On 10/9/2025 11:53 AM, Nuno Das Neves wrote: > On 10/6/2025 4:08 PM, Easwar Hariharan wrote: >> Use a better mapping of hypervisor status codes to errno values and >> disambiguate the catch-all -EIO value. While here, remove the duplicate >> INVALID_LP_INDEX and INVALID_REGISTER_VALUES hypervisor status entries. >> > > To be honest, in retrospect the idea of 'translating' the hypercall error > codes is a bit pointless. hv_result_to_errno() allows the hypercall helper > functions to be a bit cleaner, but that's about it. When debugging you > almost always want to know the actual hypercall error code. Translating > it imperfectly is often useless, and at worst creates a red > herring/obfuscates the true source of the error.
I feel like you're thinking from the perspective of Microsoft engineers working with the hypervisor. IMHO, the translation is useful for the rest of the kernel to understand and possibly handle differently the possible errors. EBUSY for example is meant to indicate that the operation is already done, or that the target (device/hypervisor) is busy with something else. Timeouts may be handled by a retry, perhaps with a backoff period. > > With that in mind, updating the errno mappings to be more accurate feels > like unnecessary churn. It might even be better to remove the errno mappings > altogether and just translate HV_STATUS_SUCCESS to 0 and any other error > to -EIO or some other 'signal' error code to make it more obvious that > a *hypercall* error occurred and not some other Linux error. We'd still > want to keep the table in some form because it's also used for the error > strings. IMHO, an arbitrarily chosen 'signal' error code wouldn't tell support or the rest of the community that a hypercall error occurred, without a print or trace closest to the hypercall location. > > The cleanup removing the duplicates in the table is welcome. > > Nuno > <snip>
