Re: [PATCH 0/9] drm/xen-front: Add support for Xen PV display frontend

From: Oleksandr Andrushchenko
Date: Tue Feb 27 2018 - 07:40:57 EST


Please find some more clarifications on VirtIO use with Xen
(I would like to thank Xen community for helping with this)

1. Possible security issues - VirtIO devices are PCI bus masters, thus
allowing real device (running, for example, in untrusted driver domain)
to get control over guest's memory by writing to its memory

2. VirtIO currently uses GFNs written into the shared ring, without Xen
grants support. This will require generic grant-mapping/sharing layer
to be added to VirtIO.

3. VirtIO requires QEMU PCI emulation for setting up a device. Xen PV (and PVH)
domains don't use QEMU for platform emulation in order to reduce attack surface.
(PVH is in the process of gaining PCI config space emulation though, but it is
optional, not a requirement)

4. Most of the PV drivers a guest uses at the moment are Xen PV drivers, e.g. net,
block, console, so only virtio-gpu will require QEMU to run.
Although this use case would work on x86 it will require additional changes
to get this running on ARM, which is my target platform.

Thank you,
Oleksandr

On 02/26/2018 10:21 AM, Oleksandr Andrushchenko wrote:
**

*Hi, all!*

*

Last *Friday* some concerns on #dri-devel were raised wrt "yet

another driver" for Xen and why not virtio-gpu. Let me highlight

on why we need a new paravirtualized driver for Xen and why we

can't just use virtio. Hope this helps the communities (both Xen

and DRI) to have better understanding of this work and our motivation.


Disclaimer: some or all of the below may sound weak argument or

not 100% correct, so any help on clarifying the below is more

than welcome ;)


1. First of all, we are targeting ARM embedded use-cases and for

ARM we do not use QEMU [1]: "...Xen on ARM is not just a straight

1:1 port of x86 Xen... Xen on ARM does not need QEMU because it does

not do any emulation. It accomplishes the goal by exploiting

virtualization support in hardware as much as possible and using

paravirtualized interfaces for IO."


That being said it is still possible to run virtio-gpu and Xen+QEMU: [2]


In this case QEMU can be used for device virtualization, e.g. network,

block, console. But these already exist as Xen para-virtualized drivers

again eliminating the need for QEMU: typical ARM system runs para-virtualized

drivers for network, block, console etc.


2. virtio-gpu requires PCI/MMIO emulation

virtio-gpu (virtio-gpu-pci) require virtio-pci, but para-virtualized device

drivers do not need this.


3. No need for 3d/virgl.

There are use-cases which either do not use OpenGL at all or will use

custom virtualization solutions allowing sharing of a real GPU with guest,

e.g. vGPU approach.


4. More freedom for buffer allocation.

As of now virtio-gpu is only capable of allocating buffers via TTM, while

there are use-cases where we need to have more freedom:

for systems which do not provide IOMMU support, but having specific

requirements for display buffers, it is possible to allocate such buffers

at backend side and share those with the frontend driver.

For example, if host domain is 1:1 mapped and has DRM/GPU hardware expecting

physically contiguous memory (in PA, not IPA), this allows implementing

zero-copying use-cases.


5. Zero-copying support at backend side

Having native Xen implementation allows implementing zero-copying use-cases

on backend side with the help of supporting driver DRM driver [3] which we

hope to upstream as well (it is not yet ready in terms of code cleanup).


6. QEMU backends for virtio-gpu cannot be used as is, e.g. guest displays

could be just a part of the final user experience. Thus, a QEMU backend

must be modified to interact, for example, with Automotive Grade Linux

display manager. So, QEMU part needs modifications.

In our use-case we have a backend which supports multi-touch and guest

display(s) and running either as a weston client (which is not supported

by QEMU at the moment?) or KMS/DRM client. This allows us to enable much

more use-cases**without the need to run QEMU.

*

*Thank you,*

**Oleksandr Andrushchenko*
*

*
*

*[1] https://wiki.xen.org/wiki/Xen_ARM_with_Virtualization_Extensions_whitepaper*

*

[2] https://elinux.org/R-Car/Virtualization

[3] https://github.com/xen-troops/linux/blob/ces2018/drivers/gpu/drm/xen/xen_drm_zcopy_drv.c


*


On 02/21/2018 10:03 AM, Oleksandr Andrushchenko wrote:
From: Oleksandr Andrushchenko <oleksandr_andrushchenko@xxxxxxxx>

Hello!

This patch series adds support for Xen [1] para-virtualized
frontend display driver. It implements the protocol from
include/xen/interface/io/displif.h [2].
Accompanying backend [3] is implemented as a user-space application
and its helper library [4], capable of running as a Weston client
or DRM master.
Configuration of both backend and frontend is done via
Xen guest domain configuration options [5].

*******************************************************************************
* Driver limitations
*******************************************************************************
 1. Configuration options 1.1 (contiguous display buffers) and 2 (backend
ÂÂÂÂ allocated buffers) below are not supported at the same time.

 2. Only primary plane without additional properties is supported.

 3. Only one video mode supported which resolution is configured via XenStore.

 4. All CRTCs operate at fixed frequency of 60Hz.

*******************************************************************************
* Driver modes of operation in terms of display buffers used
*******************************************************************************
 Depending on the requirements for the para-virtualized environment, namely
 requirements dictated by the accompanying DRM/(v)GPU drivers running in both
 host and guest environments, number of operating modes of para-virtualized
 display driver are supported:
ÂÂ - display buffers can be allocated by either frontend driver or backend
ÂÂ - display buffers can be allocated to be contiguous in memory or not

 Note! Frontend driver itself has no dependency on contiguous memory for
ÂÂÂÂÂÂÂ its operation.

*******************************************************************************
* 1. Buffers allocated by the frontend driver.
*******************************************************************************

 The below modes of operation are configured at compile-time via
 frontend driver's kernel configuration.

 1.1. Front driver configured to use GEM CMA helpers
ÂÂÂÂÂÂ This use-case is useful when used with accompanying DRM/vGPU driver in
ÂÂÂÂÂÂ guest domain which was designed to only work with contiguous buffers,
ÂÂÂÂÂÂ e.g. DRM driver based on GEM CMA helpers: such drivers can only import
ÂÂÂÂÂÂ contiguous PRIME buffers, thus requiring frontend driver to provide
ÂÂÂÂÂÂ such. In order to implement this mode of operation para-virtualized
ÂÂÂÂÂÂ frontend driver can be configured to use GEM CMA helpers.

 1.2. Front driver doesn't use GEM CMA
ÂÂÂÂÂÂ If accompanying drivers can cope with non-contiguous memory then, to
ÂÂÂÂÂÂ lower pressure on CMA subsystem of the kernel, driver can allocate
ÂÂÂÂÂÂ buffers from system memory.

 Note! If used with accompanying DRM/(v)GPU drivers this mode of operation
ÂÂÂ may require IOMMU support on the platform, so accompanying DRM/vGPU
ÂÂÂ hardware can still reach display buffer memory while importing PRIME
ÂÂÂ buffers from the frontend driver.

*******************************************************************************
* 2. Buffers allocated by the backend
*******************************************************************************

 This mode of operation is run-time configured via guest domain configuration
 through XenStore entries.

 For systems which do not provide IOMMU support, but having specific
 requirements for display buffers it is possible to allocate such buffers
 at backend side and share those with the frontend.
 For example, if host domain is 1:1 mapped and has DRM/GPU hardware expecting
 physically contiguous memory, this allows implementing zero-copying
 use-cases.


I would like to thank at least, but not at last the following
people/communities who helped this driver to happen ;)

1. My team at EPAM for continuous support
2. Xen community for answering tons of questions on different
modes of operation of the driver with respect to virtualized
environment.
3. Rob Clark for "GEM allocation for para-virtualized DRM driver" [6]
4. Maarten Lankhorst for "Atomic driver and old remove FB behavior" [7]
5. Ville SyrjÃlà for "Questions on page flips and atomic modeset" [8]

Thank you,
Oleksandr Andrushchenko

P.S. There are two dependencies for this driver limiting some of the
use-cases which are on review now:
1. "drm/simple_kms_helper: Add {enable|disable}_vblank callback support" [9]
2. "drm/simple_kms_helper: Fix NULL pointer dereference with no active CRTC" [10]

[1] https://wiki.xen.org/wiki/Paravirtualization_(PV)#PV_IO_Drivers
[2] https://elixir.bootlin.com/linux/v4.16-rc2/source/include/xen/interface/io/displif.h
[3] https://github.com/xen-troops/displ_be
[4] https://github.com/xen-troops/libxenbe
[5] https://xenbits.xen.org/gitweb/?p=xen.git;a=blob;f=docs/man/xl.cfg.pod.5.in;h=a699367779e2ae1212ff8f638eff0206ec1a1cc9;hb=refs/heads/master#l1257
[6] https://lists.freedesktop.org/archives/dri-devel/2017-March/136038.html
[7] https://www.spinics.net/lists/dri-devel/msg164102.html
[8] https://www.spinics.net/lists/dri-devel/msg164463.html
[9] https://patchwork.freedesktop.org/series/38073/
[10] https://patchwork.freedesktop.org/series/38139/

Oleksandr Andrushchenko (9):
ÂÂ drm/xen-front: Introduce Xen para-virtualized frontend driver
ÂÂ drm/xen-front: Implement Xen bus state handling
ÂÂ drm/xen-front: Read driver configuration from Xen store
ÂÂ drm/xen-front: Implement Xen event channel handling
ÂÂ drm/xen-front: Implement handling of shared display buffers
ÂÂ drm/xen-front: Introduce DRM/KMS virtual display driver
ÂÂ drm/xen-front: Implement KMS/connector handling
ÂÂ drm/xen-front: Implement GEM operations
ÂÂ drm/xen-front: Implement communication with backend

 drivers/gpu/drm/Kconfig | 2 +
 drivers/gpu/drm/Makefile | 1 +
 drivers/gpu/drm/xen/Kconfig | 30 ++
 drivers/gpu/drm/xen/Makefile | 17 +
 drivers/gpu/drm/xen/xen_drm_front.c | 712 ++++++++++++++++++++++++++++
 drivers/gpu/drm/xen/xen_drm_front.h | 154 ++++++
 drivers/gpu/drm/xen/xen_drm_front_cfg.c | 84 ++++
 drivers/gpu/drm/xen/xen_drm_front_cfg.h | 45 ++
 drivers/gpu/drm/xen/xen_drm_front_conn.c | 125 +++++
 drivers/gpu/drm/xen/xen_drm_front_conn.h | 35 ++
 drivers/gpu/drm/xen/xen_drm_front_drv.c | 294 ++++++++++++
 drivers/gpu/drm/xen/xen_drm_front_drv.h | 73 +++
 drivers/gpu/drm/xen/xen_drm_front_evtchnl.c | 399 ++++++++++++++++
 drivers/gpu/drm/xen/xen_drm_front_evtchnl.h | 89 ++++
 drivers/gpu/drm/xen/xen_drm_front_gem.c | 360 ++++++++++++++
 drivers/gpu/drm/xen/xen_drm_front_gem.h | 46 ++
 drivers/gpu/drm/xen/xen_drm_front_gem_cma.c | 93 ++++
 drivers/gpu/drm/xen/xen_drm_front_kms.c | 299 ++++++++++++
 drivers/gpu/drm/xen/xen_drm_front_kms.h | 30 ++
 drivers/gpu/drm/xen/xen_drm_front_shbuf.c | 430 +++++++++++++++++
 drivers/gpu/drm/xen/xen_drm_front_shbuf.h | 80 ++++
 21 files changed, 3398 insertions(+)
 create mode 100644 drivers/gpu/drm/xen/Kconfig
 create mode 100644 drivers/gpu/drm/xen/Makefile
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front.c
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front.h
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_cfg.c
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_cfg.h
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_conn.c
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_conn.h
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_drv.c
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_drv.h
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_evtchnl.c
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_evtchnl.h
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_gem.c
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_gem.h
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_gem_cma.c
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_kms.c
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_kms.h
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_shbuf.c
 create mode 100644 drivers/gpu/drm/xen/xen_drm_front_shbuf.h