Re: applied: [PATCH docs/manager/qemu-server v5 00/21] Add API and UI for custom CPU models

[Arthur Bied-Charreton <a.bied-charreton@proxmox.com>]

On Mon, May 18, 2026 at 12:27:09PM +0200, Thomas Lamprecht wrote:
> On Fri, 15 May 2026 11:28:17 +0200, Arthur Bied-Charreton wrote:
> > This picks up and extends an old series [0] by Stefan Reiter.
> > 
> > This series adds a full CRUD API and a UI editor for custom CPU models
> > which allows users to manage them in the Datacenter interface rather
> > than editing /etc/pve/virtual-guest/cpu-models.conf manually.
> > 
> > It also improves on the existing VM CPU flags selector by providing a
> > list of nodes supporting each flag to help gauge cluster compatibility.
> > 
> > [...]
> 
> Applied, thanks!
> 
> [qemu-server]:
> 
> [01/10] cpu config: rename CPU models config path variable
>         commit: 6cd77495450e586095ba1bfae09cd061599f894a
> [02/10] cpu flags: move cpu flags-related utilities to their own module
>         commit: 71c42d38600595c7ea89901c0cb22529e74e09e3
> [03/10] cpu flags: compare against JSON::true when querying supported flags
>         commit: 8508fe8d987164cb771fa5df11589ca25563e549
> [04/10] cpu flags: normalize CPU flags to QEMU's format
>         commit: 52a3cfa08c9e45156a882d33d49bda8f481c62dd
> [05/10] cpu flags: add helper querying CPU flags with nodes supporting them
>         commit: aba97f99dddd65d254af5a402b48f0a5444a046c
> [06/10] cpu config: rename custom CPU model config loader
>         commit: ccb2c4787048c2610207cdcd7033f3eb6091a1c0
> [07/10] cpu config: add helpers to lock and write config
>         commit: 4483cfbee4ccee4b68ec1bd781085b4376acfcf3
> [08/10] cpu: register standard option for CPU format
>         commit: fcfe1c139e028f80244f8956881a7e3df75162bc
> [09/10] api: cpu flags: improve flags list returned by endpoint
>         commit: 4154e623527ba6a1e8df8a87f28fdd3b7343920d
> [10/10] custom cpu models: avoid redundant config load
>         commit: 9d8e6f7af6d0e97e19f56fae37d74923467bb86b
> 
> [pve-manager]:
> 
> [01/10] cluster: reorder imports
>         commit: c50ec74c3f10c42d493ee84d761b70a3dc49851a
> [02/10] cluster: makefile: reorder perl sources and align backslashes
>         commit: e2ab0188a97772cc380cb974bc6afdda37c221e8
> [03/10] api: add endpoint querying available CPU flags cluster-wide
>         commit: 6ed37d55cb2a59855eb90350486f7a19c13547e4
> [04/10] api: add CRUD handlers for custom CPU models
>         commit: db32d51ec9f58a8fab5f5c16e33bc2c4ac022f88
> [05/10] ui: cpu model selector: allow filtering out custom models
>         commit: b3a515d25c276db64678d20661a5a395eaa603db
> [06/10] ui: add basic custom CPU model editor
>         commit: e67867c9668b6724e32caa6a78ca83881501328d
> [07/10] ui: cpu flags selector: add CPU flag editor for custom models
>         commit: 5589e4007457a751a16c81c38a93110a1ddedbab
> [08/10] ui: cpu flags selector: allow filtering out flags supported on 0 nodes
>         commit: c5e48bb514eab33a03efbbfac94654c3c67c37f2
> [09/10] ui: cpu flags selector: add search bar for large lists of flags
>         commit: 41edb0b46326777a2b5c74b7e4c258a7f02d0161
> [10/10] ui: group custom CPU with resource mappings
>         commit: 48f151ad6b24c7803087194cd987ef532cb3c56b
> 
> [pve-docs]:
> 
> [1/1] qm: add anchor to "CPU Type" section
>       commit: 2cfbddac74d9227284fb7b25310137a2fc202eae
> 
> A few notes from the fix-ups I made while applying, in case useful for future
> submissions. Two UI patterns stood out as worth naming explicitly.
> 
> When a backend property gets a UI, make the column header, form-field label,
> API description, and docs anchor converge on one name. The custom CPU panel had
> 'Phys-Bits' / 'Physical Address Bits' / 'phys-bits' across column / form / API,
> 'Hidden' where users would say 'Hide Hypervisor', and 'HyperV-Vendor' (column)
> vs 'Hyper-V Vendor' (form label). Each small mismatch costs a few seconds at
> every support ticket or docs-grep.
> 
> Co-locate filter controls with the data they filter, and make their defaults
> context-aware. The flag editor originally had the accel toggle in a row below
> the form and a 'supported by at least one node' checkbox under the grid - three
> filter widgets in three different bands. Folding accel and the new per-node
> multi-select into the grid's top toolbar next to the search field gives one
> scannable filter band right above the data, with the blanket toggle remaining
> below as a quick-disable. That blanket filter was also default-on, which on a
> single-node cluster (where supported-on is empty for all flags) hides every
> flag - defaults like that should be on for end-user editing but off for
> cluster-wide curation where they actively hurt.
> 
> The rest are smaller, mostly one-liners:
> 
> - Tighten CRUD input schemas at the schema layer rather than trusting callers.
>   cputype accepted any string until I enforced pve-configid + maxLength=40,
>   re-verified after stripping the optional 'custom-' prefix, and rejected the
>   reserved name 'delete' (collides with PUT's delete= parameter).
> - Prefer //= over = in property-merge helpers so a caller passing a stricter
>   schema isn't silently overwritten by the default; add_cpu_json_properties was
>   overwriting our stricter cputype format.
> - Sibling endpoints should agree on return shape - the per-node cpu-flags
>   endpoint had 'description' as required while the new cluster-wide one can
>   omit it; mark optional to match.
> - Any non-form widget inside a form panel needs submitValue: false +
>   isFormField: false. The CRUD Create dialog POSTed 'combo-NNNN-inputEl=...'
>   until the toolbar widgets were marked.
> - Selection-bound buttons: Proxmox.button.Button with disabled: true auto-binds
>   to the grid's selModel; the bare Edit button was active with no row selected
>   and on an empty list.
> - 'nested-virt' is a PVE shorthand resolved at VM start - whitelist it in
>   supported-on filters; x86-64-vN are PVE-internal abstract profiles
>   - tag with 'abstract' so the VM picker can hide them.
> - 'Sys.Audit on /nodes' was misused as a 'trusted user' proxy for custom-CPU
>   assignment; replaced with Mapping.{Audit,Use,Modify} on /mapping/cpu/<name>
>   (matching PCI/USB/Dir) for both assignment and the listing endpoint.
> - Perl style: 'keys $hash->%*' over 'keys %{$hash}' (the latter flattened by
>   accident here); compare to JSON::true directly, not the string '1'; optional
>   $errmsg on lock helpers matching the Mapping::* pattern lets callers surface
>   useful contention messages.
> 
> In ProcessorEdit I also dropped a duplicate KVM checkbox and added a
> hint pointing at the new per-node accel filter.

Thanks a lot for the detailed feedback and the fixups, I will keep these
patterns in mind for future submissions.