Re: [PATCH v10 20/24] uapi: Add ARM SCMI Telemetry definitions
Fayssal Benmlih <[email protected]>
| Newsgroups | org.kernel.vger.arm-scmi,org.infradead.lists.linux-arm-kernel,org.kernel.vger.linux-doc,org.kernel.vger.linux-kernel |
|---|---|
| Message-ID | <[email protected]> |
Hi Cristian, The ioctl directions, BATCH_READ encoding, signed status description and batch partial-failure text address several of my V7 comments. A few UAPI contract details remain unclear. The scmi_tlm_batch documentation now says that each ioctl enforces an appropriate upper bound, but neither a common ABI limit nor the exact per-command limits are defined. Userspace still cannot determine which nonzero batch sizes are guaranteed to be accepted. Please define an ABI limit or document the limit for every command, for example whether it is the enumerated number of DEs, groups, SHMTIs or UUIDs. The capacity semantics of list operations also need to be explicit. The zero-item query is now documented, but the UAPI does not say what happens when a nonzero num_items is smaller or larger than the available resource count, or whether num_items reports the required count on -ENOSPC. Please define one consistent capacity/result-count contract for all list operations. The event structure still does not document subscription ownership and lifetime: whether the subscription belongs to the open telemetry fd, whether it survives closing that fd or the eventfd, and who is allowed to cancel a cookie. Likewise, scmi_tlm_shmti_info describes len as the size used with mmap(), but does not define whether the required length is PAGE_ALIGN(offset + len), or the permitted offset, protection and sharing modes. These details become permanent ABI once the interface is merged, so they should be defined in the UAPI rather than left to the current implementation. Thanks, Fayçal