[PATCH 0/7] amd-iommu debug patches for 2.6.31

[PATCH 0/7] amd-iommu debug patches for 2.6.31

[Joerg Roedel]

This patchset contains updates to the debugging code in the AMD IOMMU
driver for Linux. It introduces two additional features:

  * A new command line parameter to let the driver dump the parsed
    contents of the IVRS ACPI table to dmesg. This is a debug feature to
    make it easier to find the bug for problems reported to me
  * A new Kconfig entry to enable IOMMU stress testing code. I made this
    Kconfig entry generic because other IOMMU implementations may pick
    this up as well.

Please review.

diffstat:

 arch/x86/Kconfig.debug                 |    8 +++
 arch/x86/include/asm/amd_iommu_types.h |    6 ++
 arch/x86/kernel/amd_iommu.c            |   10 ++-
 arch/x86/kernel/amd_iommu_init.c       |  109 ++++++++++++++++++++++++++++++++
 4 files changed, 129 insertions(+), 4 deletions(-)

shortlog:

Joerg Roedel (7):
      amd-iommu: add amd_iommu_dump parameter
      amd-iommu: add dump for iommus described in ivrs table
      amd-iommu: print ivhd information to dmesg when requested
      amd-iommu: print ivmd information to dmesg when requested
      amd-iommu: move protection domain printk to dump code
      x86/iommu: add IOMMU_STRESS Kconfig entry
      amd-iommu: disable device isolation with CONFIG_IOMMU_STRESS

[PATCH 1/7] amd-iommu: add amd_iommu_dump parameter

[Joerg Roedel]

[ impact: add parameter to dump the ACPI table for AMD IOMMU ]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    6 ++++++
 arch/x86/kernel/amd_iommu_init.c       |   10 ++++++++++
 2 files changed, 16 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index 95c8cd9..89dfb37 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -194,6 +194,12 @@
 #define PD_DMA_OPS_MASK		(1UL << 0) /* domain used for dma_ops */
 #define PD_DEFAULT_MASK		(1UL << 1) /* domain is a default dma_ops
 					      domain for an IOMMU */
+extern bool amd_iommu_dump;
+#define DUMP_printk(format, arg...)					\
+	do {								\
+		if (amd_iommu_dump)						\
+			printk(KERN_INFO "AMD IOMMU: " format, ## arg);	\
+	} while(0);
 
 /*
  * This structure contains generic data for  IOMMU protection domains
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index 8c0be09..57fb7a7 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -115,6 +115,8 @@ struct ivmd_header {
 	u64 range_length;
 } __attribute__((packed));
 
+bool amd_iommu_dump;
+
 static int __initdata amd_iommu_detected;
 
 u16 amd_iommu_last_bdf;			/* largest PCI device id we have
@@ -1211,6 +1213,13 @@ void __init amd_iommu_detect(void)
  *
  ****************************************************************************/
 
+static int __init parse_amd_iommu_dump(char *str)
+{
+	amd_iommu_dump = true;
+
+	return 1;
+}
+
 static int __init parse_amd_iommu_options(char *str)
 {
 	for (; *str; ++str) {
@@ -1235,5 +1244,6 @@ static int __init parse_amd_iommu_size_options(char *str)
 	return 1;
 }
 
+__setup("amd_iommu_dump", parse_amd_iommu_dump);
 __setup("amd_iommu=", parse_amd_iommu_options);
 __setup("amd_iommu_size=", parse_amd_iommu_size_options);
-- 
1.6.3.1

[PATCH 2/7] amd-iommu: add dump for iommus described in ivrs table

[Joerg Roedel]

[ impact: get information about iommus described in the ivrs acpi table ]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu_init.c |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index 57fb7a7..2816590 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -748,6 +748,15 @@ static int __init init_iommu_all(struct acpi_table_header *table)
 		h = (struct ivhd_header *)p;
 		switch (*p) {
 		case ACPI_IVHD_TYPE:
+
+			DUMP_printk("IOMMU: device: %02x:%02x.%01x cap: %04x "
+				    "seg: %d flags: %01x info %04x\n",
+				    PCI_BUS(h->devid), PCI_SLOT(h->devid),
+				    PCI_FUNC(h->devid), h->cap_ptr,
+				    h->pci_seg, h->flags, h->info);
+			DUMP_printk("       mmio-addr: %016llx\n",
+				    h->mmio_phys);
+
 			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
 			if (iommu == NULL)
 				return -ENOMEM;
-- 
1.6.3.1

[PATCH 3/7] amd-iommu: print ivhd information to dmesg when requested

[Joerg Roedel]

[ impact: be able to show in dmesg whats in the acpi ivhd fields ]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu_init.c |   73 ++++++++++++++++++++++++++++++++++++++
 1 files changed, 73 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index 2816590..fe3e645 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -598,32 +598,83 @@ static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
 	p += sizeof(struct ivhd_header);
 	end += h->length;
 
+
 	while (p < end) {
 		e = (struct ivhd_entry *)p;
 		switch (e->type) {
 		case IVHD_DEV_ALL:
+
+			DUMP_printk("  DEV_ALL\t\t\t first devid: %02x:%02x.%x"
+				    " last device %02x:%02x.%x flags: %02x\n",
+				    PCI_BUS(iommu->first_device),
+				    PCI_SLOT(iommu->first_device),
+				    PCI_FUNC(iommu->first_device),
+				    PCI_BUS(iommu->last_device),
+				    PCI_SLOT(iommu->last_device),
+				    PCI_FUNC(iommu->last_device),
+				    e->flags);
+
 			for (dev_i = iommu->first_device;
 					dev_i <= iommu->last_device; ++dev_i)
 				set_dev_entry_from_acpi(iommu, dev_i,
 							e->flags, 0);
 			break;
 		case IVHD_DEV_SELECT:
+
+			DUMP_printk("  DEV_SELECT\t\t\t devid: %02x:%02x.%x "
+				    "flags: %02x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags);
+
 			devid = e->devid;
 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
 			break;
 		case IVHD_DEV_SELECT_RANGE_START:
+
+			DUMP_printk("  DEV_SELECT_RANGE_START\t "
+				    "devid: %02x:%02x.%x flags: %02x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags);
+
 			devid_start = e->devid;
 			flags = e->flags;
 			ext_flags = 0;
 			alias = false;
 			break;
 		case IVHD_DEV_ALIAS:
+
+			DUMP_printk("  DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
+				    "flags: %02x devid_to: %02x:%02x.%x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags,
+				    PCI_BUS(e->ext >> 8),
+				    PCI_SLOT(e->ext >> 8),
+				    PCI_FUNC(e->ext >> 8));
+
 			devid = e->devid;
 			devid_to = e->ext >> 8;
 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
 			amd_iommu_alias_table[devid] = devid_to;
 			break;
 		case IVHD_DEV_ALIAS_RANGE:
+
+			DUMP_printk("  DEV_ALIAS_RANGE\t\t "
+				    "devid: %02x:%02x.%x flags: %02x "
+				    "devid_to: %02x:%02x.%x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags,
+				    PCI_BUS(e->ext >> 8),
+				    PCI_SLOT(e->ext >> 8),
+				    PCI_FUNC(e->ext >> 8));
+
 			devid_start = e->devid;
 			flags = e->flags;
 			devid_to = e->ext >> 8;
@@ -631,17 +682,39 @@ static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
 			alias = true;
 			break;
 		case IVHD_DEV_EXT_SELECT:
+
+			DUMP_printk("  DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
+				    "flags: %02x ext: %08x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags, e->ext);
+
 			devid = e->devid;
 			set_dev_entry_from_acpi(iommu, devid, e->flags,
 						e->ext);
 			break;
 		case IVHD_DEV_EXT_SELECT_RANGE:
+
+			DUMP_printk("  DEV_EXT_SELECT_RANGE\t devid: "
+				    "%02x:%02x.%x flags: %02x ext: %08x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags, e->ext);
+
 			devid_start = e->devid;
 			flags = e->flags;
 			ext_flags = e->ext;
 			alias = false;
 			break;
 		case IVHD_DEV_RANGE_END:
+
+			DUMP_printk("  DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid));
+
 			devid = e->devid;
 			for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
 				if (alias)
-- 
1.6.3.1

[PATCH 4/7] amd-iommu: print ivmd information to dmesg when requested

[Joerg Roedel]

[ impact: be able to show in dmesg whats in the acpi ivmd fields ]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu_init.c |   11 +++++++++++
 1 files changed, 11 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index fe3e645..b90a78c 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -983,6 +983,7 @@ static int __init init_exclusion_range(struct ivmd_header *m)
 static int __init init_unity_map_range(struct ivmd_header *m)
 {
 	struct unity_map_entry *e = 0;
+	char *s;
 
 	e = kzalloc(sizeof(*e), GFP_KERNEL);
 	if (e == NULL)
@@ -991,13 +992,16 @@ static int __init init_unity_map_range(struct ivmd_header *m)
 	switch (m->type) {
 	default:
 	case ACPI_IVMD_TYPE:
+		s = "IVMD_TYPEi\t\t\t";
 		e->devid_start = e->devid_end = m->devid;
 		break;
 	case ACPI_IVMD_TYPE_ALL:
+		s = "IVMD_TYPE_ALL\t\t";
 		e->devid_start = 0;
 		e->devid_end = amd_iommu_last_bdf;
 		break;
 	case ACPI_IVMD_TYPE_RANGE:
+		s = "IVMD_TYPE_RANGE\t\t";
 		e->devid_start = m->devid;
 		e->devid_end = m->aux;
 		break;
@@ -1006,6 +1010,13 @@ static int __init init_unity_map_range(struct ivmd_header *m)
 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
 	e->prot = m->flags >> 1;
 
+	DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
+		    " range_start: %016llx range_end: %016llx flags: %x\n", s,
+		    PCI_BUS(e->devid_start), PCI_SLOT(e->devid_start),
+		    PCI_FUNC(e->devid_start), PCI_BUS(e->devid_end),
+		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
+		    e->address_start, e->address_end, m->flags);
+
 	list_add_tail(&e->list, &amd_iommu_unity_map);
 
 	return 0;
-- 
1.6.3.1

[PATCH 5/7] amd-iommu: move protection domain printk to dump code

[Joerg Roedel]

[ impact: only printk protection-domain relations if requested ]

This information is only helpful for debugging. Don't print it anymore
unless explicitly requested.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   10 ++++++----
 1 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index a97db99..3356599 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1009,8 +1009,9 @@ static int device_change_notifier(struct notifier_block *nb,
 		if (!dma_domain)
 			dma_domain = iommu->default_dom;
 		attach_device(iommu, &dma_domain->domain, devid);
-		printk(KERN_INFO "AMD IOMMU: Using protection domain %d for "
-		       "device %s\n", dma_domain->domain.id, dev_name(dev));
+		DUMP_printk(KERN_INFO "AMD IOMMU: Using protection domain "
+			    "%d for device %s\n",
+			    dma_domain->domain.id, dev_name(dev));
 		break;
 	case BUS_NOTIFY_UNBIND_DRIVER:
 		if (!domain)
@@ -1133,8 +1134,9 @@ static int get_device_resources(struct device *dev,
 			dma_dom = (*iommu)->default_dom;
 		*domain = &dma_dom->domain;
 		attach_device(*iommu, *domain, *bdf);
-		printk(KERN_INFO "AMD IOMMU: Using protection domain %d for "
-				"device %s\n", (*domain)->id, dev_name(dev));
+		DUMP_printk(KERN_INFO "AMD IOMMU: Using protection domain "
+				"%d for device %s\n",
+				(*domain)->id, dev_name(dev));
 	}
 
 	if (domain_for_device(_bdf) == NULL)
-- 
1.6.3.1

[PATCH 6/7] x86/iommu: add IOMMU_STRESS Kconfig entry

[Joerg Roedel]

[ impact: add a Kconfig option to enable stress-test code im iommu drivers ]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
---
 arch/x86/Kconfig.debug |    8 ++++++++
 1 files changed, 8 insertions(+), 0 deletions(-)

diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
index 5865712..33fac6b 100644
--- a/arch/x86/Kconfig.debug
+++ b/arch/x86/Kconfig.debug
@@ -159,6 +159,14 @@ config IOMMU_DEBUG
 	  options. See Documentation/x86_64/boot-options.txt for more
 	  details.
 
+config IOMMU_STRESS
+	bool "Enable IOMMU stress-test mode"
+	---help---
+	  This option disables various optimizations in IOMMU related
+	  code to do real stress testing of the IOMMU code. This option
+	  will cause a performance drop and should only be enabled for
+	  testing.
+
 config IOMMU_LEAK
 	bool "IOMMU leak tracing"
 	depends on IOMMU_DEBUG && DMA_API_DEBUG
-- 
1.6.3.1

[PATCH 7/7] amd-iommu: disable device isolation with CONFIG_IOMMU_STRESS

[Joerg Roedel]

[ impact: test for race conditions in dma_ops related code ]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu_init.c |    6 ++++++
 1 files changed, 6 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index b90a78c..6694112 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -124,8 +124,14 @@ u16 amd_iommu_last_bdf;			/* largest PCI device id we have
 LIST_HEAD(amd_iommu_unity_map);		/* a list of required unity mappings
 					   we find in ACPI */
 unsigned amd_iommu_aperture_order = 26; /* size of aperture in power of 2 */
+
+#ifdef CONFIG_IOMMU_STRESS
+bool amd_iommu_isolate = false;
+#else
 bool amd_iommu_isolate = true;		/* if true, device isolation is
 					   enabled */
+#endif
+
 bool amd_iommu_unmap_flush;		/* if true, flush on every unmap */
 
 LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
-- 
1.6.3.1