", base,
size, marker, name);
{
MutexLocker ml(CodeObservers::mutex());
(*file_write)(buffer, strlen(buffer), out_file_);
}
}
private:
void* out_file_;
DISALLOW_COPY_AND_ASSIGN(PerfCodeObserver);
};
// Code observer that generates a JITDUMP[1] file that can be interpreted by
// perf-inject to generate ELF images for JIT generated code objects, which
// allows both perf-report and perf-annotate to recognize them.
//
// Usage:
//
// $ perf record -k mono dart --generate-perf-jitdump benchmark.dart
// $ perf inject -j -i perf.data -o perf.data.jitted
// $ perf report -i perf.data.jitted
//
// [1] see linux/tools/perf/Documentation/jitdump-specification.txt for
// JITDUMP binary format.
class JitDumpCodeObserver : public CodeObserver {
public:
JitDumpCodeObserver() : pid_(getpid()) {
char* const filename = OS::SCreate(nullptr, "/tmp/jit-%" Pd ".dump", pid_);
const int fd = open(filename, O_CREAT | O_TRUNC | O_RDWR, 0666);
free(filename);
if (fd == -1) {
return;
}
// Map JITDUMP file, this mapping will be recorded by perf. This allows
// perf-inject to find this file later.
const long page_size = sysconf(_SC_PAGESIZE); // NOLINT(runtime/int)
if (page_size == -1) {
close(fd);
return;
}
mapped_ =
mmap(nullptr, page_size, PROT_READ | PROT_EXEC, MAP_PRIVATE, fd, 0);
if (mapped_ == nullptr) {
close(fd);
return;
}
mapped_size_ = page_size;
out_file_ = fdopen(fd, "w+");
if (out_file_ == nullptr) {
close(fd);
return;
}
// Buffer the output to avoid high IO overheads - we are going to be
// writing all JIT generated code out.
setvbuf(out_file_, nullptr, _IOFBF, 2 * MB);
// Disable code write protection and vm isolate write protection, because
// calling mprotect on the pages filled with JIT generated code objects
// confuses perf.
FLAG_write_protect_code = false;
FLAG_write_protect_vm_isolate = false;
#if !defined(DART_PRECOMPILED_RUNTIME)
// Enable code comments.
FLAG_code_comments = true;
#endif
// Write JITDUMP header.
WriteHeader();
}
~JitDumpCodeObserver() {
if (mapped_ != nullptr) {
munmap(mapped_, mapped_size_);
mapped_ = nullptr;
}
if (out_file_ != nullptr) {
fclose(out_file_);
out_file_ = nullptr;
}
}
virtual bool IsActive() const {
return FLAG_generate_perf_jitdump && (out_file_ != nullptr);
}
virtual void Notify(const char* name,
uword base,
uword prologue_offset,
uword size,
bool optimized,
const CodeComments* comments) {
MutexLocker ml(CodeObservers::mutex());
const char* marker = optimized ? "*" : "";
char* buffer = OS::SCreate(Thread::Current()->zone(), "%s%s", marker, name);
const size_t name_length = strlen(buffer);
WriteDebugInfo(base, comments);
CodeLoadEvent ev;
ev.event = BaseEvent::kLoad;
ev.size = sizeof(ev) + (name_length + 1) + size;
ev.time_stamp = OS::GetCurrentMonotonicTicks();
ev.process_id = getpid();
ev.thread_id = syscall(SYS_gettid);
ev.vma = base;
ev.code_address = base;
ev.code_size = size;
ev.code_id = code_id_++;
WriteFully(&ev, sizeof(ev));
WriteFully(buffer, name_length + 1);
WriteFully(reinterpret_cast<void*>(base), size);
}
private:
struct Header {
const uint32_t magic = 0x4A695444;
const uint32_t version = 1;
const uint32_t size = sizeof(Header);
uint32_t elf_mach_target;
const uint32_t reserved = 0xDEADBEEF;
uint32_t process_id;
uint64_t time_stamp;
const uint64_t flags = 0;
};
struct BaseEvent {
enum Event {
kLoad = 0,
kMove = 1,
kDebugInfo = 2,
kClose = 3,
kUnwindingInfo = 4
};
uint32_t event;
uint32_t size;
uint64_t time_stamp;
};
struct CodeLoadEvent : BaseEvent {
uint32_t process_id;
uint32_t thread_id;
uint64_t vma;
uint64_t code_address;
uint64_t code_size;
uint64_t code_id;
};
struct DebugInfoEvent : BaseEvent {
uint64_t address;
uint64_t entry_count;
// DebugInfoEntry entries[entry_count_];
};
struct DebugInfoEntry {
uint64_t address;
int32_t line_number;
int32_t column;
// Followed by nul-terminated name.
};
// ELF machine architectures
// From linux/include/uapi/linux/elf-em.h
static const uint32_t EM_386 = 3;
static const uint32_t EM_X86_64 = 62;
static const uint32_t EM_ARM = 40;
static const uint32_t EM_AARCH64 = 183;
static const uint32_t EM_RISCV = 243;
static uint32_t GetElfMachineArchitecture() {
#if TARGET_ARCH_IA32
return EM_386;
#elif TARGET_ARCH_X64
return EM_X86_64;
#elif TARGET_ARCH_ARM
return EM_ARM;
#elif TARGET_ARCH_ARM64
return EM_AARCH64;
#elif TARGET_ARCH_RISCV32 || TARGET_ARCH_RISCV64
return EM_RISCV;
#else
UNREACHABLE();
return 0;
#endif
}
#if ARCH_IS_64_BIT
static const int kElfHeaderSize = 0x40;
#else
static const int kElfHeaderSize = 0x34;
#endif
void WriteDebugInfo(uword base, const CodeComments* comments) {
if (comments == nullptr || comments->Length() == 0) {
return;
}
// Open the comments file for the given code object.
// Note: for some reason we can't emit all comments into a single file
// the mapping between PCs and lines goes out of sync (might be
// perf-annotate bug).
char* comments_file_name =
OS::SCreate(nullptr, "/tmp/jit-%" Pd "-%" Pd ".cmts", pid_, code_id_);
const intptr_t filename_length = strlen(comments_file_name);
FILE* comments_file = fopen(comments_file_name, "w");
setvbuf(comments_file, nullptr, _IOFBF, 2 * MB);
// Count the number of DebugInfoEntry we are going to emit: one
// per PC.
intptr_t entry_count = 0;
for (uint64_t i = 0, len = comments->Length(); i < len;) {
const intptr_t pc_offset = comments->PCOffsetAt(i);
while (i < len && comments->PCOffsetAt(i) == pc_offset) {
i++;
}
entry_count++;
}
DebugInfoEvent info;
info.event = BaseEvent::kDebugInfo;
info.time_stamp = OS::GetCurrentMonotonicTicks();
info.address = base;
info.entry_count = entry_count;
info.size = sizeof(info) +
entry_count * (sizeof(DebugInfoEntry) + filename_length + 1);
const int32_t padding = Utils::RoundUp(info.size, 8) - info.size;
info.size += padding;
// Write out DebugInfoEvent record followed by entry_count DebugInfoEntry
// records.
WriteFully(&info, sizeof(info));
intptr_t line_number = 0; // Line number within comments_file.
for (intptr_t i = 0, len = comments->Length(); i < len;) {
const intptr_t pc_offset = comments->PCOffsetAt(i);
while (i < len && comments->PCOffsetAt(i) == pc_offset) {
line_number += WriteLn(comments_file, comments->CommentAt(i));
i++;
}
DebugInfoEntry entry;
entry.address = base + pc_offset + kElfHeaderSize;
entry.line_number = line_number;
entry.column = 0;
WriteFully(&entry, sizeof(entry));
WriteFully(comments_file_name, filename_length + 1);
}
// Write out the padding.
const char padding_bytes[8] = {0};
WriteFully(padding_bytes, padding);
fclose(comments_file);
free(comments_file_name);
}
void WriteHeader() {
Header header;
header.elf_mach_target = GetElfMachineArchitecture();
header.process_id = getpid();
header.time_stamp = OS::GetCurrentTimeMicros();
WriteFully(&header, sizeof(header));
}
// Returns number of new-lines written.
intptr_t WriteLn(FILE* f, const char* comment) {
fputs(comment, f);
fputc('\n', f);
intptr_t line_count = 1;
while ((comment = strstr(comment, "\n")) != nullptr) {
line_count++;
}
return line_count;
}
void WriteFully(const void* buffer, size_t size) {
const char* ptr = static_cast<const char*>(buffer);
while (size > 0) {
const size_t written = fwrite(ptr, 1, size, out_file_);
if (written == 0) {
UNREACHABLE();
break;
}
size -= written;
ptr += written;
}
}
const intptr_t pid_;
FILE* out_file_ = nullptr;
void* mapped_ = nullptr;
long mapped_size_ = 0; // NOLINT(runtime/int)
intptr_t code_id_ = 0;
DISALLOW_COPY_AND_ASSIGN(JitDumpCodeObserver);
};
#endif // !PRODUCT
const char* OS::Name() {
return "linux";
}
intptr_t OS::ProcessId() {
return static_cast<intptr_t>(getpid());
}
static bool LocalTime(int64_t seconds_since_epoch, tm* tm_result) {
time_t seconds = static_cast<time_t>(seconds_since_epoch);
if (seconds != seconds_since_epoch) return false;
struct tm* error_code = localtime_r(&seconds, tm_result);
return error_code != NULL;
}
const char* OS::GetTimeZoneName(int64_t seconds_since_epoch) {
tm decomposed;
bool succeeded = LocalTime(seconds_since_epoch, &decomposed);
// If unsuccessful, return an empty string like V8 does.
return (succeeded && (decomposed.tm_zone != NULL)) ? decomposed.tm_zone : "";
}
int OS::GetTimeZoneOffsetInSeconds(int64_t seconds_since_epoch) {
tm decomposed;
bool succeeded = LocalTime(seconds_since_epoch, &decomposed);
// Even if the offset was 24 hours it would still easily fit into 32 bits.
// If unsuccessful, return zero like V8 does.
return succeeded ? static_cast<int>(decomposed.tm_gmtoff) : 0;
}
int OS::GetLocalTimeZoneAdjustmentInSeconds() {
// TODO(floitsch): avoid excessive calls to tzset?
tzset();
// Even if the offset was 24 hours it would still easily fit into 32 bits.
// Note that Unix and Dart disagree on the sign.
return static_cast<int>(-timezone);
}
int64_t OS::GetCurrentTimeMillis() {
return GetCurrentTimeMicros() / 1000;
}
int64_t OS::GetCurrentTimeMicros() {
// gettimeofday has microsecond resolution.
struct timeval tv;
if (gettimeofday(&tv, NULL) < 0) {
UNREACHABLE();
return 0;
}
return (static_cast<int64_t>(tv.tv_sec) * 1000000) + tv.tv_usec;
}
int64_t OS::GetCurrentMonotonicTicks() {
struct timespec ts;
if (clock_gettime(CLOCK_MONOTONIC, &ts) != 0) {
UNREACHABLE();
return 0;
}
// Convert to nanoseconds.
int64_t result = ts.tv_sec;
result *= kNanosecondsPerSecond;
result += ts.tv_nsec;
return result;
}
int64_t OS::GetCurrentMonotonicFrequency() {
return kNanosecondsPerSecond;
}
int64_t OS::GetCurrentMonotonicMicros() {
int64_t ticks = GetCurrentMonotonicTicks();
ASSERT(GetCurrentMonotonicFrequency() == kNanosecondsPerSecond);
return ticks / kNanosecondsPerMicrosecond;
}
int64_t OS::GetCurrentThreadCPUMicros() {
struct timespec ts;
if (clock_gettime(CLOCK_THREAD_CPUTIME_ID, &ts) != 0) {
UNREACHABLE();
return -1;
}
int64_t result = ts.tv_sec;
result *= kMicrosecondsPerSecond;
result += (ts.tv_nsec / kNanosecondsPerMicrosecond);
return result;
}
int64_t OS::GetCurrentThreadCPUMicrosForTimeline() {
return OS::GetCurrentThreadCPUMicros();
}
// TODO(5411554): May need to hoist these architecture dependent code
// into a architecture specific file e.g: os_ia32_linux.cc
intptr_t OS::ActivationFrameAlignment() {
#if defined(TARGET_ARCH_IA32) || defined(TARGET_ARCH_X64) || \
defined(TARGET_ARCH_ARM64) || defined(TARGET_ARCH_RISCV32) || \
defined(TARGET_ARCH_RISCV64)
const int kMinimumAlignment = 16;
#elif defined(TARGET_ARCH_ARM)
const int kMinimumAlignment = 8;
#else
#error Unsupported architecture.
#endif
intptr_t alignment = kMinimumAlignment;
// TODO(5411554): Allow overriding default stack alignment for
// testing purposes.
// Flags::DebugIsInt("stackalign", &alignment);
ASSERT(Utils::IsPowerOfTwo(alignment));
ASSERT(alignment >= kMinimumAlignment);
return alignment;
}
int OS::NumberOfAvailableProcessors() {
return sysconf(_SC_NPROCESSORS_ONLN);
}
void OS::Sleep(int64_t millis) {
int64_t micros = millis * kMicrosecondsPerMillisecond;
SleepMicros(micros);
}
void OS::SleepMicros(int64_t micros) {
struct timespec req; // requested.
struct timespec rem; // remainder.
int64_t seconds = micros / kMicrosecondsPerSecond;
micros = micros - seconds * kMicrosecondsPerSecond;
int64_t nanos = micros * kNanosecondsPerMicrosecond;
req.tv_sec = seconds;
req.tv_nsec = nanos;
while (true) {
int r = nanosleep(&req, &rem);
if (r == 0) {
break;
}
// We should only ever see an interrupt error.
ASSERT(errno == EINTR);
// Copy remainder into requested and repeat.
req = rem;
}
}
// TODO(regis): Function called only from the simulator.
void OS::DebugBreak() {
__builtin_trap();
}
DART_NOINLINE uintptr_t OS::GetProgramCounter() {
return reinterpret_cast<uintptr_t>(
__builtin_extract_return_addr(__builtin_return_address(0)));
}
void OS::Print(const char* format, ...) {
va_list args;
va_start(args, format);
VFPrint(stdout, format, args);
va_end(args);
}
void OS::VFPrint(FILE* stream, const char* format, va_list args) {
vfprintf(stream, format, args);
fflush(stream);
}
char* OS::SCreate(Zone* zone, const char* format, ...) {
va_list args;
va_start(args, format);
char* buffer = VSCreate(zone, format, args);
va_end(args);
return buffer;
}
char* OS::VSCreate(Zone* zone, const char* format, va_list args) {
// Measure.
va_list measure_args;
va_copy(measure_args, args);
intptr_t len = Utils::VSNPrint(NULL, 0, format, measure_args);
va_end(measure_args);
char* buffer;
if (zone != nullptr) {
buffer = zone->Alloc<char>(len + 1);
} else {
buffer = reinterpret_cast<char*>(malloc(len + 1));
}
ASSERT(buffer != NULL);
// Print.
va_list print_args;
va_copy(print_args, args);
Utils::VSNPrint(buffer, len + 1, format, print_args);
va_end(print_args);
return buffer;
}
bool OS::StringToInt64(const char* str, int64_t* value) {
ASSERT(str != NULL && strlen(str) > 0 && value != NULL);
int32_t base = 10;
char* endptr;
int i = 0;
if (str[0] == '-') {
i = 1;
} else if (str[0] == '+') {
i = 1;
}
if ((str[i] == '0') && (str[i + 1] == 'x' || str[i + 1] == 'X') &&
(str[i + 2] != '\0')) {
base = 16;
}
errno = 0;
if (base == 16) {
// Unsigned 64-bit hexadecimal integer literals are allowed but
// immediately interpreted as signed 64-bit integers.
*value = static_cast<int64_t>(strtoull(str, &endptr, base));
} else {
*value = strtoll(str, &endptr, base);
}
return ((errno == 0) && (endptr != str) && (*endptr == 0));
}
void OS::RegisterCodeObservers() {
#ifndef PRODUCT
if (FLAG_generate_perf_events_symbols) {
CodeObservers::Register(new PerfCodeObserver);
}
if (FLAG_generate_perf_jitdump) {
CodeObservers::Register(new JitDumpCodeObserver);
}
#endif // !PRODUCT
}
void OS::PrintErr(const char* format, ...) {
va_list args;
va_start(args, format);
VFPrint(stderr, format, args);
va_end(args);
}
void OS::Init() {}
void OS::Cleanup() {}
void OS::PrepareToAbort() {}
void OS::Abort() {
PrepareToAbort();
abort();
}
void OS::Exit(int code) {
exit(code);
}
} // namespace dart
#endif // defined(DART_HOST_OS_LINUX)