log☇︎
117 entries in 1.184s
BingoBoingo: asciilifeform: Sure, but if we print a MIPS down the line sim-devices may be the way to do it instead of DMA everywhere
asciilifeform: yer still stuck with certain idjicies of iron (e.g. dma) but then when finally bake new iron, you have a ready compiler & userland to plant on it, that actually behaves.
a111: 113 results for "dma", http://btcbase.org/log-search?q=dma
asciilifeform: !#s dma
a111: Logged on 2018-10-06 14:47 mircea_popescu: anyway, to put some meat on the bones of "works with cca 2015 vintage drivers, and them only." : incorrect vintage drivers manage to literally fuck the chip, i have never seen dma errors before in teh logs, now have i seen chips refusing to turn back on and in general hardware level spew, "wrong mode d3", "error -5" etc.
mircea_popescu: anyway, to put some meat on the bones of "works with cca 2015 vintage drivers, and them only." : incorrect vintage drivers manage to literally fuck the chip, i have never seen dma errors before in teh logs, now have i seen chips refusing to turn back on and in general hardware level spew, "wrong mode d3", "error -5" etc. ☟︎
asciilifeform: ( recall the dma thread, for instance. 9000+ hacks around the retardation of von neumann box. )
asciilifeform: mircea_popescu: in principle. in practice, very difficult ( in asciilifeform's understanding, entirely impractical ) to avoid your model being poisoned at the heart by the retardation of the iron ( dma, interrupts, whole orchestra of horrors that prevent anything like effectual isolation of processes , esp in kernel, and -- more insidiously -- prevent sane model of computation even in head )
asciilifeform: von neumann machine offers no sane means of asynchrony. ( hence idjit kludges like dma, little cpus in erry kbd/mouse/nic etc )
a111: Logged on 2016-09-17 16:40 asciilifeform: and the hardware ~is~ shit. boot up one of these (if you can actually get it to boot.) and say hello to 1 fps graphics, disks without dma (you don't know what these feel like until trying personally), nic that works when the moon is full strictly, etc.
ckang: then for a minute I was exploring using the ESP32's cryptography accelerator with QSPI and DMA
mircea_popescu: wget WILL time the netcard for you, the netcard has dma, that's the wholew story.
asciilifeform: that is the monumental scale of an idiocy like copying sparc, dma, winblowz toolchain.
asciilifeform: in j. von n.'s model, there is (ideally. in actual life, all kinds of 'unprincipled exceptions' like dma) 1 prime mover, what today we call 'cpu'.
a111: Logged on 2018-01-05 18:45 asciilifeform: ( upstack : dma, interrupts, pipeline, instruction reorderer, 'hyperthreading', multiple buses, 'bridges' -- all are epicycles ( hey mircea_popescu ! ) from vonneumannism , where instructions 'push' (unrelated to stack concept) outputs, rather than 'pull' inputs as they oughta )
asciilifeform: ( upstack : dma, interrupts, pipeline, instruction reorderer, 'hyperthreading', multiple buses, 'bridges' -- all are epicycles ( hey mircea_popescu ! ) from vonneumannism , where instructions 'push' (unrelated to stack concept) outputs, rather than 'pull' inputs as they oughta ) ☟︎
a111: 99 results for "dma", http://btcbase.org/log-search?q=dma
asciilifeform: !#s dma
asciilifeform: on the other hand, pipeline idea per se was a mistake; same kind of failure to invent dataflowism as dma
asciilifeform: mircea_popescu: we did, possibly in the dma thread
mircea_popescu: ofcoruse such a thing can not exist until we actually make tmsr-computer without dma
mircea_popescu: apparently it can be formulated so as to dissolve in dmf/dma etc
a111: Logged on 2017-08-30 13:03 asciilifeform: ( and obsessing over cpu, in a machine arch where nic has dma -- is tunnel vision. )
asciilifeform: ( and obsessing over cpu, in a machine arch where nic has dma -- is tunnel vision. ) ☟︎
asciilifeform: spyked: didja miss the part where it runs on no modern nic ? or the thread where ALL modern nics are built to use dma, which gives the nic vendor access to every byte of your ram ?
spyked: I'm still in the process of grokking the DMA+interrupts discussion.
a111: 89 results for "dma", http://btcbase.org/log-search?q=dma
asciilifeform: !#s dma
asciilifeform: btw at some point somebody will ask, 'why not use existing gadget with dma, e.g. usb3, to diddle memory'. so answer, intel et al are implementing 'iommu', so that only nsa-approved diddles go through dma controller. plus os can see if it has usb3 etc hanging off
asciilifeform: the 'nic can dma, haha' thing evaporates.
mircea_popescu: so what, no dma ?
mircea_popescu: no, have motherboard design where the "dma" is a specified bank, and the cpu ram is a different bank ; and let it call whichever it specifically wants.
mircea_popescu: seems rather like the whole nic-dma combo is being fortified as a wolf's last stand. i guess they expect ot lose cpus (to the chinese).
asciilifeform: if your box still has such things as intel NICs, DMA, pci.
Framedragger: aha! that'd be something. i mean, still full-on DMA, but yeah, k.
asciilifeform: trinque: the flaw is that you gotta support a megatonne of liquishit for even nic to work -- dma, page tables, etc
asciilifeform: where, for example, do i buy a nic that doesn't dma ? or a cpu that works with it.
asciilifeform: over which you can then dma, etc.
asciilifeform: put pci, dma, usb, etc. to bed, for good.
asciilifeform: ( pretty sure we had 'the' dma thread not long ago )
asciilifeform: ditto, e.g., dma.
asciilifeform: where in this mix do we get dma tho.
asciilifeform: mircea_popescu: (whole point of dma controller)
mircea_popescu: asciilifeform actually are dma calls cpu blocking ?!
jhvh1: ben_vulpes: {'city': 'Wilmington', 'region_name': 'DE', 'area_code': 302, 'longitude': -75.597, 'country_code3': 'USA', 'latitude': 39.56450000000001, 'postal_code': '19893', 'dma_code': 504, 'country_code': 'US', 'country_name': 'United States'}
jhvh1: pete_dushenski: {'city': 'Wilmington', 'region_name': 'DE', 'area_code': 302, 'longitude': -75.5664, 'country_code3': 'USA', 'latitude': 39.75919999999999, 'postal_code': '19880', 'dma_code': 504, 'country_code': 'US', 'country_name': 'United States'}
jhvh1: shinohai: {'city': 'Denton', 'region_name': 'TX', 'area_code': 940, 'longitude': -97.1264, 'country_code3': 'USA', 'latitude': 33.183400000000006, 'postal_code': '76205', 'dma_code': 623, 'country_code': 'US', 'country_name': 'United States'}
asciilifeform: this goes back to the 'dma' thread. large machine registers are an ad-hoc and clumsy form of parallelism.
asciilifeform: you can lose dma if you're willing to lose nic and video.
mircea_popescu: because dma and interrupts ; and bad memory addressing.
asciilifeform: interrupts, dma, are relics from dark age when there was 1 cpu and it cost most of what the machine cost.
a111: Logged on 2017-01-19 17:49 asciilifeform: the two major caltrops re 'iron lisps on x86' are 1) interrupts 2) dma
mircea_popescu: http://btcbase.org/log/2017-01-19#1605218 << actually non-interrupt / non-dma (and possibly tagged ram) is about the reason to make own fpga cpu ☝︎
asciilifeform: trinque: tldr: interrupts and dma.
asciilifeform: the two major caltrops re 'iron lisps on x86' are 1) interrupts 2) dma ☟︎
a111: Logged on 2016-12-02 19:56 trinque: how do I currently know by looking at my database that no item in it was altered by somebody spraying bits into my server via network card -> dma
trinque: how do I currently know by looking at my database that no item in it was altered by somebody spraying bits into my server via network card -> dma ☟︎
asciilifeform: given nic dma and other wonders, i would not even sign under this.
thestringpuller: DMA to this address and out comes sound
Framedragger: ("truly shady" except for, you know, the whole shitty stack incl cpu, dma etc)
mircea_popescu: ith an Intel key. It's unlikely that this is ever replaced by something open source. Firmware that runs on an ARC core inside the chipset. It runs entirely out-of-band with the main CPU. It has DMA access to the entire system memory and can access the networking adapters in a way transparent to the OS (separate MAC and IP)."
asciilifeform: all dma-enabled (because pc arch doesn't actually offer any other way to move Gb/sec)
asciilifeform: phf: unfortunately is is ~not~ possible to do without, e.g., disk dma.
asciilifeform: and the hardware ~is~ shit. boot up one of these (if you can actually get it to boot.) and say hello to 1 fps graphics, disks without dma (you don't know what these feel like until trying personally), nic that works when the moon is full strictly, etc. ☟︎
ben_vulpes: what dma
a111: 2 results for "dma bad", http://btcbase.org/log-search?q=dma%20bad
thestringpuller: $s dma bad
a111: 1 results for "why is dma bad", http://btcbase.org/log-search?q=why%20is%20dma%20bad
thestringpuller: $s why is dma bad
a111: Logged on 2016-07-22 12:33 mircea_popescu: http://btcbase.org/log/2016-07-22#1508541 << the dma thing is mostly a sufferance of the desktop designs yea
a111: Logged on 2016-07-22 06:48 mats: fwiw, i don't believe the radio in the iphone has dma
mircea_popescu: http://btcbase.org/log/2016-07-22#1508541 << the dma thing is mostly a sufferance of the desktop designs yea ☝︎
mats: fwiw, i don't believe the radio in the iphone has dma
a111: Logged on 2016-07-17 21:46 asciilifeform: writing 'open' whatevers for a machine where a closed radio has dma to the whole bus, is a laugh
asciilifeform: the whole dma thing is a riot.
asciilifeform: ditto 'open' os for a pc where closed gpu has dma.
asciilifeform: writing 'open' whatevers for a machine where a closed radio has dma to the whole bus, is a laugh ☟︎
asciilifeform: rings have 0 to do with it. in pc arch, every device on the bus can issue dma cycles and clobber ram.
asciilifeform: on x86, or any other platform with dma, all drivers are 'part of the kernel' whether honestly stated or not.
asciilifeform: 'a rich peripheral interface with DMA.....'
asciilifeform: (e.g., most nics have a - documented or not - provision for dma)
asciilifeform: mats: only thing is, i can do arbitrary dma on just about any box where i can access ~any~ pci device's standard mmio space - no need to screw with the pci config regs
asciilifeform: (can i talk to the nic? does it have dma?)
thestringpuller: even on the gameboy the dma controllers are annoying as fuck
thestringpuller: Yea I also figure you aren't a fan of DMA
assbot: 27 results for 'dma' : http://s.b-a.link/?q=dma
ascii_field: !s dma
ascii_field: ditto dma.
mats: anyway... [2015-04-02 16:59:20] <asciilifeform> mats: i can trivially make mincemeat of this by issuing raw dma cycles. << UDEREF is about exploitation, not constraining the kernel
asciilifeform: (and, optionally, disk dma req. complete)
asciilifeform: mircea_popescu: the nic knows how to dma, recall.
asciilifeform: http://log.bitcoin-assets.com/?date=19-08-2015#1243805 << 'even root'. what nonsense is this. root isn't root if he can't, e.g., operate a dma-capable device on the bus and flip arbitrary bits ☝︎
asciilifeform: example of interesting seekrit: 'magic packet that turns an intel nic to arbitrary-dma whatever into/out of ram'
asciilifeform: i will add, mainly for n00bz in audience, that this nic - like all modern nics - does dma.
asciilifeform: on a machine with von neumann arch, and even things like dma, no less
mircea_popescu: but you could also wipe the 16kb if you never called the thing, via DMA of the time
mircea_popescu: because really, trying to rewrite it without dma ... might as well start over.
mircea_popescu: dma is actually one of the principal reasons i agree with alf's "current computing stack must die"
mats: yay DMA
assbot: Logged on 02-04-2015 20:59:21; asciilifeform: mats: i can trivially make mincemeat of this by issuing raw dma cycles.