Add support for PAE (#97)

[pwmarcz]

Physical memory is still limited to 32-bit addresses, but systems that enable PAE should work now.

Ubuntu 12

I managed to boot the Ubuntu 12 ISO. However, it takes a lot of time to start and is too slow to do anything...

(During the boot, it complains about soft lockup - CPU #0 stuck for 23s!. Apparently it happens during the inintrd decompression unlzma, because it's pretty slow. However, if you just wait, it should go through).

Serenity OS (#299)

PAE also should unlock Serenity OS. I tried it, but there are still some issues. After applying some workarounds, I managed to boot it and it looks like it initializes network, mouse, and starts various processes. However, graphics fail to display. I'll post an update later.

[copy]

This is where the 64-bit address is read from, is that correct?

Could you change it to something like:

        let page_table_entry = read_aligned64(page_table_addr);
        dbg_assert!(page_table_entry & 0xFFFF_FFFF_0000_0000 == 0, "Unsupported: Physical 64-bit address");
        return (page_table_addr as i32, page_table_entry as i32);

[pwmarcz]

Done.

There's no read_aligned64 and I wasn't sure what the convention for the argument should be. I went with "byte address >> 2", for consistency with read_aligned32 (using "byte address >> 3" sounded like a more confusing option).

The highest bit of PDE and PTE (but not the highest entry, PDPTE) is execute-disable and can be set. I added comments about it.

[copy]

Brilliant work, thanks. Only change I'd suggest is to assert when a 64-bit address is encountered, so we don't silently cut them off in debug mode.

By the way, I got an older version of SerenityOS working (September 2020). Will add to the website once this is merged.

Newer versions seem to fail with an issue similar to #597.

[pwmarcz]

Nice! I haven't done anything with Serenity OS yet, but I started working on PAE because I hoped to run it here.

[copy]

Thanks!

[copy]

@pwmarcz Out of curiosity, what workaround were you referring to?

PAE also should unlock Serenity OS. I tried it, but there are still some issues. After applying some workarounds, I managed to boot it and it looks like it initializes network, mouse, and starts various processes.

[pwmarcz]

@copy Here's what happened:

• I couldn't figure out how to boot from the QEMU image, but the extlinux one worked
• Serenity crashed (kernel panic on serial console) when initializing a pseudo-filesystem and adding information about ACPI; it's a null pointer exception since the ACPI subsystem failed to initialize. I commented out that part:

--- a/Kernel/Firmware/ACPI/Parser.cpp
+++ b/Kernel/Firmware/ACPI/Parser.cpp
@@ -69,6 +69,9 @@ UNMAP_AFTER_INIT ACPISysFSComponent::ACPISysFSComponent(NonnullOwnPtr<KString> t
 
 UNMAP_AFTER_INIT void ACPISysFSDirectory::find_tables_and_register_them_as_components()
 {
+    if (!ACPI::Parser::the())
+        return;
+
     NonnullRefPtrVector<SysFSComponent> components;
     size_t ssdt_count = 0;
     ACPI::Parser::the()->enumerate_static_tables([&](StringView signature, PhysicalAddress p_table, size_t length) {

• Then, apparently it did not like the fact that the emulated machine only has primary IDE channel, and got stuck trying to initialize the secondary IDE channel (IDEChannel: reset failed, busy flag on master stuck). When the secondary IDE device is absent, the port that Serenity looks at (IDE control) is treated by v86 as unmapped and hardcoded to return 0xFF.

  Unfortunately enabling the secondary IDE channel (i.e. adding a cdrom device, or an "empty" device) didn't work, I think extlinux failed to boot.

  The following workaround (in v86) worked for me:

--- a/src/cpu.js
+++ b/src/cpu.js
@@ -847,6 +847,8 @@ CPU.prototype.init = function(settings, device_bus)
         if(settings.cdrom)
         {
             this.devices.cdrom = new IDEDevice(this, settings.cdrom, undefined, true, ide_device_count++, device_bus);
+        } else {
+            this.io.register_read(0x0376, null, () => 0);
         }
 
         this.devices.pit = new PIT(this, device_bus);

• After that, Serenity still did not like v86's hard disk. I noticed that it tries to access it through addresses reported by PCI, and when you disable that and use the "standard" addresses, it works.

--- a/Kernel/Storage/ATA/IDEController.cpp
+++ b/Kernel/Storage/ATA/IDEController.cpp
@@ -133,9 +133,9 @@ UNMAP_AFTER_INIT void IDEController::initialize(bool force_pio)
     if (!is_bus_master_capable())
         force_pio = true;
 
-    auto bar0 = PCI::get_BAR0(pci_address());
+    auto bar0 = 0; //PCI::get_BAR0(pci_address());
     auto primary_base_io = (bar0 == 0x1 || bar0 == 0) ? IOAddress(0x1F0) : IOAddress(bar0 & (~1));
-    auto bar1 = PCI::get_BAR1(pci_address());
+    auto bar1 = 0; //PCI::get_BAR1(pci_address());
     auto primary_control_io = (bar1 == 0x1 || bar1 == 0) ? IOAddress(0x3F6) : IOAddress(bar1 & (~1));
     auto bar2 = PCI::get_BAR2(pci_address());
     auto secondary_base_io = (bar2 == 0x1 || bar2 == 0) ? IOAddress(0x170) : IOAddress(bar2 & (~1));

• At this point, I have an apparently working installation (according to serial logs), except for graphics - nothing appears on screen and the mode doesn't change.

I don't know anything about IDE or PCI, so it's hard for me to say if the above is the fault of v86 or Serenity.