README

This code implements an X86 legacy bios.  It is intended to be
compiled using standard gnu tools (eg, gas and gcc).

To build, one should be able to run "make" in the main directory.  The
resulting file "out/bios.bin" contains the processed bios image.


Testing of images:

To test the bios under bochs, one will need to instruct bochs to use
the new bios image.  Use the 'romimage' option - for example:

bochs -q 'floppya: 1_44=myfdimage.img' 'romimage: file=out/bios.bin'

To test under qemu, one will need to create a directory with all the
bios images and then overwrite the main bios image.  For example:

cp /usr/share/qemu/*.bin mybiosdir/
cp out/bios.bin mybiosdir/
cp out/*.aml mybiosdir/

Once this is setup, one can instruct qemu to use the newly created
directory for rom images.  For example:

qemu -L mybiosdir/ -fda myfdimage.img


Overview of files:

The src/ directory contains the bios source code.  Several of the
files are compiled twice - once for 16bit mode and once for 32bit
mode.  (The build system will remove code that is not needed for a
particular mode.)

The tools/ directory contains helper utilities for manipulating and
building the final rom.

The out/ directory is created by the build process - it contains all
temporary and final files.


Build overview:

The 16bit code is compiled via gcc to assembler (file out/ccode.16.s).
The gcc "-fwhole-program" and "-ffunction-sections -fdata-sections"
options are used to optimize the process so that gcc can efficiently
compile and discard unneeded code.  (In the code, one can use the
macros 'VISIBLE16' and 'VISIBLE32FLAT' to instruct a symbol to be
outputted in 16bit and 32bit mode respectively.)

This resulting assembler code is pulled into romlayout.S.  The gas
option ".code16gcc" is used prior to including the gcc generated
assembler - this option enables gcc to generate valid 16 bit code.

The post code (post.c) is entered, via the function handle_post(), in
32bit mode.  The 16bit post vector (in romlayout.S) transitions the
cpu into 32 bit mode before calling the post.c code.

In the last step of compilation, the 32 bit code is merged into the 16
bit code so that one binary file contains both.  Currently, both 16bit
and 32bit code will be located in the memory at 0xe0000-0xfffff.


GCC 16 bit limitations:

Although the 16bit code is compiled with gcc, developers need to be
aware of the environment.  In particular, global variables _must_ be
treated specially.

The code has full access to stack variables and general purpose
registers.  The entry code in romlayout.S will push the original
registers on the stack before calling the C code and then pop them off
(including any required changes) before returning from the interrupt.
Changes to CS, DS, and ES segment registers in C code is also safe.
Changes to other segment registers (SS, FS, GS) need to be restored
manually.

Stack variables (and pointers to stack variables) work as they
normally do in standard C code.

However, variables stored outside the stack need to be accessed via
the GET_VAR and SET_VAR macros (or one of the helper macros described
below).  This is due to the 16bit segment nature of the X86 cpu when
it is in "real mode".  The C entry code will set DS and SS to point to
the stack segment.  Variables not on the stack need to be accessed via
an explicit segment register.  Any other access requires altering one
of the other segment registers (usually ES) and then accessing the
variable via that segment register.

There are three low-level ways to access a remote variable:
GET/SET_VAR, GET/SET_FARVAR, and GET/SET_FLATPTR.  The first set takes
an explicit segment descriptor (eg, "CS") and offset.  The second set
will take a segment id and offset, set ES to the segment id, and then
make the access via the ES segment.  The last method is similar to the
second, except it takes a pointer that would be valid in 32-bit flat
mode instead of a segment/offset pair.

Most BIOS variables are stored in global variables, the "BDA", or
"EBDA" memory areas.  Because this is common, three sets of helper
macros (GET/SET_GLOBAL, GET/SET_BDA, and GET/SET_EBDA) are available
to simplify these accesses.

Global variables defined in the C code can be read in 16bit mode if
the variable declaration is marked with VAR16, VAR16VISIBLE,
VAR16EXPORT, or VAR16FIXED.  The GET_GLOBAL macro will then allow read
access to the variable.  Global variables are stored in the 0xf000
segment.  Because the f-segment is marked read-only during run-time,
the 16bit code is not permitted to change the value of 16bit variables
(use of the SET_GLOBAL macro from 16bit mode will cause a link error).
Code running in 32bit mode can not access variables with VAR16, but
can access variables marked with VAR16VISIBLE, VAR16EXPORT,
VAR16FIXED, or with no marking at all.  The 32bit code can use the
GET/SET_GLOBAL macros, but they are not required.


GCC 16 bit stack limitations:

Another limitation of gcc is its use of 32-bit temporaries.  Gcc will
allocate 32-bits of space for every variable - even if that variable
is only defined as a 'u8' or 'u16'.  If one is not careful, using too
much stack space can break old DOS applications.

There does not appear to be explicit documentation on the minimum
stack space available for bios calls.  However, Freedos has been
observed to call into the bios with less than 150 bytes available.

Note that the post code and boot code (irq 18/19) do not have a stack
limitation because the entry points for these functions transition the
cpu to 32bit mode and reset the stack to a known state.  Only the
general purpose 16-bit service entry points are affected.

There are some ways to reduce stack usage: making sure functions are
tail-recursive often helps, reducing the number of parameters passed
to functions often helps, sometimes reordering variable declarations
helps, inlining of functions can sometimes help, and passing of packed
structures can also help.  It is also possible to transition to/from
an extra stack stored in the EBDA using the stack_hop helper function.

Some useful stats: the overhead for the entry to a bios handler that
takes a 'struct bregs' is 42 bytes of stack space (6 bytes from
interrupt insn, 32 bytes to store registers, and 4 bytes for call
insn).  An entry to an ISR handler without args takes 30 bytes (6 + 20
+ 4).


Debugging the bios:

The bios will output information messages to a special debug port.
Under qemu, one can view these messages by adding '-chardev
stdio,id=seabios -device isa-debugcon,iobase=0x402,chardev=seabios' to
the qemu command line.  Once this is done, one should see status
messages on the console.

The gdb-server mechanism of qemu is also useful.  One can use gdb with
qemu to debug system images.  To use this, add '-s -S' to the qemu
command line.  For example:

qemu -L mybiosdir/ -fda myfdimage.img -s -S

Then, in another session, run gdb with either out/rom16.o (to debug
bios 16bit code) or out/rom32.o (to debug bios 32bit code).  For
example:

gdb out/rom16.o

Once in gdb, use the command "target remote localhost:1234" to have
gdb connect to qemu.  See the qemu documentation for more information
on using gdb and qemu in this mode.  Note that gdb seems to get
breakpoints confused when the cpu is in 16-bit real mode.  This makes
stepping through the program difficult (though 'step instruction'
still works).  Also, one may need to set 16bit break points at both
the cpu address and memory address (eg, break *0x1234 ; break
*0xf1234).