| Filename | Latest commit message | Latest commit date |
|---|---|---|
3b26575b4a
Three things, and the first one is the freeze.
**The loader jumped out from under itself.** A machine with a lot of RAM
stopped at "Starting the kernel" with the splash still up: no panic, no reset,
nothing. The firmware's own exception handler had the answer the whole time,
once a machine was found that printed it:
X64 Exception Type - 0E(#PF - Page-Fault)
ExceptionData - 0x10 I:1 P:0
RIP - 0000000140004DEE <- the loader's own code, at 5.4 GB
RAX - 0000000000001000 <- the page tables it had just installed
RCX - 00000000001000F0 <- the kernel it was about to jump to
I:1 P:0 is an instruction fetch from a page that is not present. The loader
built page tables covering four gigabytes, loaded cr3 -- and the firmware had
placed the loader image *above* four gigabytes, so the next instruction fetch
was unmapped. It reproduces exactly on `-m 8G` and never on `-m 1G`, which is
why every test here passed.
The PDPT is one page whether four entries are filled or all 512, so all 512
are: 512 GB of identity map, replaced by kvmalloc() a few hundred instructions
later.
**And 3 GB was never the limit, it was a circle.** PHYSTOP_MAX was 0xC0000000
because kvmalloc() mapped exactly four 1 GB pages, and kvmalloc mapped four
because that was the cap -- each holding the other up, throwing away thirteen
gigabytes of a sixteen gigabyte machine. kvmalloc now maps as many gigabytes as
the machine has RAM and tells ioremap where the direct map stopped, instead of
ioremap assuming it stopped at four.
Freeing memory becomes two passes, and the ordering is forced rather than
chosen: kinit1() runs *before* kvmalloc because kvmalloc calls kalloc for its
own page tables, and freeing a page means writing into it -- so kinit1 can only
touch what the loader mapped. kinit_high() adds the rest afterwards.
`phystop` and the new `lowtop` are now two different numbers: the DMA arena
stays below 4 GB because a device is not obliged to reach the top of memory,
and an RTL8139's buffer address register is thirty-two bits wide.
**And the 0 MB.** The System Information screen reported "0 MB usable" on a
machine with sixteen gigabytes in it, because memory_mb() had its own 16 KB
guess at how big a memory map gets, and returned zero when GetMemoryMap said
EFI_BUFFER_TOO_SMALL. That is the same bug as the previous commit's, in a
second copy, and this one failed silently. There is one buffer and one
map_read() now.
Verified at 1 GB, 8 GB and 16 GB: all three boot, no traps, no panics, socktest
and memfdtest green in each. 8 GB and 16 GB did not boot at all before -- and
16 GB now reports 17408 MB with "mem: 13312 MB above 4 GB added to the
allocator", where the old kernel would have used three. Both display
configurations still clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
||
| .. | ||
| common | ||
| uefi | ||
| README.md | ||
Booting SchOS
One firmware path:
boot/
uefi/ the loader: a 64-bit PE application
common/ the user interface, shared with nothing else now
There were two. boot/bios/ held an MBR, an El Torito CD stub and a real-mode
stage 2, and it was a separate directory rather than #ifdefs because the two
shared almost nothing: a BIOS loader lives in 16-bit real mode and talks to the
firmware through software interrupts, while UEFI hands a loader long mode with
paging already on and a table of function pointers to call.
That asymmetry is what removed it. Everything since roughly 2012 boots UEFI,
CSM — the firmware-side BIOS emulation that let legacy loaders run at all — has
been going away since Intel dropped support for it in 2020, and the cost of the
second path was not the code but the tax on everything that touched it: stage 2
had a 65-sector ceiling that it silently overflowed, and the resolution menu
could not be shared because VBE mode setting is INT 10h and needs real mode.
The contract below survives it. It was written to let a second loader exist, and it is worth keeping in that shape: it is what the kernel expects, stated once, rather than a description of whoever happens to be calling.
What the loader must hand the kernel
Whichever path runs, the kernel expects the same things. This is the contract; a second loader is correct exactly when it satisfies it.
| Where | Notes | |
|---|---|---|
| Kernel image | loaded per its ELF program headers | entry at entry64 |
| CPU state | see below -- the two paths differ here | |
| Memory map | E820_MAP_ADDR, 24-byte entries, count at E820_COUNT_ADDR |
kernel/arch/x86_64/e820.c reads it |
| Framebuffer | VBE_INFO_ADDR |
address, pitch, width, height, depth |
| Verbose flag | BOOT_VERBOSE_FLAG_ADDR |
set by the F8 menu |
The addresses are defined once, in include/memlayout.h and include/schboot.h.
The kernel does not care which firmware produced them.
The entry point
UEFI hands a loader long mode with paging already on. Dropping back to
protected mode only to climb out again is a mode switch whose failure mode is a
triple fault with no console, so the kernel publishes an entry that takes the
machine as it already is.
kernel/arch/x86_64/entry.S carries a sch64_header — magic SCH64BOO,
8-byte aligned within the first 16 KB, followed by a 64-bit physical address —
found by scanning. include/schboot.h holds the magic so loader and kernel
cannot disagree.
The 32-bit multiboot entry is still in entry.S and now has no caller: the
BIOS loader was the only thing that used it. It is left in place rather than
removed in the same change that removed the loader, because deleting kernel
entry code is its own piece of work with its own way of failing silently.
What that entry expects:
- long mode, interrupts off
- CR3 pointing at tables that identity-map the low 4 GB and map the same
4 GB at
0xffff800000000000.entry64low's first instruction is amovabsto a high address, so identity alone triple faults immediately. - RBX = the framebuffer physical address, and never 1 — that value marks an
application processor in
entry64high.
The kernel loads its own GDT at that entry (entry64uefi), rather than the
loader doing it. A loader arriving in long mode is still on the firmware's
descriptors, and the first thing the kernel does that depends on its own
selectors takes a #GP against a GDT the firmware no longer owns — it
surfaced as unexpected trap 13 ... cs=38, a long way from the cause.
e820.c reads a BIOS memory map. UEFI hands out a different structure
(EFI_MEMORY_DESCRIPTOR, from GetMemoryMap), so the UEFI loader has to
translate it into the same 24-byte entries rather than inventing a second
format for the kernel to learn.
What booting looks like now
make esp stages EFI/BOOT/BOOTX64.EFI and schos/kernel, and QEMU is
pointed at OVMF with that directory as a virtual FAT volume — make qemu.
make iso builds a disc with a single El Torito entry pointing at a FAT image
holding the same two files.
Neither is an installable disk. A real GPT image with a real FAT partition is the installer's job, and until it exists there is nothing here that writes to a disk a machine could boot from unaided.
Secure Boot is not addressed either: the loader is unsigned, so a machine enforcing it will refuse to run this.