| blueCmd | sending 0x1 as address through ioctl made my kernel crash and the simulator exit | 01:34 |
|---|---|---|
| blueCmd | complaining about access to non aligned address (and rightly so) - but crashing sounds a bit hard | 01:35 |
| blueCmd | harsh* | 01:35 |
| blueCmd | http://0ecb8fa744969b07.paste.se/ < I will provide a better test case and simulator output later | 01:37 |
| -!- X-Scale` is now known as X-Scale | 07:25 | |
| blueCmd | would it be unreasonable to implement some sort of crypto acceleration in the simulator? emulate some HW and then let the host do the calculations | 14:00 |
| blueCmd | doing SSH handshakes takes a couple of minutes right now :P | 14:00 |
| stekern | that could be pretty cool, then you could implement it in real hw as well | 14:02 |
| blueCmd | yes | 14:02 |
| stekern | I imagine that ssh is pretty slow there as well | 14:02 |
| blueCmd | my hope is that muxed control session will be "fast" | 14:03 |
| blueCmd | we will see in a minute or two :) | 14:03 |
| stekern | how is ssh calculation abstracted? | 14:05 |
| blueCmd | there are two operations in dropbear that takes 90 or 95 of the time | 14:05 |
| stekern | i.e. how hard would it be to make stuff realise that there's hw-accel for it? | 14:06 |
| blueCmd | aha | 14:06 |
| blueCmd | no idea :) | 14:06 |
| blueCmd | http://www.vm.ibm.com/devpages/spera/MG_OpenSSH.pdf looking at that atm | 14:06 |
| stekern | ideally it would be a completely decoupled block from the cpu (i.e. seperate wb-core) | 14:06 |
| blueCmd | yeah | 14:07 |
| stekern | where does ssh acceleration normally take place? | 14:08 |
| blueCmd | no idea | 14:08 |
| stekern | (I'm a bit interested for unrelated stuff at dayjob) | 14:08 |
| blueCmd | the idea came to me just now, haven't really done any research :P | 14:09 |
| blueCmd | openssh 4.4 has this in changelog: | 14:09 |
| blueCmd | - Add optional support for OpenSSL hardware accelerators (engines), enabled using the --with-ssl-engine configure option. | 14:09 |
| blueCmd | padlock is an HW engine apparently. VIA C3 has some sort of CPU with HW assisted crypto | 14:10 |
| stekern | ok, cool, then there's probably at least some abstraction layer for it | 14:10 |
| stekern | then there's of course the question at what level things should be accelerated | 14:12 |
| stekern | I'm not very familiar with the algorithms involved, but I assume operations in the calculations can be accelerated at a lower level | 14:13 |
| blueCmd | sha and hashes should be pretty straight forward, no? | 14:14 |
| blueCmd | generation of keys | 14:15 |
| stekern | sounds correct | 14:15 |
| blueCmd | f = g^y mod p, that operation takes a lot of time as of now | 14:16 |
| blueCmd | (it's in a comment, I don't know what any of those variables are or how big they are) | 14:16 |
| stekern | juliusb did a proof-of-concept sha256 hw-accelerator showoff for a presentation iirc | 14:17 |
| blueCmd | http://linux.die.net/man/3/dh_openssl | 14:20 |
| blueCmd | there you have the bn_mod_exp - that's the one | 14:20 |
| blueCmd | so overwriding that somehow and it would work! I see now way this can fail | 14:21 |
| blueCmd | hah, joking aside - I might just look into this, sounds fun | 14:22 |
| _franck_ | jeremybennett: newlib already uses NOP_EXIT_SILENT | 16:12 |
| _franck_ | that's why I did this change | 16:13 |
| jeremybennett | Ah - OK | 16:33 |
| _franck_ | is that something I did wrong while compiling the toolchain? : http://pastebin.com/xRiWeiT7 | 16:38 |
| _franck_ | or I need to track this down, until it works ? :) | 16:38 |
| _franck_ | compiling with -static cause this | 16:39 |
| jeremybennett | OpenRISC engineers might be interested in this EE Times article... http://www.eetimes.com//design/eda-design/4406001/Android-processor-IP-is-royalty-free?cid=NL_EETimesDaily | 19:46 |
| stekern | I don't get the headline, what does android have anything to do with it? | 21:07 |
| _franck_ | blueCmd: are you around ? | 23:27 |
| blueCmd | _franck_: now! | 23:31 |
| * blueCmd just got home | 23:32 | |
| _franck_ | /opt/or1k-toolchain/lib/gcc/or1k-elf/4.8.0/../../../../or1k-elf/bin/ld: error in /opt/or1k-toolchain/lib/gcc/or1k-elf/4.8.0/crtend.o(.eh_frame); no .eh_frame_hdr table will be created. | 23:33 |
| _franck_ | reading this: http://sourceware.org/bugzilla/show_bug.cgi?id=4454 | 23:33 |
| _franck_ | it looks like it is related to tc-or1k | 23:33 |
| _franck_ | and since you've touched it, I was wondering if you know what this warning is ? | 23:34 |
| blueCmd | _franck_: yes, I know that warning | 23:34 |
| blueCmd | hm, what version of gcc and binutils are you using? | 23:34 |
| _franck_ | you mean the host gcc ? | 23:36 |
| blueCmd | well, your or1k-elf-gcc should have an ./libgcc/config/or1k/t-crtstuff file | 23:37 |
| blueCmd | that says CRTSTUFF_T_CFLAGS += -fno-dwarf2-cfi-asm | 23:37 |
| blueCmd | which should fix that warning | 23:37 |
| _franck_ | oups, my bad my repo wasn't up to date.... | 23:39 |
| blueCmd | np ;) | 23:41 |
| blueCmd | omg | 23:53 |
| blueCmd | test isspace: is 'c' a space? YES | 23:53 |
| blueCmd | I love eglibc | 23:54 |
| * asm is not a dwarf | 23:55 | |
| blueCmd | asm: :D | 23:56 |
| asm | ;) | 23:58 |
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!