On Fri, 03 Jun 2022 14:14:59 +0000 Simon Ser <contact@xxxxxxxxxxx> wrote: > Hi, > > Please, read this thread: > https://lists.freedesktop.org/archives/dri-devel/2020-March/thread.html#259615 > > It has a lot of information about the pitfalls of cursor hotspot and > other things done by VM software. > > In particular: since the driver will ignore the KMS cursor plane > position set by user-space, I don't think it's okay to just expose > without opt-in from user-space (e.g. with a DRM_CLIENT_CAP). > > cc wayland-devel and Pekka for user-space feedback. > > On Thursday, June 2nd, 2022 at 17:42, Zack Rusin <zack@xxxxxxx> wrote: > > > - all userspace code needs to hardcore a list of drivers which require > > hotspots because there's no way to query from drm "does this driver > > require hotspot" > > Can you elaborate? I'm not sure I understand what you mean here. > Hi Zack and everyone, I would like to try to reboot this discussion and explain where I come from. Maybe I have misunderstood something. I have no fundamental objection to adding KMS properties for cursor hotspot, but I do want to make sure the design is fully thought out and not simply copied from legacy KMS, because atomic KMS with universal planes is not like legacy KMS. To my understanding from the past discussions, the fundamental reason why you are proposing hotspot properties is that when one is running a desktop inside a VM and looking at it through a VM viewer application, the pointer cursor is misplaced: the hotspot is not where the end user sees it. This you never mentioned in any of the patches nor in the cover letter. I can only guess that this misplacement could be the reason why some display servers have deny-listed paravirtualized drivers. While your goal is to get paravirtualized drivers out of the deny-lists, we must first understand why they got there in the first place. Why are pointer cursors misplaced on paravirtualized drivers? It is because the paravirtualized drivers or VM viewers do *not* place the cursor plane at the CRTC_X, CRTC_Y position in the guest CRTC area. This is obvious: if CRTC_X, CRTC_Y were honoured, there would be no misplacement. Instead, the VM stack plays clever tricks with cursor planes. I have understood only one of those tricks, and it goes something like this. To improve hand-eye coordination, that is to reduce the hand-to-eye response time a.k.a latency, the VM guest KMS driver relays the cursor plane separately to the VM viewer application. The VM viewer application presents the cursor plane content by pushing them to the host window system as the pointer cursor. This means the host window system will be autonomously moving the cursor plane image around, completely disregarding what the guest KMS client programmed into CRTC_X, CRTC_Y. When this works, it is a huge improvement in user experience. I believe this is called "seamless pointer" or such. When it doesn't work, the cursor is misplaced or even completely arbitrary guest windows start flying around as if they were the pointer cursor. In legacy KMS, cursors have always been very special and had their own special UAPI with even hotspot information. There paravirtualized drivers got away with these clever tricks because no-one bothered putting anything but actual cursor images on the (one) cursor plane. Then comes KMS universal planes concept. All planes are assumed roughly equal, apart from what KMS properties tell userspace about them. The plane type primary/overlay/cursor is still kept, but only because it helps userspace find a KMS configuration that the driver accepts at all. Hardware may not be able to light up a CRTC without at least one primary plane, for example. Atomic KMS requires universal planes. The atomic KMS UAPI contract says, that a plane is positioned at CRTC_X, CRTC_Y inside the respective CRTC. I do not know of any documented exceptions to this. Therefore, an atomic driver that does not show a cursor plane at the programmed CRTC_X, CRTC_Y is violating the UAPI contract. See https://www.kernel.org/doc/html/latest/gpu/drm-kms.html#standard-plane-properties and how "cursor" plane type makes no exceptions, and how CRTC_X and CRTC_Y are defined. Also note that all hardware drivers and VKMS apparently follow the contract. Given this UAPI contract, it is very easy for userspace to make the conclusion that a cursor plane is just another plane it can use for whatever it wants. Weston and wlroots indeed leverage this, putting also normal windows and other stuff to the cursor plane when they happen to fit. If a paravirtualized driver commandeers the cursor plane display position, a possible result is that arbitrary windows start flying around unexpectedly. Therefore we have two problems: - paravirtualized drivers commandeering the cursor plane - VM software not able to implement "seamless pointer" correctly To my understanding, this patch series concerns only the latter problem but not the former problem. I believe the solutions to both are related and need to be considered together. How do we stop paravirtualized drivers from commandeering their cursor plane when guest userspace does not expect it? How do we still make "seamless pointer" possible when the guest userspace is prepared for cursor plane commandeering? These are the questions that need answers. To me, getting paravirtualized drivers off display server deny-lists is a consequence, a secondary goal. The primary goal must be to fix why the drivers ended up in deny-lists in the first place (and I am only assuming that this is he reason, so maybe there are other reasons?). I believe the solution has two parts: - The guest KMS driver needs to know whether the guest userspace is prepared for the cursor plane being commandeered. If userspace does not indicate it is prepared for it, commandeering must not happen. - Cursor hotspot needs new KMS properties, and a KMS client must set them if the KMS client indicates it is prepared for cursor plane commandeering. How to exactly do those can be discussed after we can agree on what problems we are solving. There are further problems with cursor plane commandeering. The 2020 email thread Simon linked to discusses the problem of pointer devices: if VM guest userspace takes pointer input from multiple sources, how will the VM stack know which virtual input device, if any, should drive the cursor plane position? To me the answer to this question seems it could be intimately tied to the first problem: commandeering the cursor plane is allowed only if guest userspace tells the guest KMS driver which input device the cursor plane shall be related to. If no input device is indicated, then commandeering must not happen. I can understand if people do not want to tackle this question, because it probably has not been a problem yet. Ignoring it would be unfortunate though, because that would seem to imply that VM software still needs to keep some heuristics for when commandeering the cursor plane is desired. I could also talk about multiple cursors, but I guess that goes really too far. Which root problems do you want to solve exactly? Thanks, pq
Attachment:
pgpQUVt0W2nbz.pgp
Description: OpenPGP digital signature
