[enjoyerofslop]

What and why

Fixes #1361 for builds made after it.

Symptom: the provisional profile's heap check reports "the heap request did not fit; set mem_reserve" and the build finishes. The first run still stops at SetHeapSize OOM: req 0x18ce00000 > cap 0xb8000000, then CRASH sig=10 (SIGBUS) in nninitStartup, on the heap it never got. Found on The Legend of Zelda: Ocarina of Time (Switch 2, 0400486024DEC000, update v131072, SDK 22.2.8).

Cause: sometimes the block has less than 1 GB left for the alias region beside the requested heap region. In that case heap_signals (dm/games/profile_guess.py) grew mem_reserve to 256 MB + 1 GB + the stack region + the request, as if the alias region were 1 GB, but it set only mem_reserve. The heap region is what follows the profile's alias region and stacks (hle_svc.c mem_init), so any alias region above 1 GB came off the heap region.

• The Switch 2 guess for a game with a system resource is a 4.375 GB alias region (0x118000000).
• Ocarina of Time asks for 0x18ce00000, and the check set mem_reserve 0x1f0000000.
• That leaves the heap region at 0x1f0000000 − 0x10000000 − 0x118000000 − 0x10000000 = 0xb8000000.

Change: the grown block keeps the profile's alias region, at least 1 GB as before:
mem_reserve = 256 MB + max(mem_alias, 1 GB) + mem_stack + the request rounded up to 128 MB.

• A profile whose alias region is 1 GB or less gets the same keys as before.
• The branch that shrinks the alias region when the block has room (Tears of the Kingdom's 0x98000000) is unchanged.
• Ocarina of Time's check now sets 0x2c8000000.

I kept the alias region rather than setting it to 1 GB: Ocarina of Time maps it up to +1.15 GB within two minutes of its title movie.

Reference entry: The heap check grows the block around the profile's alias region. The symptom index is regenerated.

Second commit, CI only: lint-python has been red on main since dd6ce5b, which added three more rule['cuts'][0] = … lines to test_nso_match.py (6 pyright reportIndexIssue findings against the baseline's 3). nso_match.recipe() is now annotated -> dict, as patch_recipe() beside it already is. All 6 findings go and the baseline shrinks by 3 (lint_python.py --update); no behavior changes. It's a separate commit, so it can be dropped if main fixes this first.

Checked

• Tests:
  • python3 -m unittest dm/tools/tests/test_nso_match.py: 27 tests OK after the second commit, and lint_python.py --against main then reports 0 new findings.
  • python3 -m unittest dm/tools/tests/test_profile_guess.py: 31 tests OK. The new HeapSignals covers three cases: the Ocarina layout, a small alias region (old keys) and Tears of the Kingdom's shrink branch. Its Ocarina case fails on the old code with the 0xb8000000 heap region.
  • dm/tools/tests/test_docs.py and dm/tools/tests/test_nsp_incremental.py pass.
  • Not run: make -C dm/host/tests (no host code changed), so test_memory_profile.py's heap probe, which takes the other branch, was not run either.
• Game run: The Legend of Zelda: Ocarina of Time, 0400486024DEC000, update v131072.
  • heap_signals run on the real heap-check log gives mem_reserve 0x2c8000000 and leaves mem_alias at 0x118000000.
  • A host built from this branch with those keys logs [mem] heap 6.20 GB (heap region 6.25 GB) and reaches nnMain.
  • The boot then stops at the NPLN / gRPC nn::diag abort at sdk+0xb8fe8. That is a separate problem, which the eventfd change in [WIP] The Legend of Zelda: Ocarina of Time (Switch 2, 0400486024DEC000, v0, SDK 22.2.8) runs on the Mac through the title movie into Kokiri Forest #1299 addresses; with it applied, the game plays its title movie and keeps running.
  • Not checked: a fresh provisional build running the heap check end to end. The keys come from heap_signals on that run's log.
• Devices and OS: M5 Max (Mac17,6), macOS 27.0.1.
• regress.py: not run. The change only touches what a provisional profile's heap check writes. Hand-made profiles, the host and the shader chain are untouched.

Docs

• The finding is in its docs/reference/memory.md entry, and python3 dm/tools/docs_symptoms.py was run (the symptom index)
• No game files, keys or data extracted from a game in the change or the attachments

🤖 Generated with Claude Code

[bogdanandreifilip]

Thank you very much !

[seedds]

Landed as e9d1ef2 (lint) and eab0aaa (the fix), with your code and diagnosis and a Co-authored-by trailer. Thank you.

Three text changes on top, because #1373 predates e9d71e2, which put the heap and stacks before the alias region:

• The test docstring and the memory.md entry now state the formula instead of the old region order. The formula itself holds on both layouts.
• Since e9d71e2 the Mac's alias region is 64 GB whatever the profile says, so keeping the 4.375 GB alias and cutting it to 1 GB give the same layout on the Mac. The comment and the entry now say that it matters only to the region check and the iPhone, where a block this size is over the limit either way.
• The recovery advice now says to delete the provisional profile. Deleting only the # heap check: line re-solves from the old block and gives a 1 GB alias region.

Checked on main: the real hle_svc.c compiled with the old keys logs SetHeapSize OOM: req 0x18ce00000 > cap 0xb8000000, and with the new ones grants the request ([mem] heap 6.20 GB (heap region 6.25 GB)). The host memory tests pass. No game was booted on the current layout.