v0.5.0 2003/06/xx KCC <kccheng@soliton.com.tw>

* change pcitg_t from a black box pointer to black box structure.
* fix free_irq(..., NULL), the 2nd arg should be &minor.

v0.4.0 2003/06/09 KCC <kccheng@soliton.com.tw>

* minor bug fix
* improve pcitg_irq_handler() by using "dev_id" as minor number or
  index to boards[].  Now the IRQ latency is ~8usec (it's ~10usec for
  v0.3.0).
* change "struct pcitg_t" to "struct pcitg".  g++ seems doesn't like
  "typedef struct pcitg_t { ... } *pcitg_t";

v0.3.0 2003/05/20 KCC <kccheng@soliton.com.tw>

* update doc
* remove /proc/pcitg/boardXY entry which has no hardware associated.
* bug fix for driver cleanup
* remove unecessary board scan in irq_handler() to improve signal latency
  from ~> 50us to ~> 25us at the test host (RH9).
* implement pcitg_in() API.
* implement pcitg_mbox_in() and pcitg_mbox_out() APIs.

v0.2.0 2003/05/09 KCC <kccheng@soliton.com.tw>

* define PCITG_OK, PCITG_ERROR, PCITG_NULL so both
  Linux and Windows code can have same macro from condition checking

* add both pcitg_set_irqmask() and pci_get_irqmask()
* remove pcitg_in2()/pcitg_out2()
* bug fix for driver's irq_handler 
* define side effect of pcitg_connect(), if pcitg_sig_handler == NULL, the
  IRQ will be disabled.
 
v0.1.0 2003/05/08 KCC <kccheng@soliton.com.tw>

* Modify function prototype of signal handler by adding another
  argument for IRQ vector which is passed by kernel's kill_proc_info()
  call.  Both board id (lower 16 bits) and vector (upper 16 bits)
  are combined together as "info->si_int" and pass to user space's
  signal handler.

* Implement pcitg_slot_name() which is just empty call in v0.0.0.

v0.0.0 2003/05/06 KCC <kccheng@soliton.com.tw>

* PCITG 1st release, tested on RH8.0 & RH9
