tree 6ebc0f53611df980db7069fa77ed0f6438864f46
parent bfa89261e59df17dc365936fda026bcf85f606a9
author Anthony Rocha <anthony.rocha@amd.com> 1782854868 -0700
committer Anthony Rocha <116300062+rusty1968@users.noreply.github.com> 1784209367 -0700

ast10x0-hace: fix HACE KAT stack overflow (HardFault)

The HACE SHA-2/HMAC KAT test (hace_sha256_test) HardFaulted with an
unaligned UsageFault (wild pointer 0x191818f6) at the sha256 stream-9000
case. Root cause: run_hace_sha2_kats builds a large single stack frame
that overflowed the pigweed-default 2 KiB kernel bootstrap stack into
adjacent .bss (INPUT_BUF); the first case that writes that 4 KiB buffer
(stream-9000) clobbered the live frame and corrupted a saved pointer.
The one-shot cases passed only because they never write INPUT_BUF.

Two parts:
- hmac.rs: move the large HMAC working buffers (key, ipad, opad, and the
  1 KiB msg) off the stack into a single .ram_nc static (HmacNcBufs), so
  HaceHmacCtx is just a few scalars (~20 bytes). This removes the dominant
  per-op stack pressure from the back-to-back HMAC cases. Remaining
  copy_from_slice calls are also converted to zip-copies.
- config.rs: raise KernelConfigInterface::KERNEL_STACK_SIZE_BYTES from the
  2 KiB default to 16 KiB, giving the bootstrap/idle kernel stacks ample
  headroom for the crypto KAT workloads. The AST10x0 has 768 KiB SRAM so
  the extra RAM is negligible.

hace_sha256_test and no_panics_test both pass.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>