tleleu / rpms / kernel

Forked from rpms/kernel 3 years ago
Clone
Blob Blame History Raw
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Laura Abbott <labbott@redhat.com>
Date: Mon, 20 May 2019 22:21:02 -0400
Subject: [PATCH] iommu/arm-smmu: workaround DMA mode issues

Message-id: <20190520222102.19488-1-labbott@redhat.com>
Patchwork-id: 259215
O-Subject: [ARK INTERNAL PATCH] iommu/arm-smmu: workaround DMA mode issues
Bugzilla:
RH-Acked-by: Mark Langsdorf <mlangsdo@redhat.com>
RH-Acked-by: Mark Salter <msalter@redhat.com>

From: Mark Salter <msalter@redhat.com>

Rebased for v5.2-rc1

	Bugzilla: 1652259
	Brew: https://brewweb.engineering.redhat.com/brew/taskinfo?taskID=19244562
	Upstream status: RHEL only.

	rhel8 commit 65feb1ed0ec9a088a63a90d46c0f7563ac96ad0f
	Author: Mark Salter <msalter@redhat.com>
	Date:   Wed Nov 21 17:15:59 2018 +0100

	    [iommu] iommu/arm-smmu: workaround DMA mode issues

	    Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1624077
	    Brew: https://brewweb.engineering.redhat.com/brew/taskinfo?taskID=18112820
	    Testing: Verified iommu.passthrough=1 no longer needed on gigabyte platforms.
	    Upstream Status: RHEL-only

	    In RHEL_ALT 7.5 we carried a RHEL-only patch which forced the arm smmuv2
	    into bypass mode due to performance issues on CN88xx. This was intended
	    to be a temporary hack until the issues were resolved. Another vendor
	    had issues with the iommu in bypass mode so we reverted the RHEL-only
	    patch so that iommu is in DMA mode by default (upstream default).

	    It turns on that there are remaining SMMU DMA mode issues on Gigabyte
	    platformws with CN88xx cpus. The problem manifests itself by pcie
	    card drivers failing to initialize the cards when SMMU is in DMA mode.
	    The root cause has not been determined yet, but looks likely to be
	    a hw or firmware issue. This patch forces bypass mode for Gigabyte
	    platforms. CN88xx isn't officially supported in RHEL but we have a
	    lot of them being used internally for testing, so I think we want
	    this to support that use case in RHEL8.

	    Signed-off-by: Mark Salter <msalter@redhat.com>
	    Signed-off-by: Herton R. Krzesinski <herton@redhat.com>

	Acked-by: Mark Salter <msalter@redhat.com>
	Acked-by: Donald Dutile <ddutile@redhat.com>

Upstream Status: RHEL only
Signed-off-by: Laura Abbott <labbott@redhat.com>
---
 drivers/iommu/iommu.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index 609bd25bf154..c701912b13ab 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -7,6 +7,7 @@
 #define pr_fmt(fmt)    "iommu: " fmt

 #include <linux/device.h>
+#include <linux/dmi.h>
 #include <linux/kernel.h>
 #include <linux/bug.h>
 #include <linux/types.h>
@@ -2849,3 +2850,24 @@ int iommu_sva_get_pasid(struct iommu_sva *handle)
 	return ops->sva_get_pasid(handle);
 }
 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid);
+
+#ifdef CONFIG_ARM64
+static int __init iommu_quirks(void)
+{
+	const char *vendor, *name;
+
+	vendor = dmi_get_system_info(DMI_SYS_VENDOR);
+	name = dmi_get_system_info(DMI_PRODUCT_NAME);
+
+	if (vendor &&
+	    (strncmp(vendor, "GIGABYTE", 8) == 0 && name &&
+	     (strncmp(name, "R120", 4) == 0 ||
+	      strncmp(name, "R270", 4) == 0))) {
+		pr_warn("Gigabyte %s detected, force iommu passthrough mode", name);
+		iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
+	}
+
+	return 0;
+}
+arch_initcall(iommu_quirks);
+#endif
-- 
2.28.0