Hi Folks,
There are a few requirements that I would like to ensure are documented in various revised documentation. I'm curious to know whether you'd like the current in-kernel documentation to include things at the level of "GICv3 use requires that every processor have a Processor Device in the DSDT". Is that too much detail for the kernel documentation?
Jon.
On Tue, Apr 26, 2016 at 11:38:07PM -0400, Jon Masters wrote:
Hi Folks,
There are a few requirements that I would like to ensure are documented in various revised documentation. I'm curious to know whether you'd like the current in-kernel documentation to include things at the level of "GICv3 use requires that every processor have a Processor Device in the DSDT". Is that too much detail for the kernel documentation?
If this is about catching oversights and mistakes (as seems to be the case for the example), having {boot,run}time checks in the kernel is much more likely to have an impact, especially if there is a helpful diagnostic.
Otherewise, this kind of requirement, if anything, belongs in the ACPI spec. If it's in the ACPI spec, having it in the kernel is redundant. If it's not in the ACPI spec, it will be an uphill struggle to convince people to implement Linux-flavoured ACPI rather than generic, standard ACPI.
Thanks, Mark.
On 04/27/2016 05:25 AM, Mark Rutland wrote:
On Tue, Apr 26, 2016 at 11:38:07PM -0400, Jon Masters wrote:
Hi Folks,
There are a few requirements that I would like to ensure are documented in various revised documentation. I'm curious to know whether you'd like the current in-kernel documentation to include things at the level of "GICv3 use requires that every processor have a Processor Device in the DSDT". Is that too much detail for the kernel documentation?
If this is about catching oversights and mistakes (as seems to be the case for the example), having {boot,run}time checks in the kernel is much more likely to have an impact, especially if there is a helpful diagnostic.
Otherewise, this kind of requirement, if anything, belongs in the ACPI spec. If it's in the ACPI spec, having it in the kernel is redundant. If it's not in the ACPI spec, it will be an uphill struggle to convince people to implement Linux-flavoured ACPI rather than generic, standard ACPI.
Thanks, Mark.
I would agree with Mark that this is more appropriate to the ACPI spec where it can be fairly unambiguously described -- for example, something along the lines of "for every processor ID used in an MADT subtable, there must be a corresponding ACPI Device object with _HID ACPI0007...".
That being said, why should this be required? MADT subtables may have the info needed.
-
Al Stone -
Jon Masters -
Mark Rutland