add timing dither to ticktimer

add a little bit of timing dither to one of the core delay routines
this hardens the OS against glitch attacks by making the timing of
concurrent tasks less predictable
This commit is contained in:
bunnie 2025-12-14 16:05:12 +08:00
parent 6635770530
commit 074b0429d3
2 changed files with 62 additions and 1 deletions

View file

@ -27,6 +27,7 @@ xous-semver = "0.1.5"
utralib = { version = "0.1.26", optional = true, default-features = false }
loader = { path = "../../loader", optional = true }
bao1x-api = { path = "../../libs/bao1x-api", optional = true }
rand_chacha = { version = "0.3.1", optional = true }
# TODO: fix this
#[target.'cfg(target_arch = "arm")'.dependencies]
@ -43,7 +44,7 @@ renode = ["utralib/renode", "susres/renode", "watchdog"]
board-dabao = []
board-baosec = []
bao1x = ["utralib/bao1x", "utralib/std", "bao1x-api"]
bao1x = ["utralib/bao1x", "utralib/std", "bao1x-api", "rand_chacha"]
hosted-baosec = []
susres = []

View file

@ -12,6 +12,11 @@ use log::{error, info};
mod platform;
use platform::implementation::*;
use platform::*;
#[cfg(feature = "bao1x")]
use rand_chacha::{
ChaCha8Rng,
rand_core::{RngCore, SeedableRng},
};
#[cfg(not(any(target_arch = "arm", feature = "bao1x")))]
use susres::SuspendOrder;
@ -61,6 +66,26 @@ fn main() -> ! {
susres::Susres::new(Some(SuspendOrder::Last), &xns, api::Opcode::SuspendResume as u32, sr_cid)
.expect("couldn't create suspend/resume object");
#[cfg(feature = "bao1x")]
let xns = xous_names::XousNames::new().unwrap();
#[cfg(feature = "bao1x")]
let wakeup_cid = xns
.request_connection(bao1x_api::SERVER_NAME_BAO1X_HAL)
.expect("Couldn't connect to bao1x HAL server");
#[cfg(feature = "bao1x")]
let mut csprng = {
let mut seed = [0u8; 32];
for chunk in seed.chunks_mut(16) {
let random = xous::syscall::create_server_id().unwrap();
for (dst, src) in chunk.chunks_mut(4).zip(random.to_array()) {
dst.copy_from_slice(&src.to_ne_bytes());
}
}
ChaCha8Rng::from_seed(seed)
};
#[cfg(feature = "bao1x")]
let mut rand_cache = csprng.next_u64();
// A list of all sleep requests in the system, sorted by the time at which it
// expires. That is, if a request comes in to sleep for 1000 ms, and the ticktimer
// is currently at 900, the Request will be `1900`.
@ -133,6 +158,24 @@ fn main() -> ! {
// but this definitely would need to be done for e.g. Memory messages).
core::mem::forget(msg_opt.take());
// add some small timing jitter every time a sleep call is made. This helps
// improve robustness against glitch attacks while running in OS mode.
#[cfg(feature = "bao1x")]
{
// Probably, a short delay is sufficient to randomize execution from glitches
const DELAY_LOG2: u32 = 2;
let i = (rand_cache as u32) & ((1 << DELAY_LOG2) - 1);
for _ in 0..(i * 2) {
bao1x_api::bollard!(4);
}
rand_cache >>= DELAY_LOG2;
// quite intentionally, this can terminate "early": doesn't always take the same
// number of iterations to trigger a re-seed. This contributes a
// bit to execution timing randomization without adding too much overhead.
if rand_cache == 0 {
rand_cache = csprng.next_u64();
}
}
// let timeout_queue = timeout_heap.entry(msg.sender.pid()).or_default();
ticktimer.recalculate_sleep(
&mut sleep_heap,
@ -194,6 +237,23 @@ fn main() -> ! {
ticktimer.suspend();
#[cfg(not(any(target_arch = "arm", feature = "bao1x")))]
susres.suspend_until_resume(_token).expect("couldn't execute suspend/resume");
#[cfg(feature = "bao1x")]
{
log::info!("waiting on good morning");
// Do not use the HAL-Service API to avoid creating a circular dependency on the HAL crate
xous::send_message(
wakeup_cid,
xous::Message::new_blocking_scalar(
bao1x_api::HalOpcode::GoodNight as usize,
0,
0,
0,
0,
),
)
.unwrap();
log::info!("good morning!");
}
ticktimer.resume();
}),