[sdcc-devel] [PATCH] Support for autovector interrupts

[Johannes S. <js...@co...>]

Hi,

The attached patch adds a #pragma NOIV to sdcc, which
prevents sdcc from generating irq vector table entries
for functions declared as "interrupt".
The patch is agains sdcc CVS from a few days ago.

Here's why we need it:

We are writing a firmware for a DVB-T - USB receiver based
on the Cypress EZ-USB FX2 (DVB == digital video broadcast).
All software and hardware schematics are GPLed and hosted
on linuxtv.org CVS (CVS module dvb-hw).

The EZ-USB FX2 has a 8051 core derived from the Dallas DS80C320,
plus FIFOs and interface hardware for handling USB-2.0
high speed data transfers. To reduce processing overhead
in the USB irq handler for finding out which of the 27 possible
sources caused the USB irq, the EZ-USB FX2 has an autovector
feature which works like this:

- in the irq vector table there is a
    LJMP 0xHHLL
- the hardware replaces the low byte of the jump target address
  with an autovector number (which is a multiple of 4)
- somewhere, beginning on a page boundary, there is a
  secondary irq vector table with LJMPs to the real
  irq handlers; this table is set up manually;
  the high byte of this table's address is put in the
  LJMP in the irq vector table

The individual USB irq handlers still have to be
declared as
  #pragma NOIV
  void foobar_irq(void) interrupt 0;

The #pragma NOIV is modeled after the Keil tool's 8051 C compiler
autovector support from the EZ-USB FX2 developer's kit.
http://www.keil.com/dd/chip/3205.htm

You can get the sources for our project (there isn't much
software yet, just some basics) via CVS from
cvs -d :pserver:ano...@cv...:/cvs/linuxtv co dvb-hw
(login with empty password).

See also http://linuxtv.org/cgi-bin/cvsweb.cgi/.

The EZ-USB FX2 data sheet is available online from
http://www.cypress.com/cfuploads/img/products/38-08012.pdf (~500K)
http://www.cypress.com/cfuploads/support/developer_kits/FX2_TechRefManual.pdf (~5.5MB)


I would be glad if you would accept my patch for sdcc. Or if you have
better ideas about how to support the autovector irqs, please
let me know.


Best regards,
Johannes