[PATCH] intel gpu passthrough: Expose vendor specific pci cap on host bridge.

[PATCH] intel gpu passthrough: Expose vendor specific pci cap on host bridge.

[Jean Guyader]

[-- Attachment #1: Type: text/plain, Size: 349 bytes --]


Some versions of the Windows Intel GPU driver expect the vendor
PCI capability to be there on the host bridge config space when
passing through a Intel GPU.

Signed-off-by: Jean Guyader <jean.guyader@eu.citrix.com>
---
 hw/pt-graphics.c |   51 ++++++++++++++++++++++++++++++++++++++++++++++-----
 1 files changed, 46 insertions(+), 5 deletions(-)


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-intel-gpu-passthrough-Expose-vendor-specific-pci-cap.patch --]
[-- Type: text/x-patch; name="0001-intel-gpu-passthrough-Expose-vendor-specific-pci-cap.patch", Size: 2551 bytes --]

diff --git a/hw/pt-graphics.c b/hw/pt-graphics.c
index fec7390..1120639 100644
--- a/hw/pt-graphics.c
+++ b/hw/pt-graphics.c
@@ -60,6 +60,44 @@ void igd_pci_write(PCIDevice *pci_dev, uint32_t config_addr, uint32_t val, int l
     }
 }
 
+#define PCI_INTEL_VENDOR_CAP            0x34
+#define PCI_INTEL_VENDOR_CAP_TYPE       0x09
+static uint32_t igd_pci_read_vendor_cap(PCIDevice *pci_dev, uint32_t config_addr, int len,
+                                        uint32_t val)
+{
+    struct pci_dev *pci_dev_host_bridge = pt_pci_get_dev(0, 0, 0);
+    uint32_t vendor_cap = 0;
+    uint32_t cap_type = 0;
+    uint32_t cap_size = 0;
+
+    vendor_cap = pt_pci_host_read(pci_dev_host_bridge, PCI_INTEL_VENDOR_CAP, 1);
+    if (config_addr == PCI_INTEL_VENDOR_CAP)
+    {
+        /* Only do the following for vendor specific caps (0x09) */
+        if (cap_type != PCI_INTEL_VENDOR_CAP_TYPE)
+            return 0;
+
+        if (vendor_cap == 0)
+            return 0;
+    }
+
+    cap_type = pt_pci_host_read(pci_dev_host_bridge, vendor_cap, 1);
+    cap_size = pt_pci_host_read(pci_dev_host_bridge, vendor_cap + 2, 1);
+
+    /* Remove the next capability link */
+    if (config_addr == vendor_cap + 1)
+        return 0;
+
+    if (config_addr >= vendor_cap &&
+            config_addr + len < vendor_cap + cap_size)
+    {
+        return pt_pci_host_read(pci_dev_host_bridge, config_addr, len);
+    }
+
+    /* -1, this function doesn't deal with this config space offset */
+    return -1;
+}
+
 uint32_t igd_pci_read(PCIDevice *pci_dev, uint32_t config_addr, int len)
 {
     struct pci_dev *pci_dev_host_bridge = pt_pci_get_dev(0, 0, 0);
@@ -82,14 +120,17 @@ uint32_t igd_pci_read(PCIDevice *pci_dev, uint32_t config_addr, int len)
         case 0xa4:        /* SNB: graphics base of stolen memory */
         case 0xa8:        /* SNB: base of GTT stolen memory */
             val = pt_pci_host_read(pci_dev_host_bridge, config_addr, len);
-#ifdef PT_DEBUG_PCI_CONFIG_ACCESS
-            PT_LOG_DEV(pci_dev, "addr=%x len=%x val=%x\n",
-                    config_addr, len, val);
-#endif
             break;
         default:
-            val = pci_default_read_config(pci_dev, config_addr, len);
+            val = igd_pci_read_vendor_cap(pci_dev, config_addr, len, val);
+            if (val == -1)
+                val = pci_default_read_config(pci_dev, config_addr, len);
+
     }
+#ifdef PT_DEBUG_PCI_CONFIG_ACCESS
+    PT_LOG_DEV(pci_dev, "addr=%x len=%x val=%x\n",
+            config_addr, len, val);
+#endif
     return val;
 }
 

[-- Attachment #3: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: [PATCH] intel gpu passthrough: Expose vendor specific pci cap on host bridge.

[Jean Guyader]

On 16/01 02:40, Jean Guyader wrote:
> 
> Some versions of the Windows Intel GPU driver expect the vendor
> PCI capability to be there on the host bridge config space when
> passing through a Intel GPU.
> 
> Signed-off-by: Jean Guyader <jean.guyader@eu.citrix.com>
From: Ross Philipson <Ross.Philipson@citrix.com>

Jean

Re: [PATCH] intel gpu passthrough: Expose vendor specific pci cap on host bridge.

[Ross Philipson]

> -----Original Message-----
> From: Jean Guyader [mailto:jean.guyader@eu.citrix.com]
> Sent: Monday, January 16, 2012 10:05 AM
> To: xen-devel@lists.xensource.com
> Cc: allen.m.kay@intel.com; Ian Jackson; Jean Guyader; Ross Philipson
> Subject: Re: [PATCH] intel gpu passthrough: Expose vendor specific pci
> cap on host bridge.
> 
> On 16/01 02:40, Jean Guyader wrote:
> >
> > Some versions of the Windows Intel GPU driver expect the vendor PCI
> > capability to be there on the host bridge config space when passing
> > through a Intel GPU.
> >
> > Signed-off-by: Jean Guyader <jean.guyader@eu.citrix.com>
> From: Ross Philipson <Ross.Philipson@citrix.com>
> 
> Jean
Acked-by: Ross Philipson <Ross.Philipson@citrix.com>

Re: [PATCH] intel gpu passthrough: Expose vendor specificpci cap on host bridge.

[djmagee]

This always fails because it checks cap_type before it's set.  Moving the pt_pci_host_read call before the check gets rid of these messages in qemu log:

pcilib: sysfs_read: read failed: Invalid argument

-----Original Message-----
From: xen-devel-bounces@lists.xensource.com [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Jean Guyader
Sent: Monday, January 16, 2012 9:40 AM
To: xen-devel@lists.xensource.com
Cc: Ian.Jackson@eu.citrix.com; allen.m.kay@intel.com; Jean Guyader
Subject: [Xen-devel] [PATCH] intel gpu passthrough: Expose vendor specificpci cap on host bridge.


Some versions of the Windows Intel GPU driver expect the vendor PCI capability to be there on the host bridge config space when passing through a Intel GPU.

Signed-off-by: Jean Guyader <jean.guyader@eu.citrix.com>
---
 hw/pt-graphics.c |   51 ++++++++++++++++++++++++++++++++++++++++++++++-----
 1 files changed, 46 insertions(+), 5 deletions(-)

Re: [PATCH] intel gpu passthrough: Expose vendor specificpci cap on host bridge.

[Jean Guyader]

On 17/01 02:50, djmagee@mageenet.net wrote:
> This always fails because it checks cap_type before it's set.  Moving the pt_pci_host_read call before the check gets rid of these messages in qemu log:
> 
> pcilib: sysfs_read: read failed: Invalid argument
> 

This error is due to a missing patch that should be applied before this one.
I'll resend the patch within a serie that will include the extra patch.

Thanks for your review.

Jean