--- Log for 15.09.120 Server: card.freenode.net Channel: #rockbox --- Nick: logbot_ Version: Dancer V4.16 Started: 17 days and 9 hours ago 00.01.04 Quit ac_laptop (Ping timeout: 260 seconds) 00.01.09 Join t0mato [0] (~t0mato@193.32.127.155) 00.46.29 *** Saving seen data "./dancer.seen" 01.33.05 Join atsampson [0] (~ats@cartman.offog.org) 01.41.39 Join t0mato7 [0] (~t0mato@193.32.127.158) 01.43.04 Quit t0mato (Ping timeout: 240 seconds) 01.43.05 Nick t0mato7 is now known as t0mato (~t0mato@193.32.127.158) 01.48.11 Join johnb5 [0] (~johnb2@p5b3afe8e.dip0.t-ipconnect.de) 02.06.23 Quit johnb5 (Ping timeout: 258 seconds) 02.15.00 Join johnb5 [0] (~johnb2@p5b3afe8e.dip0.t-ipconnect.de) 02.28.35 Quit johnb5 (Ping timeout: 240 seconds) 02.31.44 Join petur [0] (~petur@205.153.122.11) 02.31.44 Quit petur (Changing host) 02.31.44 Join petur [0] (~petur@rockbox/developer/petur) 02.46.30 *** Saving seen data "./dancer.seen" 03.12.50 Quit St3ak (Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net) 03.15.19 Join pamaury [0] (~pamaury@rockbox/developer/pamaury) 03.16.15 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 03.50.04 Quit pamaury (Ping timeout: 244 seconds) 04.21.10 Join johnb5 [0] (~johnb2@p5b3afe8e.dip0.t-ipconnect.de) 04.28.07 Join pamaury [0] (~pamaury@maths.r-prg.net.univ-paris7.fr) 04.28.07 Quit pamaury (Changing host) 04.28.07 Join pamaury [0] (~pamaury@rockbox/developer/pamaury) 04.46.33 *** Saving seen data "./dancer.seen" 04.57.44 Quit johnb5 (Ping timeout: 258 seconds) 05.19.03 Join johnb5 [0] (~johnb2@p5b3afe8e.dip0.t-ipconnect.de) 05.26.34 Quit Acou_Bass (Ping timeout: 260 seconds) 05.31.55 Join Acou_Bass [0] (~eddie@cpc97736-bolt17-2-0-cust152.10-3.cable.virginm.net) 05.32.59 Quit johnb5 (Ping timeout: 240 seconds) 06.45.04 Quit St3ak (Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net) 06.46.35 *** Saving seen data "./dancer.seen" 06.51.21 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 06.53.38 # _bilgus: Good references: https://www.kernel.org/doc/htmldocs/writing_musb_glue_layer, and linux/drivers/usb/musb/{jz4740.c,musbhsdma.c} 06.54.22 # Ingenic used a bone-stock MUSB HDRC implementation, along with the Inventra DMA implementation. 07.35.41 Join johnb5 [0] (~johnb2@p5b3afe8e.dip0.t-ipconnect.de) 07.37.52 Quit St3ak (Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net) 07.38.19 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 07.49.32 Quit johnb5 (Ping timeout: 256 seconds) 07.50.45 Quit St3ak (Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net) 07.51.56 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 07.52.21 Quit St3ak (Client Quit) 07.53.44 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 08.46.36 *** Saving seen data "./dancer.seen" 09.22.53 # almost have it. it's successfully probing the mass storage device but failing to mount 09.38.30 Quit St3ak (Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net) 09.39.31 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 09.40.33 Quit St3ak (Client Quit) 09.41.49 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 09.50.27 Quit St3ak (Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net) 09.52.00 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 09.52.18 Join t0mato8 [0] (~t0mato@193.32.127.158) 09.52.38 Quit t0mato (Read error: Connection reset by peer) 09.52.38 Nick t0mato8 is now known as t0mato (~t0mato@193.32.127.158) 09.54.06 Join massiveH [0] (~massiveH@ool-18e4e82f.dyn.optonline.net) 09.56.32 Join amdj` [0] (~aaron@freenode/staff/atheme.amdj) 09.56.32 Quit amdj (Killed (wilhelm.freenode.net (Nickname regained by services))) 09.56.32 Nick amdj` is now known as amdj (~aaron@freenode/staff/atheme.amdj) 09.56.38 Quit pamaury (Quit: Konversation terminated!) 09.57.21 Join marex-cloud_ [0] (sid137234@gateway/web/irccloud.com/x-dhieptucikesrfgl) 09.59.27 Join ParkerR_ [0] (ParkerR@znc.withg.org) 10.00.13 Quit ParkerR (Ping timeout: 272 seconds) 10.00.14 Quit marex-cloud (Ping timeout: 272 seconds) 10.00.15 Quit Rondom (Ping timeout: 272 seconds) 10.00.15 Quit WakiMiko (Ping timeout: 272 seconds) 10.00.15 Quit Tsesarevich (Ping timeout: 272 seconds) 10.01.12 Join Rondom [0] (~rondom@modo.nonmodosedetiam.net) 10.07.15 Join WakiMiko [0] (~WakiMiko@unaffiliated/wakimiko) 10.07.41 Join Tsesarevich [0] (Tsesarevic@fluxbuntu/founder/joejaxx) 10.21.09 # ok, got it! DMA is operational on the SEND side. firing up sysbench to stress it a bit.. 10.30.59 Quit St3ak (Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net) 10.32.02 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 10.34.17 Join St3ak` [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 10.35.17 Quit St3ak (Remote host closed the connection) 10.42.57 # _bilgus: g#2746 10.44.31 # benchmarks apparently unaffected. Hmm. 10.46.37 *** Saving seen data "./dancer.seen" 11.05.47 Quit massiveH (Quit: Leaving) 11.06.19 # the RX side is going to be the same basic rework structure but with more complex rules. 11.11.03 # <_bilgus> haha thats funny that kernel.org thing is what I was reading last night talk about serendipitous 11.18.19 # <_bilgus> speachy very nice the removal of magic numbers helps alot 11.22.38 # it also makes it clear what's needed to better genericize the DMA -- we can have eight different endpoints with DMA at once. 11.23.38 # at least this driver is arguably useful beyond the jz4760, should we find another SoC that uses the same DMA engine 11.23.49 # s/DMA/DMA+USB/ 11.30.33 Quit dweeber (Read error: Connection reset by peer) 11.44.46 # <_bilgus> so thats odd, the reads / writes are still within that 5% range but the seq write halved in latency and Rand r/w doubled 11.45.07 # I'd have expected no change at all in writes. 11.45.35 # since only device->host transfers were DMA-enabled. 11.46.04 # I do need to re-enable the logging and see what kind of transfer sizes we're actually using. 11.47.14 # (and if mode1 DMA transfers are being used at all..) 11.48.05 # <_bilgus> I'm guessing its bypassing your code 11.52.14 # <_bilgus> ((unsigned long)ptr & 4 == 0) 11.53.02 # whoops, that should be &3 11.54.23 # no change in sequential read. 12.09.25 # huh, I think it's using DMA only for the first transfer, and PIO for the remainder. 12.17.02 # okay I must be having a major brain fart but that (ptr & 3 == 0) test is responsible for skipping dma from ever being used. but based on the log messags of the addresses used, that test is always true.. 12.18.21 # (the buffer always seems to be 16-byte aligned) 12.26.21 # <_bilgus> well removing it makes it now want to connect on usb plug 12.26.35 # clearly badness remains. 12.26.38 # <_bilgus> not* 12.34.16 # <_bilgus> hmm it looks like transfer speeds drop in half after a ROLO 12.35.45 # <_bilgus> maybe just a coincidence i'll restart and try again 12.35.51 Join ZincAlloy [0] (~Adium@2a02:8108:943f:d824:2c0a:df60:96e0:39b) 12.36.51 # I haven't noticed that 12.44.50 # <_bilgus> nah happening after a restart too maybe somethings up with the pc 12.45.05 Quit _bilgus (Quit: Leaving) 12.46.39 *** Saving seen data "./dancer.seen" 12.52.51 Join _bilgus [0] (~bilgus@2605:a000:1301:89bd:846e:542c:8bed:d177) 13.08.31 # <_bilgus> hmm yeah I can't get the opriginal speeds back even with the hold switch WEIRD 13.09.29 Join lebellium [0] (~lebellium@89-92-69-66.hfc.dyn.abo.bbox.fr) 13.25.20 # with the hold switch I was seeing completely identical behavior+performance. 13.26.16 # <_bilgus> I'm going to revert to head and try again 13.27.31 # just pushed a new version of the patch, with some fixes but also disables DMA completely by default now. 13.31.27 Quit Huntereb (Read error: Connection reset by peer) 13.32.33 # <_bilgus> better than whern it started even if it does still have issues 13.33.24 # gotta put it all aside again for a while. 13.33.53 # <_bilgus> I'll play with it wed night and see what I can figure out 13.37.09 # <_bilgus> ok speeds are back @ HEAD so perhaps something isn't quite right 13.37.28 # there are zero functional changes with dma disabled 13.38.11 # <_bilgus> well you'd think the hold switch would reflect that then no? 13.38.44 # I'd expect so, yeah. but please try with the most recent patch as the baseline (it compiles out the DMA code altogether) 13.39.02 # <_bilgus> ok 13.41.01 # <_bilgus> gotta head out of town, Ill be sure to do some more testing / tweaking 13.41.32 # I think I'll be able to poke at this again tonight. weather is crappy so it's indoor activities instead. 13.42.08 # the linux driver is structured very differently so it's difficult to suss out some of the more subtle sequencing issues. 13.50.22 Join sakax [0] (~r0b0t@unaffiliated/r0b0t) 13.58.14 Join johnb5 [0] (~johnb2@p5b3afe8e.dip0.t-ipconnect.de) 14.31.47 Join p3tur [0] (~petur@rockbox/developer/petur) 14.31.48 Quit petur (Read error: Connection reset by peer) 14.31.49 Quit p3tur (Remote host closed the connection) 14.46.41 *** Saving seen data "./dancer.seen" 14.51.21 Quit johnb5 (Ping timeout: 265 seconds) 15.06.44 Join johnb5 [0] (~johnb2@p5b3afe8e.dip0.t-ipconnect.de) 15.20.22 Quit johnb5 (Ping timeout: 256 seconds) 15.24.40 Quit michaelni (Quit: Leaving) 15.54.20 # got mode0 working for small transfers. big transfers (mode 1) are going into never-never land. 16.16.07 Join MrZeus_ [0] (~MrZeus@2a02:c7f:70d0:6a00:28cb:21a2:5343:b164) 16.30.40 Join michaelni [0] (~michael@213-47-68-29.cable.dynamic.surfer.at) 16.43.49 Join pamaury [0] (~pamaury@rockbox/developer/pamaury) 16.46.44 *** Saving seen data "./dancer.seen" 16.48.45 # well. looks like the hardware doesn't support large TX transfers. Feature flag in the core says it's not implemented. 16.54.07 Join ac_laptop [0] (~ac_laptop@186.2.247.129) 16.58.45 Quit St3ak` (Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net) 16.59.41 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 17.00.26 Quit St3ak (Client Quit) 17.03.27 Join St3ak [0] (~st3ak@st3ak3000.powered.by.lunarbnc.net) 17.29.34 Quit michaelni (Ping timeout: 256 seconds) 17.31.01 # _bilgus: code updated. considerably restructured. 17.35.35 Quit prg318 (Quit: ZNC 1.7.5 - https://znc.in) 17.36.36 Quit lebellium (Quit: Leaving) 17.38.36 Join prg318 [0] (~prg@deadcodersociety/prg318) 17.41.22 Quit ZincAlloy (Quit: Leaving.) 17.42.33 # still ahven't found why mode1 isn't working. max dma is 1MB, and the HW is suppsoed to just work. best guess is that there's some other setup bit missing. 17.42.42 # but that's a problem for another time. 17.42.56 Quit pamaury (Ping timeout: 244 seconds) 18.02.19 Join MrZeus__ [0] (~MrZeus@2a02:c7f:70d0:6a00:d29:2940:c454:1176) 18.10.42 Quit MrZeus_ (Ping timeout: 272 seconds) 18.22.21 Quit sakax (Quit: Leaving) 18.33.15 Join Will07 [0] (cb207702@bcfw1.barker.nsw.edu.au) 18.35.41 Part Will07 18.46.46 *** Saving seen data "./dancer.seen" 19.02.25 Join bluebrother [0] (~dom@rockbox/developer/bluebrother) 19.03.05 Join fs-bluebot [0] (~fs-bluebo@55d46982.access.ecotel.net) 19.05.10 Quit fs-bluebot_ (Ping timeout: 244 seconds) 19.06.08 Quit bluebrother^ (Ping timeout: 260 seconds) 19.23.23 Quit t0mato (Ping timeout: 240 seconds) 19.23.29 Join Soap [0] (~Soap@rockbox/staff/soap) 19.49.52 Join t0mato [0] (~t0mato@193.32.127.158) 19.58.57 Quit bluebrother (Disconnected by services) 19.59.02 Join bluebrother^ [0] (~dom@rockbox/developer/bluebrother) 19.59.07 Join fs-bluebot_ [0] (~fs-bluebo@55d46b43.access.ecotel.net) 20.01.41 Quit fs-bluebot (Ping timeout: 272 seconds) 20.13.13 # dma transferred the first block.. and the USB core seems to be ignoring it.. kicking back an underrun error for some reason 20.36.31 Quit MrZeus__ (Ping timeout: 272 seconds) 20.46.49 *** Saving seen data "./dancer.seen" 21.36.22 Join Huntereb [0] (~Huntereb@174.226.20.225) 21.59.07 Join _bilgus_ [0] (~bilgus@2605:a000:1301:89bd:846e:542c:8bed:d177) 21.59.48 Quit _bilgus (Ping timeout: 244 seconds) 22.02.13 # _bilgus_: just pushed working DMA USB TX. sequential reads jumped up about 33%, from 6.6->8.8MB/s 22.02.13 Quit _bilgus_ (Remote host closed the connection) 22.04.08 Join _bilgus_ [0] (~bilgus@2605:a000:1301:89bd:846e:542c:8bed:d177) 22.11.18 # given the substantial DMA overhead, it's debatable if it's worth using DMA for smaller tranfer sizes, but it's not like we're multitasking when operating as a USB device. 22.11.35 Quit ac_laptop (Ping timeout: 258 seconds) 22.46.51 *** Saving seen data "./dancer.seen" 23.57.31 Quit TheSeven (Ping timeout: 244 seconds) 23.57.48 Join [7] [0] (~quassel@rockbox/developer/TheSeven)