x86_64 SMP bringup #345
Labels
Comments
|
cc @iximeow |
|
switching over to ioapic is probably kind of complicated; might have to rewrite some of our interrupt code, and idk if timer 1 works with ioapic... |
hawkw
added a commit
that referenced
this issue
Oct 14, 2022
This commit starts on #345 by using the `acpi` crate to read information from the ACPI tables related to SMP. It adds an `acpi::bringup_smp` function in `arch::x86_64`, but this function doesn't *currently* actually bring up application processors. Instead, it just reads data from the ACPI tables and logs it as a way to check that we are, in fact, able to find the information we need. Additionally, I added a new `ArchInfo` struct that contains opaque architecture-specific data from the bootloader's boot info. This is passed into the `kernel_start` function, but the cross-platform kernel code doesn't consume it --- instead, it's just provided so it can be passed back into `arch_init` once we've done cross-platform setup stuff. This felt better than sticking the RSDP addr in a global variable or something... This is a first step towards #345. Future commits will actually switch to IOAPIC interrupt handling and then start the application processors.
hawkw
added a commit
that referenced
this issue
Oct 14, 2022
This commit starts on #345 by using the `acpi` crate to read information from the ACPI tables related to SMP. It adds an `acpi::bringup_smp` function in `arch::x86_64`, but this function doesn't *currently* actually bring up application processors. Instead, it just reads data from the ACPI tables and logs it as a way to check that we are, in fact, able to find the information we need. Additionally, I added a new `ArchInfo` struct that contains opaque architecture-specific data from the bootloader's boot info. This is passed into the `kernel_start` function, but the cross-platform kernel code doesn't consume it --- instead, it's just provided so it can be passed back into `arch_init` once we've done cross-platform setup stuff. This felt better than sticking the RSDP addr in a global variable or something... This is a first step towards #345. Future commits will actually switch to IOAPIC interrupt handling and then start the application processors.
hawkw
added a commit
that referenced
this issue
Oct 14, 2022
This commit starts on #345 by using the `acpi` crate to read information from the ACPI tables related to SMP. It adds an `acpi::bringup_smp` function in `arch::x86_64`, but this function doesn't *currently* actually bring up application processors. Instead, it just reads data from the ACPI tables and logs it as a way to check that we are, in fact, able to find the information we need. Additionally, I added a new `ArchInfo` struct that contains opaque architecture-specific data from the bootloader's boot info. This is passed into the `kernel_start` function, but the cross-platform kernel code doesn't consume it --- instead, it's just provided so it can be passed back into `arch_init` once we've done cross-platform setup stuff. This felt better than sticking the RSDP addr in a global variable or something... This is a first step towards #345. Future commits will actually switch to IOAPIC interrupt handling and then start the application processors.
|
APIC interrupt handling in progress here: #351 |
|
APIC interrupt handling (#351) works now!!! I think all we need to do now is |
hawkw
added a commit
that referenced
this issue
Oct 22, 2022
This branch adds support for APIC interrupt handling on x86_64 systems. In order to support Survival MultiPlayer, we need to be able to handle interrupts using the APIC rather than the PICs, because the APIC interrupt model allows multiple CPU cores to handle interrupts. This is part of the work towards SMP support (#345). This is a large change to the `hal-x86_64` crate's `interrupt` module in order to add types representing the `LocalApic` and `IoApic`, and to change the interrupt controller abstraction to switch to handling interrupts using APICs if the system has APICs. If APICs are not detected, we continue to use the PICs, but cannot support SMP. In addition, the kernel is updated to use the new HAL APIs. Several smaller changes to other parts of mycelium have been factored out from this PR to merge separately: * b99fb81 refac(hal-x86_64): move `pit` and `tsc` into `time` (#356) * 9994cd5 feat(hal-x86_64): add PIT driver (#355) * 30431a9 fix(trace): weird indentation on new spans (#354) * 91e97b7 feat(hal-x86_64): add `cpu::Msr` type for accessing MSRs (#353) We'll probably want to make additional changes to the new interrupt code in subsequent PRs, but this is enough to get mycelium to use the APICs, and demonstrate that he `maitake` timer wheel still works using the local APIC timer).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
acpicrate) (825dcf1)The text was updated successfully, but these errors were encountered: