2020-11-05 17:00:56

by Wei Liu

[permalink] [raw]
Subject: [PATCH v2 17/17] x86/hyperv: handle IO-APIC when running as root

Just like MSI/MSI-X, IO-APIC interrupts are remapped by Microsoft
Hypervisor when Linux runs as the root partition. Implement an IRQ chip
to handle mapping and unmapping of IO-APIC interrupts.

Use custom functions for mapping and unmapping ACPI GSIs. They will
issue Microsoft Hypervisor specific hypercalls on top of the native
routines.

Signed-off-by: Sunil Muthuswamy <[email protected]>
Co-Developed-by: Sunil Muthuswamy <[email protected]>
Signed-off-by: Wei Liu <[email protected]>
---
arch/x86/hyperv/hv_init.c | 13 +++
arch/x86/hyperv/irqdomain.c | 226 ++++++++++++++++++++++++++++++++++++
2 files changed, 239 insertions(+)

diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c
index a46b817b5b2a..2c2189832da7 100644
--- a/arch/x86/hyperv/hv_init.c
+++ b/arch/x86/hyperv/hv_init.c
@@ -267,10 +267,23 @@ static int hv_cpu_die(unsigned int cpu)
return 0;
}

+int hv_acpi_register_gsi(struct device *dev, u32 gsi, int trigger, int polarity);
+void hv_acpi_unregister_gsi(u32 gsi);
+
+extern int (*native_acpi_register_gsi)(struct device *dev, u32 gsi, int trigger, int polarity);
+extern void (*native_acpi_unregister_gsi)(u32 gsi);
+
static int __init hv_pci_init(void)
{
int gen2vm = efi_enabled(EFI_BOOT);

+ if (hv_root_partition) {
+ native_acpi_register_gsi = __acpi_register_gsi;
+ native_acpi_unregister_gsi = __acpi_unregister_gsi;
+ __acpi_register_gsi = hv_acpi_register_gsi;
+ __acpi_unregister_gsi = hv_acpi_unregister_gsi;
+ }
+
/*
* For Generation-2 VM, we exit from pci_arch_init() by returning 0.
* The purpose is to suppress the harmless warning:
diff --git a/arch/x86/hyperv/irqdomain.c b/arch/x86/hyperv/irqdomain.c
index 80109e3cbf8f..ae9a728589a4 100644
--- a/arch/x86/hyperv/irqdomain.c
+++ b/arch/x86/hyperv/irqdomain.c
@@ -9,6 +9,8 @@
#include <linux/pci.h>
#include <linux/irq.h>
#include <asm/mshyperv.h>
+#include <asm/apic.h>
+#include <asm/io_apic.h>

struct rid_data {
struct pci_dev *bridge;
@@ -328,3 +330,227 @@ struct irq_domain * __init hv_create_pci_msi_domain(void)
return d;
}

+/* Copied from io_apic.c */
+union entry_union {
+ struct { u32 w1, w2; };
+ struct IO_APIC_route_entry entry;
+};
+
+static int hv_unmap_ioapic_interrupt(int gsi)
+{
+ union hv_device_id device_id;
+ int ioapic, ioapic_id;
+ u8 ioapic_pin;
+ struct IO_APIC_route_entry ire;
+ union entry_union eu;
+ struct hv_interrupt_entry entry;
+
+ ioapic = mp_find_ioapic(gsi);
+ ioapic_pin = mp_find_ioapic_pin(ioapic, gsi);
+ ioapic_id = mpc_ioapic_id(ioapic);
+ ire = ioapic_read_entry(ioapic, ioapic_pin);
+
+ eu.entry = ire;
+
+ /*
+ * Polarity may have been set by us, but Hyper-V expects the exact same
+ * entry. See the mapping routine.
+ */
+ eu.entry.polarity = 0;
+
+ memset(&entry, 0, sizeof(entry));
+ entry.source = HV_INTERRUPT_SOURCE_IOAPIC;
+ entry.ioapic_rte.low_uint32 = eu.w1;
+ entry.ioapic_rte.high_uint32 = eu.w2;
+
+ device_id.as_uint64 = 0;
+ device_id.device_type = HV_DEVICE_TYPE_IOAPIC;
+ device_id.ioapic.ioapic_id = (u8)ioapic_id;
+
+ return hv_unmap_interrupt(device_id.as_uint64, &entry) & HV_HYPERCALL_RESULT_MASK;
+}
+
+static int hv_map_ioapic_interrupt(int ioapic_id, int trigger, int vcpu, int vector,
+ struct hv_interrupt_entry *out_entry)
+{
+ unsigned long flags;
+ struct hv_input_map_device_interrupt *input;
+ struct hv_output_map_device_interrupt *output;
+ union hv_device_id device_id;
+ struct hv_device_interrupt_descriptor *intr_desc;
+ u16 status;
+
+ device_id.as_uint64 = 0;
+ device_id.device_type = HV_DEVICE_TYPE_IOAPIC;
+ device_id.ioapic.ioapic_id = (u8)ioapic_id;
+
+ local_irq_save(flags);
+ input = *this_cpu_ptr(hyperv_pcpu_input_arg);
+ output = *this_cpu_ptr(hyperv_pcpu_output_arg);
+ memset(input, 0, sizeof(*input));
+ intr_desc = &input->interrupt_descriptor;
+ input->partition_id = hv_current_partition_id;
+ input->device_id = device_id.as_uint64;
+ intr_desc->interrupt_type = HV_X64_INTERRUPT_TYPE_FIXED;
+ intr_desc->target.vector = vector;
+ intr_desc->vector_count = 1;
+
+ if (trigger)
+ intr_desc->trigger_mode = HV_INTERRUPT_TRIGGER_MODE_LEVEL;
+ else
+ intr_desc->trigger_mode = HV_INTERRUPT_TRIGGER_MODE_EDGE;
+
+ __set_bit(vcpu, (unsigned long *)&intr_desc->target.vp_mask);
+
+ status = hv_do_rep_hypercall(HVCALL_MAP_DEVICE_INTERRUPT, 0, 0, input, output) &
+ HV_HYPERCALL_RESULT_MASK;
+ local_irq_restore(flags);
+
+ *out_entry = output->interrupt_entry;
+
+ return status;
+}
+
+static unsigned int hv_ioapic_startup_irq(struct irq_data *data)
+{
+ u16 status;
+ struct IO_APIC_route_entry ire;
+ u32 vector;
+ struct irq_cfg *cfg;
+ int ioapic;
+ u8 ioapic_pin;
+ int ioapic_id;
+ int gsi;
+ union entry_union eu;
+ struct cpumask *affinity;
+ int cpu, vcpu;
+ struct hv_interrupt_entry entry;
+ struct mp_chip_data *mp_data = data->chip_data;
+
+ gsi = data->irq;
+ cfg = irqd_cfg(data);
+ affinity = irq_data_get_effective_affinity_mask(data);
+ cpu = cpumask_first_and(affinity, cpu_online_mask);
+ vcpu = hv_cpu_number_to_vp_number(cpu);
+
+ vector = cfg->vector;
+
+ ioapic = mp_find_ioapic(gsi);
+ ioapic_pin = mp_find_ioapic_pin(ioapic, gsi);
+ ioapic_id = mpc_ioapic_id(ioapic);
+ ire = ioapic_read_entry(ioapic, ioapic_pin);
+
+ /*
+ * Always try unmapping. We do not have visibility into which whether
+ * an IO-APIC has been mapped or not. We can't use chip_data because it
+ * already points to mp_data.
+ *
+ * We don't use retarget interrupt hypercalls here because Hyper-V
+ * doens't allow root to change the vector or specify VPs outside of
+ * the set that is initially used during mapping.
+ */
+ status = hv_unmap_ioapic_interrupt(gsi);
+
+ if (!(status == HV_STATUS_SUCCESS || status == HV_STATUS_INVALID_PARAMETER)) {
+ pr_debug("%s: unexpected unmap status %d\n", __func__, status);
+ return -1;
+ }
+
+ status = hv_map_ioapic_interrupt(ioapic_id, ire.trigger, vcpu, vector, &entry);
+
+ if (status != HV_STATUS_SUCCESS) {
+ pr_err("%s: map hypercall failed, status %d\n", __func__, status);
+ return -1;
+ }
+
+ /* Update the entry in mp_chip_data. It is used in other places. */
+ mp_data->entry = *(struct IO_APIC_route_entry *)&entry.ioapic_rte;
+
+ /* Sync polarity -- Hyper-V's returned polarity is always 0... */
+ mp_data->entry.polarity = ire.polarity;
+
+ eu.w1 = entry.ioapic_rte.low_uint32;
+ eu.w2 = entry.ioapic_rte.high_uint32;
+ ioapic_write_entry(ioapic, ioapic_pin, eu.entry);
+
+ return 0;
+}
+
+static void hv_ioapic_mask_irq(struct irq_data *data)
+{
+ mask_ioapic_irq(data);
+}
+
+static void hv_ioapic_unmask_irq(struct irq_data *data)
+{
+ unmask_ioapic_irq(data);
+}
+
+static int hv_ioapic_set_affinity(struct irq_data *data,
+ const struct cpumask *mask, bool force)
+{
+ /*
+ * We only update the affinity mask here. Programming the hardware is
+ * done in irq_startup.
+ */
+ return ioapic_set_affinity(data, mask, force);
+}
+
+void hv_ioapic_ack_level(struct irq_data *irq_data)
+{
+ /*
+ * Per email exchange with Hyper-V team, all is needed is write to
+ * LAPIC's EOI register. They don't support directed EOI to IO-APIC.
+ * Hyper-V handles it for us.
+ */
+ apic_ack_irq(irq_data);
+}
+
+struct irq_chip hv_ioapic_chip __read_mostly = {
+ .name = "HV-IO-APIC",
+ .irq_startup = hv_ioapic_startup_irq,
+ .irq_mask = hv_ioapic_mask_irq,
+ .irq_unmask = hv_ioapic_unmask_irq,
+ .irq_ack = irq_chip_ack_parent,
+ .irq_eoi = hv_ioapic_ack_level,
+ .irq_set_affinity = hv_ioapic_set_affinity,
+ .irq_retrigger = irq_chip_retrigger_hierarchy,
+ .irq_get_irqchip_state = ioapic_irq_get_chip_state,
+ .flags = IRQCHIP_SKIP_SET_WAKE,
+};
+
+
+int (*native_acpi_register_gsi)(struct device *dev, u32 gsi, int trigger, int polarity);
+void (*native_acpi_unregister_gsi)(u32 gsi);
+
+int hv_acpi_register_gsi(struct device *dev, u32 gsi, int trigger, int polarity)
+{
+ int irq = gsi;
+
+#ifdef CONFIG_X86_IO_APIC
+ irq = native_acpi_register_gsi(dev, gsi, trigger, polarity);
+ if (irq < 0) {
+ pr_err("native_acpi_register_gsi failed %d\n", irq);
+ return irq;
+ }
+
+ if (trigger) {
+ irq_set_status_flags(irq, IRQ_LEVEL);
+ irq_set_chip_and_handler_name(irq, &hv_ioapic_chip,
+ handle_fasteoi_irq, "ioapic-fasteoi");
+ } else {
+ irq_clear_status_flags(irq, IRQ_LEVEL);
+ irq_set_chip_and_handler_name(irq, &hv_ioapic_chip,
+ handle_edge_irq, "ioapic-edge");
+ }
+#endif
+ return irq;
+}
+
+void hv_acpi_unregister_gsi(u32 gsi)
+{
+#ifdef CONFIG_X86_IO_APIC
+ (void)hv_unmap_ioapic_interrupt(gsi);
+ native_acpi_unregister_gsi(gsi);
+#endif
+}
--
2.20.1


2020-11-12 16:58:42

by Vitaly Kuznetsov

[permalink] [raw]
Subject: Re: [PATCH v2 17/17] x86/hyperv: handle IO-APIC when running as root

Wei Liu <[email protected]> writes:

> Just like MSI/MSI-X, IO-APIC interrupts are remapped by Microsoft
> Hypervisor when Linux runs as the root partition. Implement an IRQ chip
> to handle mapping and unmapping of IO-APIC interrupts.
>
> Use custom functions for mapping and unmapping ACPI GSIs. They will
> issue Microsoft Hypervisor specific hypercalls on top of the native
> routines.
>
> Signed-off-by: Sunil Muthuswamy <[email protected]>
> Co-Developed-by: Sunil Muthuswamy <[email protected]>
> Signed-off-by: Wei Liu <[email protected]>
> ---
> arch/x86/hyperv/hv_init.c | 13 +++
> arch/x86/hyperv/irqdomain.c | 226 ++++++++++++++++++++++++++++++++++++
> 2 files changed, 239 insertions(+)
>
> diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c
> index a46b817b5b2a..2c2189832da7 100644
> --- a/arch/x86/hyperv/hv_init.c
> +++ b/arch/x86/hyperv/hv_init.c
> @@ -267,10 +267,23 @@ static int hv_cpu_die(unsigned int cpu)
> return 0;
> }
>
> +int hv_acpi_register_gsi(struct device *dev, u32 gsi, int trigger, int polarity);
> +void hv_acpi_unregister_gsi(u32 gsi);
> +
> +extern int (*native_acpi_register_gsi)(struct device *dev, u32 gsi, int trigger, int polarity);
> +extern void (*native_acpi_unregister_gsi)(u32 gsi);
> +
> static int __init hv_pci_init(void)
> {
> int gen2vm = efi_enabled(EFI_BOOT);
>
> + if (hv_root_partition) {
> + native_acpi_register_gsi = __acpi_register_gsi;
> + native_acpi_unregister_gsi = __acpi_unregister_gsi;
> + __acpi_register_gsi = hv_acpi_register_gsi;
> + __acpi_unregister_gsi = hv_acpi_unregister_gsi;
> + }
> +
> /*
> * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
> * The purpose is to suppress the harmless warning:
> diff --git a/arch/x86/hyperv/irqdomain.c b/arch/x86/hyperv/irqdomain.c
> index 80109e3cbf8f..ae9a728589a4 100644
> --- a/arch/x86/hyperv/irqdomain.c
> +++ b/arch/x86/hyperv/irqdomain.c
> @@ -9,6 +9,8 @@
> #include <linux/pci.h>
> #include <linux/irq.h>
> #include <asm/mshyperv.h>
> +#include <asm/apic.h>
> +#include <asm/io_apic.h>
>
> struct rid_data {
> struct pci_dev *bridge;
> @@ -328,3 +330,227 @@ struct irq_domain * __init hv_create_pci_msi_domain(void)
> return d;
> }
>
> +/* Copied from io_apic.c */
> +union entry_union {
> + struct { u32 w1, w2; };
> + struct IO_APIC_route_entry entry;
> +};
> +
> +static int hv_unmap_ioapic_interrupt(int gsi)
> +{
> + union hv_device_id device_id;
> + int ioapic, ioapic_id;
> + u8 ioapic_pin;
> + struct IO_APIC_route_entry ire;
> + union entry_union eu;
> + struct hv_interrupt_entry entry;
> +
> + ioapic = mp_find_ioapic(gsi);
> + ioapic_pin = mp_find_ioapic_pin(ioapic, gsi);
> + ioapic_id = mpc_ioapic_id(ioapic);
> + ire = ioapic_read_entry(ioapic, ioapic_pin);
> +
> + eu.entry = ire;
> +
> + /*
> + * Polarity may have been set by us, but Hyper-V expects the exact same
> + * entry. See the mapping routine.
> + */
> + eu.entry.polarity = 0;
> +
> + memset(&entry, 0, sizeof(entry));
> + entry.source = HV_INTERRUPT_SOURCE_IOAPIC;
> + entry.ioapic_rte.low_uint32 = eu.w1;
> + entry.ioapic_rte.high_uint32 = eu.w2;
> +
> + device_id.as_uint64 = 0;
> + device_id.device_type = HV_DEVICE_TYPE_IOAPIC;
> + device_id.ioapic.ioapic_id = (u8)ioapic_id;
> +
> + return hv_unmap_interrupt(device_id.as_uint64, &entry) & HV_HYPERCALL_RESULT_MASK;
> +}
> +
> +static int hv_map_ioapic_interrupt(int ioapic_id, int trigger, int vcpu, int vector,
> + struct hv_interrupt_entry *out_entry)
> +{
> + unsigned long flags;
> + struct hv_input_map_device_interrupt *input;
> + struct hv_output_map_device_interrupt *output;
> + union hv_device_id device_id;
> + struct hv_device_interrupt_descriptor *intr_desc;
> + u16 status;
> +
> + device_id.as_uint64 = 0;
> + device_id.device_type = HV_DEVICE_TYPE_IOAPIC;
> + device_id.ioapic.ioapic_id = (u8)ioapic_id;
> +
> + local_irq_save(flags);
> + input = *this_cpu_ptr(hyperv_pcpu_input_arg);
> + output = *this_cpu_ptr(hyperv_pcpu_output_arg);
> + memset(input, 0, sizeof(*input));
> + intr_desc = &input->interrupt_descriptor;
> + input->partition_id = hv_current_partition_id;
> + input->device_id = device_id.as_uint64;
> + intr_desc->interrupt_type = HV_X64_INTERRUPT_TYPE_FIXED;
> + intr_desc->target.vector = vector;
> + intr_desc->vector_count = 1;
> +
> + if (trigger)
> + intr_desc->trigger_mode = HV_INTERRUPT_TRIGGER_MODE_LEVEL;
> + else
> + intr_desc->trigger_mode = HV_INTERRUPT_TRIGGER_MODE_EDGE;
> +
> + __set_bit(vcpu, (unsigned long *)&intr_desc->target.vp_mask);
> +
> + status = hv_do_rep_hypercall(HVCALL_MAP_DEVICE_INTERRUPT, 0, 0, input, output) &
> + HV_HYPERCALL_RESULT_MASK;
> + local_irq_restore(flags);
> +
> + *out_entry = output->interrupt_entry;
> +
> + return status;
> +}
> +
> +static unsigned int hv_ioapic_startup_irq(struct irq_data *data)
> +{
> + u16 status;
> + struct IO_APIC_route_entry ire;
> + u32 vector;
> + struct irq_cfg *cfg;
> + int ioapic;
> + u8 ioapic_pin;
> + int ioapic_id;
> + int gsi;
> + union entry_union eu;
> + struct cpumask *affinity;
> + int cpu, vcpu;
> + struct hv_interrupt_entry entry;
> + struct mp_chip_data *mp_data = data->chip_data;
> +
> + gsi = data->irq;
> + cfg = irqd_cfg(data);
> + affinity = irq_data_get_effective_affinity_mask(data);
> + cpu = cpumask_first_and(affinity, cpu_online_mask);
> + vcpu = hv_cpu_number_to_vp_number(cpu);
> +
> + vector = cfg->vector;
> +
> + ioapic = mp_find_ioapic(gsi);
> + ioapic_pin = mp_find_ioapic_pin(ioapic, gsi);
> + ioapic_id = mpc_ioapic_id(ioapic);
> + ire = ioapic_read_entry(ioapic, ioapic_pin);
> +
> + /*
> + * Always try unmapping. We do not have visibility into which whether
> + * an IO-APIC has been mapped or not. We can't use chip_data because it
> + * already points to mp_data.
> + *
> + * We don't use retarget interrupt hypercalls here because Hyper-V
> + * doens't allow root to change the vector or specify VPs outside of
> + * the set that is initially used during mapping.
> + */
> + status = hv_unmap_ioapic_interrupt(gsi);
> +
> + if (!(status == HV_STATUS_SUCCESS || status == HV_STATUS_INVALID_PARAMETER)) {
> + pr_debug("%s: unexpected unmap status %d\n", __func__, status);
> + return -1;

Nit: the function returns 'unsigned int' but I see other 'irq_startup'
routines return negative values too, however, they tend to returd
'-ESOMETHING' so maybe -EFAULT here?

> + }
> +
> + status = hv_map_ioapic_interrupt(ioapic_id, ire.trigger, vcpu, vector, &entry);
> +
> + if (status != HV_STATUS_SUCCESS) {
> + pr_err("%s: map hypercall failed, status %d\n", __func__, status);
> + return -1;

and here.

> + }
> +
> + /* Update the entry in mp_chip_data. It is used in other places. */
> + mp_data->entry = *(struct IO_APIC_route_entry *)&entry.ioapic_rte;
> +
> + /* Sync polarity -- Hyper-V's returned polarity is always 0... */
> + mp_data->entry.polarity = ire.polarity;
> +
> + eu.w1 = entry.ioapic_rte.low_uint32;
> + eu.w2 = entry.ioapic_rte.high_uint32;
> + ioapic_write_entry(ioapic, ioapic_pin, eu.entry);
> +
> + return 0;
> +}
> +
> +static void hv_ioapic_mask_irq(struct irq_data *data)
> +{
> + mask_ioapic_irq(data);
> +}
> +
> +static void hv_ioapic_unmask_irq(struct irq_data *data)
> +{
> + unmask_ioapic_irq(data);
> +}
> +
> +static int hv_ioapic_set_affinity(struct irq_data *data,
> + const struct cpumask *mask, bool force)
> +{
> + /*
> + * We only update the affinity mask here. Programming the hardware is
> + * done in irq_startup.
> + */
> + return ioapic_set_affinity(data, mask, force);
> +}
> +
> +void hv_ioapic_ack_level(struct irq_data *irq_data)
> +{
> + /*
> + * Per email exchange with Hyper-V team, all is needed is write to
> + * LAPIC's EOI register. They don't support directed EOI to IO-APIC.
> + * Hyper-V handles it for us.
> + */
> + apic_ack_irq(irq_data);
> +}
> +
> +struct irq_chip hv_ioapic_chip __read_mostly = {
> + .name = "HV-IO-APIC",
> + .irq_startup = hv_ioapic_startup_irq,
> + .irq_mask = hv_ioapic_mask_irq,
> + .irq_unmask = hv_ioapic_unmask_irq,
> + .irq_ack = irq_chip_ack_parent,
> + .irq_eoi = hv_ioapic_ack_level,
> + .irq_set_affinity = hv_ioapic_set_affinity,
> + .irq_retrigger = irq_chip_retrigger_hierarchy,
> + .irq_get_irqchip_state = ioapic_irq_get_chip_state,
> + .flags = IRQCHIP_SKIP_SET_WAKE,
> +};
> +
> +
> +int (*native_acpi_register_gsi)(struct device *dev, u32 gsi, int trigger, int polarity);
> +void (*native_acpi_unregister_gsi)(u32 gsi);
> +
> +int hv_acpi_register_gsi(struct device *dev, u32 gsi, int trigger, int polarity)
> +{
> + int irq = gsi;
> +
> +#ifdef CONFIG_X86_IO_APIC
> + irq = native_acpi_register_gsi(dev, gsi, trigger, polarity);
> + if (irq < 0) {
> + pr_err("native_acpi_register_gsi failed %d\n", irq);
> + return irq;
> + }
> +
> + if (trigger) {
> + irq_set_status_flags(irq, IRQ_LEVEL);
> + irq_set_chip_and_handler_name(irq, &hv_ioapic_chip,
> + handle_fasteoi_irq, "ioapic-fasteoi");
> + } else {
> + irq_clear_status_flags(irq, IRQ_LEVEL);
> + irq_set_chip_and_handler_name(irq, &hv_ioapic_chip,
> + handle_edge_irq, "ioapic-edge");
> + }
> +#endif
> + return irq;
> +}
> +
> +void hv_acpi_unregister_gsi(u32 gsi)
> +{
> +#ifdef CONFIG_X86_IO_APIC
> + (void)hv_unmap_ioapic_interrupt(gsi);
> + native_acpi_unregister_gsi(gsi);
> +#endif
> +}

--
Vitaly

2020-11-13 16:05:20

by Wei Liu

[permalink] [raw]
Subject: Re: [PATCH v2 17/17] x86/hyperv: handle IO-APIC when running as root

On Thu, Nov 12, 2020 at 05:56:41PM +0100, Vitaly Kuznetsov wrote:
[...]
> > +static unsigned int hv_ioapic_startup_irq(struct irq_data *data)
> > +{
> > + u16 status;
> > + struct IO_APIC_route_entry ire;
> > + u32 vector;
> > + struct irq_cfg *cfg;
> > + int ioapic;
> > + u8 ioapic_pin;
> > + int ioapic_id;
> > + int gsi;
> > + union entry_union eu;
> > + struct cpumask *affinity;
> > + int cpu, vcpu;
> > + struct hv_interrupt_entry entry;
> > + struct mp_chip_data *mp_data = data->chip_data;
> > +
> > + gsi = data->irq;
> > + cfg = irqd_cfg(data);
> > + affinity = irq_data_get_effective_affinity_mask(data);
> > + cpu = cpumask_first_and(affinity, cpu_online_mask);
> > + vcpu = hv_cpu_number_to_vp_number(cpu);
> > +
> > + vector = cfg->vector;
> > +
> > + ioapic = mp_find_ioapic(gsi);
> > + ioapic_pin = mp_find_ioapic_pin(ioapic, gsi);
> > + ioapic_id = mpc_ioapic_id(ioapic);
> > + ire = ioapic_read_entry(ioapic, ioapic_pin);
> > +
> > + /*
> > + * Always try unmapping. We do not have visibility into which whether
> > + * an IO-APIC has been mapped or not. We can't use chip_data because it
> > + * already points to mp_data.
> > + *
> > + * We don't use retarget interrupt hypercalls here because Hyper-V
> > + * doens't allow root to change the vector or specify VPs outside of
> > + * the set that is initially used during mapping.
> > + */
> > + status = hv_unmap_ioapic_interrupt(gsi);
> > +
> > + if (!(status == HV_STATUS_SUCCESS || status == HV_STATUS_INVALID_PARAMETER)) {
> > + pr_debug("%s: unexpected unmap status %d\n", __func__, status);
> > + return -1;
>
> Nit: the function returns 'unsigned int' but I see other 'irq_startup'
> routines return negative values too, however, they tend to returd
> '-ESOMETHING' so maybe -EFAULT here?
>

The return type should've been int instead. That's what the function
signature in struct irq_chip looks like.

> > + }
> > +
> > + status = hv_map_ioapic_interrupt(ioapic_id, ire.trigger, vcpu, vector, &entry);
> > +
> > + if (status != HV_STATUS_SUCCESS) {
> > + pr_err("%s: map hypercall failed, status %d\n", __func__, status);
> > + return -1;
>
> and here.
>

-EINVAL would be more appropriate in both cases.

Wei.

> > + }
> > +
> > + /* Update the entry in mp_chip_data. It is used in other places. */
> > + mp_data->entry = *(struct IO_APIC_route_entry *)&entry.ioapic_rte;
> > +
> > + /* Sync polarity -- Hyper-V's returned polarity is always 0... */
> > + mp_data->entry.polarity = ire.polarity;
> > +
> > + eu.w1 = entry.ioapic_rte.low_uint32;
> > + eu.w2 = entry.ioapic_rte.high_uint32;
> > + ioapic_write_entry(ioapic, ioapic_pin, eu.entry);
> > +
> > + return 0;
> > +}
> > +
> > +static void hv_ioapic_mask_irq(struct irq_data *data)
> > +{
> > + mask_ioapic_irq(data);
> > +}
> > +
> > +static void hv_ioapic_unmask_irq(struct irq_data *data)
> > +{
> > + unmask_ioapic_irq(data);
> > +}
> > +
> > +static int hv_ioapic_set_affinity(struct irq_data *data,
> > + const struct cpumask *mask, bool force)
> > +{
> > + /*
> > + * We only update the affinity mask here. Programming the hardware is
> > + * done in irq_startup.
> > + */
> > + return ioapic_set_affinity(data, mask, force);
> > +}
> > +
> > +void hv_ioapic_ack_level(struct irq_data *irq_data)
> > +{
> > + /*
> > + * Per email exchange with Hyper-V team, all is needed is write to
> > + * LAPIC's EOI register. They don't support directed EOI to IO-APIC.
> > + * Hyper-V handles it for us.
> > + */
> > + apic_ack_irq(irq_data);
> > +}
> > +
> > +struct irq_chip hv_ioapic_chip __read_mostly = {
> > + .name = "HV-IO-APIC",
> > + .irq_startup = hv_ioapic_startup_irq,
> > + .irq_mask = hv_ioapic_mask_irq,
> > + .irq_unmask = hv_ioapic_unmask_irq,
> > + .irq_ack = irq_chip_ack_parent,
> > + .irq_eoi = hv_ioapic_ack_level,
> > + .irq_set_affinity = hv_ioapic_set_affinity,
> > + .irq_retrigger = irq_chip_retrigger_hierarchy,
> > + .irq_get_irqchip_state = ioapic_irq_get_chip_state,
> > + .flags = IRQCHIP_SKIP_SET_WAKE,
> > +};
> > +
> > +
> > +int (*native_acpi_register_gsi)(struct device *dev, u32 gsi, int trigger, int polarity);
> > +void (*native_acpi_unregister_gsi)(u32 gsi);
> > +
> > +int hv_acpi_register_gsi(struct device *dev, u32 gsi, int trigger, int polarity)
> > +{
> > + int irq = gsi;
> > +
> > +#ifdef CONFIG_X86_IO_APIC
> > + irq = native_acpi_register_gsi(dev, gsi, trigger, polarity);
> > + if (irq < 0) {
> > + pr_err("native_acpi_register_gsi failed %d\n", irq);
> > + return irq;
> > + }
> > +
> > + if (trigger) {
> > + irq_set_status_flags(irq, IRQ_LEVEL);
> > + irq_set_chip_and_handler_name(irq, &hv_ioapic_chip,
> > + handle_fasteoi_irq, "ioapic-fasteoi");
> > + } else {
> > + irq_clear_status_flags(irq, IRQ_LEVEL);
> > + irq_set_chip_and_handler_name(irq, &hv_ioapic_chip,
> > + handle_edge_irq, "ioapic-edge");
> > + }
> > +#endif
> > + return irq;
> > +}
> > +
> > +void hv_acpi_unregister_gsi(u32 gsi)
> > +{
> > +#ifdef CONFIG_X86_IO_APIC
> > + (void)hv_unmap_ioapic_interrupt(gsi);
> > + native_acpi_unregister_gsi(gsi);
> > +#endif
> > +}
>
> --
> Vitaly
>

2020-11-13 16:08:28

by Wei Liu

[permalink] [raw]
Subject: Re: [PATCH v2 17/17] x86/hyperv: handle IO-APIC when running as root

On Fri, Nov 13, 2020 at 04:01:58PM +0000, Wei Liu wrote:
> On Thu, Nov 12, 2020 at 05:56:41PM +0100, Vitaly Kuznetsov wrote:
> [...]
> > > +static unsigned int hv_ioapic_startup_irq(struct irq_data *data)
> > > +{
> > > + u16 status;
> > > + struct IO_APIC_route_entry ire;
> > > + u32 vector;
> > > + struct irq_cfg *cfg;
> > > + int ioapic;
> > > + u8 ioapic_pin;
> > > + int ioapic_id;
> > > + int gsi;
> > > + union entry_union eu;
> > > + struct cpumask *affinity;
> > > + int cpu, vcpu;
> > > + struct hv_interrupt_entry entry;
> > > + struct mp_chip_data *mp_data = data->chip_data;
> > > +
> > > + gsi = data->irq;
> > > + cfg = irqd_cfg(data);
> > > + affinity = irq_data_get_effective_affinity_mask(data);
> > > + cpu = cpumask_first_and(affinity, cpu_online_mask);
> > > + vcpu = hv_cpu_number_to_vp_number(cpu);
> > > +
> > > + vector = cfg->vector;
> > > +
> > > + ioapic = mp_find_ioapic(gsi);
> > > + ioapic_pin = mp_find_ioapic_pin(ioapic, gsi);
> > > + ioapic_id = mpc_ioapic_id(ioapic);
> > > + ire = ioapic_read_entry(ioapic, ioapic_pin);
> > > +
> > > + /*
> > > + * Always try unmapping. We do not have visibility into which whether
> > > + * an IO-APIC has been mapped or not. We can't use chip_data because it
> > > + * already points to mp_data.
> > > + *
> > > + * We don't use retarget interrupt hypercalls here because Hyper-V
> > > + * doens't allow root to change the vector or specify VPs outside of
> > > + * the set that is initially used during mapping.
> > > + */
> > > + status = hv_unmap_ioapic_interrupt(gsi);
> > > +
> > > + if (!(status == HV_STATUS_SUCCESS || status == HV_STATUS_INVALID_PARAMETER)) {
> > > + pr_debug("%s: unexpected unmap status %d\n", __func__, status);
> > > + return -1;
> >
> > Nit: the function returns 'unsigned int' but I see other 'irq_startup'
> > routines return negative values too, however, they tend to returd
> > '-ESOMETHING' so maybe -EFAULT here?
> >
>
> The return type should've been int instead. That's what the function
> signature in struct irq_chip looks like.

Actually it is unsigned int. Oh well.

Wei.