Re: [PATCH 6.0 20/20] fbdev/core: Remove remove_conflicting_pci_framebuffers()

From: Thomas Zimmermann
Date: Tue Nov 01 2022 - 06:34:47 EST


(cc: Alex Williamson)

Hi

Am 01.11.22 um 09:42 schrieb Boris V.:
On 24/10/2022 13:31, Greg Kroah-Hartman wrote:
From: Thomas Zimmermann <tzimmermann@xxxxxxx>

commit 9d69ef1838150c7d87afc1a87aa658c637217585 upstream.

Remove remove_conflicting_pci_framebuffers() and implement similar
functionality in aperture_remove_conflicting_pci_device(), which was
the only caller. Removes an otherwise unused interface and streamlines
the aperture helper. No functional changes.

Signed-off-by: Thomas Zimmermann <tzimmermann@xxxxxxx>
Reviewed-by: Javier Martinez Canillas <javierm@xxxxxxxxxx>
Link: https://patchwork.freedesktop.org/patch/msgid/20220718072322.8927-5-tzimmermann@xxxxxxx
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
  drivers/video/aperture.c         |   30 ++++++++++++++----------
  drivers/video/fbdev/core/fbmem.c |   48 ---------------------------------------
  include/linux/fb.h               |    2 -
  3 files changed, 18 insertions(+), 62 deletions(-)

--- a/drivers/video/aperture.c
+++ b/drivers/video/aperture.c
@@ -335,30 +335,36 @@ EXPORT_SYMBOL(aperture_remove_conflictin
   */
  int aperture_remove_conflicting_pci_devices(struct pci_dev *pdev, const char *name)
  {
+    bool primary = false;
      resource_size_t base, size;
      int bar, ret;
-    /*
-     * WARNING: Apparently we must kick fbdev drivers before vgacon,
-     * otherwise the vga fbdev driver falls over.
-     */
-#if IS_REACHABLE(CONFIG_FB)
-    ret = remove_conflicting_pci_framebuffers(pdev, name);
-    if (ret)
-        return ret;
+#ifdef CONFIG_X86
+    primary = pdev->resource[PCI_ROM_RESOURCE].flags & IORESOURCE_ROM_SHADOW;
  #endif
-    ret = vga_remove_vgacon(pdev);
-    if (ret)
-        return ret;
      for (bar = 0; bar < PCI_STD_NUM_BARS; ++bar) {
          if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM))
              continue;
+
          base = pci_resource_start(pdev, bar);
          size = pci_resource_len(pdev, bar);
-        aperture_detach_devices(base, size);
+        ret = aperture_remove_conflicting_devices(base, size, primary, name);
+        if (ret)
+            break;
      }
+    if (ret)
+        return ret;
+
+    /*
+     * WARNING: Apparently we must kick fbdev drivers before vgacon,
+     * otherwise the vga fbdev driver falls over.
+     */
+    ret = vga_remove_vgacon(pdev);
+    if (ret)
+        return ret;
+
      return 0;
  }
--- a/drivers/video/fbdev/core/fbmem.c
+++ b/drivers/video/fbdev/core/fbmem.c
@@ -1788,54 +1788,6 @@ int remove_conflicting_framebuffers(stru
  EXPORT_SYMBOL(remove_conflicting_framebuffers);
  /**
- * remove_conflicting_pci_framebuffers - remove firmware-configured framebuffers for PCI devices
- * @pdev: PCI device
- * @name: requesting driver name
- *
- * This function removes framebuffer devices (eg. initialized by firmware)
- * using memory range configured for any of @pdev's memory bars.
- *
- * The function assumes that PCI device with shadowed ROM drives a primary
- * display and so kicks out vga16fb.
- */
-int remove_conflicting_pci_framebuffers(struct pci_dev *pdev, const char *name)
-{
-    struct apertures_struct *ap;
-    bool primary = false;
-    int err, idx, bar;
-
-    for (idx = 0, bar = 0; bar < PCI_STD_NUM_BARS; bar++) {
-        if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM))
-            continue;
-        idx++;
-    }
-
-    ap = alloc_apertures(idx);
-    if (!ap)
-        return -ENOMEM;
-
-    for (idx = 0, bar = 0; bar < PCI_STD_NUM_BARS; bar++) {
-        if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM))
-            continue;
-        ap->ranges[idx].base = pci_resource_start(pdev, bar);
-        ap->ranges[idx].size = pci_resource_len(pdev, bar);
-        pci_dbg(pdev, "%s: bar %d: 0x%lx -> 0x%lx\n", __func__, bar,
-            (unsigned long)pci_resource_start(pdev, bar),
-            (unsigned long)pci_resource_end(pdev, bar));
-        idx++;
-    }
-
-#ifdef CONFIG_X86
-    primary = pdev->resource[PCI_ROM_RESOURCE].flags &
-                    IORESOURCE_ROM_SHADOW;
-#endif
-    err = remove_conflicting_framebuffers(ap, name, primary);
-    kfree(ap);
-    return err;
-}
-EXPORT_SYMBOL(remove_conflicting_pci_framebuffers);
-
-/**
   *    register_framebuffer - registers a frame buffer device
   *    @fb_info: frame buffer info structure
   *
--- a/include/linux/fb.h
+++ b/include/linux/fb.h
@@ -615,8 +615,6 @@ extern ssize_t fb_sys_write(struct fb_in
  /* drivers/video/fbmem.c */
  extern int register_framebuffer(struct fb_info *fb_info);
  extern void unregister_framebuffer(struct fb_info *fb_info);
-extern int remove_conflicting_pci_framebuffers(struct pci_dev *pdev,
-                           const char *name);
  extern int remove_conflicting_framebuffers(struct apertures_struct *a,
                         const char *name, bool primary);
  extern int fb_prepare_logo(struct fb_info *fb_info, int rotate);





Hello,

this patch seems to disable console/framebuffer when vfio-pci is used.
I hava 2 nvidia GPUs one is used for host and other is passed through to VM.

Vfio uses this helper to unload the driver before passing it to a VM AFAIU. But unless you're using nouveau, you're on your own.

Best regards
Thomas

Now after this patch, when vfio-pci module is loaded with parameter ids=10de:2486,10de:228b,
console is lost/frozen, last message is that vfio-pci module was loaded and then there is no more output.
This PCI IDs (10de:2486,10de:228b) are for secondary GPU, primary/boot GPU is used for host and boot messages are displayed on primary/boot GPU.

Using dmesg I see this messages after vfio-pci is loaded:

[    3.993601] VFIO - User Level meta-driver version: 0.3
[    4.020239] Console: switching to colour dummy device 80x25
[    4.020335] vfio-pci 0000:1a:00.0: vgaarb: changed VGA decodes: olddecodes=io+mem,decodes=none:owns=none
[    4.020722] vfio_pci: add [10de:2486[ffffffff:ffffffff]] class 0x000000/00000000
[    4.116616] vfio_pci: add [10de:228b[ffffffff:ffffffff]] class 0x000000/00000000

I guess the problem here is "Console: switching to colour dummy device 80x25", but I don't know why this happens.
Last working kernel is 6.0.3, after upgrading to 6.0.4 (and 6.0.5, 6.0.6), console is no longer working.
By git bisecting it seems bad commit is af9ac541e88390d97b01d5e8c77309d2637c1d4c.


--
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