1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
|
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 b6858adc4f17..4c2da9f17514 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>
@@ -2838,3 +2839,24 @@ 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
--
2.26.2
|
