↑back Search ←Prev date Next date→ Show only urls | (Click on time to select a line by its url) |
Who | What | When |
---|---|---|
*** | freakazoo has quit IRC (Ping timeout: 246 seconds) | [04:12] |
...................................................................................... (idle for 7h5mn) | ||
bbrezillon | hverkuil: I know it's too late for 5.2 but I'd like to send a v5 of the rk-mpeg2 patchset
is that a problem? should I wait after the merge window? | [11:17] |
..... (idle for 21mn) | ||
hverkuil | bbrezillon: it's fine to post it, but it won't be processed until after the merge window. | [11:38] |
bbrezillon | hverkuil: that's fine
just want to get that out of my TODO list :) | [11:39] |
.................. (idle for 1h27mn) | ||
hverkuil | mchehab: pinchartl: I did some tests with the virtual drivers how well they handle an unbind while streaming.
vivid and vim2m are doing this correctly. vicodec hangs, so I will need to figure out why. vimc crashes. shuah: something to look at ^^^ In the vicodec case I am 99% certain that it is a driver bug. I suspect the same for vimc, but there may be core lifetime issues involved as well. | [13:06] |
pinchartl | hverkuil: so I wasn't completely wrong :-)
$ lsmod | grep 'vimc ' vimc 16384 0 while streaming... | [13:09] |
hverkuil | Hmm, the vimc subdevs should increment the main vimc driver module count.
The refcount of vimc_capture goes to 1 when streaming, so that's correct. | [13:12] |
..... (idle for 24mn) | ||
Each loaded component gets a pointer to the top-level vimc data, so vimc_ent_sd_register should increment the vimc module refcount and decrement in vimc_ent_sd_unregister.
loaded -> bound But something probably should also be done when you start to capture, otherwise the e.g. vimc_sensor would still be 0, meaning that you can rmmod the sensor while streaming. | [13:37] | |
shuah | hverkuil: I saw your email about vimc from yesterday. Will take a look
I assume the case "start streaming and while streaming run unbind"? | [13:47] |
hverkuil | unbind isn't even needed, rmmod works as well... | [13:51] |
shuah | okay - thanks | [13:52] |
....... (idle for 31mn) | ||
hverkuil | pinchartl: fixed the vicodec issue with unbind. Typical driver bug. | [14:23] |
shuah | hverkuil: media dev allocator increments the refcount for the other driver - that is the way ensure rmmod does nothing when module is in use. | [14:26] |
mchehab | hverkuil: your fixes to vicodec seem OK on my eyes. Those media drivers still using devm_* almost certainly have lifetime management issues | [14:36] |
hverkuil | yeah, that seemed a good idea at the time, until we realized that that doesn't play nice with unbind. | [14:37] |
mchehab | yep | [14:37] |
hverkuil | We're hardly the only subsystem with this problem, I suspect it's everywhere. | [14:38] |
mchehab | I suspect they can still be fixed to keep working with devm, but that may require creating a virtual device with would be the parent of the V4L2 device
(not that I think it would worth the effort - explicit management is IMO the best way) | [14:39] |
shuah | hverkuil: is there a specific driver you want me to take a look for this "streaming - rmmmod" problem? | [14:41] |
hverkuil | shuah: you mean e.g. vimc vs vimc_sensor? | [14:42] |
shuah | right | [14:42] |
hverkuil | vimc certainly crashes.
I haven't done any real testing with the other vimc drivers. | [14:42] |
shuah | okay - thanks. I will try with vimc | [14:42] |
hverkuil | FYI: I'm leaving in a few minutes. | [14:44] |
*** | benjiG has left | [14:58] |
prabhakarlad has left | [15:08] | |
.............................................................. (idle for 5h6mn) | ||
sailus | jmondi: Buona sera! | [20:14] |
.................. (idle for 1h25mn) | ||
*** | ntrrgc has quit IRC (*.net *.split)
al has quit IRC (*.net *.split) koike has quit IRC (*.net *.split) | [21:39] |
...... (idle for 25mn) | ||
pfallenop has quit IRC (Ping timeout: 250 seconds) | [22:08] |
↑back Search ←Prev date Next date→ Show only urls | (Click on time to select a line by its url) |