9c6c51
From 8996236421b4adf090ac721e8d525953deb9687a Mon Sep 17 00:00:00 2001
9c6c51
Message-Id: <8996236421b4adf090ac721e8d525953deb9687a@dist-git>
9c6c51
From: Daniel Henrique Barboza <danielhb413@gmail.com>
9c6c51
Date: Fri, 3 May 2019 13:54:52 +0200
9c6c51
Subject: [PATCH] qemu_domain: NVLink2 bridge detection function for PPC64
9c6c51
9c6c51
The NVLink2 support in QEMU implements the detection of NVLink2
9c6c51
capable devices by verifying the attributes of the VFIO mem region
9c6c51
QEMU allocates for the NVIDIA GPUs. To properly allocate an
9c6c51
adequate amount of memLock, Libvirt needs this information before
9c6c51
a QEMU instance is even created, thus querying QEMU is not
9c6c51
possible and opening a VFIO window is too much.
9c6c51
9c6c51
An alternative is presented in this patch. Making the following
9c6c51
assumptions:
9c6c51
9c6c51
- if we want GPU RAM to be available in the guest, an NVLink2 bridge
9c6c51
must be passed through;
9c6c51
9c6c51
- an unknown PCI device can be classified as a NVLink2 bridge
9c6c51
if its device tree node has 'ibm,gpu', 'ibm,nvlink',
9c6c51
'ibm,nvlink-speed' and 'memory-region'.
9c6c51
9c6c51
This patch introduces a helper called @ppc64VFIODeviceIsNV2Bridge
9c6c51
that checks the device tree node of a given PCI device and
9c6c51
check if it meets the criteria to be a NVLink2 bridge. This
9c6c51
new function will be used in a follow-up patch that, using the
9c6c51
first assumption, will set up the rlimits of the guest
9c6c51
accordingly.
9c6c51
9c6c51
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
9c6c51
(cherry picked from commit cc9f03801c2618102027ddc4a988193ae290b651)
9c6c51
9c6c51
https: //bugzilla.redhat.com/show_bug.cgi?id=1505998
9c6c51
Signed-off-by: Erik Skultety <eskultet@redhat.com>
9c6c51
Message-Id: <99efa386adcf68789b7c8cffda2ef24f5f47f0c4.1556884443.git.eskultet@redhat.com>
9c6c51
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
9c6c51
---
9c6c51
 src/qemu/qemu_domain.c | 30 ++++++++++++++++++++++++++++++
9c6c51
 1 file changed, 30 insertions(+)
9c6c51
9c6c51
diff --git a/src/qemu/qemu_domain.c b/src/qemu/qemu_domain.c
9c6c51
index f91de0b743..a8bc618389 100644
9c6c51
--- a/src/qemu/qemu_domain.c
9c6c51
+++ b/src/qemu/qemu_domain.c
9c6c51
@@ -9805,6 +9805,36 @@ qemuDomainUpdateCurrentMemorySize(virQEMUDriverPtr driver,
9c6c51
 }
9c6c51
 
9c6c51
 
9c6c51
+/**
9c6c51
+ * ppc64VFIODeviceIsNV2Bridge:
9c6c51
+ * @device: string with the PCI device address
9c6c51
+ *
9c6c51
+ * This function receives a string that represents a PCI device,
9c6c51
+ * such as '0004:04:00.0', and tells if the device is a NVLink2
9c6c51
+ * bridge.
9c6c51
+ */
9c6c51
+static ATTRIBUTE_UNUSED bool
9c6c51
+ppc64VFIODeviceIsNV2Bridge(const char *device)
9c6c51
+{
9c6c51
+    const char *nvlink2Files[] = {"ibm,gpu", "ibm,nvlink",
9c6c51
+                                  "ibm,nvlink-speed", "memory-region"};
9c6c51
+    size_t i;
9c6c51
+
9c6c51
+    for (i = 0; i < ARRAY_CARDINALITY(nvlink2Files); i++) {
9c6c51
+        VIR_AUTOFREE(char *) file = NULL;
9c6c51
+
9c6c51
+        if ((virAsprintf(&file, "/sys/bus/pci/devices/%s/of_node/%s",
9c6c51
+                         device, nvlink2Files[i])) < 0)
9c6c51
+            return false;
9c6c51
+
9c6c51
+        if (!virFileExists(file))
9c6c51
+            return false;
9c6c51
+    }
9c6c51
+
9c6c51
+    return true;
9c6c51
+}
9c6c51
+
9c6c51
+
9c6c51
 /**
9c6c51
  * getPPC64MemLockLimitBytes:
9c6c51
  * @def: domain definition
9c6c51
-- 
9c6c51
2.21.0
9c6c51