The I/O-APIC generates an MSI cycle with address/data bits taken from its Redirection Table Entry in some combination which used to make sense, but now is just a bunch of bits which get passed through in some seemingly arbitrary order. Instead of making IRQ remapping drivers directly frob the I/OA-PIC RTE, let them just do their job and generate an MSI message. The bit swizzling to turn that MSI message into the I/O-APIC's RTE is the same in all cases, since it's a function of the I/O-APIC hardware. The IRQ remappers have no real need to get involved with that. The only slight caveat is that the I/OAPIC is interpreting some of those fields too, and it does want the 'vector' field to be unique to make EOI work. The AMD IOMMU happens to put its IRTE index in the bits that the I/O-APIC thinks are the vector field, and accommodates this requirement by reserving the first 32 indices for the I/O-APIC. The Intel IOMMU doesn't actually use the bits that the I/O-APIC thinks are the vector field, so it fills in the 'pin' value there instead. [ tglx: Replaced the unreadably macro maze with the cleaned up RTE/msi_msg bitfields and added commentry to explain the mapping magic ]
Unified · Split
| @@ -45,12 +45,11 @@ enum irq_alloc_type { · 44 unmodified lines | ||
| 45 | 45 | }; |
| 46 | 46 | |
| 47 | 47 | struct ioapic_alloc_info { |
| 48 | − int pin; | |
| 49 | − int node; | |
| 50 | − u32 is_level : 1; | |
| 51 | − u32 active_low : 1; | |
| 52 | − u32 valid : 1; | |
| 53 | − struct IO_APIC_route_entry *entry; | |
| 48 | + int pin; | |
| 49 | + int node; | |
| 50 | + u32 is_level : 1; | |
| 51 | + u32 active_low : 1; | |
| 52 | + u32 valid : 1; | |
| 54 | 53 | }; |
| 55 | 54 | |
| 56 | 55 | struct uv_alloc_info { |
| @@ -48,6 +48,7 @@ · 47 unmodified lines | ||
| 48 | 48 | #include <linux/jiffies.h> /* time_after() */ |
| 49 | 49 | #include <linux/slab.h> |
| 50 | 50 | #include <linux/memblock.h> |
| 51 | +#include <linux/msi.h> | |
| 51 | 52 | |
| 52 | 53 | #include <asm/irqdomain.h> |
| 53 | 54 | #include <asm/io.h> |
| @@ -63,7 +64,6 @@ · 9 unmodified lines | ||
| 63 | 64 | #include <asm/setup.h> |
| 64 | 65 | #include <asm/irq_remapping.h> |
| 65 | 66 | #include <asm/hw_irq.h> |
| 66 | − | |
| 67 | 67 | #include <asm/apic.h> |
| 68 | 68 | |
| 69 | 69 | #define for_each_ioapic(idx) \ |
| @@ -1848,21 +1848,58 @@ static void ioapic_ir_ack_level(struct irq_data *irq_data) · 1778 unmodified lines | ||
| 1848 | 1848 | eoi_ioapic_pin(data->entry.vector, data); |
| 1849 | 1849 | } |
| 1850 | 1850 | |
| 1851 | +/* | |
| 1852 | + * The I/OAPIC is just a device for generating MSI messages from legacy | |
| 1853 | + * interrupt pins. Various fields of the RTE translate into bits of the | |
| 1854 | + * resulting MSI which had a historical meaning. | |
| 1855 | + * | |
| 1856 | + * With interrupt remapping, many of those bits have different meanings | |
| 1857 | + * in the underlying MSI, but the way that the I/OAPIC transforms them | |
| 1858 | + * from its RTE to the MSI message is the same. This function allows | |
| 1859 | + * the parent IRQ domain to compose the MSI message, then takes the | |
| 1860 | + * relevant bits to put them in the appropriate places in the RTE in | |
| 1861 | + * order to generate that message when the IRQ happens. | |
| 1862 | + * | |
| 1863 | + * The setup here relies on a preconfigured route entry (is_level, | |
| 1864 | + * active_low, masked) because the parent domain is merely composing the | |
| 1865 | + * generic message routing information which is used for the MSI. | |
| 1866 | + */ | |
| 1867 | +static void ioapic_setup_msg_from_msi(struct irq_data *irq_data, | |
| 1868 | + struct IO_APIC_route_entry *entry) | |
| 1869 | +{ | |
| 1870 | + struct msi_msg msg; | |
| 1871 | + | |
| 1872 | + /* Let the parent domain compose the MSI message */ | |
| 1873 | + irq_chip_compose_msi_msg(irq_data, &msg); | |
| 1874 | + | |
| 1875 | + /* | |
| 1876 | + * - Real vector | |
| 1877 | + * - DMAR/IR: 8bit subhandle (ioapic.pin) | |
| 1878 | + * - AMD/IR: 8bit IRTE index | |
| 1879 | + */ | |
| 1880 | + entry->vector = msg.arch_data.vector; | |
| 1881 | + /* Delivery mode (for DMAR/IR all 0) */ | |
| 1882 | + entry->delivery_mode = msg.arch_data.delivery_mode; | |
| 1883 | + /* Destination mode or DMAR/IR index bit 15 */ | |
| 1884 | + entry->dest_mode_logical = msg.arch_addr_lo.dest_mode_logical; | |
| 1885 | + /* DMAR/IR: 1, 0 for all other modes */ | |
| 1886 | + entry->ir_format = msg.arch_addr_lo.dmar_format; | |
| 1887 | + /* | |
| 1888 | + * DMAR/IR: index bit 0-14. | |
| 1889 | + * | |
| 1890 | + * All other modes have bit 0-6 of dmar_index_0_14 cleared and the | |
| 1891 | + * topmost 8 bits are destination id bit 0-7 (entry::destid_0_7). | |
| 1892 | + */ | |
| 1893 | + entry->ir_index_0_14 = msg.arch_addr_lo.dmar_index_0_14; | |
| 1894 | +} | |
| 1895 | + | |
| 1851 | 1896 | static void ioapic_configure_entry(struct irq_data *irqd) |
| 1852 | 1897 | { |
| 1853 | 1898 | struct mp_chip_data *mpd = irqd->chip_data; |
| 1854 | − struct irq_cfg *cfg = irqd_cfg(irqd); | |
| 1855 | 1899 | struct irq_pin_list *entry; |
| 1856 | 1900 | |
| 1857 | − /* | |
| 1858 | − * Only update when the parent is the vector domain, don't touch it | |
| 1859 | − * if the parent is the remapping domain. Check the installed | |
| 1860 | − * ioapic chip to verify that. | |
| 1861 | − */ | |
| 1862 | − if (irqd->chip == &ioapic_chip) { | |
| 1863 | − mpd->entry.destid_0_7 = cfg->dest_apicid; | |
| 1864 | − mpd->entry.vector = cfg->vector; | |
| 1865 | − } | |
| 1901 | + ioapic_setup_msg_from_msi(irqd, &mpd->entry); | |
| 1902 | + | |
| 1866 | 1903 | for_each_irq_pin(entry, mpd->irq_2_pin) |
| 1867 | 1904 | __ioapic_write_entry(entry->apic, entry->pin, mpd->entry); |
| 1868 | 1905 | } |
| @@ -2919,14 +2956,23 @@ static void mp_irqdomain_get_attr(u32 gsi, struct mp_chip_data *data, · 1050 unmodified lines | ||
| 2919 | 2956 | } |
| 2920 | 2957 | } |
| 2921 | 2958 | |
| 2922 | −static void mp_setup_entry(struct irq_cfg *cfg, struct mp_chip_data *data, | |
| 2923 | − struct IO_APIC_route_entry *entry) | |
| 2959 | +/* | |
| 2960 | + * Configure the I/O-APIC specific fields in the routing entry. | |
| 2961 | + * | |
| 2962 | + * This is important to setup the I/O-APIC specific bits (is_level, | |
| 2963 | + * active_low, masked) because the underlying parent domain will only | |
| 2964 | + * provide the routing information and is oblivious of the I/O-APIC | |
| 2965 | + * specific bits. | |
| 2966 | + * | |
| 2967 | + * The entry is just preconfigured at this point and not written into the | |
| 2968 | + * RTE. This happens later during activation which will fill in the actual | |
| 2969 | + * routing information. | |
| 2970 | + */ | |
| 2971 | +static void mp_preconfigure_entry(struct mp_chip_data *data) | |
| 2924 | 2972 | { |
| 2973 | + struct IO_APIC_route_entry *entry = &data->entry; | |
| 2974 | + | |
| 2925 | 2975 | memset(entry, 0, sizeof(*entry)); |
| 2926 | − entry->delivery_mode = apic->delivery_mode; | |
| 2927 | − entry->dest_mode_logical = apic->dest_mode_logical; | |
| 2928 | − entry->destid_0_7 = cfg->dest_apicid; | |
| 2929 | − entry->vector = cfg->vector; | |
| 2930 | 2976 | entry->is_level = data->is_level; |
| 2931 | 2977 | entry->active_low = data->active_low; |
| 2932 | 2978 | /* |
| @@ -2939,11 +2985,10 @@ static void mp_setup_entry(struct irq_cfg *cfg, struct mp_chip_data *data, · 6 unmodified lines | ||
| 2939 | 2985 | int mp_irqdomain_alloc(struct irq_domain *domain, unsigned int virq, |
| 2940 | 2986 | unsigned int nr_irqs, void *arg) |
| 2941 | 2987 | { |
| 2942 | − int ret, ioapic, pin; | |
| 2943 | − struct irq_cfg *cfg; | |
| 2944 | − struct irq_data *irq_data; | |
| 2945 | − struct mp_chip_data *data; | |
| 2946 | 2988 | struct irq_alloc_info *info = arg; |
| 2989 | + struct mp_chip_data *data; | |
| 2990 | + struct irq_data *irq_data; | |
| 2991 | + int ret, ioapic, pin; | |
| 2947 | 2992 | unsigned long flags; |
| 2948 | 2993 | |
| 2949 | 2994 | if (!info || nr_irqs > 1) |
| @@ -2961,7 +3006,6 @@ int mp_irqdomain_alloc(struct irq_domain *domain, unsigned int virq, · 11 unmodified lines | ||
| 2961 | 3006 | if (!data) |
| 2962 | 3007 | return -ENOMEM; |
| 2963 | 3008 | |
| 2964 | − info->ioapic.entry = &data->entry; | |
| 2965 | 3009 | ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, info); |
| 2966 | 3010 | if (ret < 0) { |
| 2967 | 3011 | kfree(data); |
| @@ -2975,23 +3019,20 @@ int mp_irqdomain_alloc(struct irq_domain *domain, unsigned int virq, · 7 unmodified lines | ||
| 2975 | 3019 | irq_data->chip_data = data; |
| 2976 | 3020 | mp_irqdomain_get_attr(mp_pin_to_gsi(ioapic, pin), data, info); |
| 2977 | 3021 | |
| 2978 | − cfg = irqd_cfg(irq_data); | |
| 2979 | 3022 | add_pin_to_irq_node(data, ioapic_alloc_attr_node(info), ioapic, pin); |
| 2980 | 3023 | |
| 2981 | − local_irq_save(flags); | |
| 2982 | − if (info->ioapic.entry) | |
| 2983 | − mp_setup_entry(cfg, data, info->ioapic.entry); | |
| 3024 | + mp_preconfigure_entry(data); | |
| 2984 | 3025 | mp_register_handler(virq, data->is_level); |
| 3026 | + | |
| 3027 | + local_irq_save(flags); | |
| 2985 | 3028 | if (virq < nr_legacy_irqs()) |
| 2986 | 3029 | legacy_pic->mask(virq); |
| 2987 | 3030 | local_irq_restore(flags); |
| 2988 | 3031 | |
| 2989 | 3032 | apic_printk(APIC_VERBOSE, KERN_DEBUG |
| 2990 | − "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i Dest:%d)\n", | |
| 2991 | − ioapic, mpc_ioapic_id(ioapic), pin, cfg->vector, | |
| 2992 | − virq, data->is_level, data->active_low, | |
| 2993 | − cfg->dest_apicid); | |
| 2994 | − | |
| 3033 | + "IOAPIC[%d]: Preconfigured routing entry (%d-%d -> IRQ %d Level:%i ActiveLow:%i)\n", | |
| 3034 | + ioapic, mpc_ioapic_id(ioapic), pin, virq, | |
| 3035 | + data->is_level, data->active_low); | |
| 2995 | 3036 | return 0; |
| 2996 | 3037 | } |
| 2997 | 3038 | |
| @@ -3669,7 +3669,6 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, · 3668 unmodified lines | ||
| 3669 | 3669 | int devid, int index, int sub_handle) |
| 3670 | 3670 | { |
| 3671 | 3671 | struct irq_2_irte *irte_info = &data->irq_2_irte; |
| 3672 | − struct IO_APIC_route_entry *entry; | |
| 3673 | 3672 | struct amd_iommu *iommu = amd_iommu_rlookup_table[devid]; |
| 3674 | 3673 | |
| 3675 | 3674 | if (!iommu) |
| @@ -3683,17 +3682,6 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, · 7 unmodified lines | ||
| 3683 | 3682 | |
| 3684 | 3683 | switch (info->type) { |
| 3685 | 3684 | case X86_IRQ_ALLOC_TYPE_IOAPIC: |
| 3686 | − /* Setup IOAPIC entry */ | |
| 3687 | − entry = info->ioapic.entry; | |
| 3688 | − info->ioapic.entry = NULL; | |
| 3689 | − memset(entry, 0, sizeof(*entry)); | |
| 3690 | − entry->vector = index; | |
| 3691 | − entry->is_level = info->ioapic.is_level; | |
| 3692 | − entry->active_low = info->ioapic.active_low; | |
| 3693 | − /* Mask level triggered irqs. */ | |
| 3694 | − entry->masked = info->ioapic.is_level; | |
| 3695 | − break; | |
| 3696 | − | |
| 3697 | 3685 | case X86_IRQ_ALLOC_TYPE_HPET: |
| 3698 | 3686 | case X86_IRQ_ALLOC_TYPE_PCI_MSI: |
| 3699 | 3687 | case X86_IRQ_ALLOC_TYPE_PCI_MSIX: |
| @@ -40,7 +40,6 @@ static int hyperv_ir_set_affinity(struct irq_data *data, · 39 unmodified lines | ||
| 40 | 40 | { |
| 41 | 41 | struct irq_data *parent = data->parent_data; |
| 42 | 42 | struct irq_cfg *cfg = irqd_cfg(data); |
| 43 | − struct IO_APIC_route_entry *entry; | |
| 44 | 43 | int ret; |
| 45 | 44 | |
| 46 | 45 | /* Return error If new irq affinity is out of ioapic_max_cpumask. */ |
| @@ -51,9 +50,6 @@ static int hyperv_ir_set_affinity(struct irq_data *data, · 4 unmodified lines | ||
| 51 | 50 | if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE) |
| 52 | 51 | return ret; |
| 53 | 52 | |
| 54 | − entry = data->chip_data; | |
| 55 | − entry->destid_0_7 = cfg->dest_apicid; | |
| 56 | − entry->vector = cfg->vector; | |
| 57 | 53 | send_cleanup_vector(cfg); |
| 58 | 54 | |
| 59 | 55 | return 0; |
| @@ -89,20 +85,6 @@ static int hyperv_irq_remapping_alloc(struct irq_domain *domain, · 29 unmodified lines | ||
| 89 | 85 | |
| 90 | 86 | irq_data->chip = &hyperv_ir_chip; |
| 91 | 87 | |
| 92 | − /* | |
| 93 | − * If there is interrupt remapping function of IOMMU, setting irq | |
| 94 | − * affinity only needs to change IRTE of IOMMU. But Hyper-V doesn't | |
| 95 | − * support interrupt remapping function, setting irq affinity of IO-APIC | |
| 96 | − * interrupts still needs to change IO-APIC registers. But ioapic_ | |
| 97 | − * configure_entry() will ignore value of cfg->vector and cfg-> | |
| 98 | − * dest_apicid when IO-APIC's parent irq domain is not the vector | |
| 99 | − * domain.(See ioapic_configure_entry()) In order to setting vector | |
| 100 | − * and dest_apicid to IO-APIC register, IO-APIC entry pointer is saved | |
| 101 | − * in the chip_data and hyperv_irq_remapping_activate()/hyperv_ir_set_ | |
| 102 | − * affinity() set vector and dest_apicid directly into IO-APIC entry. | |
| 103 | − */ | |
| 104 | − irq_data->chip_data = info->ioapic.entry; | |
| 105 | − | |
| 106 | 88 | /* |
| 107 | 89 | * Hypver-V IO APIC irq affinity should be in the scope of |
| 108 | 90 | * ioapic_max_cpumask because no irq remapping support. |
| @@ -119,22 +101,9 @@ static void hyperv_irq_remapping_free(struct irq_domain *domain, · 10 unmodified lines | ||
| 119 | 101 | irq_domain_free_irqs_common(domain, virq, nr_irqs); |
| 120 | 102 | } |
| 121 | 103 | |
| 122 | −static int hyperv_irq_remapping_activate(struct irq_domain *domain, | |
| 123 | − struct irq_data *irq_data, bool reserve) | |
| 124 | −{ | |
| 125 | − struct irq_cfg *cfg = irqd_cfg(irq_data); | |
| 126 | − struct IO_APIC_route_entry *entry = irq_data->chip_data; | |
| 127 | − | |
| 128 | − entry->destid_0_7 = cfg->dest_apicid; | |
| 129 | − entry->vector = cfg->vector; | |
| 130 | − | |
| 131 | − return 0; | |
| 132 | −} | |
| 133 | − | |
| 134 | 104 | static const struct irq_domain_ops hyperv_ir_domain_ops = { |
| 135 | 105 | .alloc = hyperv_irq_remapping_alloc, |
| 136 | 106 | .free = hyperv_irq_remapping_free, |
| 137 | − .activate = hyperv_irq_remapping_activate, | |
| 138 | 107 | }; |
| 139 | 108 | |
| 140 | 109 | static int __init hyperv_prepare_irq_remapping(void) |
| @@ -1280,9 +1280,9 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, · 1279 unmodified lines | ||
| 1280 | 1280 | int index, int sub_handle) |
| 1281 | 1281 | { |
| 1282 | 1282 | struct irte *irte = &data->irte_entry; |
| 1283 | − struct IO_APIC_route_entry *entry; | |
| 1284 | 1283 | |
| 1285 | 1284 | prepare_irte(irte, irq_cfg->vector, irq_cfg->dest_apicid); |
| 1285 | + | |
| 1286 | 1286 | switch (info->type) { |
| 1287 | 1287 | case X86_IRQ_ALLOC_TYPE_IOAPIC: |
| 1288 | 1288 | /* Set source-id of interrupt request */ |
| @@ -1293,39 +1293,20 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, · 4 unmodified lines | ||
| 1293 | 1293 | irte->trigger_mode, irte->dlvry_mode, |
| 1294 | 1294 | irte->avail, irte->vector, irte->dest_id, |
| 1295 | 1295 | irte->sid, irte->sq, irte->svt); |
| 1296 | − | |
| 1297 | − entry = info->ioapic.entry; | |
| 1298 | − info->ioapic.entry = NULL; | |
| 1299 | − memset(entry, 0, sizeof(*entry)); | |
| 1300 | − entry->ir_index_15 = !!(index & 0x8000); | |
| 1301 | − entry->ir_format = true; | |
| 1302 | − entry->ir_index_0_14 = index & 0x7fff; | |
| 1303 | − /* | |
| 1304 | − * IO-APIC RTE will be configured with virtual vector. | |
| 1305 | − * irq handler will do the explicit EOI to the io-apic. | |
| 1306 | − */ | |
| 1307 | − entry->vector = info->ioapic.pin; | |
| 1308 | − entry->is_level = info->ioapic.is_level; | |
| 1309 | − entry->active_low = info->ioapic.active_low; | |
| 1310 | − /* Mask level triggered irqs. */ | |
| 1311 | − entry->masked = info->ioapic.is_level; | |
| 1296 | + sub_handle = info->ioapic.pin; | |
| 1312 | 1297 | break; |
| 1313 | − | |
| 1314 | 1298 | case X86_IRQ_ALLOC_TYPE_HPET: |
| 1299 | + set_hpet_sid(irte, info->devid); | |
| 1300 | + break; | |
| 1315 | 1301 | case X86_IRQ_ALLOC_TYPE_PCI_MSI: |
| 1316 | 1302 | case X86_IRQ_ALLOC_TYPE_PCI_MSIX: |
| 1317 | − if (info->type == X86_IRQ_ALLOC_TYPE_HPET) | |
| 1318 | − set_hpet_sid(irte, info->devid); | |
| 1319 | − else | |
| 1320 | − set_msi_sid(irte, msi_desc_to_pci_dev(info->desc)); | |
| 1321 | − | |
| 1322 | − fill_msi_msg(&data->msi_entry, index, sub_handle); | |
| 1303 | + set_msi_sid(irte, msi_desc_to_pci_dev(info->desc)); | |
| 1323 | 1304 | break; |
| 1324 | − | |
| 1325 | 1305 | default: |
| 1326 | 1306 | BUG_ON(1); |
| 1327 | 1307 | break; |
| 1328 | 1308 | } |
| 1309 | + fill_msi_msg(&data->msi_entry, index, sub_handle); | |
| 1329 | 1310 | } |
| 1330 | 1311 | |
| 1331 | 1312 | static void intel_free_irq_resources(struct irq_domain *domain, |