x86_64 virtio:get irqnum support for x86_64 virtio dev #97
Conversation
| pub unsafe fn new() -> Self { | ||
| Acpi { | ||
| rsdp: AcpiTables::search_for_rsdp_bios(LocalAcpiHandler).unwrap(), | ||
| aml_context: AmlContext::new(Box::new(LocalAmlHandler), DebugVerbosity::None), |
There was a problem hiding this comment.
Can we eliminate the dependency of the crate aml to avoid alloc in axhal?
There was a problem hiding this comment.
Sorry, we can't for now. Because both acpi and aml crate rely on alloc. Crate acpi is used to parse ACPI table and Crate aml is used to invoke ACPI method. Both are need to finish the job of getting irq routing from PCI device pin to APIC pin.
Although the developers of acpi and aml crate offer another crate rsdp for environment without heap alloctors, it only finds and parses RSDP and cannot parse the data and methods in ACPI table.
Crate aml is hard to eliminate since we need to invoke ACPI method _PRT to get irq routing from PCI device pin to APIC pin.
However, if we accept a legacy way, we may get irq number of PCI device without ACPI completely.
It seems that PCI device irq pin connects to two APIC pins in many motherboards (e.g. in current arceos qemu environment, virtio-net PCI device pinA is connected to APIC pin both 11 and 22) , one (e.g. 11) is for compatibility with PIC, and can be readed from PCI configuration space under IRQ Line field.
Since arceos disables PIC and uses APIC only, I think it is better to use the pin specialized for APIC (which needs to get from ACPI) and neglect the pin for PIC.
By the way, using the PIC one(e.g. 11) has another (small) programming problem.
Current virtio_drivers crate on github, which is responsible for parsing virtio PCI configuration space, doesn't have a method to read IRQ Line field. We may need to add this method if we choose the legacy solution.
Any suggestions are welcome and appreciated.😁
There was a problem hiding this comment.
Can we drop the legacy IRQ support and use MSI instead?
There was a problem hiding this comment.
Yes, we can. It seems to be a much better solution. I will try it later.
Can we drop the legacy IRQ support and use MSI instead?
This PR
Since the virtio framework in the main branch does not currently support interrupts, this PR provides the
get_pci_irq_vectorfunction for fetching x86_64 device interrupt numbers.