| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | |
| 3 | #include <linux/cpuhotplug.h> |
| 4 | #include <linux/cpumask.h> |
| 5 | #include <linux/slab.h> |
| 6 | #include <linux/mm.h> |
| 7 | |
| 8 | #include <asm/apic.h> |
| 9 | |
| 10 | #include "local.h" |
| 11 | |
| 12 | #define apic_cluster(apicid) ((apicid) >> 4) |
| 13 | |
| 14 | /* |
| 15 | * __x2apic_send_IPI_mask() possibly needs to read |
| 16 | * x86_cpu_to_logical_apicid for all online cpus in a sequential way. |
| 17 | * Using per cpu variable would cost one cache line per cpu. |
| 18 | */ |
| 19 | static u32 *x86_cpu_to_logical_apicid __read_mostly; |
| 20 | |
| 21 | static DEFINE_PER_CPU(cpumask_var_t, ipi_mask); |
| 22 | static DEFINE_PER_CPU_READ_MOSTLY(struct cpumask *, cluster_masks); |
| 23 | |
| 24 | static int x2apic_acpi_madt_oem_check(char *oem_id, char *oem_table_id) |
| 25 | { |
| 26 | return x2apic_enabled(); |
| 27 | } |
| 28 | |
| 29 | static void x2apic_send_IPI(int cpu, int vector) |
| 30 | { |
| 31 | u32 dest = x86_cpu_to_logical_apicid[cpu]; |
| 32 | |
| 33 | /* x2apic MSRs are special and need a special fence: */ |
| 34 | weak_wrmsr_fence(); |
| 35 | __x2apic_send_IPI_dest(dest, vector, APIC_DEST_LOGICAL); |
| 36 | } |
| 37 | |
| 38 | static void |
| 39 | __x2apic_send_IPI_mask(const struct cpumask *mask, int vector, int apic_dest) |
| 40 | { |
| 41 | unsigned int cpu, clustercpu; |
| 42 | struct cpumask *tmpmsk; |
| 43 | unsigned long flags; |
| 44 | u32 dest; |
| 45 | |
| 46 | /* x2apic MSRs are special and need a special fence: */ |
| 47 | weak_wrmsr_fence(); |
| 48 | local_irq_save(flags); |
| 49 | |
| 50 | tmpmsk = this_cpu_cpumask_var_ptr(ipi_mask); |
| 51 | cpumask_copy(tmpmsk, mask); |
| 52 | /* If IPI should not be sent to self, clear current CPU */ |
| 53 | if (apic_dest != APIC_DEST_ALLINC) |
| 54 | __cpumask_clear_cpu(smp_processor_id(), tmpmsk); |
| 55 | |
| 56 | /* Collapse cpus in a cluster so a single IPI per cluster is sent */ |
| 57 | for_each_cpu(cpu, tmpmsk) { |
| 58 | struct cpumask *cmsk = per_cpu(cluster_masks, cpu); |
| 59 | |
| 60 | dest = 0; |
| 61 | for_each_cpu_and(clustercpu, tmpmsk, cmsk) |
| 62 | dest |= x86_cpu_to_logical_apicid[clustercpu]; |
| 63 | |
| 64 | if (!dest) |
| 65 | continue; |
| 66 | |
| 67 | __x2apic_send_IPI_dest(dest, vector, APIC_DEST_LOGICAL); |
| 68 | /* Remove cluster CPUs from tmpmask */ |
| 69 | cpumask_andnot(tmpmsk, tmpmsk, cmsk); |
| 70 | } |
| 71 | |
| 72 | local_irq_restore(flags); |
| 73 | } |
| 74 | |
| 75 | static void x2apic_send_IPI_mask(const struct cpumask *mask, int vector) |
| 76 | { |
| 77 | __x2apic_send_IPI_mask(mask, vector, APIC_DEST_ALLINC); |
| 78 | } |
| 79 | |
| 80 | static void |
| 81 | x2apic_send_IPI_mask_allbutself(const struct cpumask *mask, int vector) |
| 82 | { |
| 83 | __x2apic_send_IPI_mask(mask, vector, APIC_DEST_ALLBUT); |
| 84 | } |
| 85 | |
| 86 | static u32 x2apic_calc_apicid(unsigned int cpu) |
| 87 | { |
| 88 | return x86_cpu_to_logical_apicid[cpu]; |
| 89 | } |
| 90 | |
| 91 | static void init_x2apic_ldr(void) |
| 92 | { |
| 93 | struct cpumask *cmsk = this_cpu_read(cluster_masks); |
| 94 | |
| 95 | BUG_ON(!cmsk); |
| 96 | |
| 97 | cpumask_set_cpu(smp_processor_id(), cmsk); |
| 98 | } |
| 99 | |
| 100 | /* |
| 101 | * As an optimisation during boot, set the cluster_mask for all present |
| 102 | * CPUs at once, to prevent each of them having to iterate over the others |
| 103 | * to find the existing cluster_mask. |
| 104 | */ |
| 105 | static void prefill_clustermask(struct cpumask *cmsk, unsigned int cpu, u32 cluster) |
| 106 | { |
| 107 | int cpu_i; |
| 108 | |
| 109 | for_each_present_cpu(cpu_i) { |
| 110 | struct cpumask **cpu_cmsk = &per_cpu(cluster_masks, cpu_i); |
| 111 | u32 apicid = apic->cpu_present_to_apicid(cpu_i); |
| 112 | |
| 113 | if (apicid == BAD_APICID || cpu_i == cpu || apic_cluster(apicid) != cluster) |
| 114 | continue; |
| 115 | |
| 116 | if (WARN_ON_ONCE(*cpu_cmsk == cmsk)) |
| 117 | continue; |
| 118 | |
| 119 | BUG_ON(*cpu_cmsk); |
| 120 | *cpu_cmsk = cmsk; |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | static int alloc_clustermask(unsigned int cpu, u32 cluster, int node) |
| 125 | { |
| 126 | struct cpumask *cmsk = NULL; |
| 127 | unsigned int cpu_i; |
| 128 | |
| 129 | /* |
| 130 | * At boot time, the CPU present mask is stable. The cluster mask is |
| 131 | * allocated for the first CPU in the cluster and propagated to all |
| 132 | * present siblings in the cluster. If the cluster mask is already set |
| 133 | * on entry to this function for a given CPU, there is nothing to do. |
| 134 | */ |
| 135 | if (per_cpu(cluster_masks, cpu)) |
| 136 | return 0; |
| 137 | |
| 138 | if (system_state < SYSTEM_RUNNING) |
| 139 | goto alloc; |
| 140 | |
| 141 | /* |
| 142 | * On post boot hotplug for a CPU which was not present at boot time, |
| 143 | * iterate over all possible CPUs (even those which are not present |
| 144 | * any more) to find any existing cluster mask. |
| 145 | */ |
| 146 | for_each_possible_cpu(cpu_i) { |
| 147 | u32 apicid = apic->cpu_present_to_apicid(cpu_i); |
| 148 | |
| 149 | if (apicid != BAD_APICID && apic_cluster(apicid) == cluster) { |
| 150 | cmsk = per_cpu(cluster_masks, cpu_i); |
| 151 | /* |
| 152 | * If the cluster is already initialized, just store |
| 153 | * the mask and return. There's no need to propagate. |
| 154 | */ |
| 155 | if (cmsk) { |
| 156 | per_cpu(cluster_masks, cpu) = cmsk; |
| 157 | return 0; |
| 158 | } |
| 159 | } |
| 160 | } |
| 161 | /* |
| 162 | * No CPU in the cluster has ever been initialized, so fall through to |
| 163 | * the boot time code which will also populate the cluster mask for any |
| 164 | * other CPU in the cluster which is (now) present. |
| 165 | */ |
| 166 | alloc: |
| 167 | cmsk = kzalloc_node(sizeof(*cmsk), GFP_KERNEL, node); |
| 168 | if (!cmsk) |
| 169 | return -ENOMEM; |
| 170 | per_cpu(cluster_masks, cpu) = cmsk; |
| 171 | prefill_clustermask(cmsk, cpu, cluster); |
| 172 | |
| 173 | return 0; |
| 174 | } |
| 175 | |
| 176 | static int x2apic_prepare_cpu(unsigned int cpu) |
| 177 | { |
| 178 | u32 phys_apicid = apic->cpu_present_to_apicid(cpu); |
| 179 | u32 cluster = apic_cluster(phys_apicid); |
| 180 | u32 logical_apicid = (cluster << 16) | (1 << (phys_apicid & 0xf)); |
| 181 | int node = cpu_to_node(cpu); |
| 182 | |
| 183 | x86_cpu_to_logical_apicid[cpu] = logical_apicid; |
| 184 | |
| 185 | if (alloc_clustermask(cpu, cluster, node) < 0) |
| 186 | return -ENOMEM; |
| 187 | |
| 188 | if (!zalloc_cpumask_var_node(&per_cpu(ipi_mask, cpu), GFP_KERNEL, node)) |
| 189 | return -ENOMEM; |
| 190 | |
| 191 | return 0; |
| 192 | } |
| 193 | |
| 194 | static int x2apic_dead_cpu(unsigned int dead_cpu) |
| 195 | { |
| 196 | struct cpumask *cmsk = per_cpu(cluster_masks, dead_cpu); |
| 197 | |
| 198 | if (cmsk) |
| 199 | cpumask_clear_cpu(dead_cpu, cmsk); |
| 200 | free_cpumask_var(per_cpu(ipi_mask, dead_cpu)); |
| 201 | return 0; |
| 202 | } |
| 203 | |
| 204 | static int x2apic_cluster_probe(void) |
| 205 | { |
| 206 | u32 slots; |
| 207 | |
| 208 | if (!x2apic_mode) |
| 209 | return 0; |
| 210 | |
| 211 | slots = max_t(u32, L1_CACHE_BYTES/sizeof(u32), nr_cpu_ids); |
| 212 | x86_cpu_to_logical_apicid = kcalloc(slots, sizeof(u32), GFP_KERNEL); |
| 213 | if (!x86_cpu_to_logical_apicid) |
| 214 | return 0; |
| 215 | |
| 216 | if (cpuhp_setup_state(CPUHP_X2APIC_PREPARE, "x86/x2apic:prepare", |
| 217 | x2apic_prepare_cpu, x2apic_dead_cpu) < 0) { |
| 218 | pr_err("Failed to register X2APIC_PREPARE\n"); |
| 219 | kfree(x86_cpu_to_logical_apicid); |
| 220 | x86_cpu_to_logical_apicid = NULL; |
| 221 | return 0; |
| 222 | } |
| 223 | init_x2apic_ldr(); |
| 224 | return 1; |
| 225 | } |
| 226 | |
| 227 | static struct apic apic_x2apic_cluster __ro_after_init = { |
| 228 | |
| 229 | .name = "cluster x2apic", |
| 230 | .probe = x2apic_cluster_probe, |
| 231 | .acpi_madt_oem_check = x2apic_acpi_madt_oem_check, |
| 232 | |
| 233 | .dest_mode_logical = true, |
| 234 | |
| 235 | .disable_esr = 0, |
| 236 | |
| 237 | .init_apic_ldr = init_x2apic_ldr, |
| 238 | .cpu_present_to_apicid = default_cpu_present_to_apicid, |
| 239 | |
| 240 | .max_apic_id = UINT_MAX, |
| 241 | .x2apic_set_max_apicid = true, |
| 242 | .get_apic_id = x2apic_get_apic_id, |
| 243 | |
| 244 | .calc_dest_apicid = x2apic_calc_apicid, |
| 245 | |
| 246 | .send_IPI = x2apic_send_IPI, |
| 247 | .send_IPI_mask = x2apic_send_IPI_mask, |
| 248 | .send_IPI_mask_allbutself = x2apic_send_IPI_mask_allbutself, |
| 249 | .send_IPI_allbutself = x2apic_send_IPI_allbutself, |
| 250 | .send_IPI_all = x2apic_send_IPI_all, |
| 251 | .send_IPI_self = x2apic_send_IPI_self, |
| 252 | .nmi_to_offline_cpu = true, |
| 253 | |
| 254 | .read = native_apic_msr_read, |
| 255 | .write = native_apic_msr_write, |
| 256 | .eoi = native_apic_msr_eoi, |
| 257 | .icr_read = native_x2apic_icr_read, |
| 258 | .icr_write = native_x2apic_icr_write, |
| 259 | }; |
| 260 | |
| 261 | apic_driver(apic_x2apic_cluster); |