On Wednesday, 11 December 2019 20:16:52 CET Paul Fertser wrote: > Hey Ben, > > On Wed, Dec 11, 2019 at 10:06:26AM -0800, Ben Greear wrote: > > On 12/11/19 6:44 AM, Paul Fertser wrote: > > > According to many bugreports [0][1][2] the default ath10k-ct kernel > ... > > And also if you want to just have the makefile pass a -DBUILD_ATH10K_SMALL > > or something > > like that and #ifdef code in the ath10k-ct driver, then I'd apply that > > patch to ath10k-ct > > so that you don't need the patches. > > I am offering my patch to the OpenWrt maintainers as kind of a > stop-gap measure to get ath10k-ct working for the release (or in any > way they think is appropriate). Another approach they can choose is to > select the upstream ath10k for those devices. Otherwise some > previously supported boards will require manual intervention to get > WiFi working after an upgrade. > > Regarding your fwcfg idea, I am not sure it will work as it seems the > PCI initialisation is happening before fwcfg is parsed and applied. > > Adding a Kconfig option is another possibility. > > But what do you think about an additional module parameter, wouldn't > it be the cleanest solution in the long run? > > BTW, according to the git logs the patches were initially added by > Christian Lamparter, so I hope he can clarify the situation a > bit. Probably there were some performance tests executed back than to > measure the impact. > Heh no. These patches come up in discussions time and time again. And I would rather see them being removed alltogether. What I can tell you is that the Idea of limiting ath10k memory thirst came from Qualcomm itself.
If you look on the ML you can find the old posts like: <https://www.mail-archive.com/lede-dev@lists.infradead.org/msg04738.html> And for reference: Here's a independent bootlog (from pepe2k/Piotr Dymacz no less) with the "Low Memory System" messages for the RT-AC58U: <https://gist.github.com/pepe2k/eba2766f05ccf4e089347c531c49848b> >From what I remember Sven Eckelmann also measured the impact from the patches on the performance and posted his results to the OpenWrt ML (google will find them). I think for this to have any chance of moving forward you'll need to pressure your ODMs and if that doesn't work: Go with a different WIFI chip vendor that supports low memory devices, or add more RAM. >From what I can tell, Qualcomm nowadays gets what they want "for free" and for the past four-five years, they certainly didn't feel pressured to add "low memory" support to ath10k. Cheers, Christian _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
