saymrwulf/betrusted-curve25519-dalek-source
1
0
Fork 0
mirror of https://github.com/saymrwulf/betrusted-curve25519-dalek-source.git synced 2026-09-08 21:00:38 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

cleanup to refer to local job primitives

Browse source
This commit is contained in:
bunnie 2024-03-11 18:15:19 +08:00
parent ebcc3702d3
commit 4c58a5166f
1 changed files with 310 additions and 366 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

190
curve25519-dalek/src/montgomery.rs
View file

@ -329,7 +329,7 @@ impl ProjectivePoint {
pub fn as_affine(&self) -> MontgomeryPoint { pub fn as_affine(&self) -> MontgomeryPoint {
//TODO: consider making this a seperate feature. Something like "panic_on_sw_eval" which would //TODO: consider making this a seperate feature. Something like "panic_on_sw_eval" which would
//be ameniable to upstreaming //be ameniable to upstreaming
#[cfg(all(not(test),curve25519_dalek_backend = "u32e_backend"))] // due to issue https://github.com/rust-lang/rust/issues/59168, you will have to manually comment this out when running a test on the full system and not just this crate. #[cfg(all(not(test), curve25519_dalek_backend = "u32e_backend"))] // due to issue https://github.com/rust-lang/rust/issues/59168, you will have to manually comment this out when running a test on the full system and not just this crate.
log::warn!("sw as_affine being used - check for build config errors!"); log::warn!("sw as_affine being used - check for build config errors!");
let u = &self.U * &self.W.invert(); let u = &self.U * &self.W.invert();
MontgomeryPoint(u.as_bytes()) MontgomeryPoint(u.as_bytes())
@ -463,25 +463,23 @@ impl ProjectivePoint {
mul %31, %29, %21 mul %31, %29, %21
fin // finish execution fin // finish execution
); );
let mut engine = engine_25519::Engine25519::new();
let mut job = engine_25519::Job { use crate::backend::serial::u32e::*;
id: None, ensure_engine();
ucode: [0; 1024], let mut ucode_hw: &'static mut [u32] = unsafe {
uc_len: mcode.len() as u32, core::slice::from_raw_parts_mut(ENGINE_MEM.unwrap().as_mut_ptr() as *mut u32, 1024)
uc_start: 0, };
window: Some(0), let rf_hw: &mut [u32] = unsafe {
rf: [0; engine_25519::RF_SIZE_IN_U32], core::slice::from_raw_parts_mut(
(ENGINE_MEM.unwrap().as_mut_ptr() as usize + RF_U8_BASE) as *mut u32,
TOTAL_RF_SIZE_IN_U32,
)
}; };
for (&src, dst) in mcode.iter().zip(job.ucode.iter_mut()) {
*dst = src as u32;
}
copy_to_rf(self.U.as_bytes(), 29, &mut job.rf);
copy_to_rf(self.W.as_bytes(), 30, &mut job.rf);
// start the run copy_to_rf(self.U.as_bytes(), 29, rf_hw, 0);
let result_rf = engine.spawn_job(job).expect("couldn't run engine job"); copy_to_rf(self.W.as_bytes(), 30, rf_hw, 0);
MontgomeryPoint(copy_from_rf(31, &result_rf)) MontgomeryPoint(run_job(&mut ucode_hw, &rf_hw, &mcode, 0))
} }
} }
@ -541,27 +539,6 @@ pub(crate) fn differential_add_and_double(
Q.W = t17; // W_{Q'} = U_D * 4 (W_P U_Q - U_P W_Q)^2 Q.W = t17; // W_{Q'} = U_D * 4 (W_P U_Q - U_P W_Q)^2
} }
#[cfg(curve25519_dalek_backend = "u32e_backend")]
fn copy_to_rf(bytes: [u8; 32], register: usize, rf: &mut [u32; engine_25519::RF_SIZE_IN_U32]) {
use core::convert::TryInto;
for (byte, rf_dst) in bytes.chunks_exact(4).zip(rf[register * 8..(register+1)*8].iter_mut()) {
*rf_dst = u32::from_le_bytes(byte.try_into().expect("chunks_exact failed us"));
}
}
#[cfg(curve25519_dalek_backend = "u32e_backend")]
fn copy_from_rf(register: usize, rf: &[u32; engine_25519::RF_SIZE_IN_U32]) -> [u8; 32] {
let mut ret: [u8; 32] = [0; 32];
for (src, dst) in rf[register*8 .. (register+1)*8].iter().zip(ret.chunks_exact_mut(4).into_iter()) {
for (&src_byte, dst_byte) in src.to_le_bytes().iter().zip(dst.iter_mut()) {
*dst_byte = src_byte;
}
}
ret
}
#[allow(dead_code)] // absorbed into mul, but might be useful later on as a subroutine for something else #[allow(dead_code)] // absorbed into mul, but might be useful later on as a subroutine for something else
#[cfg(curve25519_dalek_backend = "u32e_backend")] #[cfg(curve25519_dalek_backend = "u32e_backend")]
pub(crate) fn differential_add_and_double( pub(crate) fn differential_add_and_double(
@ -650,39 +627,36 @@ pub(crate) fn differential_add_and_double(
fin // finish execution fin // finish execution
); );
let mut engine = engine_25519::Engine25519::new(); use crate::backend::serial::u32e::*;
ensure_engine();
let mut job = engine_25519::Job { let mut ucode_hw: &'static mut [u32] = unsafe {
id: None, core::slice::from_raw_parts_mut(ENGINE_MEM.unwrap().as_mut_ptr() as *mut u32, 1024)
ucode: [0; 1024], };
uc_len: mcode.len() as u32, let rf_hw: &mut [u32] = unsafe {
uc_start: 0, core::slice::from_raw_parts_mut(
window: Some(0), (ENGINE_MEM.unwrap().as_mut_ptr() as usize + RF_U8_BASE) as *mut u32,
rf: [0; engine_25519::RF_SIZE_IN_U32], TOTAL_RF_SIZE_IN_U32,
)
}; };
for (&src, dst) in mcode.iter().zip(job.ucode.iter_mut()) {
*dst = src;
}
// P.U in %20 // P.U in %20
// P.W in %21 // P.W in %21
// Q.U in %22 // Q.U in %22
// Q.W in %23 // Q.W in %23
// affine_PmQ in %24 // affine_PmQ in %24
copy_to_rf(P.U.as_bytes(), 20, &mut job.rf); copy_to_rf(P.U.as_bytes(), 20, rf_hw, 0);
copy_to_rf(P.W.as_bytes(), 21, &mut job.rf); copy_to_rf(P.W.as_bytes(), 21, rf_hw, 0);
copy_to_rf(Q.U.as_bytes(), 22, &mut job.rf); copy_to_rf(Q.U.as_bytes(), 22, rf_hw, 0);
copy_to_rf(Q.W.as_bytes(), 23, &mut job.rf); copy_to_rf(Q.W.as_bytes(), 23, rf_hw, 0);
copy_to_rf(affine_PmQ.as_bytes(), 24, &mut job.rf); copy_to_rf(affine_PmQ.as_bytes(), 24, rf_hw, 0);
// start the run // start the run
let result_rf = engine.spawn_job(job).expect("couldn't run engine job"); run_job(&mut ucode_hw, &rf_hw, &mcode, 0);
P.U = FieldElement::from_bytes(&copy_from_rf(20, &result_rf)); P.U = FieldElement::from_bytes(&copy_from_rf(20, &rf_hw, 0));
P.W = FieldElement::from_bytes(&copy_from_rf(21, &result_rf)); P.W = FieldElement::from_bytes(&copy_from_rf(21, &rf_hw, 0));
Q.U = FieldElement::from_bytes(&copy_from_rf(22, &result_rf)); Q.U = FieldElement::from_bytes(&copy_from_rf(22, &rf_hw, 0));
Q.W = FieldElement::from_bytes(&copy_from_rf(23, &result_rf)); Q.W = FieldElement::from_bytes(&copy_from_rf(23, &rf_hw, 0));
} }
define_mul_assign_variants!(LHS = MontgomeryPoint, RHS = Scalar); define_mul_assign_variants!(LHS = MontgomeryPoint, RHS = Scalar);
@ -705,10 +679,12 @@ impl Mul<&Scalar> for &MontgomeryPoint {
/// Given `self` \\( = u\_0(P) \\), and a `Scalar` \\(n\\), return \\( u\_0([n]P) \\). /// Given `self` \\( = u\_0(P) \\), and a `Scalar` \\(n\\), return \\( u\_0([n]P) \\).
#[cfg(curve25519_dalek_backend = "u32e_backend")] #[cfg(curve25519_dalek_backend = "u32e_backend")]
fn mul(self, scalar: &Scalar) -> MontgomeryPoint { fn mul(self, scalar: &Scalar) -> MontgomeryPoint {
use crate::backend::serial::u32e::*;
log::debug!("hw mont"); log::debug!("hw mont");
// Algorithm 8 of Costello-Smith 2017 // Algorithm 8 of Costello-Smith 2017
let affine_u = FieldElement::from_bytes(&self.0); let affine_u = FieldElement::from_bytes(&self.0);
let mut x0 = ProjectivePoint::identity(); let x0 = ProjectivePoint::identity();
let x1 = ProjectivePoint { let x1 = ProjectivePoint {
U: affine_u, U: affine_u,
W: FieldElement::ONE, W: FieldElement::ONE,
@ -716,10 +692,6 @@ impl Mul<&Scalar> for &MontgomeryPoint {
// for now, prefer to use the fully-accelerated version where this code is local to the server // for now, prefer to use the fully-accelerated version where this code is local to the server
// instead of transmitting it every call with the data...gives about a 2x performance speedup // instead of transmitting it every call with the data...gives about a 2x performance speedup
if false {
#[cfg(not(test))] // due to issue https://github.com/rust-lang/rust/issues/59168, you will have to manually comment this out when running a test on the full system and not just this crate.
log::warn!("wrong multiply being used!");
let mcode = assemble_engine25519!( let mcode = assemble_engine25519!(
start: start:
// P.U in %20 // P.U in %20
@ -983,72 +955,44 @@ impl Mul<&Scalar> for &MontgomeryPoint {
mul %31, %29, %21 mul %31, %29, %21
fin // finish execution fin // finish execution
); );
let mut engine = engine_25519::Engine25519::new();
let mut job = engine_25519::Job { let window = 0;
id: None, ensure_engine();
ucode: [0; 1024], let mut ucode_hw: &'static mut [u32] = unsafe {
uc_len: mcode.len() as u32, core::slice::from_raw_parts_mut(ENGINE_MEM.unwrap().as_mut_ptr() as *mut u32, 1024)
uc_start: 0, };
window: Some(0), let mut rf_hw: &mut [u32] = unsafe {
rf: [0; engine_25519::RF_SIZE_IN_U32], core::slice::from_raw_parts_mut(
(ENGINE_MEM.unwrap().as_mut_ptr() as usize + RF_U8_BASE) as *mut u32,
TOTAL_RF_SIZE_IN_U32,
)
}; };
for (&src, dst) in mcode.iter().zip(job.ucode.iter_mut()) { copy_to_rf(x0.U.as_bytes(), 25, &mut rf_hw, window);
*dst = src as u32; copy_to_rf(x0.W.as_bytes(), 26, &mut rf_hw, window);
} copy_to_rf(x1.U.as_bytes(), 27, &mut rf_hw, window);
copy_to_rf(x1.W.as_bytes(), 28, &mut rf_hw, window);
copy_to_rf(affine_u.as_bytes(), 24, &mut rf_hw, window);
copy_to_rf(scalar.bytes, 31, &mut rf_hw, window);
copy_to_rf(
[
254, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00,
],
19,
&mut rf_hw,
window,
); // 254 as loop counter
copy_to_rf(x0.U.as_bytes(), 25, &mut job.rf); MontgomeryPoint(run_job(&mut ucode_hw, &rf_hw, &mcode, window))
copy_to_rf(x0.W.as_bytes(), 26, &mut job.rf);
copy_to_rf(x1.U.as_bytes(), 27, &mut job.rf);
copy_to_rf(x1.W.as_bytes(), 28, &mut job.rf);
copy_to_rf(affine_u.as_bytes(), 24, &mut job.rf);
copy_to_rf(scalar.bytes, 31, &mut job.rf);
// load the number 254 into the loop index register
copy_to_rf([
254, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
], 19, &mut job.rf);
// start the run
let result_rf = engine.spawn_job(job).expect("couldn't run engine job");
if false { // unmerged affine path
x0.U = FieldElement::from_bytes(&copy_from_rf(25, &result_rf));
x0.W = FieldElement::from_bytes(&copy_from_rf(26, &result_rf));
//Note: is seems this TODO has been handled as ProjectivePoint's as_affine already
//has an accelerated version. Should this be removed or is there further
//optimization work to be done. If so, what is that work?
// TODO: optimize this relatively innocuous looking call.
// this consumes about 100ms runtime -- need to implement this using
// curve25519 acceleration!
x0.as_affine()
} else {
MontgomeryPoint(copy_from_rf(31, &result_rf))
}
} else {
let mut engine = engine_25519::Engine25519::new();
let job = engine_25519::MontgomeryJob {
x0_u: x0.U.as_bytes(),
x0_w: x0.W.as_bytes(),
x1_u: x1.U.as_bytes(),
x1_w: x1.W.as_bytes(),
affine_u: affine_u.as_bytes(),
scalar: scalar.bytes,
};
MontgomeryPoint(engine.montgomery_job(job).expect("couldn't run montgomery multiply job"))
}
} }
/// Given `self` \\( = u\_0(P) \\), and a `Scalar` \\(n\\), return \\( u\_0(\[n\]P) \\) /// Given `self` \\( = u\_0(P) \\), and a `Scalar` \\(n\\), return \\( u\_0(\[n\]P) \\)
#[cfg(not(curve25519_dalek_backend = "u32e_backend"))] #[cfg(not(curve25519_dalek_backend = "u32e_backend"))]
fn mul(self, scalar: &Scalar) -> MontgomeryPoint { fn mul(self, scalar: &Scalar) -> MontgomeryPoint {
// TODO: consider feature "panic_on_sw_eval" // TODO: consider feature "panic_on_sw_eval"
#[cfg(all(not(test),curve25519_dalek_backend = "u32e_backend"))] // due to issue https://github.com/rust-lang/rust/issues/59168, you will have to manually comment this out when running a test on the full system and not just this crate. #[cfg(all(not(test), curve25519_dalek_backend = "u32e_backend"))] // due to issue https://github.com/rust-lang/rust/issues/59168, you will have to manually comment this out when running a test on the full system and not just this crate.
log::warn!("sw montgomery multiply being used - check for build config errors!"); log::warn!("sw montgomery multiply being used - check for build config errors!");
// We multiply by the integer representation of the given Scalar. By scalar invariant #1, // We multiply by the integer representation of the given Scalar. By scalar invariant #1,
// the MSB is 0, so we can skip it. // the MSB is 0, so we can skip it.