| 00:16.11 | *** join/#edev skynet-2000 (~skynet-20@unaffiliated/skynet2000) |
| 00:27.26 | *** join/#edev djerome (~djerome@ip24-251-139-203.ph.ph.cox.net) |
| 01:37.29 | *** mode/#edev [+o prpplague] by ChanServ |
| 04:05.06 | *** join/#edev mbuf (user@nat/redhat/x-zdvoljxtxgobdmeg) |
| 04:20.51 | *** join/#edev skynet-2000 (~skynet-20@unaffiliated/skynet2000) |
| 04:30.28 | *** join/#edev skynet-2000 (~skynet-20@unaffiliated/skynet2000) |
| 04:52.14 | *** join/#edev War2 (~war2@unaffiliated/war2) |
| 06:28.02 | *** join/#edev JoeLlama (~snork@unaffiliated/joellama) |
| 09:15.38 | *** join/#edev sndcrb (~sndcrb@dsl-hkibrasgw2-fef8de00-126.dhcp.inet.fi) |
| 09:26.07 | *** join/#edev nullpuppy (~nullpuppy@freematrix/staff/nullpuppy) |
| 09:36.48 | *** join/#edev nullpupp1 (~nullpuppy@freematrix/staff/nullpuppy) |
| 09:44.05 | *** join/#edev florian (~fuchs@Maemo/community/contributor/florian) |
| 10:32.21 | *** join/#edev madsy (~madsy@188.113.88.119) |
| 10:32.22 | *** join/#edev madsy (~madsy@fu/coder/madsy) |
| 10:47.19 | *** join/#edev g1powermac (~g1powerma@unaffiliated/g1powermac) |
| 14:24.38 | *** join/#edev SterNiX (~LessIsMor@unaffiliated/nu253r/x-0655220) |
| 15:02.39 | *** join/#edev GPSFan (~kenm@64.92.145.112) |
| 15:02.39 | *** mode/#edev [+o GPSFan] by ChanServ |
| 15:15.02 | *** join/#edev dijenerate (~dijenerat@173.225.251.175) |
| 16:09.10 | *** join/#edev SterNiX (~LessIsMor@unaffiliated/nu253r/x-0655220) |
| 16:16.41 | *** join/#edev bgamari (~ben@gamari.physics.umass.edu) |
| 16:19.20 | *** join/#edev CosmicPenguin (~nobody@soa.codeaurora.org) |
| 16:19.20 | *** mode/#edev [+o CosmicPenguin] by ChanServ |
| 16:31.09 | *** join/#edev landley (~landley@140.242.26.2) |
| 16:31.13 | landley | Morning. |
| 16:32.32 | *** join/#edev risca (~risca@wnpgmb0412w-ds01-123-237.dynamic.mtsallstream.net) |
| 16:34.46 | g1powermac | hey landley |
| 16:37.42 | landley | g1powermac: how goes? |
| 16:38.28 | landley | The 3.2 kernel build is very subtly broken on arm. It uses the host "size" instead of the target "size"... which my build filters out during the airlock step. :) |
| 16:47.47 | g1powermac | ahh, nice ;-) |
| 16:48.34 | g1powermac | is doing well, been reading up on lots of algorithm topics lately, mostly related to data mining and machine learning |
| 16:48.37 | g1powermac | you? |
| 16:49.01 | landley | I still haven't managed to watch the video on how modern batteries work. |
| 16:49.46 | landley | Trying really hard to get this darn u-boot feature shoehorned in today. |
| 16:49.58 | landley | The massive overcomplication of TI's uboot variant is mind boggling. |
| 16:50.06 | landley | It doesn't need to do half this stuff. |
| 16:50.26 | landley | I'm waiting for them to add a syscall interface and make it run apps. |
| 16:50.27 | *** join/#edev prpplague (~danders@nat/ti/x-ishikkqbfheknckv) |
| 16:50.35 | g1powermac | did you promptly blame prpplague just because he works at TI? ;-) |
| 16:50.43 | landley | No. |
| 16:50.48 | g1powermac | speaking of which ;-) |
| 16:50.50 | prpplague | greetings earthlings |
| 16:50.54 | landley | Hiya prpplague. |
| 16:51.02 | prpplague | what's cookin? |
| 16:51.04 | landley | The ti fork of uboot for the netra is horrific. |
| 16:51.11 | landley | It doesn't need to do half this stuff. |
| 16:52.59 | landley | prpplague: on, and the 3.2 kernel developed a really subtle build-time bug on arm. |
| 16:53.27 | landley | It's using the host "size" binary for the link instead of the target one. My airlock step caught it, by which I mean "my build broke because size wasn't in the path". |
| 16:53.45 | prpplague | landley: ahh interesting! |
| 16:54.06 | landley | Daniel Robbins (founder of gentoo, current maintainer of funtoo) gave me a login to a really fire breathing 16-way server with 48 gigs of ram that does builds in under a minute. |
| 16:54.13 | landley | So bisecting isn't as bad as it could be. :) |
| 16:54.20 | landley | But I'm mostly working on day job stuff with u-boot. |
| 16:55.26 | prpplague | landley: ahh dandy |
| 16:55.30 | landley | prpplague: while you're at it, why the hell is CONFIG_OMAP_SERIAL split across arch/arm/mach-omap2/serial.c and drivers/serial/omap-serial.c? |
| 16:55.35 | landley | I mean "what is TI smoking"? |
| 16:55.35 | prpplague | is trying to finish up some demo stuff for CES |
| 16:55.40 | landley | Question I ask a lot... |
| 16:55.53 | landley | Today's the deadline for CELF paper proposals. Trying to figure out if I want to apply. |
| 16:56.11 | landley | It would be my first time since the linux foundation destroyed CELF. I feel a bit bad about interacting with the corpse. |
| 16:56.19 | prpplague | landley: the root cause is that many of the TI developers are not trained/experienced in developing for the open source community |
| 16:56.19 | landley | But Tim Bird seems to still be involved, so... |
| 16:56.26 | landley | prpplague: I noticed. |
| 16:56.39 | landley | I interviewed there. Seemed like a lot of great guys absolutely imprisoned in a bureaucratic nightmare. |
| 16:56.51 | landley | Apparently they do just enough defense contract work to screw up the rest of the company pretty badly. |
| 16:57.04 | prpplague | landley: i am not going to ELF in feb, and i from what i hear, not many people |
| 16:57.13 | landley | Not that they could quite come out and _say_ this... |
| 16:57.27 | prpplague | landley: well it is more from the aspect that they had been doing closed source for phones and such |
| 16:57.31 | landley | prpplague: yeah, but now that I've written off OLS too I'm not really going to _any_ conferences. |
| 16:57.39 | prpplague | landley: yea |
| 16:57.55 | prpplague | landley: wmat and i have been joking around with the idea of an embedded conference |
| 16:58.07 | prpplague | brb, need coffee |
| 16:58.09 | landley | prpplague: I co-founded penguicon and founded linucon. |
| 16:58.19 | landley | I've been toying around with the idea of doing a new convention myself. :) |
| 16:58.33 | landley | Those two were "combination linux expo and science fiction convention". Science fiction, real technology. |
| 16:58.43 | prpplague | landley: yea, my thought was that if we actually decided to do it, we'd get you involved since neither of us have experience with conference organization |
| 16:59.16 | landley | At our first penguicon, we had Terry Pratchett talking about his boyhood experiments soldering together a bunch of chips he pulled out of speaking clocks and such to make a giant board that gave meteorological reports in the morning and everything. |
| 16:59.24 | landley | There is an mp3 of this on archive.org. |
| 16:59.41 | landley | prpplague: where would you be likely to hold it? |
| 16:59.57 | landley | Me, I've been pondering Austin, but Penguicon is up in michigan and I've never lived there. |
| 17:00.27 | landley | (My co-founder Tracy Worcester did. She chaired Confluence 19100 back in 2000, so she handled the fannish side of things.) |
| 17:00.39 | g1powermac | mutters something about cheap conference halls available here in KY, but no one would dare to enter the state. . . |
| 17:02.26 | landley | So the OMAP serial is an 8250 with extensive bug workarounds? (Reading the driver. "Empty RX fifo read causes an abort". Golf claps, everyone, golf claps!) |
| 17:04.51 | prpplague | landley: we did not get as far as deciding on a location |
| 17:05.24 | prpplague | landley: yea, basically, but there are some additional high speed features that haven't been merged mainline yet |
| 17:06.17 | landley | prpplague: you're in dallas, I'm in Austin, I have no idea where wmat is... |
| 17:06.31 | landley | g1powermac: you don't want a conference hall, you want a hotel with attached function space. |
| 17:07.02 | landley | The way science fiction conventions work is they barter a discount on function space for a "room block", I.E. a commitment to fill a certain number of hotel rooms for a certain number of nights. |
| 17:07.41 | landley | The general rule of thumb is if you can fill the entire hotel, you get the function space for free. If you fill half the hotel, you get a 50% discount on their function space, and you can work out a sliding scale if you ask for it. |
| 17:08.32 | g1powermac | landley, got that too here |
| 17:08.53 | landley | You generally negotiate some sort of discount on the normal room rate, which the attendees have to identify themselves as belonging to the convention to get. (That way, the hotel can track how many room nights the event is bringing in.) |
| 17:09.12 | *** join/#edev JoeLlama (~snork@cpe-50-113-130-245.socal.res.rr.com) |
| 17:09.18 | *** join/#edev JoeLlama (~snork@unaffiliated/joellama) |
| 17:09.22 | landley | The ideal is to "own" the hotel, I.E. fill it up entirely with your event attendees (plus the airline rooms which are always there, but that's a side issue). |
| 17:09.55 | landley | Pilots and flight attendants on layover away from home never cause trouble, they either sleep through anything or join in. :) |
| 17:10.40 | landley | Once you own the hotel, nobody else can complain about anything you do during the con (nobody there to get upset about noise or hall costumes or the wiccan ceremony in the lobby). |
| 17:11.35 | landley | And the hotel staff gets really friendly to you, because a hotel manager's 100% completion job performance is to fill the hotel. It usually doesn't happen. When it happens, they score a "perfect" on their job performance (assuming you don't do too much damage), and the rest of the staff is happy if management is happy. |
| 17:12.16 | landley | So if you only fill up 25% of the hotel negotiating things like keeping the pool open 24 hours is hard, but when you own the hotel "ask and ye shall receive". |
| 17:12.57 | g1powermac | you may not get complaints from people inside the hotel, but once the locals get wind of the 'penguin worship', you're screwed ;-) |
| 17:13.12 | prpplague | landley: interesting |
| 17:13.16 | landley | For example, http://www.youtube.com/watch?v=w2mj-Sq2oeo happened the first year we owned our hotel, never heard a peep from the hotel staff about it... |
| 17:13.30 | landley | Of course it took our even to year 4 to get there... |
| 17:13.44 | landley | g1powermac: in Austin? Once again, they'd join in... |
| 17:14.52 | g1powermac | hehe, no, here in KY :-) |
| 17:15.01 | landley | Another bone you can throw a hotel in your negotiations is holding a banquet. The catering/facilities people tend to have their own budget, with their own goals to meet. |
| 17:15.29 | *** join/#edev risca (~risca@wnpgmb0412w-ds01-123-237.dynamic.mtsallstream.net) |
| 17:15.38 | g1powermac | prpplague knows what I'm talking about :-) |
| 17:16.31 | landley | Telling them you'd like them to cater an event (and then seating a guest of honor at each table, and selling separate tickets for the remaining ~5 seats at the table), makes their eyes light up... |
| 17:18.37 | landley | You need a bunch of different skills to run a con. Hotel negotiation is one of the anchor skills. That's your main cost center, and most of your logistical issues. |
| 17:19.10 | landley | Ops (operations) is another, that's a sort of catch all actually running the infrastructure at the con. |
| 17:20.16 | landley | Guest liason is another: inviting people and handling their travel and scheduling and green room access and communication and such. (Generally one dedicated person per guest as their "minion", and they report to the person in charge of all guests.) |
| 17:20.34 | landley | Registration is another: handling the tickets. |
| 17:21.10 | landley | Treasurer: handling the money, bank account. Between those two somebody should understand credit cards (both online and at the door, in Linucon's case it was Mirell.) |
| 17:21.42 | landley | Advertising. Sometimes this is two positions: online and local. That's its own whole thing. |
| 17:21.54 | landley | Volunteer wrangler. |
| 17:21.55 | landley | Security. |
| 17:22.04 | landley | Art show. Con Suite. Green Room. |
| 17:23.16 | landley | You generally have "concom" (the people handling all the above), and then you have "staff" (the people who report to them), and then you have panelists (people who speak on the panels), volunteers (attendees who want a discount or crash space, or just to help out), and general attendees. |
| 17:23.26 | landley | Either at the top (or underneath it all, depending) is the con chair. |
| 17:24.01 | *** join/#edev SterNiX (~LessIsMor@unaffiliated/nu253r/x-0655220) |
| 17:24.36 | landley | Oh, webmaster is a concom position these days: server admin, run the mailing lists, blog and twitter... |
| 17:25.20 | landley | Heh, forgot "programming". That's a concom position, the person who schedules (and staffs!) all the panels. |
| 17:25.39 | landley | I'll stop now... |
| 17:33.11 | prpplague | landley: yea, i think we'd want to target more hardcore embedded developers with lots of hands-on stuff |
| 17:33.22 | landley | prpplague: I'm all for it. :) |
| 17:37.59 | landley | Yay, jay's emailing me the relevant documentation for the chip! |
| 17:42.59 | *** join/#edev Guest35886 (~LessIsMor@2.176.167.120) |
| 17:55.08 | *** join/#edev anoNxeRo (~LessIsMor@unaffiliated/nu253r/x-0655220) |
| 17:59.22 | *** join/#edev florian (~fuchs@sign-4db6aec1.pool.mediaWays.net) |
| 17:59.22 | *** join/#edev florian (~fuchs@Maemo/community/contributor/florian) |
| 18:15.37 | *** join/#edev skynet-2000 (~skynet-20@unaffiliated/skynet2000) |
| 19:05.34 | Mirell | Woah. |
| 19:05.38 | Mirell | Backlog of convention related chat. |
| 19:06.34 | *** join/#edev florian (~fuchs@eduroam.uni-siegen.de) |
| 19:06.34 | *** join/#edev florian (~fuchs@Maemo/community/contributor/florian) |
| 19:17.16 | landley | Mirell: you have been warned I wanna do another con in Austin someday. |
| 19:18.01 | landley | I keep trying to attach to an existing one and push, but armadillocon and worldcon are both well-established to the point of being extremely set in their ways. They don't like trying anything new. |
| 19:18.54 | landley | Maybe sxsw could get new stuff attached to it, but I suspect we'd get lost in the noise. That sucker's huge. |
| 19:42.47 | *** join/#edev skynet-2000 (~skynet-20@unaffiliated/skynet2000) |
| 20:01.47 | *** join/#edev anoNxeRo (~LessIsMor@unaffiliated/nu253r/x-0655220) |
| 20:01.55 | *** join/#edev skynet-2000 (~skynet-20@99-62-100-172.lightspeed.tukrga.sbcglobal.net) |
| 20:02.00 | *** join/#edev skynet-2000 (~skynet-20@unaffiliated/skynet2000) |
| 20:13.08 | Mirell | Also, expensive. |
| 20:21.44 | *** join/#edev Matth_FA (~Matth_FA@mtl93-14-78-241-192-235.fbx.proxad.net) |
| 20:33.40 | calculus | landley: looks like the hexagon cpu has landed in mainline 3.2 kernel |
| 20:33.56 | landley | calculus: yup, I was crowing about this a while ago. |
| 20:34.09 | landley | Did they give any useful toolchain links, or do we still have to fish a fork out of code aurora? |
| 20:35.23 | landley | Mirell: yup. I still think south of $50 for the weekend is a sweet spot for hobbyist attendance, but it would be nice if we could find a way to charge professionals' employers more, ala http://www.joelonsoftware.com/articles/CamelsandRubberDuckies.html |
| 20:35.30 | landley | calculus: do you have a test system? |
| 20:35.54 | landley | The real hardware is in most android phones (any based on qualcomm's snapdragon, anyway), but we never got a useful _bootloader_ for it. |
| 20:36.21 | landley | Although you might be able to make one by disassembling the binary blob they feed to the multimedia codec, since that's hexagon code and runs on the hexagon. :) |
| 20:36.34 | landley | And last I checked was not encrypted or anything... |
| 20:41.32 | *** mode/#edev [+o landley] by ChanServ |
| 20:45.27 | *** join/#edev skynet-2000 (~Skynet-20@unaffiliated/skynet2000) |
| 20:45.46 | sorphin | sighs. FAIL on Mastercard job. If I don't find something by March/End of March, I'm fscked. |
| 20:45.49 | landley | For penguicon they pretty much decided market segmentation was beyond them, charged everybody the cheap rate, and then got corporate sponsorships. Lots of technical conferences do that. |
| 20:46.23 | landley | The problem is some organizations think "if it's that cheap it can't be worth anything", and don't want to spend $2k on travel expenses for a $50 conference. |
| 20:46.49 | landley | Plus we hold it over the weekend rather than during the week, which is optimized for hobbyists (attending on their own time) instead of professionals (attending on their boss's dime). |
| 20:47.13 | Mirell | Ayup |
| 20:47.14 | landley | But really, there's buckets of professional conferences, and they all suck. |
| 20:47.20 | landley | From my point of view, anyway. |
| 20:47.26 | Mirell | The four day conference is an idea. |
| 20:47.35 | Mirell | Thursday is "PROFESSIONAL TRAINING FOR $$$$" |
| 20:47.41 | Mirell | For the business folks |
| 20:47.45 | Mirell | And to convince bosses |
| 20:47.50 | landley | Tutorials, yup. |
| 20:47.52 | Mirell | Friday-Sunday is the regular hobbyist conference |
| 20:48.07 | landley | Friday afternoon through sunday, anyway. You can do more tutorials friday morning. |
| 20:48.22 | landley | The hobbyists can just about take off work early and show up around 3. |
| 20:49.26 | *** join/#edev skynet-2000 (~Skynet-20@99-62-100-172.lightspeed.tukrga.sbcglobal.net) |
| 20:49.26 | *** join/#edev skynet-2000 (~Skynet-20@unaffiliated/skynet2000) |
| 20:49.26 | calculus | landley: nope, no test system |
| 20:49.54 | *** part/#edev sorphin (~dstagner@50-76-61-17-ip-static.hfc.comcastbusiness.net) |
| 20:49.54 | landley | There's a similar "don't run sunday too late" for you out-of-towners who want to head to the airport or drive back home. Sunday generally doesn't run past 4pm officially, after that's the dead dog party and general hangout. |
| 20:50.07 | landley | calculus: qualcomm was trying to add support to QEMU. I dunno if they ever managed to make it work, though. |
| 20:51.30 | landley | Alas, looks like there wasn't much. Here was some sniffing around: http://www.mail-archive.com/qemu-devel@nongnu.org/msg35984.html |
| 20:53.02 | calculus | which is why I like SCALE so much, it is mostly Saturday/Sunday with tutorials Friday (and possibly Thursday) |
| 20:53.07 | calculus | also it is cheap |
| 20:53.15 | roxfan | is hexagon the same thing as qdsp? |
| 20:53.40 | landley | roxfan: hexagon is qdsp6 |
| 20:53.47 | roxfan | oh nice |
| 20:53.53 | landley | The snapdragon actually has 4 processors in it. |
| 20:54.06 | roxfan | did they release any public datasheets too? |
| 20:54.20 | landley | Two of them are arm, and armv7 called "scorpion" and an old armv5 that's the boot processor and does some low-level radio stuff after it fires up Android on scorpion. |
| 20:54.58 | landley | qdsp6 is the multimedia coprocessor that does mp4 encoding/decoding and mp3 encoding/decoding and during phone calls does some signal processing and all sorts of stuff, but it's basically running a realtime binary blob under their own in-house RTOS. |
| 20:55.28 | landley | (That's what they ported Linux to. In theory, it allows them to ditch scorpion and not only save power and chip real estate, but not have to pay license fees to arm. It's quite a fire breathing chip in many respects.) |
| 20:55.56 | landley | The fourth processor is an old qdsp4, it's back before they added lots of general purpose CPU instructions to it, and is really just a DSP. It does signal processing like the armv5 does. |
| 20:56.14 | landley | roxfan: dunno. |
| 20:56.35 | landley | All qualcomm's open source stuff goes through two layers of "OH GOD GET IT OFF ME" isolation to protect qualcomm's laywers' egos. |
| 20:56.43 | landley | The first layer is "QUIC", Qualcomm Innovation Center. |
| 20:57.01 | landley | http://www.quicinc.com/ |
| 20:57.28 | landley | They're who to ask. (Note: quic doesn't actually exist, it's a polite fiction, but there's probably some HR people assigned to it who can route you to the right guys.) |
| 20:57.53 | landley | The second layer is the "code aurora forum", a partnership between qualcomm and qualcomm to try to pretend that it's a general purpose industry consortium. |
| 20:58.21 | landley | https://www.codeaurora.org/ |
| 20:58.28 | landley | That's where they host all their git stuff. |
| 20:59.02 | Mirell | Parternship between Qualcomm and .. Qualcomm? |
| 20:59.17 | landley | Note: they outsourced the actual "we got it to work, now please get it upstream" work of the Linux port to Linutronix, so Thomas Gleixner's guys are going to know buckets about this... but are probably tied up in huge DMAs. |
| 20:59.36 | landley | We did mange to at least mail 'em a couple comet boards while were were there. (A hexagon development board.) |
| 20:59.47 | landley | Mirell: They got somebody else to nominally be involved, Intel I think. |
| 20:59.54 | landley | But really: it's just a second layer of wrapper for qualcomm. |
| 21:00.29 | landley | The thing about Qualcomm is that the lawyers are driving, because they're better than engineering at lying about revenue. Qualcomm's quarterly revenue statements are a polite fiction. |
| 21:00.50 | landley | All the revenue from patent licensing gets credited to legal, but the costs of coming UP with that IP are deducted from engineering. |
| 21:01.00 | SpeedEvil | <PROTECTED> |
| 21:01.01 | Mirell | CDMA patent expiring and all |
| 21:01.05 | landley | Thus legal _seems_ to have three times the net profit of actuall making stuff. |
| 21:01.18 | landley | Mirell: the lawyers are terrified of that, yes. They're scrambling to come up with replacements. |
| 21:01.19 | SpeedEvil | Actually - looks at the box - I haven't unwrapped it, and it arrived last week. |
| 21:01.25 | SpeedEvil | (nokia lumia 700) |
| 21:01.27 | landley | meanwhile engineering is going "maybe we'll get some power back within the company". |
| 21:02.01 | landley | SpeedEvil: ask how you run hexagon linux on your snapdragon device. |
| 21:02.13 | SpeedEvil | I would guess not. |
| 21:02.19 | landley | You need a bootloader. |
| 21:02.27 | SpeedEvil | I assume it's signed bootloader. |
| 21:02.30 | landley | No, I mean agitate on linux-kernel. "How do I run this on this hardware?" |
| 21:02.35 | landley | SpeedEvil: not that I know of. |
| 21:02.39 | SpeedEvil | hmm. |
| 21:03.00 | SpeedEvil | It would be a lot more interesting to me if I could run linux on it. |
| 21:03.15 | landley | The vendor might have added that, but qualcomm was mostly leaning on the fact that the instruction set of the hexagon was only available if you got their dev kit under NDA. (Even though it was based on gcc.) |
| 21:03.17 | SpeedEvil | But then that raises the fun issues of the add-on hardware past the CPU |
| 21:03.27 | SpeedEvil | Interesting. |
| 21:03.31 | landley | They _technically_ comply by posting toolchains on code aurora forum, but you have to know where to look for 'em. |
| 21:04.17 | SpeedEvil | A fairly fundamental issue is that I have no windows stuff. |
| 21:04.23 | landley | https://www.codeaurora.org/patches/quic/hexagon/ |
| 21:04.31 | landley | SpeedEvil: the toolchains run on linux. :) |
| 21:04.31 | SpeedEvil | So doing anything from that side is tricky. |
| 21:04.47 | SpeedEvil | Sure - but tehre isn't a flasher |
| 21:04.51 | SpeedEvil | For linux |
| 21:05.07 | SpeedEvil | So I'd need to RE the windows one, presumably. |
| 21:05.16 | SpeedEvil | Or try to go the JTAG route. |
| 21:05.25 | SpeedEvil | I assume qualcomm doesn't release datasheets? |
| 21:05.26 | landley | SpeedEvil: you know how the open source androids for the Nexus S and such have a "binary blob" they need to feed to the hardware to make the multimedia and phone parts work? |
| 21:05.34 | SpeedEvil | yes. |
| 21:05.41 | landley | That binary blob is fed to the hexagon, which then gets kicked to start running it. |
| 21:05.56 | landley | You can disassemble it with the hexagon objdump. :) |
| 21:06.00 | SpeedEvil | neat |
| 21:06.18 | landley | The _other_ little piece of complexity is that the MMU in the hexagon is fake (software mmu). |
| 21:06.27 | SpeedEvil | uclinux! :) |
| 21:06.28 | landley | The hexagon just has TLBs, which get loaded by a page fault routine. |
| 21:06.55 | SpeedEvil | The hexagon is the 'DSP' bit? |
| 21:07.04 | landley | (And it hasn't got enough TLB entries for series stressing, rev 3 or so of qdsp6 increases the number of TLBs but the shipping nexus hardware was rev 2, or something like that.) |
| 21:07.06 | SpeedEvil | dsp/radio |
| 21:07.10 | landley | hexagon is a DSP that grew legs. |
| 21:07.43 | landley | The current stuff is a 6 stage pipeline clocked at 600 mhz. each pipeline stage is 4 issue VLIW. |
| 21:08.01 | landley | And it has SIMD instructions so it can do a heck of a lot per clock cycle. |
| 21:08.04 | SpeedEvil | That's not a nice simple lean DSP anymore. |
| 21:08.08 | SpeedEvil | :) |
| 21:08.34 | landley | The _clever_ thing they did was decouple the pipeline stages. They have 6 independent register profiles, they treat each as a separate thread, and round robin them down the pipeline. |
| 21:08.39 | SpeedEvil | It actually does radio hardware stuff, and isn't signed? |
| 21:08.46 | landley | (With appropriate logic to completely power down a stage that's halted.) |
| 21:09.49 | landley | It's a bit like hyper-threading, but this way they have no pipeline interlocks. They just removed that logic entirely: the next/previous stages have nothing to do with each other so can't interfere/stall. |
| 21:09.58 | landley | The chip looks to linux like a 6-way SMP running at 100mhz. |
| 21:10.06 | landley | Each thread is fairly hard realtime. :) |
| 21:10.26 | SpeedEvil | Funky. Does nasty things to required memory bandwidth, I guess. |
| 21:10.45 | SpeedEvil | thinks of propellor. |
| 21:10.46 | landley | That's plus the 4 issue VLIW plus SIMD means it's really really good at multimedia stuff. |
| 21:10.55 | landley | SpeedEvil: yes and no. |
| 21:11.12 | SpeedEvil | But yes, somewhat more complex. |
| 21:11.16 | landley | Keep in mind that the chip is 600 mhz and there's 6 clock cycles between each memory fetch/write for a given thread profile. |
| 21:11.24 | SpeedEvil | Ah |
| 21:11.35 | SpeedEvil | And 600MHz isn't as fast as it used to be |
| 21:11.37 | landley | That plus your standard L1/L2 cache covers a multitude of memory access sins. |
| 21:11.53 | landley | (And if the memory isn't ready, it just sends a NOP down the pipeline and waits 6 more clock cycles for its' turn.) |
| 21:12.04 | landley | A central idea in this chip is "never do wasted work". |
| 21:12.11 | landley | Because wasted work eats power. |
| 21:12.11 | SpeedEvil | I guess if you can fit a 'frame' into L2, it's golden. |
| 21:12.50 | landley | So there's no "speculative" anything. If it's not going to keep the result, it doesn't do the calculation. |
| 21:13.03 | landley | They get the speed back with massive parallelism. |
| 21:13.08 | SpeedEvil | It's astonishing - not so much how high perforance phones have got - but how well they don't do stuff. |
| 21:13.13 | SpeedEvil | Powersaving. |
| 21:13.20 | landley | Hexagon is a really cool chip. |
| 21:13.27 | landley | It's also buried in qualcomm internal politics. |
| 21:13.39 | SpeedEvil | :/ |
| 21:13.47 | landley | Their in-house ARM chip (scorpion) has a whole team of people working on it in Raleigh. The Hexagon guys are in Austin. |
| 21:13.58 | landley | The scorpion guys are NOT happy about the idea of being rendered obsolete. |
| 21:14.03 | SpeedEvil | It would be so nice if people would just sell chips with datasheets. |
| 21:14.08 | SpeedEvil | To whoever asks. |
| 21:14.14 | landley | Thus the Linux on hexagon guys are explicitly NOT going after the phone market. |
| 21:15.03 | landley | I think to a certain extent, getting Linux for hexagon support out in the open source community was a way _around_ the company internal blocks on wider use of the chip. |
| 21:15.15 | SpeedEvil | 3G wifi APs? |
| 21:15.18 | roxfan | got a link to a blob? |
| 21:15.32 | landley | But without hobbyist test hardware, a bootloader that lets you get a USB serial console on your phone, or or QEMU support... |
| 21:15.53 | landley | roxfan: ask the cyanogenmod guys. |
| 21:16.07 | roxfan | meh |
| 21:18.13 | SpeedEvil | landley: Is the RF hardware on a seperate processor, or is that run from hexagon? |
| 21:19.47 | *** join/#edev anoNxeRo (~LessIsMor@unaffiliated/nu253r/x-0655220) |
| 21:20.53 | landley | roxfan: try this guy then, https://plus.google.com/112218872649456413744#112218872649456413744/posts |
| 21:21.14 | landley | SpeedEvil: some of it's on hexagon, some of it's on qdsp4, some of it's on the armv5. |
| 21:21.49 | landley | I just had a 6 month contract. Due to the internal politics they couldn't get the funding to renew it, so I was there through the end of October 2010. |
| 21:22.05 | *** join/#edev skynet-2000 (~Skynet-20@99-62-100-172.lightspeed.tukrga.sbcglobal.net) |
| 21:22.05 | *** join/#edev skynet-2000 (~Skynet-20@unaffiliated/skynet2000) |
| 21:22.24 | landley | I didn't really fiddle with stuff that wasn't on Hexagon (we were testing on comet boards, not phones), and it was over a year ago... |
| 21:22.59 | landley | I know hexagon handles the actual voice packetization, but I think it hands off to one of the other chips to run the radio. But I really dunno. |
| 21:23.07 | SpeedEvil | Idly wondering, as there are pretty obvious issues if you can screw with the low-level radio. |
| 21:23.44 | SpeedEvil | Nightmare scenario is someone figuring out how to do direct phone-phone comms, and the app being popular. |
| 21:26.19 | *** join/#edev skynet-2000 (~Skynet-20@unaffiliated/skynet2000) |
| 21:41.21 | CosmicPenguin | Lexar just released a 256GB CF card |
| 22:22.31 | mnemoc | uh |
| 23:15.03 | Mirell | Sigh. |
| 23:25.22 | Mirell | I give up. |
| 23:31.14 | landley | Ah, here's the hexagon reference stuff: https://developer.qualcomm.com/hexagon-processor |
| 23:31.20 | landley | They have added new public thingies. :) |
| 23:31.26 | landley | Quite a bit in the past year, looks like. |
| 23:32.08 | landley | SpeedEvil, roxfan: poke. |
| 23:32.21 | landley | Still no bootloader for android phones, and that page doesn't seem to link to the toolchain or say where to get a test environment... |
| 23:32.41 | SpeedEvil | hmm. Fun. |
| 23:32.52 | SpeedEvil | I have a windows phone though. |
| 23:32.54 | landley | Note that HexagonV2 is what's in the Nexus One. Dunno about the newer versions. |
| 23:33.27 | landley | Hexagonv4 should definitely have the insufficient TLB slots issue fixed. (I think v3 shipped before that.) But that's just a performance issue trying to do a -j 6 gcc build or similar. |
| 23:33.27 | SpeedEvil | Hence possibly less hackable than rooted android. |
| 23:33.57 | landley | SpeedEvil: A) I'm so sorry, B) WHY????? C) Is it a snapdragon SOC or some other hardware? |
| 23:34.38 | SpeedEvil | A) thanks. B) nokia was giving them out free. C) I've not investigated other than it's qualcomm and 1GHz |
| 23:34.52 | CosmicPenguin | thats a snapdragon |
| 23:34.55 | Mirell | Mmm... |
| 23:35.01 | CosmicPenguin | I don't think there are any non snapdragon windows phones |
| 23:35.04 | CosmicPenguin | maybe overseas |
| 23:35.13 | landley | Note: the HVM is a closed source binary blob that implements the MMU. (It has a TLB but not an MMU, so it walks page tables in software to reload TLB entries. Hence, software MMU.) |
| 23:35.47 | landley | I asked why they didn't open source the darn MMU blob and they went "we have multiple patents on it", and I went "software MMUs were around in the 80's, what's to patent", and (since I was talking to an engineer) they went "stuff that wasn't patented in the 80's". Sigh. |
| 23:36.42 | SpeedEvil | :/ |
| 23:37.18 | SpeedEvil | I was using 'mergemem' back in the day. |
| 23:37.25 | SpeedEvil | I had a ~6M system. |
| 23:37.39 | SpeedEvil | It was a kernel module that merged identical pages. Worked well. |
| 23:38.05 | SpeedEvil | Needless to say, this was considerablybefore vmware got a patent on the exact idea. |
| 23:38.42 | SpeedEvil | MSM8255 says wikipedia. |
| 23:41.43 | landley | One nice thing about open source: it acts as prior art. |
| 23:42.27 | landley | One bad thing about the patent system: getting it to look at prior art other than a previous patent costs an average of $2.1 million in legal feels. |
| 23:42.42 | landley | Too bad the SFLC became useless... |
| 23:43.22 | SpeedEvil | I keep meaning to mail vmware, and say 'I intend to port this module to a new kernel' |
| 23:43.39 | SpeedEvil | and see if I can get an acknowledgement from them that this is legal |
| 23:44.08 | SpeedEvil | In principle the hashing units in some processors could make it an interesting idea |
| 23:46.17 | *** join/#edev florian (~fuchs@sign-4db6aec1.pool.mediaWays.net) |
| 23:46.17 | *** join/#edev florian (~fuchs@Maemo/community/contributor/florian) |
| 23:57.24 | *** join/#edev roxfan (dunno@91.180.36.176) |