Re: [PATCH RFC 00/35] Add ACPI support for arm64 on Xen

[Julien Grall <julien.grall@linaro.org>]

Hi Parth,

On 04/02/2015 14:01, parth.dixit@linaro.org wrote:
> From: Parth Dixit <parth.dixit@linaro.org>
> This patch series adds ACPI support for arm64.

Thank you for sending the ACPI patch series. I will give a look at it 
before the connect, so we could talk about it next week.

> Xen reads static tables described by ACPI standard (5.1) from uefi
> and configures timer, gic and uart. ACPI tables are modified to
> enable PSCI, HVC and xen specific information (grant table and
> event channel interrupts) and mapped into dom0 address space.
> A skeleton device tree is created to pass memory
> (which cannot be passed using ACPI) and rsdp pointer to DOM0.
> DOM0 parses dynamic ACPI tables and configure other peripherals.
>
> - patches 1-19 add ACPI support for xen/uefi interface
> - patches 20-35 add ACPI support for xen/dom0 interface
> Last patch "xen: arm64: Add ACPI support" enables the support for ACPI
> so that individual patches do not affect working xen code base.
>
> Instructions for reproducing the setup and working code base can be found
> in the following repositories
> wiki : https://wiki.linaro.org/LEG/Engineering/Xen_boot_on_FVP_ACPI_UEFI
> linux : https://git.linaro.org/people/parth.dixit/xen-acpi-support/leg-kernel.git
> xen : https://git.linaro.org/people/parth.dixit/xen-acpi-support/xen.git
> uefi : https://git.linaro.org/people/parth.dixit/xen-acpi-support/linaro-edk2.git
>
> Few workarounds have been made to get it working, these are as follows
> 1. Xen assumes that mmio regions are defined in firmware. It reads
> this information from uefi tables and map's it to dom0 address space.

FWIW, this is part of the UEFI spec. So it's safe to use it ;).

> Right now FVP firmware on which this series has been tested
> does not contains all the mmio regions defined in ACPI supplied by
> firmware, to overcome this hard coding of mmio regions is done.

What's the status of this bug? IIRC, a bug was filled last December.

> 2. Linux is checking if mapped pages belong to ram and when this check fails(incorrectly)
> it maps RAM using ioremap, to overcome this at present this check is
> disabled in linux.

Some people were working on dropping/reworking this check. What's the 
status?

> 3. Device specific interrupt information is defined in DSDT tables in ACPI.
> Parsing this table is out of scope for xen so it uses dom0 to parse DSDT table
> and enables the interrupt dynamically when they are enabled by dom0.

It's a design choice, not a workaround ;).

> 4. Grant table region and event channel interrupts are hard coded right now.
>
> 5. RSDP ponter is exposed to dom0 by using "rsdp" node, this node is not yet defined
> in linux and its name may get changed to "linux,acpi-rsdp" or something else.

Ditto.

BTW, when do you plan to send the Linux series? Did they decided about 
the boot protocol?

Regards,

-- 
Julien Grall