[linuxdc-cvs] CVS: linux TODO.dc,1.1,1.2 ChangeLog.dc,1.8,1.9

[M. R. B. <mr...@us...>]

Update of /cvsroot/linuxdc/linux
In directory usw-pr-cvs1:/tmp/cvs-serv19846

Modified Files:
	TODO.dc ChangeLog.dc 
Log Message:
Minor cosmetic changes to PCI and BBA

Index: TODO.dc
===================================================================
RCS file: /cvsroot/linuxdc/linux/TODO.dc,v
retrieving revision 1.1
retrieving revision 1.2
diff -C2 -r1.1 -r1.2
*** TODO.dc	2001/03/13 10:28:47	1.1
--- TODO.dc	2001/03/15 05:42:55	1.2
***************
*** 1,4 ****
--- 1,20 ----
  LinuxDC list of things TODO
  
+ 2001-03-14:
+ 
+ - New problem with PCI: Ethernet program loaders such as dcload-ip initalize
+   the GAPS subsystem to a known state, therefore when the kernel is uploaded
+   via the BBA it "technically" needs no further GAPS initialization for
+   the 8139C driver to load sucessfully.  However, the latest snapshot of
+   pci_gaps.c didn't do any init of the PCI bus whatsoever, and attempts
+   to load the kernel via serial cable or a boot CD (where GAPS hasn't been
+   initialized yet) will not load the BBA properly.  The problem is that
+   interrupts aren't being sent from GAPS to the SH4's IRQ 11, so the BBA is
+   able to transmit, but unable to receive.  I've have since added Dan Potter's
+   BBA initialization routine to pci_gaps, thinking that I missed an step
+   somewhere, but interrupts still aren't being detected by the system.
+ 
+ 2001-03-13:
+ 
  - Talk to NIIBE Yutaka <gn...@m1...> about merging our changes into
    the LinuxSH kernel.  This would have the cool side effect of our code going
***************
*** 15,18 ****
--- 31,42 ----
    tested.  But overall, the cause of the screwed-up frames needs to be found
    and fixed.
+ 
+   [FIXED] The RxConfig Ring Buffer size was statically defined to 32K, even
+   though I had added a conditional to make it 16K for the DC (the GAPS PCI
+   interface only has 32K _total_ PCI DMA).  Because the 8139C thought this
+   buffer was 32K it was causing buffer overruns (resulting in broken,
+   mishapen RX frames everywhere) and networking performance was abysmal.
+   After adding the correct value to the rtl8139_rx_config constant, there are
+   no further problems with network performance.
  
  - Hmm, the FB driver is SLOW!!  I should've known, given all the warnings on

Index: ChangeLog.dc
===================================================================
RCS file: /cvsroot/linuxdc/linux/ChangeLog.dc,v
retrieving revision 1.8
retrieving revision 1.9
diff -C2 -r1.8 -r1.9
*** ChangeLog.dc	2001/03/14 21:37:19	1.8
--- ChangeLog.dc	2001/03/15 05:42:55	1.9
***************
*** 1,4 ****
--- 1,14 ----
  2001-03-14    M. R. Brown <mr...@li...>
  
+ 	* drivers/pci/pci.ids: Added the device id for the Dreamcast
+ 	Broadband Adapter.
+ 	* include/linux/pci_ids.h: Added definitions for Sega's vendor id and
+ 	the BBA's device id.
+ 	* drivers/net/8139too.c: Change the name of the BBA to Dreamcast...
+ 	rather than Sega.
+ 
+ 	* TODO.dc: Mention the current PCI bug, and the fix for the 8139C Rx
+ 	ring buffer size bug.
+ 
  	* drivers/net/8139too.c [rtl8139_rx_config]: Added a shift for
  	RX_BUF_LEN_IDX to make sure the 16k buffer was being selected (this