| ↑back Search ←Prev date Next date→ Show only urls | (Click on time to select a line by its url) |
| Who | What | When |
|---|---|---|
| mchehab | ok, thanks | [00:03] |
| ....................................................................................... (idle for 7h10mn) | ||
| *** | Tex_Nick has left "In Linux, We Trust" | [07:13] |
| ...... (idle for 27mn) | ||
| ccaione | hverkuil: unfortunately it doesn't solve the problem. In my case the cmd passed to do_video_ioctl() is still VIDIOC_DQEVENT with a wrong embedded size for struct v4l2_event. So 'case VIDIOC_DQEVENT' is missed and I lost the ioctl | [07:40] |
| alse. The userspace is issueing the ioctl with VIDIOC_DQEVENT not VIDIOC_DQEVENT32. In do_video_ioctl() I would expect to receive VIDIOC_DQEVENT32 when the userspace is 32bit. Where the conversion VIDIOC_DQEVENT->VIDIOC_DQEVENT32 is done before reaching do_video_ioctl()? | [07:52] | |
| .................................................... (idle for 4h19mn) | ||
| nohous | hverkuil: ping | [12:11] |
| .................. (idle for 1h25mn) | ||
| sailus | hverkuil: 655e9780ab913a3a06d4a164d55e3b755524186d is wrong.
The right solution would have been to use __packed | [13:36] |
| Albeit fixing that now could prove troublesome. | [13:42] | |
| ccaione | sailus: I don't think that with __packed the problem is solved, since what's happening here is that struct timespec definition differs in userspace and kernelspace. It's not that it's just wrongly packed. | [13:55] |
| sailus | ccaione: No, not quite.
I think the original problem came about with control events. Aarch64 and arm appear to have similar alignment for the data types involved, unlike x86 and x86-64. And the "fix" probably fixed it for aarch64 but broke x86 compat at the same time. Is compat even relevant for aarch64 btw.? | [14:00] |
| ccaione | I'm on aarch64 | [14:01] |
| sailus | Can you run 32-bit arm binaries on aarch64 machines?
Ah. Interesting. I though the fix was especially intended for aarch64. | [14:01] |
| ccaione | uhm, in what case do_video_ioctl() receive a ioctl with cmd == VIDIOC_DQEVENT32? | [14:02] |
| sailus | It *should* happen when you run a 32-bit program on a 64-bit machine.
If it doesn't, it's a bug (as pinchartl noted). | [14:03] |
| ccaione | the problem is that I don't see the code path for which the ioctl called from the 32-bit program using VIDIOC_DQEVENT, arrives in do_video_ioctl() as VIDIOC_DQEVENT32 | [14:05] |
| sailus | ccaione: Can you run 32-bit arm binaries on aarch64? | [14:06] |
| ccaione | sailus: yes | [14:06] |
| ..... (idle for 23mn) | ||
| sailus | ccaione: I'll try to take a look at the issue, but right now I have other urgent matters to handle. | [14:29] |
| ccaione | sailus: np. Thank you. | [14:33] |
| .... (idle for 15mn) | ||
| sailus | ccaione: Is the size of timespec different (i.e. sizeof(struct timespec) on 32-bit arm and sizeof(compat_timespec) on aarch64) different?
What are the values? I wonder if you're getting the wrong definition of __kernel_time_t for some reason. That'd be long long, i.e. 64 bits... | [14:48] |
| ccaione | sailus: IIRC of the 32-bit app side I have 8, on the 64-bit kernel side 16. | [14:58] |
| sailus | ccaione: How does your struct compat_timespec look like?
I can't figure out how that could be 16 bytes, it's a struct of two s32's. | [15:12] |
| ccaione | sailus: I don't have the board with me to check run-time. For sure they were different. | [15:15] |
| sailus | ccaione: No need to. Just taking a glance to the source would be enough.
It's in arch/arm64/include/asm/compat.h . | [15:16] |
| ccaione | sailus: on my kernel https://paste.fedoraproject.org/351828/01287711/ | [15:19] |
| sailus | compat_time_t ?
It's in the same file. | [15:23] |
| ccaione | s32 | [15:24] |
| sailus | And... that'd be 16 bytes on aarch64?
I have to say I don't know aarch64 at all but it looks very strange to me. | [15:24] |
| ccaione | I'm not following you here since in my test compat_timespec is never used, always the full struct timespec is used instead
userspace uses sizeof(v4l2_event) to encode the size portion of the cmd and inside v4l2_event we have struct timespec | [15:25] |
| sailus | Yes. | [15:27] |
| ccaione | this generate a cmd value for the VIDIOC_DQEVENT ioctl. For the kernel the cmd value associated with VIDIOC_DQEVENT is different | [15:27] |
| sailus | But the size was 16 bytes in the kernel, right? | [15:27] |
| ccaione | since for the kernel sizeof(v4l2_event) is different | [15:27] |
| sailus | Yes, in the kernel it should match VIDIOC_DQEVENT32. | [15:28] |
| ccaione | but how? if the userspace is doing a ioctl on VIDIOC_DQEVENT, how we receive VIDIOC_DQEVENT32 on the kernel side?
I'm missing this stpe | [15:28] |
| sailus | If you have a 32-bit user space binary that runs on a 64-bit kernel, this is expected.
The ABI is different, and the memory layout of struct v4l2_event on arm32 should match that of struct v4l2_event32 on aarch64. struct v4l2_event is different on aarch64. That's the point of the compat IOCTLs. | [15:29] |
| ccaione | agree, but still I don't see the codepath that changes my VIDIOC_DQEVENT ioctl into a VIDIOC_DQEVENT32 | [15:30] |
| sailus | VIDIOC_DQEVENT on arm is exactly the same as VIDIOC_DQEVENT32 on aarch64. At least that's the intent.
(And I presume that's not the case for you, for some reason.) | [15:31] |
| ccaione | I guess that's the case also on my setup. But let me going through the code step by step. 1) my app issues an ioctl with VIDIOC_DQEVENT. In the value for the cmd is embedded the sizeof(struct v4l2_event). So for exaple cmd == 0xaaaaaaaa
2) this ioctl is passed in the kernel all the way to v4l2_compat_ioctl32() 3) from here to do_video_ioctl() cmd is still 0xaaaaaaaa hoooo, wait, I see you point now. So in theory VIDIOC_DQEVENT32 == 0xaaaaaaaa and then we change cmd = VIDIOC_DQEVENT in the switch-case sailus: ok, you are right. For some reason in my case VIDIOC_DQEVENT on arm != VIDIOC_DQEVENT32 on aarch64 | [15:45] |
| *** | ribalda has quit IRC (Ping timeout: 260 seconds) | [15:51] |
| sailus | Indeed.
I have no access to aarch64 hardware, I presume some others do, at least Andrzej who submitted the patch changing things a bit. It'd be helpful if you could debug what's causing the difference in size. Need to leave now, might be back later. | [16:03] |
| ccaione | sailus: I need to check that and I left the hw in the office. But now I undesterstood what's going on and I can better debug this issue on monday
thanks for your help | [16:12] |
| sailus | ccaione: You're welcome! | [16:13] |
| .................................... (idle for 2h58mn) | ||
| koike | mchehab, about "[PATCH v3] [media] tpg: Export the tpg code from vivid as a module", the tpg doesn't really need vivid, anyone could write a module that uses the tpg as the video pattern generator, so it doesn't really depends on vivid. I am having a hard time to understand why "depends on VIDEO_VIVID" in the tpg's Kconfig is the best option? | [19:11] |
| pinchartl | koike: I'd use "select TPG" in the vivid Kconfig entry instead | [19:13] |
| koike | pinchartl, I am going to resend the patch using "select TPG" then
pinchartl, when you use yavta -cX -F out, do you use any specific tool to visualize the captured images? | [19:14] |
| pinchartl | I usually use raw2rgbpnm3~
raw2rgbpnm | [19:17] |
| koike | pinchartl, thank you, I am going to check that | [19:17] |
| pinchartl | it was hosted on gitorious :-/
sailus: can you push it somewhere else ? | [19:19] |
| javier__ | pinchartl: in case is useful, I found an old repo some months ago and pushed to github: https://github.com/martinezjavier/raw2rgbpnm
probably is not the last version but I've been using it without issues | [19:20] |
| pinchartl | javier__: I have 8 patches on top of that. can you give pinchartl push permissions for that repo ? | [19:22] |
| javier__ | pinchartl: done | [19:24] |
| pinchartl | pushed, thanks
you can now remove my commit access :-) (if you want) | [19:26] |
| javier__ | pinchartl: I'll keep it if you don't mind, so you can push there as a temporary repo until sailus finds a new home for that repo | [19:28] |
| pinchartl | sure, no issue
I actually have 3 more patches if you want to review them :-) | [19:29] |
| javier__ | pinchartl: sure :) | [19:31] |
| pinchartl | sent | [19:36] |
| javier__ | pinchartl: Ok, thanks | [19:36] |
| sailus | pinchartl, javier__: http://salottisipuli.retiisi.org.uk/cgi-bin/gitweb.cgi?p=~sailus/raw2rgbpnm.git;a=summary | [19:41] |
| pinchartl | sailus: thanks. can you get that indexed by google ? | [19:42] |
| sailus | It's been there all the time.
How do you do that? I could add a link to gitweb to the main page I guess. That might be actually a good idea. | [19:42] |
| pinchartl | that would be useful
and also how about git.retiisi.org.uk ? | [19:42] |
| sailus | Looks like I've changed it in 2008...
I can add an alias. | [19:43] |
| pinchartl | thanks
and let me know if you like my three patches :-) | [19:44] |
| javier__ | sailus: cool, I didn't know about that repo. I'll delete the temporary one I created at github | [19:46] |
| pinchartl | javier__: or maybe link to Sakari's repo from there
that might help getting it indexed by google sailus: you can also post an announcement on linux-media | [19:48] |
| javier__ | pinchartl: OK | [19:49] |
| sailus | Don't you think it's a bit late for an announcement? :-)
Sure I can send one. | [19:51] |
| javier__ | pinchartl: your 3 format patches looks good to me | [19:59] |
| sailus | Hmm. I even have my old GPG key here I replaced in 2011.
Time to update, huh? | [20:11] |
| There's now git.retiisi.org.uk, redirecting from root to gitweb.cgi.
And a link from my homepage. Eight years is a good interval for homepage updates, right? | [20:19] | |
| *** | awalls1 has left | [20:24] |
| pinchartl | sailus: slightly late, yes :-) but having it archived on the mailing list will improve search engine ratings a bit
sailus: I'm not sure I'd call it an interval :-) sailus: if you can configure your web server to turn http://git.retiisi.org.uk/cgi-bin/gitweb.cgi?p=~sailus/raw2rgbpnm.git into http://git.retiisi.org.uk/sailus/raw2rgbpnm it would help too | [20:34] |
| sailus | pinchartl: Done.
Well, the gitweb is now available in the root. | [20:43] |
| nohous | hverkuil: ping | [20:48] |
| pinchartl | sailus: much better, thanks | [20:52] |
| ................................ (idle for 2h39mn) | ||
| headless | nite | [23:31] |
| ↑back Search ←Prev date Next date→ Show only urls | (Click on time to select a line by its url) |