Improve target CPU feature checks (backport #18940) (#18958)

* Improve check for Apple M1 silicon under Rosetta

(cherry picked from commit 59641623d1)

# Conflicts:
#	core/src/validator.rs

* test-validator: start logging asap

(cherry picked from commit ee65ffb3c8)

* test-validator: move feature check earlier in startup

(cherry picked from commit e641f257ef)

* validator: add avx2 runtime check

(cherry picked from commit c435f7b3e3)

* validator: start logging asap

(cherry picked from commit ed8285c096)

* validator: check target CPU features earlier

(cherry picked from commit 8ed0cd0fff)

* validator: remove disused cuda config argument

(cherry picked from commit 71f6d839f9)

Co-authored-by: Trent Nelson <trent@solana.com>
This commit is contained in:
mergify[bot]
2021-07-30 06:47:22 +00:00
committed by GitHub
parent f2dc9dd96e
commit b142ef5f8b
4 changed files with 176 additions and 108 deletions

View File

@ -120,7 +120,6 @@ pub struct ValidatorConfig {
pub max_genesis_archive_unpacked_size: u64,
pub wal_recovery_mode: Option<BlockstoreRecoveryMode>,
pub poh_verify: bool, // Perform PoH verification during blockstore processing at boo
pub cuda: bool,
pub require_tower: bool,
pub tower_path: Option<PathBuf>,
pub debug_keys: Option<Arc<HashSet<Pubkey>>>,
@ -177,7 +176,6 @@ impl Default for ValidatorConfig {
max_genesis_archive_unpacked_size: MAX_GENESIS_ARCHIVE_UNPACKED_SIZE,
wal_recovery_mode: None,
poh_verify: true,
cuda: false,
require_tower: false,
tower_path: None,
debug_keys: None,
@ -302,8 +300,6 @@ impl Validator {
}
}
report_target_features();
for cluster_entrypoint in &cluster_entrypoints {
info!("entrypoint: {:?}", cluster_entrypoint);
}
@ -1419,7 +1415,26 @@ fn wait_for_supermajority(
Ok(true)
}
fn report_target_features() {
fn is_rosetta_emulated() -> bool {
#[cfg(target_os = "macos")]
{
use std::str::FromStr;
std::process::Command::new("sysctl")
.args(&["-in", "sysctl.proc_translated"])
.output()
.map_err(|_| ())
.and_then(|output| String::from_utf8(output.stdout).map_err(|_| ()))
.and_then(|stdout| u8::from_str(stdout.trim()).map_err(|_| ()))
.map(|enabled| enabled == 1)
.unwrap_or(false)
}
#[cfg(not(target_os = "macos"))]
{
false
}
}
pub fn report_target_features() {
warn!(
"CUDA is {}abled",
if solana_perf::perf_libs::api().is_some() {
@ -1429,40 +1444,47 @@ fn report_target_features() {
}
);
// We exclude Mac OS here to be compatible with computers that have Mac M1 chips.
// For these computers, one must install rust/cargo/brew etc. using Rosetta 2,
// which allows them to run software targeted for x86_64 on an aarch64.
// Hence the code below will run on these machines (target_arch="x86_64")
// if we don't exclude with target_os="macos".
//
// It's going to require more more work to get Solana building
// on Mac M1's without Rosetta,
// and when that happens we should remove this
// (the feature flag for code targeting that is target_arch="aarch64")
#[cfg(all(
any(target_arch = "x86", target_arch = "x86_64"),
not(target_os = "macos")
))]
{
if !is_rosetta_emulated() {
unsafe { check_avx() };
unsafe { check_avx2() };
}
}
// Validator binaries built on a machine with AVX support will generate invalid opcodes
// when run on machines without AVX causing a non-obvious process abort. Instead detect
// the mismatch and error cleanly.
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
#[target_feature(enable = "avx")]
unsafe fn check_avx() {
if is_x86_feature_detected!("avx") {
info!("AVX detected");
} else {
error!(
"Your machine does not have AVX support, please rebuild from source on your machine"
"Incompatible CPU detected: missing AVX support. Please build from source on the target"
);
abort();
}
}
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
unsafe fn check_avx() {}
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
#[target_feature(enable = "avx2")]
unsafe fn check_avx2() {
if is_x86_feature_detected!("avx2") {
info!("AVX2 detected");
} else {
error!(
"Incompatible CPU detected: missing AVX2 support. Please build from source on the target"
);
abort();
}
}
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
unsafe fn check_avx2() {}
// Get the activated stake percentage (based on the provided bank) that is visible in gossip
fn get_stake_percent_in_gossip(bank: &Bank, cluster_info: &ClusterInfo, log: bool) -> u64 {
let mut online_stake = 0;

View File

@ -36,7 +36,6 @@ pub fn safe_clone_config(config: &ValidatorConfig) -> ValidatorConfig {
max_genesis_archive_unpacked_size: config.max_genesis_archive_unpacked_size,
wal_recovery_mode: config.wal_recovery_mode.clone(),
poh_verify: config.poh_verify,
cuda: config.cuda,
require_tower: config.require_tower,
tower_path: config.tower_path.clone(),
debug_keys: config.debug_keys.clone(),

View File

@ -282,6 +282,112 @@ fn main() {
)
.get_matches();
let ledger_path = value_t_or_exit!(matches, "ledger_path", PathBuf);
let reset_ledger = matches.is_present("reset");
let output = if matches.is_present("quiet") {
Output::None
} else if matches.is_present("log") {
Output::Log
} else {
Output::Dashboard
};
let mut ledger_fd_lock = FdLock::new(fs::File::open(&ledger_path).unwrap());
let _ledger_lock = ledger_fd_lock.try_lock().unwrap_or_else(|_| {
println!(
"Error: Unable to lock {} directory. Check if another validator is running",
ledger_path.display()
);
exit(1);
});
if reset_ledger {
remove_directory_contents(&ledger_path).unwrap_or_else(|err| {
println!("Error: Unable to remove {}: {}", ledger_path.display(), err);
exit(1);
})
}
solana_runtime::snapshot_utils::remove_tmp_snapshot_archives(&ledger_path);
let validator_log_symlink = ledger_path.join("validator.log");
let logfile = if output != Output::Log {
let validator_log_with_timestamp = format!(
"validator-{}.log",
SystemTime::now()
.duration_since(UNIX_EPOCH)
.unwrap()
.as_millis()
);
let _ = fs::remove_file(&validator_log_symlink);
symlink::symlink_file(&validator_log_with_timestamp, &validator_log_symlink).unwrap();
Some(
ledger_path
.join(validator_log_with_timestamp)
.into_os_string()
.into_string()
.unwrap(),
)
} else {
None
};
let _logger_thread = redirect_stderr_to_file(logfile);
if !ledger_path.exists() {
fs::create_dir(&ledger_path).unwrap_or_else(|err| {
println!(
"Error: Unable to create directory {}: {}",
ledger_path.display(),
err
);
exit(1);
});
}
let mut ledger_fd_lock = FdLock::new(fs::File::open(&ledger_path).unwrap());
let _ledger_lock = ledger_fd_lock.try_lock().unwrap_or_else(|_| {
println!(
"Error: Unable to lock {} directory. Check if another validator is running",
ledger_path.display()
);
exit(1);
});
if reset_ledger {
remove_directory_contents(&ledger_path).unwrap_or_else(|err| {
println!("Error: Unable to remove {}: {}", ledger_path.display(), err);
exit(1);
})
}
solana_runtime::snapshot_utils::remove_tmp_snapshot_archives(&ledger_path);
let validator_log_symlink = ledger_path.join("validator.log");
let logfile = if output != Output::Log {
let validator_log_with_timestamp = format!(
"validator-{}.log",
SystemTime::now()
.duration_since(UNIX_EPOCH)
.unwrap()
.as_millis()
);
let _ = fs::remove_file(&validator_log_symlink);
symlink::symlink_file(&validator_log_with_timestamp, &validator_log_symlink).unwrap();
Some(
ledger_path
.join(validator_log_with_timestamp)
.into_os_string()
.into_string()
.unwrap(),
)
} else {
None
};
let _logger_thread = redirect_stderr_to_file(logfile);
solana_core::validator::report_target_features();
// TODO: Ideally test-validator should *only* allow private addresses.
let socket_addr_space = SocketAddrSpace::new(/*allow_private_addr=*/ true);
let cli_config = if let Some(config_file) = matches.value_of("config_file") {
@ -302,15 +408,6 @@ fn main() {
.unwrap_or_else(|_| (Keypair::new().pubkey(), true))
});
let ledger_path = value_t_or_exit!(matches, "ledger_path", PathBuf);
let reset_ledger = matches.is_present("reset");
let output = if matches.is_present("quiet") {
Output::None
} else if matches.is_present("log") {
Output::Log
} else {
Output::Dashboard
};
let rpc_port = value_t_or_exit!(matches, "rpc_port", u16);
let faucet_port = value_t_or_exit!(matches, "faucet_port", u16);
let slots_per_epoch = value_t!(matches, "slots_per_epoch", Slot).ok();
@ -394,59 +491,6 @@ fn main() {
None
};
if !ledger_path.exists() {
fs::create_dir(&ledger_path).unwrap_or_else(|err| {
println!(
"Error: Unable to create directory {}: {}",
ledger_path.display(),
err
);
exit(1);
});
}
let mut ledger_fd_lock = FdLock::new(fs::File::open(&ledger_path).unwrap());
let _ledger_lock = ledger_fd_lock.try_lock().unwrap_or_else(|_| {
println!(
"Error: Unable to lock {} directory. Check if another validator is running",
ledger_path.display()
);
exit(1);
});
if reset_ledger {
remove_directory_contents(&ledger_path).unwrap_or_else(|err| {
println!("Error: Unable to remove {}: {}", ledger_path.display(), err);
exit(1);
})
}
solana_runtime::snapshot_utils::remove_tmp_snapshot_archives(&ledger_path);
let validator_log_symlink = ledger_path.join("validator.log");
let logfile = if output != Output::Log {
let validator_log_with_timestamp = format!(
"validator-{}.log",
SystemTime::now()
.duration_since(UNIX_EPOCH)
.unwrap()
.as_millis()
);
let _ = fs::remove_file(&validator_log_symlink);
symlink::symlink_file(&validator_log_with_timestamp, &validator_log_symlink).unwrap();
Some(
ledger_path
.join(validator_log_with_timestamp)
.into_os_string()
.into_string()
.unwrap(),
)
} else {
None
};
let _logger_thread = redirect_stderr_to_file(logfile);
let faucet_lamports = sol_to_lamports(value_of(&matches, "faucet_sol").unwrap());
let faucet_keypair_file = ledger_path.join("faucet-keypair.json");
if !faucet_keypair_file.exists() {

View File

@ -2076,6 +2076,33 @@ pub fn main() {
.exit();
}));
let logfile = {
let logfile = matches
.value_of("logfile")
.map(|s| s.into())
.unwrap_or_else(|| format!("solana-validator-{}.log", identity_keypair.pubkey()));
if logfile == "-" {
None
} else {
println!("log file: {}", logfile);
Some(logfile)
}
};
let use_progress_bar = logfile.is_none();
let _logger_thread = redirect_stderr_to_file(logfile);
info!("{} {}", crate_name!(), solana_version::version!());
info!("Starting validator with: {:#?}", std::env::args_os());
let cuda = matches.is_present("cuda");
if cuda {
solana_perf::perf_libs::init_cuda();
enable_recycler_warming();
}
solana_core::validator::report_target_features();
let authorized_voter_keypairs = keypairs_of(&matches, "authorized_voter_keypairs")
.map(|keypairs| keypairs.into_iter().map(Arc::new).collect())
.unwrap_or_else(|| vec![identity_keypair.clone()]);
@ -2197,7 +2224,6 @@ pub fn main() {
require_tower: matches.is_present("require_tower"),
tower_path: value_t!(matches, "tower", PathBuf).ok(),
dev_halt_at_slot: value_t!(matches, "dev_halt_at_slot", Slot).ok(),
cuda: matches.is_present("cuda"),
expected_genesis_hash: matches
.value_of("expected_genesis_hash")
.map(|s| Hash::from_str(s).unwrap()),
@ -2471,25 +2497,6 @@ pub fn main() {
exit(1);
});
let logfile = {
let logfile = matches
.value_of("logfile")
.map(|s| s.into())
.unwrap_or_else(|| format!("solana-validator-{}.log", identity_keypair.pubkey()));
if logfile == "-" {
None
} else {
println!("log file: {}", logfile);
Some(logfile)
}
};
let use_progress_bar = logfile.is_none();
let _logger_thread = redirect_stderr_to_file(logfile);
info!("{} {}", crate_name!(), solana_version::version!());
info!("Starting validator with: {:#?}", std::env::args_os());
let start_progress = Arc::new(RwLock::new(ValidatorStartProgress::default()));
admin_rpc_service::run(
&ledger_path,
@ -2594,10 +2601,6 @@ pub fn main() {
solana_metrics::set_host_id(identity_keypair.pubkey().to_string());
solana_metrics::set_panic_hook("validator");
if validator_config.cuda {
solana_perf::perf_libs::init_cuda();
enable_recycler_warming();
}
solana_ledger::entry::init_poh();
solana_runtime::snapshot_utils::remove_tmp_snapshot_archives(&snapshot_output_dir);