Ezequiel Garcia <ezequiel@xxxxxxxxxxxxxxxxxxxx> writes: > A previous version of the driver didn't have that. However, under certain > stress testing it was observed that the PCIe link goes down. I still have the > traces for that: > > [..] > [21833.389031] pciehp 0000:13:01.0:pcie24: pcie_isr: intr_loc 100 > [21833.389035] pciehp 0000:13:01.0:pcie24: Data Link Layer State change > [21833.389038] pciehp 0000:13:01.0:pcie24: slot(1-5): Link Down event > [21833.389076] pciehp 0000:13:01.0:pcie24: Disabling > domain:bus:device=0000:14:00 > [21833.389078] pciehp 0000:13:01.0:pcie24: pciehp_unconfigure_device: > domain:bus:dev = 0000:14:00 > [21833.389103] TW686x 0000:14:00.0: removing > [21833.416557] TW686x 0000:14:00.0: removed > [..] > > I have no idea why the link goes down (hardware issue?), > but it's better to handle it gracefully :) Could be hw issue indeed, FWIW I have never observed this problem on my hw (which is ARM-based and includes TW6869). > I understand that on some platforms this implementation could be too > costly (it's > completely cheap on any modern x86), and I intend to provide some option > to provide "frame DMA-to-buffer" and "scatter-gather DMA". Though I'm using the driver that I posted several months ago, with SG DMA field-only mode (SG DMA engine doesn't support frame mode and while non-SG DMA (field or frame) mode is certainly possible with CMA, copying the buffers is out of the question here). I wonder if that driver works for you (which would suggest a sw bug) or if the problems are the same (hw issue - system, power, or the chip itself). BTW adding incremental patches on top of the driver I posted would seem preferable to me. It would enable us to pinpoint (and not include) changes which tend to break something. For now, it seems I'll be stuck with my version, since I have to do DMA to buffers (though I could use the frame mode). Incremental patches are generally how it is done in Linux. Perhaps we should merge my driver first and you'd add your changes on top of it (preserving the existing functionality)? This is very easy with git. -- Krzysztof Halasa Industrial Research Institute for Automation and Measurements PIAP Al. Jerozolimskie 202, 02-486 Warsaw, Poland -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
