On Tue, Nov 10 2020 at 18:50, Thomas Gleixner wrote:
> On Tue, Nov 10 2020 at 16:33, David Woodhouse wrote:
>> If I could get post-5.5 kernels to boot at all with the AMD IOMMU
>> enabled, I'd have a go at throwing that together now...
>
> It can share the dmar domain code. Let me frob something.
Not much to share there and I can't access my AMD machine at the
moment. Something like the untested below should work.
Thanks,
tglx
---
--- a/arch/x86/include/asm/apic.h
+++ b/arch/x86/include/asm/apic.h
@@ -523,7 +523,7 @@ struct msi_msg;
struct irq_cfg;
extern void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
- bool dmar);
+ bool iommu);
extern void ioapic_zap_locks(void);
--- a/arch/x86/include/asm/irqdomain.h
+++ b/arch/x86/include/asm/irqdomain.h
@@ -63,4 +63,6 @@ static inline void x86_create_pci_msi_do
#define x86_pci_msi_default_domain NULL
#endif
+struct irq_domain *x86_create_iommu_msi_domain(void);
+
#endif
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -2498,7 +2498,7 @@ int hard_smp_processor_id(void)
}
void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
- bool dmar)
+ bool iommu)
{
memset(msg, 0, sizeof(*msg));
@@ -2519,7 +2519,7 @@ void __irq_msi_compose_msg(struct irq_cf
* some hypervisors allow the extended destination ID field in bits
* 5-11 to be used, giving support for 15 bits of APIC IDs in total.
*/
- if (dmar)
+ if (iommu)
msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
--- a/arch/x86/kernel/apic/msi.c
+++ b/arch/x86/kernel/apic/msi.c
@@ -184,7 +184,8 @@ static struct msi_domain_info pci_msi_do
.handler_name = "edge",
};
-struct irq_domain * __init native_create_pci_msi_domain(void)
+static struct irq_domain *__init
+__create_pci_msi_domain(struct msi_domain_info *info, const char *name)
{
struct fwnode_handle *fn;
struct irq_domain *d;
@@ -192,21 +193,25 @@ struct irq_domain * __init native_create
if (disable_apic)
return NULL;
- fn = irq_domain_alloc_named_fwnode("PCI-MSI");
+ fn = irq_domain_alloc_named_fwnode(name);
if (!fn)
return NULL;
- d = pci_msi_create_irq_domain(fn, &pci_msi_domain_info,
- x86_vector_domain);
+ d = pci_msi_create_irq_domain(fn, info, x86_vector_domain);
if (!d) {
irq_domain_free_fwnode(fn);
- pr_warn("Failed to initialize PCI-MSI irqdomain.\n");
+ pr_warn("Failed to initialize %s irqdomain.\n", name);
} else {
d->flags |= IRQ_DOMAIN_MSI_NOMASK_QUIRK;
}
return d;
}
+struct irq_domain * __init native_create_pci_msi_domain(void)
+{
+ return __create_pci_msi_domain(&pci_msi_domain_info, "PCI-MSI");
+}
+
void __init x86_create_pci_msi_domain(void)
{
x86_pci_msi_default_domain = x86_init.irqs.create_pci_msi_domain();
@@ -247,6 +252,43 @@ struct irq_domain *arch_create_remap_msi
}
#endif
+static void __maybe_unused iommu_msi_compose_msg(struct irq_data *data,
+ struct msi_msg *msg)
+{
+ __irq_msi_compose_msg(irqd_cfg(data), msg, true);
+}
+
+#ifdef CONFIG_AMD_IOMMU
+/*
+ * Similar to the Intel IOMMU abuse below. The resulting irq domain is
+ * associated to the IOMMU pci device, so that pci_enable_msi() works.
+ */
+static struct irq_chip iommu_msi_controller = {
+ .name = "IOMMU-MSI",
+ .irq_unmask = pci_msi_unmask_irq,
+ .irq_mask = pci_msi_mask_irq,
+ .irq_ack = irq_chip_ack_parent,
+ .irq_retrigger = irq_chip_retrigger_hierarchy,
+ .irq_set_affinity = msi_set_affinity,
+ .irq_compose_msi_msg = iommu_msi_compose_msg,
+ .flags = IRQCHIP_SKIP_SET_WAKE,
+};
+
+static struct msi_domain_info iommu_msi_domain_info = {
+ .flags = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
+ MSI_FLAG_PCI_MSIX,
+ .ops = &pci_msi_domain_ops,
+ .chip = &iommu_msi_controller,
+ .handler = handle_edge_irq,
+ .handler_name = "edge",
+};
+
+struct irq_domain __init *x86_create_iommu_msi_domain(void)
+{
+ return __create_pci_msi_domain(&iommu_msi_domain_info, "IOMMU-MSI");
+}
+#endif
+
#ifdef CONFIG_DMAR_TABLE
/*
* The Intel IOMMU (ab)uses the high bits of the MSI address to contain the
@@ -254,11 +296,6 @@ struct irq_domain *arch_create_remap_msi
* case for MSIs as it would be targeting real memory above 4GiB not the
* APIC.
*/
-static void dmar_msi_compose_msg(struct irq_data *data, struct msi_msg *msg)
-{
- __irq_msi_compose_msg(irqd_cfg(data), msg, true);
-}
-
static void dmar_msi_write_msg(struct irq_data *data, struct msi_msg *msg)
{
dmar_msi_write(data->irq, msg);
@@ -271,7 +308,7 @@ static struct irq_chip dmar_msi_controll
.irq_ack = irq_chip_ack_parent,
.irq_set_affinity = msi_domain_set_affinity,
.irq_retrigger = irq_chip_retrigger_hierarchy,
- .irq_compose_msi_msg = dmar_msi_compose_msg,
+ .irq_compose_msi_msg = iommu_msi_compose_msg,
.irq_write_msi_msg = dmar_msi_write_msg,
.flags = IRQCHIP_SKIP_SET_WAKE,
};
--- a/drivers/iommu/amd/init.c
+++ b/drivers/iommu/amd/init.c
@@ -1759,6 +1759,7 @@ static const struct attribute_group *amd
static int __init iommu_init_pci(struct amd_iommu *iommu)
{
+ static struct irq_domain *msidom;
int cap_ptr = iommu->cap_ptr;
int ret;
@@ -1767,6 +1768,13 @@ static int __init iommu_init_pci(struct
if (!iommu->dev)
return -ENODEV;
+ if (!msidom) {
+ msidom = x86_create_iommu_msi_domain();
+ if (!msidom)
+ return -ENODEV;
+ }
+ dev_set_msi_domain(&iommu->dev->dev, msidom);
+
/* Prevent binding other PCI device drivers to IOMMU devices */
iommu->dev->match_driver = false;
