Re: [PATCH 1/1] acpi: Implement ACPI ERST support for guests

[Paolo Bonzini <pbonzini@redhat.com>]

On 26/10/20 21:19, Eric DeVolder wrote:
> This changeset introduces support for the ACPI Error Record
> Serialization Table, ERST.
> 
> ERST is defined in [1], and the error records are defined
> according to [2].
> 
> This changeset generates the ACPI ERST table, which OSPM
> follows to program the associated ERST device. The ERST
> device occupies 8KiB of address space, with the first 8 bytes
> containing registers, and the remainder being an exchange
> buffer for reading/writing error records.
> 
> The ERST device contains two registers, ACTION and VALUE
> registers, following closely the ERST operations.
> 
> The ERST device only examines the record for the signature 'CPER',
> its record identifier, and bounds-checks the length against the size
> of the the CPER record header; otherwise all other record fields
> and payload are ignored.
> 
> There are two options for this device:
>   -global acpi-erst.size=X
>   -global acpi-erst.filename=Y
> The size X, if not specified, defaults to 64KiB, and must be within
> 64KiB and 1MiB.
> The filename Y, if not specified, defaults to "acpi-erst.backing".
> The ERST backing storage is not mapped into the guest address space,
> just the 8KiB programming area is mapped into the guest.
> 
> This has been utilized as a backend for the Linux pstore feature.
> 
> [1] ACPI 6.3 Specification, 18.3 Error Serialization
>      https://uefi.org/sites/default/files/resources/ACPI_6_3_final_Jan30.pdf
> 
> [2] UEFI 2.8 Specification, Appendix N Common Platform Error Record
>      https://uefi.org/sites/default/files/resources/UEFI_Spec_2_8_final.pdf

In addition to what Igor said, this really needs a test for the device 
(using qtest).  It doesn't have to be super fine-grained, but at least a 
smoke test that the action and value registers do what they mean for a 
sample error recording operation.

Paolo