Skip to contentSkip to conversation

iommu/amd: IVMD handling cleanup and ivmd= boot override

PATCH In review Suravee Suthikulpanit 4 participants 9 messages Apr 24 – Oct 1

Subsystems (3)
lore
Share & tools
lore

Suravee Suthikulpanit

author reviewer
#0 lore raw

v2 cover iommu/amd: IVMD handling cleanup and ivmd= boot override

This series refreshes how the AMD IOMMU driver models IVRS IVMD entries after the hardware exclusion range registers were repurposed.

The first patch renames the associated MMIO offsets to reflect completion-wait store base/limit.

The second patch drops the workaround that forced IVMD exclusion ranges to r/w unity mappings; the driver no longer programs exclusion range registers, so that the workaround is obsolete.

The third patch centralizes IVMD flag definitions, renames the unity-map structures to ivmd_entry, stores the raw IVMD flags byte from ACPI, and derives IOMMU reserved regions (including exclusion as IOMMU_RESV_RESERVED) from those flags.

The fourth patch adds an ivmd= kernel parameter so up to four per-device IVMD-style ranges can be supplied from the command line when the IVRS table is wrong or incomplete, documented in kernel-parameters.txt.

Change from v1:
(https://lore.kernel.org/linux-iommu/[email protected]/)

* Rebase on top of patch "[PATCH] iommu/amd: Remove dead code for exclusion ranges in IVMD" that was submitted by Yu Zhang <[email protected]>.
(https://lore.kernel.org/all/[email protected]/)

* Patch 1: Modified to only rename the marco.

* Patch 3: Reintroduce logic to parse IVMD exclusion flag to allow reserve region of type IOMMU_RESV_RESERVED.

* Patch 4:
- Update the sscan function to scan start, length, flags in hex format only.
- Fix apply_ivmd_cmdline_entries() to not free the cmd, which points to

    the early_ivmd_cmdline_map[]  on failure path.

Thanks,
Suravee

Suravee Suthikulpanit (4):
iommu/amd: Drop unused global exclusion range fields and init iommu/amd: Do not convert IVMD exclusion range to unity range iommu/amd: Clean up and simplify IVMD entry handling iommu/amd: Introduce boot option ivmd=seg:bus:dev.fun,start,size,flags

 .../admin-guide/kernel-parameters.txt         |  16 ++
 drivers/iommu/amd/amd_iommu_types.h           |  36 ++--
 drivers/iommu/amd/init.c                      | 159 ++++++++++++++----
 drivers/iommu/amd/iommu.c                     |  27 ++-
 4 files changed, 187 insertions(+), 51 deletions(-)
Signature · 1 line
2.34.1
author reviewer
#1 lore raw

PATCH v2 1/4 iommu/amd: Drop unused global exclusion range fields and init

Remove unused struct amd_iommu.exclusion_start/exclusion_length, the MMIO_EXCL_* flag masks, and iommu_set_exclusion_range(). These struct amd_iommu fields were never assigned, so early_enable_iommu() never programmed the hardware exclusion registers through this path.

Please note that the exclusion range registers have been repurposed to SNP completion-wait store base / limit registers. So, rename MMIO_EXCL_BASE/LIMIT via to MMIO_COMPL_STORE_BASE/LIMIT instead.

---
 drivers/iommu/amd/amd_iommu_types.h | 4 ++--
 drivers/iommu/amd/init.c            | 6 +++---
 2 files changed, 5 insertions(+), 5 deletions(-)
drivers/iommu/amd/amd_iommu_types.h +2 -2
index a481d8cbd053..6b0f1b05aa47 100644
@@ -55,8 +55,8 @@
 #define MMIO_CMD_BUF_OFFSET     0x0008
 #define MMIO_EVT_BUF_OFFSET     0x0010
 #define MMIO_CONTROL_OFFSET     0x0018
-#define MMIO_EXCL_BASE_OFFSET   0x0020
-#define MMIO_EXCL_LIMIT_OFFSET  0x0028
+#define MMIO_COMPL_STORE_BASE_OFFSET   0x0020
+#define MMIO_COMPL_STORE_LIMIT_OFFSET  0x0028
 #define MMIO_EXT_FEATURES	0x0030
 #define MMIO_PPR_LOG_OFFSET	0x0038
 #define MMIO_GA_LOG_BASE_OFFSET	0x00e0
drivers/iommu/amd/init.c +3 -3
index a33bf3f5ed8a..866249d3673e 100644
@@ -364,14 +364,14 @@ static void iommu_set_cwwb_range(struct amd_iommu *iommu)
 	 * Re-purpose Exclusion base/limit registers for Completion wait
 	 * write-back base/limit.
 	 */
-	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
+	memcpy_toio(iommu->mmio_base + MMIO_COMPL_STORE_BASE_OFFSET,
 		    &entry, sizeof(entry));
 
 	/* Note:
 	 * Default to 4 Kbytes, which can be specified by setting base
 	 * address equal to the limit address.
 	 */
-	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
+	memcpy_toio(iommu->mmio_base + MMIO_COMPL_STORE_LIMIT_OFFSET,
 		    &entry, sizeof(entry));
 }
 
@@ -991,7 +991,7 @@ static int __init remap_or_alloc_cwwb_sem(struct amd_iommu *iommu)
 		 * completion wait buffer (CWB) address. Read and re-use it.
 		 */
 		pr_info_once("Re-using CWB buffers from the previous kernel\n");
-		paddr = readq(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET) & PM_ADDR_MASK;
+		paddr = readq(iommu->mmio_base + MMIO_COMPL_STORE_BASE_OFFSET) & PM_ADDR_MASK;
 		iommu->cmd_sem = iommu_memremap(paddr, PAGE_SIZE);
 		if (!iommu->cmd_sem)
 			return -ENOMEM;
Signature · 1 line
2.34.1
author reviewer
#2 lore raw

PATCH v2 2/4 iommu/amd: Do not convert IVMD exclusion range to unity range

In reply to Suravee Suthikulpanit

Original problem was there could be multiple IVMD exclusion entries conflicting over a single exclusion range supported in older Linux kernel. The fix was first introduced in commit 387caf0b759a ("iommu/amd:
Treat per-device exclusion ranges as r/w unity-mapped regions") and in commit 0bbe4ced53e3 ("iommu/amd: Fix the overwritten field in IVMD header") , where IVMD exclusions are converted to separate unity_map_entry objects in pci_seg->unity_map so amd_iommu_get_resv_regions() treated every such entry as IOMMU_RESV_DIRECT with R/W.

Since the AMD IOMMU driver no longer support the Exclusion Range base/limit registers and rely on Linux kernel to exclude the specified IOVA ranges from being mapped this logic is no longer needed.

Remove the special case that forced e->prot to RW-UNITY mapping (i.e. IVMD_FLAG_IW | IVMD_FLAG_IR | IVMD_FLAG_UNITY) whenever IVMD_FLAG_EXCL_RANGE was set.

Fixes 0bbe4ced53e3 ("iommu/amd: Fix the overwritten field in IVMD header")
Cc Baoquan He <[email protected]>
Cc Adrian Huang <[email protected]>
Cc Jerry Snitselaar <[email protected]>
---
 drivers/iommu/amd/init.c | 10 ----------
 1 file changed, 10 deletions(-)
drivers/iommu/amd/init.c +0 -10
index 866249d3673e..2e854f4639a7 100644
@@ -2644,16 +2644,6 @@ 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;
 
-	/*
-	 * Treat per-device exclusion ranges as r/w unity-mapped regions
-	 * since some buggy BIOSes might lead to the overwritten exclusion
-	 * range (exclusion_start and exclusion_length members). This
-	 * happens when there are multiple exclusion ranges (IVMD entries)
-	 * defined in ACPI table.
-	 */
-	if (m->flags & IVMD_FLAG_EXCL_RANGE)
-		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
-
 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
 		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
 		    " flags: %x\n", s, m->pci_seg,
Signature · 1 line
2.34.1
Reviewed-by on v2 2/4
#3 lore raw

RE: [PATCH v2 2/4] iommu/amd: Do not convert IVMD exclusion range to unity range

Quoted 60 lines

-----Original Message-----
From: Suravee Suthikulpanit <[email protected]>
Sent: Friday, April 24, 2026 5:22 PM
To: [email protected]; [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Suravee Suthikulpanit <[email protected]>; Baoquan He <[email protected]>; Adrian Huang12 <[email protected]>; Jerry Snitselaar <[email protected]>
Subject: [External] [PATCH v2 2/4] iommu/amd: Do not convert IVMD exclusion range to unity range

Original problem was there could be multiple IVMD exclusion entries conflicting over a single exclusion range supported in older Linux kernel. The fix was first introduced in commit 387caf0b759a ("iommu/amd:
Treat per-device exclusion ranges as r/w unity-mapped regions") and in commit 0bbe4ced53e3 ("iommu/amd: Fix the overwritten field in IVMD header") , where IVMD exclusions are converted to separate unity_map_entry objects in pci_seg->unity_map so amd_iommu_get_resv_regions() treated every such entry as IOMMU_RESV_DIRECT with R/W.

Since the AMD IOMMU driver no longer support the Exclusion Range base/limit registers and rely on Linux kernel to exclude the specified IOVA ranges from being mapped this logic is no longer needed.

Remove the special case that forced e->prot to RW-UNITY mapping (i.e. IVMD_FLAG_IW | IVMD_FLAG_IR | IVMD_FLAG_UNITY) whenever IVMD_FLAG_EXCL_RANGE was set.

Fixes 0bbe4ced53e3 ("iommu/amd: Fix the overwritten field in IVMD

header")

Cc Baoquan He <[email protected]>
Cc Adrian Huang <[email protected]>
Cc Jerry Snitselaar <[email protected]>
---
 drivers/iommu/amd/init.c | 10 ----------
 1 file changed, 10 deletions(-)
drivers/iommu/amd/init.c index renamed from drivers/iommu/amd/init.c +0 -0

866249d3673e..2e854f4639a7 100644

drivers/iommu/amd/init.c +0 -10
@@ -2644,16 +2644,6 @@ 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;
 
-	/*
-	 * Treat per-device exclusion ranges as r/w unity-mapped regions
-	 * since some buggy BIOSes might lead to the overwritten exclusion
-	 * range (exclusion_start and exclusion_length members). This
-	 * happens when there are multiple exclusion ranges (IVMD entries)
-	 * defined in ACPI table.
-	 */
-	if (m->flags & IVMD_FLAG_EXCL_RANGE)
-		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
-
 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
 		    "%04x:%02x:%02x.%x range_start: %016llx range_end:
Quoted 2 lines

%016llx"
" flags: %x\n", s, m->pci_seg,

Looks good to me.

author reviewer
#4 lore raw

PATCH v2 3/4 iommu/amd: Clean up and simplify IVMD entry handling

In reply to Suravee Suthikulpanit

Clean up IVMD ACPI flags and mask in amd_iommu_types.h, and drop the IOMMU_PROT_* / IOMMU_UNITY_MAP_FLAG_EXCL_RANGE, which only existed to carry IVMD semantics.

Rename unity_map_entry and unity_map to ivmd_entry and ivmd_entry_map, store the raw IVMD flags byte from the ACPI table instead of shifting protection into a separate int, and derive IOMMU reserved regions from those flags in amd_iommu_get_resv_regions(). Also, rename struct list_head unity_map to ivmd_entry_map along with renaming the unity-map helper functions.

Lastyly, add error message when handling invalid IVMD entry.

---
 drivers/iommu/amd/amd_iommu_types.h | 18 +++++++------
 drivers/iommu/amd/init.c            | 39 +++++++++++++----------------
 drivers/iommu/amd/iommu.c           | 27 ++++++++++++++++----
 3 files changed, 49 insertions(+), 35 deletions(-)
drivers/iommu/amd/amd_iommu_types.h +10 -8
index 6b0f1b05aa47..9321b7cc200b 100644
@@ -375,9 +375,11 @@
 /* DTE[128:179] | DTE[184:191] */
 #define DTE_DATA2_INTR_MASK	~GENMASK_ULL(55, 52)
 
-#define IOMMU_PROT_MASK 0x03
-#define IOMMU_PROT_IR 0x01
-#define IOMMU_PROT_IW 0x02
+#define IVMD_FLAG_UNITY		BIT(0)
+#define IVMD_FLAG_IR		BIT(1)
+#define IVMD_FLAG_IW		BIT(2)
+#define IVMD_FLAG_EXCL		BIT(3)
+#define IVMD_FLAG_MASK		GENMASK_U32(3, 0)
 
 /* IOMMU capabilities */
 #define IOMMU_CAP_IOTLB   24
@@ -613,11 +615,11 @@ struct amd_iommu_pci_seg {
 	u16 *alias_table;
 
 	/*
-	 * A list of required unity mappings we find in ACPI. It is not locked
+	 * A list of required IVMD entry we find in ACPI. It is not locked
 	 * because as runtime it is only read. It is created at ACPI table
 	 * parsing time.
 	 */
-	struct list_head unity_map;
+	struct list_head ivmd_entry_map;
 };
 
 /*
@@ -873,7 +875,7 @@ struct ivhd_dte_flags {
 /*
  * One entry for unity mappings parsed out of the ACPI table.
  */
-struct unity_map_entry {
+struct ivmd_entry {
 	struct list_head list;
 
 	/* starting device id this entry is used for (including) */
@@ -886,8 +888,8 @@ struct unity_map_entry {
 	/* end address to unity map (including) */
 	u64 address_end;
 
-	/* required protection */
-	int prot;
+	/* IVMD flags for the entry */
+	u8 flags;
 };
 
 /*
drivers/iommu/amd/init.c +17 -22
index 2e854f4639a7..4b62bb89a12c 100644
@@ -70,11 +70,6 @@
 #define IVHD_FLAG_RESPASSPW_EN_MASK     0x04
 #define IVHD_FLAG_ISOC_EN_MASK          0x08
 
-#define IVMD_FLAG_EXCL_RANGE            0x08
-#define IVMD_FLAG_IW                    0x04
-#define IVMD_FLAG_IR                    0x02
-#define IVMD_FLAG_UNITY_MAP             0x01
-
 #define ACPI_DEVFLAG_INITPASS           0x01
 #define ACPI_DEVFLAG_EXTINT             0x02
 #define ACPI_DEVFLAG_NMI                0x04
@@ -134,7 +129,7 @@ struct ivhd_entry {
 
 /*
  * An AMD IOMMU memory definition structure. It defines things like exclusion
- * ranges for devices and regions that should be unity mapped.
+ * ranges, unity mapping for devices and regions.
  */
 struct ivmd_header {
 	u8 type;
@@ -1708,7 +1703,7 @@ static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id,
 
 	pci_seg->id = id;
 	init_llist_head(&pci_seg->dev_data_list);
-	INIT_LIST_HEAD(&pci_seg->unity_map);
+	INIT_LIST_HEAD(&pci_seg->ivmd_entry_map);
 	list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
 
 	if (alloc_dev_table(pci_seg))
@@ -2301,8 +2296,8 @@ static int __init amd_iommu_init_pci(void)
 	}
 
 	/*
-	 * Order is important here to make sure any unity map requirements are
-	 * fulfilled. The unity mappings are created and written to the device
+	 * Order is important here to make sure any ivmd map requirements are
+	 * fulfilled. The ivmd mappings are created and written to the device
 	 * table during the iommu_init_pci() call.
 	 *
 	 * After that we call init_device_table_dma() to make sure any
@@ -2588,28 +2583,28 @@ static int iommu_init_irq(struct amd_iommu *iommu)
  *
  * The next functions belong to the third pass of parsing the ACPI
  * table. In this last pass the memory mapping requirements are
- * gathered (like exclusion and unity mapping ranges).
+ * gathered (like exclusion and ivmd mapping ranges).
  *
  ****************************************************************************/
 
-static void __init free_unity_maps(void)
+static void __init free_ivmd_maps(void)
 {
-	struct unity_map_entry *entry, *next;
+	struct ivmd_entry *entry, *next;
 	struct amd_iommu_pci_seg *p, *pci_seg;
 
 	for_each_pci_segment_safe(pci_seg, p) {
-		list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
+		list_for_each_entry_safe(entry, next, &pci_seg->ivmd_entry_map, list) {
 			list_del(&entry->list);
 			kfree(entry);
 		}
 	}
 }
 
-/* called for unity map ACPI definition */
-static int __init init_unity_map_range(struct ivmd_header *m,
-				       struct acpi_table_header *ivrs_base)
+/* called for ivmd map ACPI definition */
+static int __init init_ivmd_map_range(struct ivmd_header *m,
+				      struct acpi_table_header *ivrs_base)
 {
-	struct unity_map_entry *e = NULL;
+	struct ivmd_entry *e = NULL;
 	struct amd_iommu_pci_seg *pci_seg;
 	char *s;
 
@@ -2642,7 +2637,7 @@ static int __init init_unity_map_range(struct ivmd_header *m,
 	}
 	e->address_start = PAGE_ALIGN(m->range_start);
 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
-	e->prot = m->flags >> 1;
+	e->flags = m->flags;
 
 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
 		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
@@ -2653,7 +2648,7 @@ static int __init init_unity_map_range(struct ivmd_header *m,
 		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
 		    e->address_start, e->address_end, m->flags);
 
-	list_add_tail(&e->list, &pci_seg->unity_map);
+	list_add_tail(&e->list, &pci_seg->ivmd_entry_map);
 
 	return 0;
 }
@@ -2669,8 +2664,8 @@ static int __init init_memory_definitions(struct acpi_table_header *table)
 
 	while (p < end) {
 		m = (struct ivmd_header *)p;
-		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
-			init_unity_map_range(m, table);
+		if (m->flags & (IVMD_FLAG_UNITY | IVMD_FLAG_EXCL))
+			init_ivmd_map_range(m, table);
 
 		p += m->length;
 	}
@@ -3115,7 +3110,7 @@ static bool __init check_ioapic_information(void)
 static void __init free_dma_resources(void)
 {
 	amd_iommu_pdom_id_destroy();
-	free_unity_maps();
+	free_ivmd_maps();
 }
 
 static void __init ivinfo_init(void *ivrs)
drivers/iommu/amd/iommu.c +22 -5
index 4e7f5b993c65..b26d29cb2d66 100644
@@ -3050,7 +3050,7 @@ static void amd_iommu_get_resv_regions(struct device *dev,
 				       struct list_head *head)
 {
 	struct iommu_resv_region *region;
-	struct unity_map_entry *entry;
+	struct ivmd_entry *entry;
 	struct amd_iommu *iommu;
 	struct amd_iommu_pci_seg *pci_seg;
 	int devid, sbdf;
@@ -3063,25 +3063,42 @@ static void amd_iommu_get_resv_regions(struct device *dev,
 	iommu = get_amd_iommu_from_dev(dev);
 	pci_seg = iommu->pci_seg;
 
-	list_for_each_entry(entry, &pci_seg->unity_map, list) {
+	list_for_each_entry(entry, &pci_seg->ivmd_entry_map, list) {
 		int type, prot = 0;
 		size_t length;
 
 		if (devid < entry->devid_start || devid > entry->devid_end)
 			continue;
 
+		/*
+		 * IVMD_FLAG_UNITY, IVMD_FLAG_IR, IVMD_FLAG_IW are ignored if
+		 * IVMD_FLAG_EXCL is set.
+		 */
+		if ((entry->flags & IVMD_FLAG_EXCL) &&
+		    (entry->flags & (IVMD_FLAG_UNITY | IVMD_FLAG_IR | IVMD_FLAG_IW)))
+			pr_err(FW_BUG "%s: Invalid IVMD flags combination: %#02x for sbdf=%#x. Ignore IW/IR/Unity flags\n",
+			       __func__, entry->flags, sbdf);
+
+		if (entry->flags & ~IVMD_FLAG_MASK)
+			pr_err("%s: Unknown flags: %#02x for sbdf=%#x\n",
+			       __func__, entry->flags & ~IVMD_FLAG_MASK, sbdf);
+
 		type   = IOMMU_RESV_DIRECT;
 		length = entry->address_end - entry->address_start;
-		if (entry->prot & IOMMU_PROT_IR)
+		if (entry->flags & IVMD_FLAG_IR)
 			prot |= IOMMU_READ;
-		if (entry->prot & IOMMU_PROT_IW)
+		if (entry->flags & IVMD_FLAG_IW)
 			prot |= IOMMU_WRITE;
+		if (entry->flags & IVMD_FLAG_EXCL) {
+ 			type = IOMMU_RESV_RESERVED;
+			prot = 0;
+		}
 
 		region = iommu_alloc_resv_region(entry->address_start,
 						 length, prot, type,
 						 GFP_KERNEL);
 		if (!region) {
-			dev_err(dev, "Out of memory allocating dm-regions\n");
+			pr_err("%s: Out of memory allocating reserved regions for sbdf=%#x\n", __func__, sbdf);
 			return;
 		}
 		list_add_tail(&region->list, head);
Signature · 1 line
2.34.1
author reviewer
#5 lore raw

PATCH v2 4/4 iommu/amd: Introduce boot option ivmd=seg:bus:dev.fun,start,size,flags

In reply to Suravee Suthikulpanit

IVRS table contains IVMD blocks, which allow firmware to specify memory usage requirements to communicate to system software based on its needs or on hardware characteristics. Each IVMD entry may be per-device, range of devices.

Some BIOS specify incorrect or missing IVMD entry. Introduce a new ivmd boot option to allow user to specify up-to 4 per-device IVMD entries at boot time. The entries are stored during driver initialization, and will be added to the per-segment ivmd_entry_map so that they can be included during struct iommu_ops.get_resv_regions.

---
 .../admin-guide/kernel-parameters.txt         |  16 +++
 drivers/iommu/amd/amd_iommu_types.h           |  14 +++
 drivers/iommu/amd/init.c                      | 104 +++++++++++++++++-
 3 files changed, 133 insertions(+), 1 deletion(-)
Documentation/admin-guide/kernel-parameters.txt +16 -0
index 03a550630644..e680a258008f 100644
@@ -2897,6 +2897,22 @@ Kernel parameters
 			  PCI device ID 00:14.5, write the parameter as:
 				ivrs_acpihid[0001:00:14.5]=AMD0020:0
 
+	ivmd		[HW,X86-64]
+			Supplement IVMD unity mapping or exclusion ranges from
+			the kernel command line (in addition to the IVRS ACPI
+			table). May be specified multiple times.
+
+			Form:
+				ivmd=segment:bus:dev.fn,start,length,flags
+
+			start and length are byte counts (decimal or 0x hex).
+			flags is the IVMD flags byte (UNITY, IR, IW, EXCL bits
+			per the AMD IOMMU specification). Use segment 0 for
+			PCI segment zero.
+
+			Example:
+				ivmd=0000:00:01.0,0xe0000000,0x100000,0xb
+
 	js=		[HW,JOY] Analog joystick
 			See Documentation/input/joydev/joystick.rst.
 
drivers/iommu/amd/amd_iommu_types.h +14 -0
index 9321b7cc200b..4f3387dd7fab 100644
@@ -892,6 +892,20 @@ struct ivmd_entry {
 	u8 flags;
 };
 
+/*
+ * IVMD-style ranges from ivrs_ivmd=... ; applied in
+ * init_memory_definitions() once per-PCI-segment
+ * ivmd_entry_map exists.
+ */
+struct ivmd_cmdline {
+	struct list_head list;
+	u16 pci_seg;
+	u16 devid;
+	u64 range_start;
+	u64 range_length;
+	u8 flags;
+};
+
 /*
  * Data structures for device handling
  */
drivers/iommu/amd/init.c +103 -1

138 lines not on this page · raw message

Signature · 1 line
2.34.1
author reviewer
#6 lore raw
In reply to Suravee Suthikulpanit

Hi All,

Are there any other concerns on this series?

Thanks
Suravee

Quoted 51 lines · Suravee Suthikulpanit

This series refreshes how the AMD IOMMU driver models IVRS IVMD entries after the hardware exclusion range registers were repurposed.

The first patch renames the associated MMIO offsets to reflect completion-wait store base/limit.

The second patch drops the workaround that forced IVMD exclusion ranges to r/w unity mappings; the driver no longer programs exclusion range registers, so that the workaround is obsolete.

The third patch centralizes IVMD flag definitions, renames the unity-map structures to ivmd_entry, stores the raw IVMD flags byte from ACPI, and derives IOMMU reserved regions (including exclusion as IOMMU_RESV_RESERVED) from those flags.

The fourth patch adds an ivmd= kernel parameter so up to four per-device IVMD-style ranges can be supplied from the command line when the IVRS table is wrong or incomplete, documented in kernel-parameters.txt.

Change from v1:
(https://lore.kernel.org/linux-iommu/[email protected]/)

* Rebase on top of patch "[PATCH] iommu/amd: Remove dead code for exclusion ranges in IVMD" that was submitted by Yu Zhang <[email protected]>.
(https://lore.kernel.org/all/[email protected]/)

* Patch 1: Modified to only rename the marco.

* Patch 3: Reintroduce logic to parse IVMD exclusion flag to allow reserve region of type IOMMU_RESV_RESERVED.

* Patch 4:
- Update the sscan function to scan start, length, flags in hex format only.
- Fix apply_ivmd_cmdline_entries() to not free the cmd, which points to

     the early_ivmd_cmdline_map[]  on failure path.

Thanks,
Suravee

Suravee Suthikulpanit (4):
iommu/amd: Drop unused global exclusion range fields and init iommu/amd: Do not convert IVMD exclusion range to unity range iommu/amd: Clean up and simplify IVMD entry handling iommu/amd: Introduce boot option ivmd=seg:bus:dev.fun,start,size,flags

  .../admin-guide/kernel-parameters.txt         |  16 ++
  drivers/iommu/amd/amd_iommu_types.h           |  36 ++--
  drivers/iommu/amd/init.c                      | 159 ++++++++++++++----
  drivers/iommu/amd/iommu.c                     |  27 ++-
  4 files changed, 187 insertions(+), 51 deletions(-)
maintainer
#7 lore raw

Hi Suravee,

Quoted 1 line · Suravee

Are there any other concerns on this series?

There are some valid checkpatch warnings, can you please fix them and send a v3?

Thanks,

Joerg

Tested-by on the v2 series
#8 lore raw
In reply to Suravee Suthikulpanit

Hi Suravee,

This series removes the IOMMU_RESV_DIRECT regions that block VFIO passthrough on my Gigabyte B850 AORUS ELITE WIFI7 (BIOS F11, 05/21/2026; same problem on F7), Ryzen 9 9950X3D, RTX 5080.

The IVRS table on this board has three IVMD type 0x22 entries, all with flags 0x08 (exclusion range only, no unity/IR/IW bits), each covering devid 0000:00:00.0 - 0000:0f:1f.7:

  IVMD 0x22 flags 0x08  0x8aa34000 len 0x1000
  IVMD 0x22 flags 0x08  0x8aa33000 len 0x1000
  IVMD 0x22 flags 0x08  0x95967000 len 0x1000

Without the series (stock 7.2.6), init_unity_map_range() converts these to r/w unity maps, so every IOMMU group on buses 0-15 (27 groups) gets them as IOMMU_RESV_DIRECT. The GPU's group (01:00.0 + 01:00.1) shows:

  0x000000008aa33000 0x000000008aa34fff direct
  0x0000000095967000 0x0000000095967fff direct
  0x00000000fee00000 0x00000000feefffff msi
  0x000000fd00000000 0x000000ffffffffff reserved

and VFIO assignment of the GPU failed (tested on F7) with "Firmware has requested this device have a 1:1 IOMMU mapping, rejecting configuring the device without a 1:1 mapping. Contact your platform vendor."

With v2 1-4 applied on top of 7.2.8 (they apply cleanly), the same group shows:

  0x000000008aa33000 0x000000008aa34fff reserved
  0x0000000095967000 0x0000000095967fff reserved
  0x00000000fee00000 0x00000000feefffff msi
  0x000000fd00000000 0x000000ffffffffff reserved

No group on the system reports a direct region any more. The host runs with the default DMA-FQ domain on every group except the iGPU (identity), and there are no AMD-Vi IO_PAGE_FAULT or FW_BUG messages in the log. The ivmd= parameter was not exercised. I have not yet assigned the GPU to a VM, so the VFIO ownership claim itself is untested on the patched kernel; the check above is of reserved regions only.

This matches the IVMD pattern in the amd_iommu=relax_unity report from March [1] (Gigabyte AM5, type 0x22 over devid 0x0000-0x0fff). That report didn't include the IVMD flags (Vasant asked for them); on this board they are exclusion-only (0x08). If those boards are the same, this series should fix them too without relax_unity. I can send the raw IVRS dump if useful.

[1] https://lore.kernel.org/all/[email protected]/

I'm happy to retest v3 when it's out.

Thanks,
Adam