--- Log opened Tue May 10 00:00:59 2016 | ||
olofk | shorne: Do you have any failed sim testcases you want me to take a look at? | 04:21 |
---|---|---|
shorne | olofk: let me upload the results. You can have a look. it looks like many are related to things like tab completion etc | 05:17 |
shorne | if you want to run the test yourself, I will uplod the commands at the top | 05:17 |
shorne | https://gist.github.com/20d5a512884e75cab0c722fa4eb9d2a3 - gdb.sum | 05:20 |
shorne | https://gist.github.com/9534262b1d682fa7d6d54bc93e3f2449 - gdb.log | 05:21 |
shorne | those are the full results | 05:21 |
shorne | hmm, seems I hit a limit on the gdb.log upload size | 05:27 |
shorne | gdb.sum is all there | 05:27 |
olofk | shorne: Do we even support ada? :) | 05:50 |
olofk | ahh.. those were marked as unsupported | 05:50 |
olofk | Is 'sleep' not supported in newlib? | 05:52 |
olofk | Looking at line 556 in gdb.sum | 05:52 |
olofk | And here we would need a special case for or1k that defines nop as l.nop https://github.com/openrisc/binutils-gdb/blob/or1k/gdb/testsuite/gdb.base/bp-permanent.c#L30 | 05:55 |
olofk | hmm... is newlib supposed to provide mman.h ? | 05:58 |
olofk | sys/mman.h | 05:58 |
olofk | musl provides it | 05:59 |
olofk | A few other arches supplies mman.h in their newlib machine dirs, but most don't | 06:01 |
olofk | Ahh.. and the reentrant wait function (_wait_r) isn't implemented in newlib | 06:17 |
olofk | Looking at the failed tests in structs.{exp,c} and realize that I don't understand a thing of this :) | 06:33 |
olofk | There were 106 fails in that test, so fixing that should improve the statistics quite a bit | 06:35 |
olofk | multiple definition of `environ' | 06:41 |
olofk | Don't know what to do about that one. Looks like several other arches do just like we do | 06:41 |
olofk | Plenty of patches for newlib :) | 06:43 |
olofk | ok, or1k is the only arch to define environ in syscalls.c. Maybe it should go somewhere else | 06:47 |
olofk | And most arches don't define it at all. What is it for? | 06:47 |
olofk | Is it for setting env variables for bare-metal programs somehow? | 06:51 |
wallento | many many questions | 07:18 |
wallento | mman I think we don't need, as we have no filesystem support | 07:19 |
wallento | environ does not make sense, maybe some old artifact | 07:19 |
wallento | it can be set, but I am wondering what for | 07:19 |
wallento | but I am not sure it can be undefined | 07:20 |
olofk | wallento: Only a few arches seem to define it | 07:33 |
olofk | But I guess you could use it to load env vars from a register or a fixed memory address. A bit like we do with device trees | 07:34 |
wallento | yep, I think so | 08:05 |
wallento | as long as we don't use such stuff, we probably don't need it | 08:05 |
olofk | I agree. Just wasn't sure if someone actually uses this already | 08:09 |
olofk | shorne: Could you paste the output from the structs test. I'm curious to know what gdb sees and what it expects | 08:12 |
olofk | ehhmm... how large is double on openrisc? | 08:17 |
olofk | Looking at the fails in callfuncs.exp. Seems to fail with "That operation is not available on integers of more than 8 bytes." on all tests using doubles | 08:21 |
olofk | Do we support double at all? | 08:21 |
olofk | I vaguely remember someone saying that we could only use float | 08:22 |
olofk | or is this a newlib build-time setting? | 08:24 |
olofk | or gcc? | 08:26 |
olofk | This is way over my head :) | 08:26 |
olofk | Hmm... where does all the warning: Register ppc not updated | 08:55 |
olofk | warning: Register sr not updated | 08:55 |
olofk | come from? | 08:55 |
olofk | I think that breaks at least some testcases | 08:55 |
olofk | Such as charset.exp where the tests expect that there isn't any output after running a certain command | 08:56 |
SMDwrk | Can anyone tell me does openrisc have multicore version? | 12:13 |
wallento | SMDwrk | 12:13 |
wallento | yes | 12:13 |
wallento | It can do SMP | 12:13 |
SMDwrk | Fully working? | 12:14 |
wallento | the mor1kx processor fully supports it with snooping cache coherency | 12:20 |
wallento | The baremetal library (newlib) supports too, Linux also | 12:20 |
shorne | olofk: thanks for going through it, let me paste that exact section. | 17:31 |
shorne | olofk: https://gist.github.com/stffrdhrn/172cca92b37202a12590fe7ff9980735#file-gdb-failures-log-L830, this all of the failure logs, not much more detail than the summary | 17:42 |
shorne | This is the actual test case | 17:52 |
shorne | https://github.com/openrisc/binutils-gdb/blob/or1k/gdb/testsuite/gdb.base/structs.exp#L227 | 17:52 |
shorne | We should make a todo list of the failures and if we want to fix, or if we should mark unsupported | 17:58 |
olofk | shorne: Agree | 18:03 |
olofk | There are probably some low-hanging fruit here, such as getting rid of the "Register {ppc,sr} not updated" and fixing some stuff in newlib | 18:05 |
olofk | But now it's time to sleep | 18:06 |
--- Log closed Wed May 11 00:00:01 2016 |
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!