[PATCHv3 3/6] Crashdump-Accepting-Active-IOMMU-Domain-Interfaces

[Bill Sumner]

---------.---------.---------.---------.---------.---------.---------.
This patch contains functions called from among the mainline code
to retrieve information from the translation tables that have been
copied into the crashdump kernel from the panicked kernel.

Most often this happens when the crashdump kernel is setting-up a new
domain for a device.  The crashdump kernel will assign to the device
(and to its new domain) the same IOVA addressing width and domain-id
that were used for this device by the panicked kernel. In the new
domain all of the IOVA addresses that were in-use by this device at
the time of the panic will be marked as "in-use" so that the crashdump
kernel will avoid re-using them.

This patch also includes one function to retrieve a bitmap of all
domain-ids in-use by the panicked kernel.  These are marked "in-use"
so that if there is a device being used by the crashdump kernel that
was not being used by the panicked kernel then that new device will
receive a fresh, unused domain-id.

The 'device_domain_values_list' is used during the operation of
duplicating (copying) the translation tables from the panicked kernel
into the crashdump kernel, to insure that all devices that were assigned
to any specific domain-id in the panicked kernel are also assigned to
that same domain-id in the crashdump kernel.  Additionally, all context
entries (there is one per device) that contain the same domain-id *must*
point to the same set of page-tables.  The 'device_domain_values_list'
insures that if the copy operation has already seen 'this' domain-id,
it will simply point the current context entry to the same top-level
page-table that has been created previously for this domain-id.

No locks are necessary for 'device_domain_values_list'.  Updates
are done only during the initialization operation of copying the
translation tables -- when only a single CPU is active.
Accesses after Linux goes multi-CPU are all read-only.

v1->v2:
Updated patch description

v2->v3:
No change

Signed-off-by: Bill Sumner <bill.sumner@xxxxxx>
---
 drivers/iommu/intel-iommu.c | 266 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 266 insertions(+)

diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
index 4172a2b..85e30e5 100644
--- a/drivers/iommu/intel-iommu.c
+++ b/drivers/iommu/intel-iommu.c
@@ -4529,3 +4529,269 @@ static int oldcopy(void *to, void *from, int size)
 	return (int) ret;
 }
 #endif /* CONFIG_CRASH_DUMP */
+#ifdef CONFIG_CRASH_DUMP
+
+
+
+/* ------------------------------------------------------------------------
+ * Interfaces for when a new domain in the new kernel needs some values
+ * from the old kernel's context entries
+ * ------------------------------------------------------------------------
+ */
+
+/* List to hold domain values found during the copy operation */
+static struct list_head *device_domain_values_list;
+
+
+/* Utility function for interface functions that follow. */
+static int
+context_get_entry(struct context_entry *context_addr,
+			struct intel_iommu *iommu, u32 bus, int devfn)
+{
+	unsigned long long q;		/* quadword scratch */
+	struct root_entry *root_phys;	/* Phys adr (root table entry) */
+	struct root_entry  root_temp;	/* Local copy of root_entry */
+	struct context_entry *context_phys;	/* Phys adr */
+
+	if (pr_dbg.domain_get)
+		pr_debug("ENTER %s B:D:F=%2.2x:%2.2x:%1.1x &context_entry:0x%llx &intel_iommu:0x%llx\n",
+			__func__, bus, devfn>>3, devfn&7,
+			(u64)context_addr, (u64)iommu);
+
+	if (bus > 255)				/* Sanity check */
+		return -5;
+	if (devfn > 255 || devfn < 0)		/* Sanity check */
+		return -6;
+
+	q = readq(iommu->reg + DMAR_RTADDR_REG);
+	pr_debug("IOMMU %d: DMAR_RTADDR_REG:0x%16.16llx\n", iommu->seq_id, q);
+	if (!q)
+		return -1;
+
+	root_phys = (struct root_entry *) q;	/* Adr(base of vector) */
+	root_phys += bus;			/* Adr(entry we want) */
+
+	oldcopy(&root_temp, root_phys, sizeof(root_temp));
+
+	pr_debug("root_temp.val:0x%llx .rsvd1:0x%llx root_phys:0x%llx\n",
+		root_temp.val, root_temp.rsvd1, (u64)root_phys);
+
+	if (!root_present(&root_temp))
+		return -2;
+
+	pr_debug("B:D:F=%2.2x:%2.2x:%1.1x root_temp.val: %llx .rsvd1: %llx\n",
+		bus, devfn >> 3, devfn & 7, root_temp.val, root_temp.rsvd1);
+
+	if (root_temp.rsvd1)			/* If (root_entry is bad) */
+		return -3;
+
+	context_phys = get_context_phys_from_root(&root_temp);
+	if (!context_phys)
+		return -4;
+
+	context_phys += devfn;			/* Adr(context_entry we want) */
+
+
+	oldcopy(context_addr, context_phys, sizeof(*context_addr));
+
+	if (pr_dbg.domain_get)
+		pr_debug("LEAVE %s returning: phys:0x%12.12llx hi:0x%16.16llx lo:0x%16.16llx\n",
+			__func__, (u64) context_phys,
+			context_addr->hi, context_addr->lo);
+	return 0;
+}
+
+
+/* Get address_width of iova for a device from old kernel (if device existed) */
+static int
+domain_get_gaw_from_old_kernel(struct intel_iommu *iommu, struct pci_dev *pdev)
+{
+	int ret;
+	struct context_entry context_temp;
+
+	if (pr_dbg.domain_get)
+		pr_debug("ENTER %s B:D:F=%2.2x:%2.2x:%1.1x iommu:%d\n",
+			__func__, pdev->bus->number,
+			pdev->devfn >> 3, pdev->devfn & 7,
+			iommu->seq_id);
+
+	ret = context_get_entry(&context_temp, iommu,
+				pdev->bus->number, pdev->devfn);
+	if (ret < 0)
+		return ret;
+
+	return (int) agaw_to_width(context_get_aw(&context_temp));
+}
+
+
+/* Get domain_id for a device from old kernel (if device existed) */
+static int
+domain_get_did_from_old_kernel(struct intel_iommu *iommu, struct pci_dev *pdev)
+{
+	int ret;
+	struct context_entry context_temp;
+
+	if (pr_dbg.domain_get)
+		pr_debug("ENTER %s B:D:F=%2.2x:%2.2x:%1.1x iommu:%d\n",
+			__func__, pdev->bus->number,
+			pdev->devfn >> 3, pdev->devfn & 7,
+			iommu->seq_id);
+
+	ret = context_get_entry(&context_temp, iommu,
+				pdev->bus->number, pdev->devfn);
+	if (ret < 0)
+		return ret;
+
+	return (int) context_get_did(&context_temp);
+}
+
+
+/* Get adr(top page_table) for a device from old kernel (if device exists) */
+static u64
+domain_get_pgd_from_old_kernel(struct intel_iommu *iommu, struct pci_dev *pdev)
+{
+	int ret;
+	struct context_entry context_temp;
+	u64 phys;
+	u64 virt;
+
+	if (pr_dbg.domain_get)
+		pr_debug("ENTER %s B:D:F=%2.2x:%2.2x:%1.1x iommu:%d\n",
+			__func__, pdev->bus->number,
+			pdev->devfn >> 3, pdev->devfn & 7,
+			iommu->seq_id);
+
+	ret = context_get_entry(&context_temp, iommu,
+				pdev->bus->number, pdev->devfn);
+	if (ret < 0)
+		return 0;
+	if (!context_get_p(&context_temp))
+		return 0;
+
+	phys = context_get_asr(&context_temp) << VTD_PAGE_SHIFT;
+	if (pr_dbg.domain_get)
+		pr_debug("%s, phys: 0x%16.16llx\n", __func__, (u64) phys);
+
+	if (!phys)
+		return 0;
+
+	virt = (u64) phys_to_virt(phys);
+	if (pr_dbg.domain_get)
+		pr_debug("%s, virt: 0x%16.16llx\n", __func__, (u64) virt);
+
+	return virt;
+}
+
+
+/* Mark IOVAs that are in-use at time of panic by a device of the old kernel.
+ * Mark IOVAs in the domain for that device in the new kernel
+ * so that all new requests from the device driver for an IOVA will avoid
+ * re-using any IOVA that was in-use by the old kernel.
+ */
+static void
+domain_get_ranges_from_old_kernel(struct dmar_domain *domain,
+				  struct intel_iommu *iommu,
+				  struct pci_dev *pdev)
+{
+	u32 bus = pdev->bus->number;
+	int devfn = pdev->devfn;
+	struct device_domain_info *i = NULL;    /* iterator for foreach */
+
+	pr_debug("ENTER %s, iommu=%d, B:D:F=%2.2x:%2.2x:%1.1x\n",
+			__func__, iommu->seq_id,
+			bus, devfn >> 3, devfn & 0x3);
+
+	list_for_each_entry(i, &device_domain_values_list[iommu->seq_id],
+				global) {
+		if (i->bus == bus && i->devfn == devfn) {
+			if (i->domain == NULL) {
+				pr_err("ERROR %s, iommu=%d, B:D:F=%2.2x:%2.2x:%1.1x\n",
+					__func__, iommu->seq_id,
+					bus, devfn >> 3, devfn & 0x3);
+
+				pr_err("FOUND B:D:F=%2.2x:%2.2x:%1.1x INFO domain-pointer is NULL\n",
+					bus, devfn >> 3, devfn & 0x3);
+				break;
+			}
+			pr_debug("FOUND B:D:F=%2.2x:%2.2x:%1.1x did:%4.4x\n",
+				bus, devfn >> 3, devfn & 0x3, i->domain->id);
+
+			copy_reserved_iova(&i->domain->iovad, &domain->iovad);
+			break;
+		}
+	}
+
+	pr_debug("LEAVE %s\n", __func__);
+}
+
+
+/* Mark domain-id's from old kernel as in-use on this iommu so that a new
+ * domain-id is allocated in the case where there is a device in the new kernel
+ * that was not in the old kernel -- and therefore a new domain-id is needed.
+ */
+static int intel_iommu_get_dids_from_old_kernel(struct intel_iommu *iommu)
+{
+	unsigned long long q;		/* quadword scratch */
+	struct root_entry *root_phys;	/* Phys(in old kernel) */
+	struct root_entry *root_temp;	/* Virt(Local copy) */
+	struct root_entry *re;		/* Loop index */
+	struct context_entry *context_phys;	/* Phys(in old kernel) */
+	struct context_entry *context_temp;	/* Virt(Local copy) */
+	struct context_entry *ce;	/* Loop index */
+	int did;			/* Each domain-id found */
+	u32 bus;			/* Index into root-entry-table */
+	u32 devfn;			/* Index into context-entry-table */
+
+	pr_debug("ENTER %s iommu:%d\n", __func__, iommu->seq_id);
+
+	q = readq(iommu->reg + DMAR_RTADDR_REG);
+	pr_debug("IOMMU %d: DMAR_RTADDR_REG:0x%16.16llx\n", iommu->seq_id, q);
+	if (!q)
+		return -ENOMEM;
+
+	root_phys = (void *)q;
+	root_temp = (struct root_entry *)alloc_pgtable_page(iommu->node);
+	if (!root_temp)
+		return -ENOMEM;
+	oldcopy(root_temp, root_phys, PAGE_SIZE);
+
+	context_temp = (struct context_entry *)alloc_pgtable_page(iommu->node);
+	if (!context_temp) {
+		free_pgtable_page(root_temp);
+		return -ENOMEM;
+	}
+
+	for (bus = 0, re = root_temp; bus < 256; bus += 1, re += 1) {
+
+		if (!root_present(re))
+			continue;
+
+		pr_debug("ROOT B:%2.2x val: %16.16llx rsvd1: %16.16llx\n",
+			bus, re->val, re->rsvd1);
+
+		if (re->rsvd1)			/* If (root_entry is bad) */
+			continue;
+
+		context_phys = get_context_phys_from_root(re);
+		if (!context_phys)
+			continue;
+
+		oldcopy(context_temp, context_phys, PAGE_SIZE);
+
+		for (devfn = 0, ce = context_temp; devfn < 512; devfn++, ce++) {
+			if (!context_get_p(ce))
+				continue;
+
+			did = context_get_did(ce);
+			set_bit(did, iommu->domain_ids);
+			pr_debug("DID B:D:F:%2.2x:%2.2x:%1.1x did:%d(0x%4.4x)\n",
+				bus, devfn >> 3, devfn & 0x7, did, did);
+		}
+
+	}
+	free_pgtable_page(root_temp);
+	free_pgtable_page(context_temp);
+	pr_debug("LEAVE %s iommu:%d\n", __func__, iommu->seq_id);
+	return 0;
+}
+#endif /* CONFIG_CRASH_DUMP */
-- 
Bill Sumner <bill.sumner@xxxxxx>

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/