Référence en anglais

Cette page technique est disponible en anglais. Les guides d’utilisation et d’installation sont traduits en français.

Shared storage budget (R04)

Both models share a 64 KiB active norcow sector. Application count limits do not allocate separate sectors. The R04 regression test fills 100 discoverable credentials, 32 HOTP accounts and the full 4096-byte serialized largeBlob array, with the FIDO PIN configured.

tests/host/fido2/test_capacity_budget.py uses unmodified python-fido2 through the real feuristic CTAPHID/CTAP implementation. Each credential has credProtect=3, hmac-secret and largeBlobKey. The maximum-fields case uses a 128-byte RP ID, 64-byte user IDs and both 64-byte account names. Each OTP has a 64-byte secret; its maximum metadata is a 32-byte issuer and a 64-byte account. Every HOTP counter is advanced and persisted.

The test writes three different full largeBlob arrays, retaining both A/B copies, then replaces a credential and an OTP record. It checks capacity refusal, preservation of existing data and reloading of the stores.

Measured with the T3B1 norcow live-item model on 18 October 2026:

Input profile	Live application bytes	Free bytes before the 1024-byte system reserve
Typical field lengths	32,768	32,752
Maximum field lengths	62,688	2,832

The model allows 65,520 bytes after the sector header. Protected values carry a 12-byte IV and 16-byte tag; public counters carry neither. T3B1 small items up to 12 stored bytes fit in one 16-byte block; large items need a header block and blocks for the value plus its validity byte. These rules come from the pinned upstream storage/storage.c and storage/norcow_blockwise.h. They conservatively bound T2B1's smaller bitwise format. Boundary examples are tested in tests/host/core/test_storage_budget.py.

tests/upy/test_capacity.py independently exercises maximum fields on the emulator's actual encrypted norcow, including twenty copy-on-write update cycles to force compaction. It passes for T2B1 and T3B1 emulators. This checks the storage implementation rather than only a Python size estimate.

The worst-case scenario leaves little room for other identities. These measurements do not promise that maximum-size OpenPGP data, four maximum PIV certificates and all AES slots also coexist with maximum-size R04 data. Those stores share the remaining space and must refuse additions safely. Hardware flash interruption and acceptance tests remain unexecuted.