Re: [PATCH 2/2] vfio/pci: Remove console drivers

From: Thomas Zimmermann
Date: Wed Jun 08 2022 - 07:12:32 EST


Hi Alex

Am 06.06.22 um 19:53 schrieb Alex Williamson:
Console drivers can create conflicts with PCI resources resulting in
userspace getting mmap failures to memory BARs. This is especially evident
when trying to re-use the system primary console for userspace drivers.
Attempt to remove all nature of conflicting drivers as part of our VGA
initialization.

First a dumb question about your use case. You want to assign a PCI graphics card to a virtual machine and need to remove the generic driver from the framebuffer?


Reported-by: Laszlo Ersek <lersek@xxxxxxxxxx>
Tested-by: Laszlo Ersek <lersek@xxxxxxxxxx>
Suggested-by: Gerd Hoffmann <kraxel@xxxxxxxxxx>
Signed-off-by: Alex Williamson <alex.williamson@xxxxxxxxxx>
---
drivers/vfio/pci/vfio_pci_core.c | 17 +++++++++++++++++
1 file changed, 17 insertions(+)

diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
index a0d69ddaf90d..e0cbcbc2aee1 100644
--- a/drivers/vfio/pci/vfio_pci_core.c
+++ b/drivers/vfio/pci/vfio_pci_core.c
@@ -13,6 +13,7 @@
#include <linux/device.h>
#include <linux/eventfd.h>
#include <linux/file.h>
+#include <linux/fb.h>
#include <linux/interrupt.h>
#include <linux/iommu.h>
#include <linux/module.h>
@@ -29,6 +30,8 @@
#include <linux/vfio_pci_core.h>
+#include <drm/drm_aperture.h>
+
#define DRIVER_AUTHOR "Alex Williamson <alex.williamson@xxxxxxxxxx>"
#define DRIVER_DESC "core driver for VFIO based PCI devices"
@@ -1793,6 +1796,20 @@ static int vfio_pci_vga_init(struct vfio_pci_core_device *vdev)
if (!vfio_pci_is_vga(pdev))
return 0;
+#if IS_REACHABLE(CONFIG_DRM)
+ drm_aperture_detach_platform_drivers(pdev);
+#endif
+
+#if IS_REACHABLE(CONFIG_FB)
+ ret = remove_conflicting_pci_framebuffers(pdev, vdev->vdev.ops->name);
+ if (ret)
+ return ret;
+#endif
+
+ ret = vga_remove_vgacon(pdev);
+ if (ret)
+ return ret;
+

You shouldn't have to copy any of the implementation of the aperture helpers.

If you call drm_aperture_remove_conflicting_pci_framebuffers() it should work correctly. The only reason why it requires a DRM driver structure as second argument is for the driver's name. [1] And that name is only used for printing an info message. [2]

The plan forward would be to drop patch 1 entirely.

For patch 2, the most trivial workaround is to instanciate struct drm_driver here and set the name field to 'vdev->vdev.ops->name'. In the longer term, the aperture helpers will be moved out of DRM and into a more prominent location. That workaround will be cleaned up then.

Alternatively, drm_aperture_remove_conflicting_pci_framebuffers() could be changed to accept the name string as second argument, but that's quite a bit of churn within the DRM code.

Best regards
Thomas

[1] https://elixir.bootlin.com/linux/v5.18.2/source/drivers/gpu/drm/drm_aperture.c#L347
[2] https://elixir.bootlin.com/linux/v5.18.2/source/drivers/video/fbdev/core/fbmem.c#L1570


ret = vga_client_register(pdev, vfio_pci_set_decode);
if (ret)
return ret;



--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Ivo Totev

Attachment: OpenPGP_signature
Description: OpenPGP digital signature