Date: Sun, 31 Mar 2013 18:24:28 -0400 From: "Adrian Chadd" <adrian.chadd@gmail.com> To: "Joshua Isom" <jrisom@gmail.com>, "Adrian Chadd" <adrian@freebsd.org> Cc: "freebsd-wireless@freebsd.org" <freebsd-wireless@freebsd.org> Subject: Re: ath not working after a motherboard and ram upgrade Message-ID: <5158b7db.897aec0a.42d0.ffffaac4@mx.google.com> In-Reply-To: <515892C4.1060002@gmail.com>
index | next in thread | previous in thread | raw e-mail
The problem isn't contigmalloc, it's making sure that it gets bounced via the local 32 bits of address space right. I'll talk with other developers and see what the deal is with 64 bit address space for 32 bit nics. Thanks, Adrian Sent from my Palm Pre on AT&T On Mar 31, 2013 3:47 PM, Joshua Isom <jrisom@gmail.com> wrote: From if_ath.c:2995: "For some situations (eg EDMA TX completion), there isn't a requirement for the ath_buf entries to be allocated." The EDMA code uses malloc, but would contigmalloc work instead? On 3/31/2013 1:38 PM, Adrian Chadd wrote: > On 31 March 2013 10:49, Joshua Isom <jrisom@gmail.com> wrote: >> I seems someone working on the kernel's already figured out how to deal with >> this. The hw.physmem sysctl can be set at boot time, with 2Gb it seems to >> be working fine. I reinstalled all the sticks, and it's working fine with >> the sysctl set to 2Gb. > > Ooooooo Yay! I mean, damn, that's busted. But yay, now we can narrow > down what the problem is! > > Try bumping it to 4GB. It should be fine at that. > > If it is, please create a PR with all the above information and I'll > start re-reviewing the DMA / buffer code to see what's going on. > > Thanks! > > > > > Adrian >help
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5158b7db.897aec0a.42d0.ffffaac4>