3.9 KiB
Concepts you may want to Google beforehand: freestanding, uint32_t, size_t
Goal: Fix miscellaneous issues with our code
The OSDev wiki has a section which describes some issues with JamesM's tutorial. Since we followed his tutorial for lessons 18-22 (interrupts through malloc), we'll need to make sure we fix any of the issues before moving on.
- Wrong CFLAGS
We add -ffreestanding
when compiling .o
files, which includes kernel_entry.o
and thus
kernel.bin
and os-image.bin
.
Before, we disabled libgcc (not libc) through the use of -nostdlib
and we didn't re-enable
it for linking. Since this is tricky, we'll delete -nostdlib
-nostdinc
was also passed to gcc, but we will need it for step 3, so let's delete it.
- kernel.c
main()
function
Modify kernel/kernel.c
and change main()
to kernel_main()
since gcc recognizes "main" as
a special keyword and we don't want to mess with that.
Change boot/kernel_entry.asm
to point to the new name accordingly.
To fix the i386-elf-ld: warning: cannot find entry symbol _start; defaulting to 0000000000001000
warning message, add a global _start;
and define the _start:
label in boot/kernel_entry.asm
.
- Reinvented datatypes
It looks like it was a bad idea to define non-standard data types like u32
and such, since
C99 introduces standard fixed-width data types like uint32_t
We need to include <stdint.h>
which works even in -ffreestanding
(but requires stdlibs)
and use those data types instead of our own, then delete them on type.h
We also delete the underscores around __asm__
and __volatile__
since they aren't needed.
- Improperly aligned
kmalloc
First, since kmalloc
uses a size parameter, we'll use the correct data type size_t
instead
of u32int_t
. size_t
should be used for all parameters which "count" stuff and cannot be
negative. Include <stddef.h>
.
We will fix our kmalloc
in the future, making it a proper memory manager and aligning data types.
For now, it will always return a new page-aligned memory block.
- Missing functions
We will implement the missing mem*
functions in following lessons
- Interrupt handlers
cli
is redundant, since we already established on the IDT entries if interrupts
are enabled within a handler using the idt_gate_t
flags.
sti
is also redundant, as iret
loads the eflags value from the stack, which contains a
bit telling whether interrupts are on or off.
In other words the interrupt handler automatically restores interrupts whether or not
interrupts were enabled before this interrupt
On cpu/isr.h
, struct registers_t
has a couple issues.
First, the alleged esp
is renamed to useless
.
The value is useless because it has to do with the current stack context, not what was interrupted.
Then, we rename useresp
to esp
We add cld
just before call isr_handler
on cpu/interrupt.asm
as suggested
by the osdev wiki.
There is a final, important issue with cpu/interrupt.asm
. The common stubs create an instance
of struct registers
on the stack and then call the C handler. But that breaks the ABI, since
the stack belongs to the called function and they may change them as they please. It is needed
to pass the struct as a pointer.
To achieve this, edit cpu/isr.h
and cpu/isr.c
and change registers_t r
into registers_t *t
,
then, instead of accessing the fields of the struct via .
, access the fields of the pointer via ->
.
Finally, in cpu/interrupt.asm
, and add a push esp
before calling both isr_handler
and
irq_handler
-- remember to also pop eax
to clear the pointer afterwards.
Both current callbacks, the timer and the keyboard, also need to be changed to use a pointer to
registers_t
.