| ↑back Search ←Prev date Next date→ Show only urls | (Click on time to select a line by its url) |
| Who | What | When |
|---|---|---|
| *** | benjiG has left | [13:58] |
| ............... (idle for 1h14mn) | ||
| ezequielg | hverkuil: ping
hverkuil: regarding your patch "tw686x-kh: specify that the DMA is 32 bits" hverkuil: can you point me at where are bounce buffers created? is that in v4l2 core or deeper? | [15:12] |
| hverkuil | ezequielg: much deeper. | [15:14] |
| ezequielg | right | [15:14] |
| hverkuil | in the low-level dma code, usually arch specific too.
Whenever I need to debug in that area it takes me at least an hour to refresh my memory how it is done. | [15:15] |
| ezequielg | so, without DMA32 flag, x86 fails to allocate memory for sg mode in tw686x. I get some "Out of IOMMU" message.
so it's quite mandatory on my setup. | [15:16] |
| hverkuil | It is mandatory for all PCI(e) cards that only do 32 bit DMA.
can the techwell do 64 bit dma? Your driver does set that flag, right? | [15:16] |
| ezequielg | no :( | [15:18] |
| hverkuil | did I miss it? Hmm. | [15:18] |
| ezequielg | http://git.infradead.org/users/ezequielg/linux/commit/05203bb93708d27e8a3642e6d37b296dafc8eb4b | [15:18] |
| hverkuil | Can you reply to my patch series containing your driver with a note that I should set that flag?
Then I'll add it when I make the final pull request. Otherwise I'll forget. | [15:19] |
| ........... (idle for 52mn) | ||
| ezequielg | hverkuil: sure. | [16:11] |
| .......... (idle for 48mn) | ||
| *** | awalls1 has left | [16:59] |
| ezequielg | hverkuil: no, the tw686x device can't do 64-bit dma, afaik. | [17:04] |
| .... (idle for 18mn) | ||
| nohous | hverkuil: what do you mean by that PCIe cards are required to only do 32b DMA access? | [17:22] |
| mchehab | hverkuil: I guess it is ok to use more than 32 bits with PCIe, but that depends on the chipset
anyway, there are some macros that evaluate that, based on the CPU (and chipset) | [17:23] |
| nohous | hverkuil: what happens if user e.g. wants the device to fill buffer passed by the USERPTR method that is physically located beyond the 32b limit? | [17:23] |
| mchehab | nohous: if the buffer is out of the DMA area range, it won't work
I guess mmap will fail | [17:24] |
| nohous | mchehab: mmap? with userptr? i feel lost | [17:24] |
| mchehab | well, the userptr buffer will be memory mapped by the Kernel
actually, I'm not sure about the current implementation at VB2... on VB1, there were a code that would warrant that the buffer would be remapped to be inside the memory range but that code could take some time... so, I don't remember if we ported it to VB2 or not | [17:25] |
| nohous | hm, nothing's as simple as it seems apparently
i believe that at least with pcie gen2, all endpoints are required to support 64b addressing | [17:28] |
| mchehab | as I said, it depends not only on the board, but also on the CPU | [17:28] |
| .... (idle for 17mn) | ||
| hverkuil | mchehab: nohous: if the DMA addressing of the device is limited to 32 bit, and you try to DMA to memory that is outside of that range, then the DMA core will allocate bounce buffers, so the device DMAs into those buffers, and that is memcpy'd to the final destination.
Which is no big deal for small buffers, but for video it is a bad idea. So if the DMA of the device is limited to 32 bit, then GFP_DMA32 should be set. Of course, that won't help if you use the USERPTR streaming mode, but for MMAP this works fine. Bounce buffers are created by lib/swiotlb.c | [17:45] |
| nohous | hverkuil: ok, interesting, thanks for insight | [17:54] |
| ............... (idle for 1h10mn) | ||
| *** | Tex_Nick has left "In Linux, We Trust" | [19:04] |
| ........................ (idle for 1h59mn) | ||
| pinchartl | mchehab: ping | [21:03] |
| mchehab | pong | [21:04] |
| pinchartl | that was fast :-)
I'd like to resubmit "[PATCH v4 3/4] media: Add type field to struct media_entity" would it make sense to quickly discuss it here ? progress over e-mail was quite slow | [21:04] |
| mchehab | we can discuss on irc, but not today | [21:05] |
| pinchartl | ok
I'll submit a new version now then | [21:05] |
| mchehab | ok | [21:05] |
| pinchartl | with the latest comments taken into account | [21:05] |
| mchehab | if the new version is ok, I'll aply, otherwise we can discuss that ;) | [21:05] |
| pinchartl | well, the ones I agree with :-)
I can live with obj_type instead of type, would you prefer that ? | [21:05] |
| mchehab | works for me
yes, obj_type sounds ok | [21:06] |
| pinchartl | ok I'll change that now and resubmit | [21:06] |
| mchehab | ok | [21:07] |
| ↑back Search ←Prev date Next date→ Show only urls | (Click on time to select a line by its url) |