[Pedro Henrique Penna (ppenna)]

Follows up #4804 from its review, for microsoft/nvx#568. When the effective CPUID that a restore recomputes differs from the snapshot's record, the E_CPU_SURFACE error now names every difference of the record, also those that only a record that no OpenVMM writes, but that the decoder admits, can have: each entry that the record lists more than once, each of its listings that differs, and a record out of order, together with the entries that differ. The tests of the error now check each register, value, and mask.

Based on feature/microvm/dev. This branch is two commits directly on e4c741e18, the head of feature/microvm/dev since #4804 merged, which microsoft/nvx's dev pins. microsoft/nvx#1039 pins this branch's head.

Why

A review of #4804 found two gaps in the comparison of a restore's effective CPUID with the snapshot's record. recorded_cpuid_differences collected the record's entries in a BTreeMap by leaf and subleaf, which keeps only the last listing of a key, and named a record out of order only when nothing else differed. decode_effective_cpuid checks the record's size and the encoding of its subleaves, but neither repeated keys nor order, so the error did not name every difference of such a record:

• A changed listing of an entry, followed by its own listing, was reported as "the snapshot lists the same entries in another order".
• A record out of order, with a changed entry, named the changed entry alone, and a restore with that entry fixed then reported the order.

The tests of the comparison also checked a flipped bit of EAX alone, and matched the differences of the topology leaves by prefix: no change of a mask alone, no EBX, ECX, or EDX, no entry with several registers, and no order of the entries that only one side has.

Each register, value, and mask (e51ab72d3)

Tests only. recorded_cpuid_names_each_register_in_which_an_entry_differs checks the whole error for a change of the value, and of the mask, of each register of an entry of a subleaf alone; for three registers of one entry, which are named in order on one line; and for an entry that differs among two entries that the snapshot lacks, followed by two entries that only the snapshot has, which are named and counted in that order.

Repeated and unordered entries (bcfc47cff)

recorded_cpuid_differences maps each leaf and subleaf to every listing of the record, in order, and compares each listing with the partition's entry, naming each that differs with the registers in which it differs. After the entries that only one side has, as before, it names each entry that the record lists more than once, with the number of its listings, and then, whatever else differs, a record that does not list its entries in order of leaf and subleaf, as EffectiveCpuid::results, whose encoding the record is, does. For a changed listing of CPUID 0x0 followed by its own:

[E_CPU_SURFACE] the effective CPUID of CPU profile intel.icelake-sp.v1 differs from the snapshot's in 2 entries: CPUID 0x0 EAX is 0x0000001b under mask 0xffffffff, and 0x0000001a under mask 0xffffffff in the snapshot; CPUID 0x0 is in the snapshot 2 times

The error keeps its code, and names the differences of a record that OpenVMM writes, which lists each entry once and in order, as before. Its count counts each item that it names, the order included, as it counted the order before. The list is never empty: the partition's entries are unique and sorted, so a record that decodes and differs from their encoding differs in an entry, in the listings of an entry, or in its order. recorded_cpuid_names_repeated_and_unordered_entries checks the whole error for an entry listed twice, changed before or after its own listing, or unchanged; for entries out of order, alone and with an entry that differs; and for an entry listed again out of order. The Guide's snapshot page says what the error names.

Validation

• On Windows, at each commit, for openvmm_core: cargo clippy --all-targets -p openvmm_core -- -D warnings, cargo doc --no-deps -p openvmm_core, cargo nextest run --profile agent -p openvmm_core (122 and 123 tests passed), and cargo xtask fmt --fix, all clean.
• With recorded_cpuid_differences of e51ab72d3 in place of bcfc47cff's, recorded_cpuid_names_repeated_and_unordered_entries fails on the review's first case: the error names "the snapshot lists the same entries in another order" instead of the changed listing and the repeated entry. The tests of e51ab72d3 pass with either.
• Every commit is signed.

[github-actions]

28 Petri tests failed (28 unstable)