<html><head><meta http-equiv="content-type" content="text/html; charset=utf-8"></head><body dir="auto">Is it better to enable SWIOTLB? Are there any advantages with SWIOTLB enabled?<div> </div><div>— Christian</div><div> <div><div id="AppleMailSignature">Sent from my iPhone</div><div> On 27. Nov 2017, at 15:53, Michel Dänzer <<a href="mailto:michel@daenzer.net">michel@daenzer.net</a>> wrote: </div><blockquote type="cite"><div>On 2017-11-27 01:17 PM, Tom St Denis wrote: <blockquote type="cite">On 27/11/17 07:02 AM, Michel Dänzer wrote: </blockquote><blockquote type="cite"><blockquote type="cite">On 2017-11-27 12:50 PM, Christian König wrote: </blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">Am 27.11.2017 um 12:02 schrieb Michel Dänzer: </blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">On 2017-11-24 05:09 PM, Michel Dänzer wrote: </blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">On 2017-11-24 03:29 PM, Christian Zigotzky wrote: </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">Hi All, </blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">I bisected today and the first bad commit is: </blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">a4dec819c8bba6365eb893a4ca88db4dd1210110 (drm/ttm: Add helper </blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">functions </blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">to populate/map in one call (v2)) [1] </blockquote></blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">It can't really be that commit, since it just adds functions but </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">doesn't </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">hook them up anywhere. Presumably it's commit </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">f7871fd19389c5f64f625a4389675d0740f0dfe4 ("drm/radeon: use new TTM </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">populate/dma map helper functions") instead, which makes the radeon </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">driver rely on ttm_populate_and_map_pages, which is just a stub </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">returning -ENOMEM when neither CONFIG_SWIOTLB nor </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">CONFIG_INTEL_IOMMU is </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">enabled. </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">I can see two possible solutions: </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">1. Make ttm_populate_and_map_pages and ttm_unmap_and_unpopulate_pages </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"> work without SWIOTLB / INTEL_IOMMU as well. </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">2. Make the drivers work without ttm_populate_and_map_pages and </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"> ttm_unmap_and_unpopulate_pages again in that case. </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">Solution 1 would be preferable. </blockquote></blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">Which solution do you want to go for? </blockquote></blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">well, can somebody please explain to me why those patches actually </blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"><blockquote type="cite">result in problems? </blockquote></blockquote></blockquote><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote><blockquote type="cite"><blockquote type="cite">I thought I did above... </blockquote></blockquote><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote><blockquote type="cite"><blockquote type="cite">Commit f7871fd19389c5f64f625a4389675d0740f0dfe4 made the radeon driver </blockquote></blockquote><blockquote type="cite"><blockquote type="cite">rely on ttm_populate_and_map_pages, which is implemented as: </blockquote></blockquote><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote><blockquote type="cite"><blockquote type="cite">static inline int ttm_populate_and_map_pages(struct device *dev, </blockquote></blockquote><blockquote type="cite"><blockquote type="cite">struct ttm_dma_tt *tt) </blockquote></blockquote><blockquote type="cite"><blockquote type="cite">{ </blockquote></blockquote><blockquote type="cite"><blockquote type="cite"> return -ENOMEM; </blockquote></blockquote><blockquote type="cite"><blockquote type="cite">} </blockquote></blockquote><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote><blockquote type="cite"><blockquote type="cite">when neither CONFIG_SWIOTLB nor CONFIG_INTEL_IOMMU is enabled. </blockquote></blockquote><blockquote type="cite"><blockquote type="cite">Previously, the driver worked fine without either of those enabled. </blockquote></blockquote><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote><blockquote type="cite"><blockquote type="cite"> </blockquote></blockquote><blockquote type="cite"> </blockquote><blockquote type="cite">I think the issue is why would you have both disabled [...] </blockquote>Doesn't matter. The drivers worked with both disabled before, now they don't. That's a regression. -- Earthling Michel Dänzer | <a href="http://www.amd.com">http://www.amd.com</a> Libre software enthusiast | Mesa and X developer </div></blockquote></div></div></body></html>