In message: <200404071245040782.0D909C4C_at_192.168.128.16> "Carlos Velasco" <freebsd_at_newipnet.com> writes: : On 06/04/2004 at 23:13 M. Warner Losh wrote: : : >: There's code that fails to properly align requests larger than 4k to : >: the appropriate boundary. Upping things to 64k just papers over that : >: problem, and tends to be somewhat wasteful of resources. : > : >Hmmm, I should make the CIS mapping be on a 64k boundary and try a : >bunch of the 'problem children' cards. That might fix a lot of : >problems... : : That's that my patch does (or pretend): : cbb0: PCI Memory allocated: 88000000 : cbb1: PCI Memory allocated: 88001000 : cis mem map 0xdc538000 (resource: 0x88010000) : pccard0: ccr_res == 88010000-880103ff, base=ff80 Cool! There's some unintended effects of doing this for ALL memory allocations on the CardBus cards. However, it is relatively easy to fix in a slightly different way. Here's the diffs I have in my tree. Do they also solve the problems for you? I've run with a 4k allocation boundary locally, but that only helps slightly. Warner --- /dell/imp/FreeBSD/src/sys/dev/pccard/pccard_cis.c Sun Oct 19 11:32:23 2003 +++ ./pccard_cis.c Wed Apr 7 09:18:42 2004 _at__at_ -126,9 +127,15 _at__at_ /* allocate some memory */ + /* + * Some reports from the field suggest that a 64k memory boundary + * helps card CIS being able to be read. Try it here and see what + * the results actually are. I'm not sure I understand why this + * would make cards work better, but it is easy enough to test. + */ rid = 0; - res = bus_alloc_resource(dev, SYS_RES_MEMORY, &rid, 0, ~0, - PCCARD_CIS_SIZE, RF_ACTIVE); + res = bus_alloc_resource(dev, SYS_RES_MEMORY, &rid, 0, ~0, + PCCARD_CIS_SIZE, RF_ACTIVE | rman_make_alignment_flags(64*1024)); if (res == NULL) { device_printf(dev, "can't alloc memory to read attributes\n"); return -1;Received on Wed Apr 07 2004 - 06:24:40 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:50 UTC