linux-kernel - [PATCH v5] PCI: vmd: Add the module param to adjust MSI mode

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <20230420070914.1383918-1-korantwork@gmail.com>
Date:   Thu, 20 Apr 2023 15:09:14 +0800
From:   korantwork@...il.com
To:     helgaas@...nel.org, nirmal.patel@...ux.intel.com,
        kbusch@...nel.org, jonathan.derrick@...ux.dev,
        lpieralisi@...nel.org
Cc:     linux-pci@...r.kernel.org, linux-kernel@...r.kernel.org,
        Xinghui Li <korantli@...cent.com>
Subject: [PATCH v5] PCI: vmd: Add the module param to adjust MSI mode

From: Xinghui Li <korantli@...cent.com>

In the past, the vmd MSI mode can only be adjusted by configuring
vmd_ids table. This patch adds another way to adjust MSI mode by
adjusting module parameter, which allows users easier to adjust the vmd
according to the I/O scenario without rebuilding driver.

- "disable_msi_bypass=0 or other values":
  Under normal circumstances, we recommend enable the VMD MSI-X bypass
  feature, which improves interrupt handling performance by avoiding
  the VMD MSI-X domain interrupt handler.

- "disable_msi_bypass=1":
  Use this when multiple NVMe devices are mounted on the same PCIe
  node with a high volume of 4K random I/O. It mitigates excessive
  pressure on the PCIe node caused by numerous interrupts from NVMe
  drives, resulting in improved I/O performance. Such as:

  In FIO 4K random test when 4 NVME(Gen4) mounted on the same PCIE port:
    - Enable bypass: read: IOPS=562k, BW=2197MiB/s, io=644GiB
    - Disable bypass: read: IOPS=1144k, BW=4470MiB/s, io=1310GiB

As not all devices support VMD MSI-X bypass, this parameter is
only applicable to devices that support the bypass function and
have already enabled it, such as VMD_28C0. Besides, this parameter
does not affect the MSI-X working mode in guest.

Signed-off-by: Xinghui Li <korantli@...cent.com>
---
 drivers/pci/controller/vmd.c | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/drivers/pci/controller/vmd.c b/drivers/pci/controller/vmd.c
index 990630ec57c6..8ee673810cbf 100644
--- a/drivers/pci/controller/vmd.c
+++ b/drivers/pci/controller/vmd.c
@@ -34,6 +34,20 @@
 #define MB2_SHADOW_OFFSET	0x2000
 #define MB2_SHADOW_SIZE		16
 
+/*
+ * The VMD disable_msi_bypass module parameter provides the alternative
+ * way to adjust MSI mode when loading vmd.ko. This parameter is only applicable
+ * to devices that both support and have enabled bypass, such as VMD_28C0.
+ * Besides, it does not affect MSI-X mode in the guest.
+ *
+ * 1: disable MSI-X bypass
+ * other values: not disable MSI-X bypass
+ */
+static int disable_msi_bypass;
+module_param(disable_msi_bypass, int, 0444);
+MODULE_PARM_DESC(disable_msi_bypass, "Whether to disable MSI-X bypass function.\n"
+	"\t\t  Only effective on the device supporting bypass, such as 28C0.");
+
 enum vmd_features {
 	/*
 	 * Device may contain registers which hint the physical location of the
@@ -875,6 +889,7 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
 			return ret;
 
 		vmd_set_msi_remapping(vmd, true);
+		dev_info(&vmd->dev->dev, "init vmd with remapping MSI-X\n");
 
 		ret = vmd_create_irq_domain(vmd);
 		if (ret)
@@ -887,6 +902,7 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
 		irq_domain_update_bus_token(vmd->irq_domain, DOMAIN_BUS_VMD_MSI);
 	} else {
 		vmd_set_msi_remapping(vmd, false);
+		dev_info(&vmd->dev->dev, "init vmd with bypass MSI-X\n");
 	}
 
 	pci_add_resource(&resources, &vmd->resources[0]);
@@ -955,6 +971,17 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
 	return 0;
 }
 
+static void vmd_config_msi_bypass_param(unsigned long *features)
+{
+	/*
+	 * Not every VMD device supports and enables bypass MSI-X.
+	 * Make sure current device has the bypass flag set.
+	 */
+	if (disable_msi_bypass == 1 &&
+	  *features & VMD_FEAT_CAN_BYPASS_MSI_REMAP)
+		*features &= ~(VMD_FEAT_CAN_BYPASS_MSI_REMAP);
+}
+
 static int vmd_probe(struct pci_dev *dev, const struct pci_device_id *id)
 {
 	unsigned long features = (unsigned long) id->driver_data;
@@ -984,6 +1011,8 @@ static int vmd_probe(struct pci_dev *dev, const struct pci_device_id *id)
 	if (err < 0)
 		goto out_release_instance;
 
+	vmd_config_msi_bypass_param(&features);
+
 	vmd->cfgbar = pcim_iomap(dev, VMD_CFGBAR, 0);
 	if (!vmd->cfgbar) {
 		err = -ENOMEM;
-- 
2.31.1