From: Denis V. <vd...@il...> - 2005-08-21 10:29:16
|
USB is likely broken - changes are very intrusive on USB side. In particular, acx100_s_create_dma_regions() is doing not fully understood things (at least by me) which seem to be related to setup of internal chip memory layout _and_ PCI-related stuff. Part of it is disabled with "#ifdef PCI" now. It is possible that I disabled too much of it for USB and USB is broken. USB testers _badly_ needed. [20050821] * Switch to wrapper functions for dealing with "tx buffers" (memory areas where we create packets for tx. For USB, it's just a part of wlandevice_t, for PCI it's a DMAable buffer pointed to by txdesc). * Completely hide nature of PCI txdescs under opaque pointer type tx_t* * acx_l_ether_to_txdesc -> acx_l_ether_to_txbuf, not using knowledge of PCI txdesc anymore. * Massive surgery on usb.c, cutting all paths into PCI code. * PCI code moved to pci.c - USB don't need these pieces anymore * acx111 and acx100 txhostdesc creation unified into single function [20050816] * PCI code switched to single-txdesc scheme * version bump to 0.2.5 [20050815] * dev_kfree_skb -> dev_kfree_skb_any [20050814] * Auto rate was reset to lowest rate by scanning code (AP beacons did it every tenth of a second!). Fixed. * USB rx is no longer uses PCI-style rx descriptor ring. tx ring elimination needs 'single-descriptor' setup to be developed for acx100 (patch exists for acx111). * Very strange sem locking problems are reported on amd64. Code which misbehaves looks fine. I do not know what's going on. Workaround: turn off preemption. [20050812] * acx100 was failing to find out radio module #, and wanted to load radio module 00. Must be fixed now. * USB: more simplifications [20050810] * USB: simplified command submission code, removed some wlandevice_t fields (now unused) [20050808] * USB changes: nuked global statics, simplified issue_cmd, shortened wlandevice_t. Added some TODOs :) ... -- vda |