// SPDX-License-Identifier: GPL-2.0+ /* * Copyright (C) 2015 Google, Inc * * Based on code from the coreboot file of the same name */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include DECLARE_GLOBAL_DATA_PTR; /* * Setting up multiprocessing * * See https://www.intel.com/content/www/us/en/intelligent-systems/intel-boot-loader-development-kit/minimal-intel-architecture-boot-loader-paper.html * * Note that this file refers to the boot CPU (the one U-Boot is running on) as * the BSP (BootStrap Processor) and the others as APs (Application Processors). * * This module works by loading some setup code into RAM at AP_DEFAULT_BASE and * telling each AP to execute it. The code that each AP runs is in * sipi_vector.S (see ap_start16) which includes a struct sipi_params at the * end of it. Those parameters are set up by the C code. * Setting up is handled by load_sipi_vector(). It inits the common block of * parameters (sipi_params) which tell the APs what to do. This block includes * microcode and the MTTRs (Memory-Type-Range Registers) from the main CPU. * There is also an ap_count which each AP increments as it starts up, so the * BSP can tell how many checked in. * * The APs are started with a SIPI (Startup Inter-Processor Interrupt) which * tells an AP to start executing at a particular address, in this case * AP_DEFAULT_BASE which contains the code copied from ap_start16. This protocol * is handled by start_aps(). * * After being started, each AP runs the code in ap_start16, switches to 32-bit * mode, runs the code at ap_start, then jumps to c_handler which is ap_init(). * This runs a very simple 'flight plan' described in mp_steps(). This sets up * the CPU and waits for further instructions by looking at its entry in * ap_callbacks[]. Note that the flight plan is only actually run for each CPU * in bsp_do_flight_plan(): once the BSP completes each flight record, it sets * mp_flight_record->barrier to 1 to allow the APs to executed the record one * by one. * * CPUS are numbered sequentially from 0 using the device tree: * * cpus { * bootph-all; * #address-cells = <1>; * #size-cells = <0>; * * cpu@0 { * bootph-all; * device_type = "cpu"; * compatible = "intel,apl-cpu"; * reg = <0>; * intel,apic-id = <0>; * }; * * cpu@1 { * device_type = "cpu"; * compatible = "intel,apl-cpu"; * reg = <1>; * intel,apic-id = <2>; * }; * * Here the 'reg' property is the CPU number and then is placed in dev_seq(cpu) * so that we can index into ap_callbacks[] using that. The APIC ID is different * and may not be sequential (it typically is if hyperthreading is supported). * * Once APs are inited they wait in ap_wait_for_instruction() for instructions. * Instructions come in the form of a function to run. This logic is in * mp_run_on_cpus() which supports running on any one AP, all APs, just the BSP * or all CPUs. The BSP logic is handled directly in mp_run_on_cpus(), by * calling the function. For the APs, callback information is stored in a * single, common struct mp_callback and a pointer to this is written to each * AP's slot in ap_callbacks[] by run_ap_work(). All APs get the message even * if it is only for one of them. When an AP notices a message it checks whether * it should call the function (see check in ap_wait_for_instruction()) and then * does so if needed. After that it sets its slot to NULL to indicate it is * done. * * While U-Boot is running it can use mp_run_on_cpus() to run code on the APs. * An example of this is the 'mtrr' command which allows reading and changing * the MTRRs on all CPUs. * * Before U-Boot exits it calls mp_park_aps() which tells all CPUs to halt by * executing a 'hlt' instruction. That allows them to be used by Linux when it * starts up. */ /* This also needs to match the sipi.S assembly code for saved MSR encoding */ struct __packed saved_msr { uint32_t index; uint32_t lo; uint32_t hi; }; /** * struct mp_flight_plan - Holds the flight plan * * @num_records: Number of flight records * @records: Pointer to each record */ struct mp_flight_plan { int num_records; struct mp_flight_record *records; }; /** * struct mp_callback - Callback information for APs * * @func: Function to run * @arg: Argument to pass to the function * @logical_cpu_number: Either a CPU number (i.e. dev_seq(cpu) or a special * value like MP_SELECT_BSP. It tells the AP whether it should process this * callback */ struct mp_callback { mp_run_func func; void *arg; int logical_cpu_number; }; /* Stores the flight plan so that APs can find it */ static struct mp_flight_plan mp_info; /* * ap_callbacks - Callback mailbox array * * Array of callback, one entry for each available CPU, indexed by the CPU * number, which is dev_seq(cpu). The entry for the main CPU is never used. * When this is NULL, there is no pending work for the CPU to run. When * non-NULL it points to the mp_callback structure. This is shared between all * CPUs, so should only be written by the main CPU. */ static struct mp_callback **ap_callbacks; static inline void barrier_wait(atomic_t *b) { while (atomic_read(b) == 0) asm("pause"); mfence(); } static inline void release_barrier(atomic_t *b) { mfence(); atomic_set(b, 1); } static inline void stop_this_cpu(void) { /* Called by an AP when it is ready to halt and wait for a new task */ for (;;) cpu_hlt(); } /* Returns 1 if timeout waiting for APs. 0 if target APs found */ static int wait_for_aps(atomic_t *val, int target, int total_delay, int delay_step) { int timeout = 0; int delayed = 0; while (atomic_read(val) != target) { udelay(delay_step); delayed += delay_step; if (delayed >= total_delay) { timeout = 1; break; } } return timeout; } static void ap_do_flight_plan(struct udevice *cpu) { int i; for (i = 0; i < mp_info.num_records; i++) { struct mp_flight_record *rec = &mp_info.records[i]; atomic_inc(&rec->cpus_entered); barrier_wait(&rec->barrier); if (rec->ap_call != NULL) rec->ap_call(cpu, rec->ap_arg); } } static int find_cpu_by_apic_id(int apic_id, struct udevice **devp) { struct udevice *dev; *devp = NULL; for (uclass_find_first_device(UCLASS_CPU, &dev); dev; uclass_find_next_device(&dev)) { struct cpu_plat *plat = dev_get_parent_plat(dev); if (plat->cpu_id == apic_id) { *devp = dev; return 0; } } return -ENOENT; } /* * By the time APs call ap_init() caching has been setup, and microcode has * been loaded */ static void ap_init(unsigned int cpu_index) { struct udevice *dev; int apic_id; int ret; /* Ensure the local apic is enabled */ enable_lapic(); apic_id = lapicid(); ret = find_cpu_by_apic_id(apic_id, &dev); if (ret) { debug("Unknown CPU apic_id %x\n", apic_id); goto done; } debug("AP: slot %d apic_id %x, dev %s\n", cpu_index, apic_id, dev ? dev->name : "(apic_id not found)"); /* * Walk the flight plan, which only returns if CONFIG_SMP_AP_WORK is not * enabled */ ap_do_flight_plan(dev); done: stop_this_cpu(); } static const unsigned int fixed_mtrrs[NUM_FIXED_MTRRS] = { MTRR_FIX_64K_00000_MSR, MTRR_FIX_16K_80000_MSR, MTRR_FIX_16K_A0000_MSR, MTRR_FIX_4K_C0000_MSR, MTRR_FIX_4K_C8000_MSR, MTRR_FIX_4K_D0000_MSR, MTRR_FIX_4K_D8000_MSR, MTRR_FIX_4K_E0000_MSR, MTRR_FIX_4K_E8000_MSR, MTRR_FIX_4K_F0000_MSR, MTRR_FIX_4K_F8000_MSR, }; static inline struct saved_msr *save_msr(int index, struct saved_msr *entry) { msr_t msr; msr = msr_read(index); entry->index = index; entry->lo = msr.lo; entry->hi = msr.hi; /* Return the next entry */ entry++; return entry; } static int save_bsp_msrs(char *start, int size) { int msr_count; int num_var_mtrrs; struct saved_msr *msr_entry; int i; msr_t msr; /* Determine number of MTRRs need to be saved */ msr = msr_read(MTRR_CAP_MSR); num_var_mtrrs = msr.lo & 0xff; /* 2 * num_var_mtrrs for base and mask. +1 for IA32_MTRR_DEF_TYPE */ msr_count = 2 * num_var_mtrrs + NUM_FIXED_MTRRS + 1; if ((msr_count * sizeof(struct saved_msr)) > size) { printf("Cannot mirror all %d msrs\n", msr_count); return -ENOSPC; } msr_entry = (void *)start; for (i = 0; i < NUM_FIXED_MTRRS; i++) msr_entry = save_msr(fixed_mtrrs[i], msr_entry); for (i = 0; i < num_var_mtrrs; i++) { msr_entry = save_msr(MTRR_PHYS_BASE_MSR(i), msr_entry); msr_entry = save_msr(MTRR_PHYS_MASK_MSR(i), msr_entry); } msr_entry = save_msr(MTRR_DEF_TYPE_MSR, msr_entry); return msr_count; } static int load_sipi_vector(atomic_t **ap_countp, int num_cpus) { struct sipi_params_16bit *params16; struct sipi_params *params; static char msr_save[512]; char *stack; ulong addr; int code_len; int size; int ret; /* Copy in the code */ code_len = ap_start16_code_end - ap_start16; debug("Copying SIPI code to %x: %d bytes\n", AP_DEFAULT_BASE, code_len); memcpy((void *)AP_DEFAULT_BASE, ap_start16, code_len); addr = AP_DEFAULT_BASE + (ulong)sipi_params_16bit - (ulong)ap_start16; params16 = (struct sipi_params_16bit *)addr; params16->ap_start = (uint32_t)ap_start; params16->gdt = (uint32_t)gd->arch.gdt; params16->gdt_limit = X86_GDT_SIZE - 1; debug("gdt = %x, gdt_limit = %x\n", params16->gdt, params16->gdt_limit); params = (struct sipi_params *)sipi_params; debug("SIPI 32-bit params at %p\n", params); params->idt_ptr = (uint32_t)x86_get_idt(); params->stack_size = CONFIG_AP_STACK_SIZE; size = params->stack_size * num_cpus; stack = memalign(4096, size); if (!stack) return -ENOMEM; params->stack_top = (u32)(stack + size); #if !defined(CONFIG_QEMU) && !defined(CONFIG_HAVE_FSP) && \ !defined(CONFIG_INTEL_MID) params->microcode_ptr = ucode_base; debug("Microcode at %x\n", params->microcode_ptr); #endif params->msr_table_ptr = (u32)msr_save; ret = save_bsp_msrs(msr_save, sizeof(msr_save)); if (ret < 0) return ret; params->msr_count = ret; params->c_handler = (uint32_t)&ap_init; *ap_countp = ¶ms->ap_count; atomic_set(*ap_countp, 0); debug("SIPI vector is ready\n"); return 0; } static int check_cpu_devices(int expected_cpus) { int i; for (i = 0; i < expected_cpus; i++) { struct udevice *dev; int ret; ret = uclass_find_device(UCLASS_CPU, i, &dev); if (ret) { debug("Cannot find CPU %d in device tree\n", i); return ret; } } return 0; } /* Returns 1 for timeout. 0 on success */ static int apic_wait_timeout(int total_delay, const char *msg) { int total = 0; if (!(lapic_read(LAPIC_ICR) & LAPIC_ICR_BUSY)) return 0; debug("Waiting for %s...", msg); while (lapic_read(LAPIC_ICR) & LAPIC_ICR_BUSY) { udelay(50); total += 50; if (total >= total_delay) { debug("timed out: aborting\n"); return -ETIMEDOUT; } } debug("done\n"); return 0; } /** * start_aps() - Start up the APs and count how many we find * * This is called on the boot processor to start up all the other processors * (here called APs). * * @num_aps: Number of APs we expect to find * @ap_count: Initially zero. Incremented by this function for each AP found * Return: 0 if all APs were set up correctly or there are none to set up, * -ENOSPC if the SIPI vector is too high in memory, * -ETIMEDOUT if the ICR is busy or the second SIPI fails to complete * -EIO if not all APs check in correctly */ static int start_aps(int num_aps, atomic_t *ap_count) { int sipi_vector; /* Max location is 4KiB below 1MiB */ const int max_vector_loc = ((1 << 20) - (1 << 12)) >> 12; if (num_aps == 0) return 0; /* The vector is sent as a 4k aligned address in one byte */ sipi_vector = AP_DEFAULT_BASE >> 12; if (sipi_vector > max_vector_loc) { printf("SIPI vector too large! 0x%08x\n", sipi_vector); return -ENOSPC; } debug("Attempting to start %d APs\n", num_aps); if (apic_wait_timeout(1000, "ICR not to be busy")) return -ETIMEDOUT; /* Send INIT IPI to all but self */ lapic_write(LAPIC_ICR2, SET_LAPIC_DEST_FIELD(0)); lapic_write(LAPIC_ICR, LAPIC_DEST_ALLBUT | LAPIC_INT_ASSERT | LAPIC_DM_INIT); debug("Waiting for 10ms after sending INIT\n"); mdelay(10); /* Send 1st SIPI */ if (apic_wait_timeout(1000, "ICR not to be busy")) return -ETIMEDOUT; lapic_write(LAPIC_ICR2, SET_LAPIC_DEST_FIELD(0)); lapic_write(LAPIC_ICR, LAPIC_DEST_ALLBUT | LAPIC_INT_ASSERT | LAPIC_DM_STARTUP | sipi_vector); if (apic_wait_timeout(10000, "first SIPI to complete")) return -ETIMEDOUT; /* Wait for CPUs to check in up to 200 us */ wait_for_aps(ap_count, num_aps, 200, 15); /* Send 2nd SIPI */ if (apic_wait_timeout(1000, "ICR not to be busy")) return -ETIMEDOUT; lapic_write(LAPIC_ICR2, SET_LAPIC_DEST_FIELD(0)); lapic_write(LAPIC_ICR, LAPIC_DEST_ALLBUT | LAPIC_INT_ASSERT | LAPIC_DM_STARTUP | sipi_vector); if (apic_wait_timeout(10000, "second SIPI to complete")) return -ETIMEDOUT; /* Wait for CPUs to check in */ if (wait_for_aps(ap_count, num_aps, 10000, 50)) { debug("Not all APs checked in: %d/%d\n", atomic_read(ap_count), num_aps); return -EIO; } return 0; } /** * bsp_do_flight_plan() - Do the flight plan on the BSP * * This runs the flight plan on the main CPU used to boot U-Boot * * @cpu: Device for the main CPU * @plan: Flight plan to run * @num_aps: Number of APs (CPUs other than the BSP) * @returns 0 on success, -ETIMEDOUT if an AP failed to come up */ static int bsp_do_flight_plan(struct udevice *cpu, struct mp_flight_plan *plan, int num_aps) { int i; int ret = 0; const int timeout_us = 100000; const int step_us = 100; for (i = 0; i < plan->num_records; i++) { struct mp_flight_record *rec = &plan->records[i]; /* Wait for APs if the record is not released */ if (atomic_read(&rec->barrier) == 0) { /* Wait for the APs to check in */ if (wait_for_aps(&rec->cpus_entered, num_aps, timeout_us, step_us)) { debug("MP record %d timeout\n", i); ret = -ETIMEDOUT; } } if (rec->bsp_call != NULL) rec->bsp_call(cpu, rec->bsp_arg); release_barrier(&rec->barrier); } return ret; } /** * get_bsp() - Get information about the bootstrap processor * * @devp: If non-NULL, returns CPU device corresponding to the BSP * @cpu_countp: If non-NULL, returns the total number of CPUs * Return: CPU number of the BSP, or -ve on error. If multiprocessing is not * enabled, returns 0 */ static int get_bsp(struct udevice **devp, int *cpu_countp) { char processor_name[CPU_MAX_NAME_LEN]; struct udevice *dev; int apic_id; int ret; cpu_get_name(processor_name); debug("CPU: %s\n", processor_name); apic_id = lapicid(); ret = find_cpu_by_apic_id(apic_id, &dev); if (ret < 0) { printf("Cannot find boot CPU, APIC ID %d\n", apic_id); return ret; } ret = cpu_get_count(dev); if (ret < 0) return log_msg_ret("count", ret); if (devp) *devp = dev; if (cpu_countp) *cpu_countp = ret; return dev_seq(dev) >= 0 ? dev_seq(dev) : 0; } /** * read_callback() - Read the pointer in a callback slot * * This is called by APs to read their callback slot to see if there is a * pointer to new instructions * * @slot: Pointer to the AP's callback slot * Return: value of that pointer */ static struct mp_callback *read_callback(struct mp_callback **slot) { dmb(); return *slot; } /** * store_callback() - Store a pointer to the callback slot * * This is called by APs to write NULL into the callback slot when they have * finished the work requested by the BSP. * * @slot: Pointer to the AP's callback slot * @val: Value to write (e.g. NULL) */ static void store_callback(struct mp_callback **slot, struct mp_callback *val) { *slot = val; dmb(); } /** * run_ap_work() - Run a callback on selected APs * * This writes @callback to all APs and waits for them all to acknowledge it, * Note that whether each AP actually calls the callback depends on the value * of logical_cpu_number (see struct mp_callback). The logical CPU number is * the CPU device's req->seq value. * * @callback: Callback information to pass to all APs * @bsp: CPU device for the BSP * @num_cpus: The number of CPUs in the system (= number of APs + 1) * @expire_ms: Timeout to wait for all APs to finish, in milliseconds, or 0 for * no timeout * Return: 0 if OK, -ETIMEDOUT if one or more APs failed to respond in time */ static int run_ap_work(struct mp_callback *callback, struct udevice *bsp, int num_cpus, uint expire_ms) { int cur_cpu = dev_seq(bsp); int num_aps = num_cpus - 1; /* number of non-BSPs to get this message */ int cpus_accepted; ulong start; int i; if (!IS_ENABLED(CONFIG_SMP_AP_WORK)) { printf("APs already parked. CONFIG_SMP_AP_WORK not enabled\n"); return -ENOTSUPP; } /* Signal to all the APs to run the func. */ for (i = 0; i < num_cpus; i++) { if (cur_cpu != i) store_callback(&ap_callbacks[i], callback); } mfence(); /* Wait for all the APs to signal back that call has been accepted. */ start = get_timer(0); do { mdelay(1); cpus_accepted = 0; for (i = 0; i < num_cpus; i++) { if (cur_cpu == i) continue; if (!read_callback(&ap_callbacks[i])) cpus_accepted++; } if (expire_ms && get_timer(start) >= expire_ms) { log(UCLASS_CPU, LOGL_CRIT, "AP call expired; %d/%d CPUs accepted\n", cpus_accepted, num_aps); return -ETIMEDOUT; } } while (cpus_accepted != num_aps); /* Make sure we can see any data written by the APs */ mfence(); return 0; } /** * ap_wait_for_instruction() - Wait for and process requests from the main CPU * * This is called by APs (here, everything other than the main boot CPU) to * await instructions. They arrive in the form of a function call and argument, * which is then called. This uses a simple mailbox with atomic read/set * * @cpu: CPU that is waiting * @unused: Optional argument provided by struct mp_flight_record, not used here * Return: Does not return */ static int ap_wait_for_instruction(struct udevice *cpu, void *unused) { struct mp_callback lcb; struct mp_callback **per_cpu_slot; if (!IS_ENABLED(CONFIG_SMP_AP_WORK)) return 0; per_cpu_slot = &ap_callbacks[dev_seq(cpu)]; while (1) { struct mp_callback *cb = read_callback(per_cpu_slot); if (!cb) { asm ("pause"); continue; } /* Copy to local variable before using the value */ memcpy(&lcb, cb, sizeof(lcb)); mfence(); if (lcb.logical_cpu_number == MP_SELECT_ALL || lcb.logical_cpu_number == MP_SELECT_APS || dev_seq(cpu) == lcb.logical_cpu_number) lcb.func(lcb.arg); /* Indicate we are finished */ store_callback(per_cpu_slot, NULL); } return 0; } static int mp_init_cpu(struct udevice *cpu, void *unused) { struct cpu_plat *plat = dev_get_parent_plat(cpu); plat->ucode_version = microcode_read_rev(); plat->device_id = gd->arch.x86_device; return device_probe(cpu); } static struct mp_flight_record mp_steps[] = { MP_FR_BLOCK_APS(mp_init_cpu, NULL, mp_init_cpu, NULL), MP_FR_BLOCK_APS(ap_wait_for_instruction, NULL, NULL, NULL), }; int mp_run_on_cpus(int cpu_select, mp_run_func func, void *arg) { struct mp_callback lcb = { .func = func, .arg = arg, .logical_cpu_number = cpu_select, }; struct udevice *dev; int num_cpus; int ret; ret = get_bsp(&dev, &num_cpus); if (ret < 0) return log_msg_ret("bsp", ret); if (cpu_select == MP_SELECT_ALL || cpu_select == MP_SELECT_BSP || cpu_select == ret) { /* Run on BSP first */ func(arg); } if (!IS_ENABLED(CONFIG_SMP_AP_WORK) || !(gd->flags & GD_FLG_SMP_READY)) { /* Allow use of this function on the BSP only */ if (cpu_select == MP_SELECT_BSP || !cpu_select) return 0; return -ENOTSUPP; } /* Allow up to 1 second for all APs to finish */ ret = run_ap_work(&lcb, dev, num_cpus, 1000 /* ms */); if (ret) return log_msg_ret("aps", ret); return 0; } static void park_this_cpu(void *unused) { stop_this_cpu(); } int mp_park_aps(void) { int ret; ret = mp_run_on_cpus(MP_SELECT_APS, park_this_cpu, NULL); if (ret) return log_ret(ret); return 0; } int mp_first_cpu(int cpu_select) { struct udevice *dev; int num_cpus; int ret; /* * This assumes that CPUs are numbered from 0. This function tries to * avoid assuming the CPU 0 is the boot CPU */ if (cpu_select == MP_SELECT_ALL) return 0; /* start with the first one */ ret = get_bsp(&dev, &num_cpus); if (ret < 0) return log_msg_ret("bsp", ret); /* Return boot CPU if requested */ if (cpu_select == MP_SELECT_BSP) return ret; /* Return something other than the boot CPU, if APs requested */ if (cpu_select == MP_SELECT_APS && num_cpus > 1) return ret == 0 ? 1 : 0; /* Try to check for an invalid value */ if (cpu_select < 0 || cpu_select >= num_cpus) return -EINVAL; return cpu_select; /* return the only selected one */ } int mp_next_cpu(int cpu_select, int prev_cpu) { struct udevice *dev; int num_cpus; int ret; int bsp; /* If we selected the BSP or a particular single CPU, we are done */ if (!IS_ENABLED(CONFIG_SMP_AP_WORK) || cpu_select == MP_SELECT_BSP || cpu_select >= 0) return -EFBIG; /* Must be doing MP_SELECT_ALL or MP_SELECT_APS; return the next CPU */ ret = get_bsp(&dev, &num_cpus); if (ret < 0) return log_msg_ret("bsp", ret); bsp = ret; /* Move to the next CPU */ assert(prev_cpu >= 0); ret = prev_cpu + 1; /* Skip the BSP if needed */ if (cpu_select == MP_SELECT_APS && ret == bsp) ret++; if (ret >= num_cpus) return -EFBIG; return ret; } int mp_init(void) { int num_aps, num_cpus; atomic_t *ap_count; struct udevice *cpu; int ret; if (IS_ENABLED(CONFIG_QFW)) { ret = qemu_cpu_fixup(); if (ret) return ret; } ret = get_bsp(&cpu, &num_cpus); if (ret < 0) { debug("Cannot init boot CPU: err=%d\n", ret); return ret; } if (num_cpus < 2) debug("Warning: Only 1 CPU is detected\n"); ret = check_cpu_devices(num_cpus); if (ret) log_warning("Warning: Device tree does not describe all CPUs. Extra ones will not be started correctly\n"); ap_callbacks = calloc(num_cpus, sizeof(struct mp_callback *)); if (!ap_callbacks) return -ENOMEM; /* Copy needed parameters so that APs have a reference to the plan */ mp_info.num_records = ARRAY_SIZE(mp_steps); mp_info.records = mp_steps; /* Load the SIPI vector */ ret = load_sipi_vector(&ap_count, num_cpus); if (ap_count == NULL) return -ENOENT; /* * Make sure SIPI data hits RAM so the APs that come up will see * the startup code even if the caches are disabled */ wbinvd(); /* Start the APs providing number of APs and the cpus_entered field */ num_aps = num_cpus - 1; ret = start_aps(num_aps, ap_count); if (ret) { mdelay(1000); debug("%d/%d eventually checked in?\n", atomic_read(ap_count), num_aps); return ret; } /* Walk the flight plan for the BSP */ ret = bsp_do_flight_plan(cpu, &mp_info, num_aps); if (ret) { debug("CPU init failed: err=%d\n", ret); return ret; } gd->flags |= GD_FLG_SMP_READY; return 0; }