Merge branches 'pm-cpufreq' and 'acpi-cppc'
authorRafael J. Wysocki <rafael.j.wysocki@intel.com>
Fri, 20 Nov 2015 00:22:10 +0000 (01:22 +0100)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Fri, 20 Nov 2015 00:22:10 +0000 (01:22 +0100)
* pm-cpufreq:
  Revert "Documentation: kernel_parameters for Intel P state driver"
  cpufreq: mediatek: fix build error
  cpufreq: intel_pstate: Add separate support for Airmont cores
  cpufreq: intel_pstate: Replace BYT with ATOM
  Revert "cpufreq: intel_pstate: Use ACPI perf configuration"
  Revert "cpufreq: intel_pstate: Avoid calculation for max/min"

* acpi-cppc:
  ACPI / CPPC: Use h/w reduced version of the PCCT structure

1  2  3 
Documentation/kernel-parameters.txt
drivers/acpi/cppc_acpi.c
drivers/cpufreq/Kconfig.arm

Simple merge
index 3c083d2cc4342049a97fae02407fb64b7e553036,0000000000000000000000000000000000000000,9361966f1b61d88bd5d689e5ce055209f0ce7a0e..6730f965b3793f25ba73125a00b18f346decf75f
mode 100644,000000,100644..100644
--- /dev/null
@@@@ -1,733 -1,0 -1,734 +1,733 @@@@
  -static int register_pcc_channel(unsigned pcc_subspace_idx)
 + /*
 +  * CPPC (Collaborative Processor Performance Control) methods used by CPUfreq drivers.
 +  *
 +  * (C) Copyright 2014, 2015 Linaro Ltd.
 +  * Author: Ashwin Chaugule <ashwin.chaugule@linaro.org>
 +  *
 +  * This program is free software; you can redistribute it and/or
 +  * modify it under the terms of the GNU General Public License
 +  * as published by the Free Software Foundation; version 2
 +  * of the License.
 +  *
 +  * CPPC describes a few methods for controlling CPU performance using
 +  * information from a per CPU table called CPC. This table is described in
 +  * the ACPI v5.0+ specification. The table consists of a list of
 +  * registers which may be memory mapped or hardware registers and also may
 +  * include some static integer values.
 +  *
 +  * CPU performance is on an abstract continuous scale as against a discretized
 +  * P-state scale which is tied to CPU frequency only. In brief, the basic
 +  * operation involves:
 +  *
 +  * - OS makes a CPU performance request. (Can provide min and max bounds)
 +  *
 +  * - Platform (such as BMC) is free to optimize request within requested bounds
 +  *   depending on power/thermal budgets etc.
 +  *
 +  * - Platform conveys its decision back to OS
 +  *
 +  * The communication between OS and platform occurs through another medium
 +  * called (PCC) Platform Communication Channel. This is a generic mailbox like
 +  * mechanism which includes doorbell semantics to indicate register updates.
 +  * See drivers/mailbox/pcc.c for details on PCC.
 +  *
 +  * Finer details about the PCC and CPPC spec are available in the ACPI v5.1 and
 +  * above specifications.
 +  */
 + 
 + #define pr_fmt(fmt)  "ACPI CPPC: " fmt
 + 
 + #include <linux/cpufreq.h>
 + #include <linux/delay.h>
 + 
 + #include <acpi/cppc_acpi.h>
 + /*
 +  * Lock to provide mutually exclusive access to the PCC
 +  * channel. e.g. When the remote updates the shared region
 +  * with new data, the reader needs to be protected from
 +  * other CPUs activity on the same channel.
 +  */
 + static DEFINE_SPINLOCK(pcc_lock);
 + 
 + /*
 +  * The cpc_desc structure contains the ACPI register details
 +  * as described in the per CPU _CPC tables. The details
 +  * include the type of register (e.g. PCC, System IO, FFH etc.)
 +  * and destination addresses which lets us READ/WRITE CPU performance
 +  * information using the appropriate I/O methods.
 +  */
 + static DEFINE_PER_CPU(struct cpc_desc *, cpc_desc_ptr);
 + 
 + /* This layer handles all the PCC specifics for CPPC. */
 + static struct mbox_chan *pcc_channel;
 + static void __iomem *pcc_comm_addr;
 + static u64 comm_base_addr;
 + static int pcc_subspace_idx = -1;
 + static u16 pcc_cmd_delay;
 + static bool pcc_channel_acquired;
 + 
 + /*
 +  * Arbitrary Retries in case the remote processor is slow to respond
 +  * to PCC commands.
 +  */
 + #define NUM_RETRIES 500
 + 
 + static int send_pcc_cmd(u16 cmd)
 + {
 +      int retries, result = -EIO;
 +      struct acpi_pcct_hw_reduced *pcct_ss = pcc_channel->con_priv;
 +      struct acpi_pcct_shared_memory *generic_comm_base =
 +              (struct acpi_pcct_shared_memory *) pcc_comm_addr;
 +      u32 cmd_latency = pcct_ss->latency;
 + 
 +      /* Min time OS should wait before sending next command. */
 +      udelay(pcc_cmd_delay);
 + 
 +      /* Write to the shared comm region. */
 +      writew(cmd, &generic_comm_base->command);
 + 
 +      /* Flip CMD COMPLETE bit */
 +      writew(0, &generic_comm_base->status);
 + 
 +      /* Ring doorbell */
 +      result = mbox_send_message(pcc_channel, &cmd);
 +      if (result < 0) {
 +              pr_err("Err sending PCC mbox message. cmd:%d, ret:%d\n",
 +                              cmd, result);
 +              return result;
 +      }
 + 
 +      /* Wait for a nominal time to let platform process command. */
 +      udelay(cmd_latency);
 + 
 +      /* Retry in case the remote processor was too slow to catch up. */
 +      for (retries = NUM_RETRIES; retries > 0; retries--) {
 +              if (readw_relaxed(&generic_comm_base->status) & PCC_CMD_COMPLETE) {
 +                      result = 0;
 +                      break;
 +              }
 +      }
 + 
 +      mbox_client_txdone(pcc_channel, result);
 +      return result;
 + }
 + 
 + static void cppc_chan_tx_done(struct mbox_client *cl, void *msg, int ret)
 + {
 +      if (ret)
 +              pr_debug("TX did not complete: CMD sent:%x, ret:%d\n",
 +                              *(u16 *)msg, ret);
 +      else
 +              pr_debug("TX completed. CMD sent:%x, ret:%d\n",
 +                              *(u16 *)msg, ret);
 + }
 + 
 + struct mbox_client cppc_mbox_cl = {
 +      .tx_done = cppc_chan_tx_done,
 +      .knows_txdone = true,
 + };
 + 
 + static int acpi_get_psd(struct cpc_desc *cpc_ptr, acpi_handle handle)
 + {
 +      int result = -EFAULT;
 +      acpi_status status = AE_OK;
 +      struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
 +      struct acpi_buffer format = {sizeof("NNNNN"), "NNNNN"};
 +      struct acpi_buffer state = {0, NULL};
 +      union acpi_object  *psd = NULL;
 +      struct acpi_psd_package *pdomain;
 + 
 +      status = acpi_evaluate_object_typed(handle, "_PSD", NULL, &buffer,
 +                      ACPI_TYPE_PACKAGE);
 +      if (ACPI_FAILURE(status))
 +              return -ENODEV;
 + 
 +      psd = buffer.pointer;
 +      if (!psd || psd->package.count != 1) {
 +              pr_debug("Invalid _PSD data\n");
 +              goto end;
 +      }
 + 
 +      pdomain = &(cpc_ptr->domain_info);
 + 
 +      state.length = sizeof(struct acpi_psd_package);
 +      state.pointer = pdomain;
 + 
 +      status = acpi_extract_package(&(psd->package.elements[0]),
 +              &format, &state);
 +      if (ACPI_FAILURE(status)) {
 +              pr_debug("Invalid _PSD data for CPU:%d\n", cpc_ptr->cpu_id);
 +              goto end;
 +      }
 + 
 +      if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
 +              pr_debug("Unknown _PSD:num_entries for CPU:%d\n", cpc_ptr->cpu_id);
 +              goto end;
 +      }
 + 
 +      if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
 +              pr_debug("Unknown _PSD:revision for CPU: %d\n", cpc_ptr->cpu_id);
 +              goto end;
 +      }
 + 
 +      if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
 +          pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
 +          pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
 +              pr_debug("Invalid _PSD:coord_type for CPU:%d\n", cpc_ptr->cpu_id);
 +              goto end;
 +      }
 + 
 +      result = 0;
 + end:
 +      kfree(buffer.pointer);
 +      return result;
 + }
 + 
 + /**
 +  * acpi_get_psd_map - Map the CPUs in a common freq domain.
 +  * @all_cpu_data: Ptrs to CPU specific CPPC data including PSD info.
 +  *
 +  *   Return: 0 for success or negative value for err.
 +  */
 + int acpi_get_psd_map(struct cpudata **all_cpu_data)
 + {
 +      int count_target;
 +      int retval = 0;
 +      unsigned int i, j;
 +      cpumask_var_t covered_cpus;
 +      struct cpudata *pr, *match_pr;
 +      struct acpi_psd_package *pdomain;
 +      struct acpi_psd_package *match_pdomain;
 +      struct cpc_desc *cpc_ptr, *match_cpc_ptr;
 + 
 +      if (!zalloc_cpumask_var(&covered_cpus, GFP_KERNEL))
 +              return -ENOMEM;
 + 
 +      /*
 +       * Now that we have _PSD data from all CPUs, lets setup P-state
 +       * domain info.
 +       */
 +      for_each_possible_cpu(i) {
 +              pr = all_cpu_data[i];
 +              if (!pr)
 +                      continue;
 + 
 +              if (cpumask_test_cpu(i, covered_cpus))
 +                      continue;
 + 
 +              cpc_ptr = per_cpu(cpc_desc_ptr, i);
 +              if (!cpc_ptr)
 +                      continue;
 + 
 +              pdomain = &(cpc_ptr->domain_info);
 +              cpumask_set_cpu(i, pr->shared_cpu_map);
 +              cpumask_set_cpu(i, covered_cpus);
 +              if (pdomain->num_processors <= 1)
 +                      continue;
 + 
 +              /* Validate the Domain info */
 +              count_target = pdomain->num_processors;
 +              if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
 +                      pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
 +              else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
 +                      pr->shared_type = CPUFREQ_SHARED_TYPE_HW;
 +              else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
 +                      pr->shared_type = CPUFREQ_SHARED_TYPE_ANY;
 + 
 +              for_each_possible_cpu(j) {
 +                      if (i == j)
 +                              continue;
 + 
 +                      match_cpc_ptr = per_cpu(cpc_desc_ptr, j);
 +                      if (!match_cpc_ptr)
 +                              continue;
 + 
 +                      match_pdomain = &(match_cpc_ptr->domain_info);
 +                      if (match_pdomain->domain != pdomain->domain)
 +                              continue;
 + 
 +                      /* Here i and j are in the same domain */
 +                      if (match_pdomain->num_processors != count_target) {
 +                              retval = -EFAULT;
 +                              goto err_ret;
 +                      }
 + 
 +                      if (pdomain->coord_type != match_pdomain->coord_type) {
 +                              retval = -EFAULT;
 +                              goto err_ret;
 +                      }
 + 
 +                      cpumask_set_cpu(j, covered_cpus);
 +                      cpumask_set_cpu(j, pr->shared_cpu_map);
 +              }
 + 
 +              for_each_possible_cpu(j) {
 +                      if (i == j)
 +                              continue;
 + 
 +                      match_pr = all_cpu_data[j];
 +                      if (!match_pr)
 +                              continue;
 + 
 +                      match_cpc_ptr = per_cpu(cpc_desc_ptr, j);
 +                      if (!match_cpc_ptr)
 +                              continue;
 + 
 +                      match_pdomain = &(match_cpc_ptr->domain_info);
 +                      if (match_pdomain->domain != pdomain->domain)
 +                              continue;
 + 
 +                      match_pr->shared_type = pr->shared_type;
 +                      cpumask_copy(match_pr->shared_cpu_map,
 +                                   pr->shared_cpu_map);
 +              }
 +      }
 + 
 + err_ret:
 +      for_each_possible_cpu(i) {
 +              pr = all_cpu_data[i];
 +              if (!pr)
 +                      continue;
 + 
 +              /* Assume no coordination on any error parsing domain info */
 +              if (retval) {
 +                      cpumask_clear(pr->shared_cpu_map);
 +                      cpumask_set_cpu(i, pr->shared_cpu_map);
 +                      pr->shared_type = CPUFREQ_SHARED_TYPE_ALL;
 +              }
 +      }
 + 
 +      free_cpumask_var(covered_cpus);
 +      return retval;
 + }
 + EXPORT_SYMBOL_GPL(acpi_get_psd_map);
 + 
-       struct acpi_pcct_subspace *cppc_ss;
 ++static int register_pcc_channel(int pcc_subspace_idx)
 + {
  -     cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
++      struct acpi_pcct_hw_reduced *cppc_ss;
 +      unsigned int len;
 + 
 +      if (pcc_subspace_idx >= 0) {
 +              pcc_channel = pcc_mbox_request_channel(&cppc_mbox_cl,
 +                              pcc_subspace_idx);
 + 
 +              if (IS_ERR(pcc_channel)) {
 +                      pr_err("Failed to find PCC communication channel\n");
 +                      return -ENODEV;
 +              }
 + 
 +              /*
 +               * The PCC mailbox controller driver should
 +               * have parsed the PCCT (global table of all
 +               * PCC channels) and stored pointers to the
 +               * subspace communication region in con_priv.
 +               */
 +              cppc_ss = pcc_channel->con_priv;
 + 
 +              if (!cppc_ss) {
 +                      pr_err("No PCC subspace found for CPPC\n");
 +                      return -ENODEV;
 +              }
 + 
 +              /*
 +               * This is the shared communication region
 +               * for the OS and Platform to communicate over.
 +               */
 +              comm_base_addr = cppc_ss->base_address;
 +              len = cppc_ss->length;
 +              pcc_cmd_delay = cppc_ss->min_turnaround_time;
 + 
 +              pcc_comm_addr = acpi_os_ioremap(comm_base_addr, len);
 +              if (!pcc_comm_addr) {
 +                      pr_err("Failed to ioremap PCC comm region mem\n");
 +                      return -ENOMEM;
 +              }
 + 
 +              /* Set flag so that we dont come here for each CPU. */
 +              pcc_channel_acquired = true;
 +      }
 + 
 +      return 0;
 + }
 + 
 + /*
 +  * An example CPC table looks like the following.
 +  *
 +  *   Name(_CPC, Package()
 +  *                   {
 +  *                   17,
 +  *                   NumEntries
 +  *                   1,
 +  *                   // Revision
 +  *                   ResourceTemplate(){Register(PCC, 32, 0, 0x120, 2)},
 +  *                   // Highest Performance
 +  *                   ResourceTemplate(){Register(PCC, 32, 0, 0x124, 2)},
 +  *                   // Nominal Performance
 +  *                   ResourceTemplate(){Register(PCC, 32, 0, 0x128, 2)},
 +  *                   // Lowest Nonlinear Performance
 +  *                   ResourceTemplate(){Register(PCC, 32, 0, 0x12C, 2)},
 +  *                   // Lowest Performance
 +  *                   ResourceTemplate(){Register(PCC, 32, 0, 0x130, 2)},
 +  *                   // Guaranteed Performance Register
 +  *                   ResourceTemplate(){Register(PCC, 32, 0, 0x110, 2)},
 +  *                   // Desired Performance Register
 +  *                   ResourceTemplate(){Register(SystemMemory, 0, 0, 0, 0)},
 +  *                   ..
 +  *                   ..
 +  *                   ..
 +  *
 +  *           }
 +  * Each Register() encodes how to access that specific register.
 +  * e.g. a sample PCC entry has the following encoding:
 +  *
 +  *   Register (
 +  *           PCC,
 +  *           AddressSpaceKeyword
 +  *           8,
 +  *           //RegisterBitWidth
 +  *           8,
 +  *           //RegisterBitOffset
 +  *           0x30,
 +  *           //RegisterAddress
 +  *           9
 +  *           //AccessSize (subspace ID)
 +  *           0
 +  *           )
 +  *   }
 +  */
 + 
 + /**
 +  * acpi_cppc_processor_probe - Search for per CPU _CPC objects.
 +  * @pr: Ptr to acpi_processor containing this CPUs logical Id.
 +  *
 +  *   Return: 0 for success or negative value for err.
 +  */
 + int acpi_cppc_processor_probe(struct acpi_processor *pr)
 + {
 +      struct acpi_buffer output = {ACPI_ALLOCATE_BUFFER, NULL};
 +      union acpi_object *out_obj, *cpc_obj;
 +      struct cpc_desc *cpc_ptr;
 +      struct cpc_reg *gas_t;
 +      acpi_handle handle = pr->handle;
 +      unsigned int num_ent, i, cpc_rev;
 +      acpi_status status;
 +      int ret = -EFAULT;
 + 
 +      /* Parse the ACPI _CPC table for this cpu. */
 +      status = acpi_evaluate_object_typed(handle, "_CPC", NULL, &output,
 +                      ACPI_TYPE_PACKAGE);
 +      if (ACPI_FAILURE(status)) {
 +              ret = -ENODEV;
 +              goto out_buf_free;
 +      }
 + 
 +      out_obj = (union acpi_object *) output.pointer;
 + 
 +      cpc_ptr = kzalloc(sizeof(struct cpc_desc), GFP_KERNEL);
 +      if (!cpc_ptr) {
 +              ret = -ENOMEM;
 +              goto out_buf_free;
 +      }
 + 
 +      /* First entry is NumEntries. */
 +      cpc_obj = &out_obj->package.elements[0];
 +      if (cpc_obj->type == ACPI_TYPE_INTEGER) {
 +              num_ent = cpc_obj->integer.value;
 +      } else {
 +              pr_debug("Unexpected entry type(%d) for NumEntries\n",
 +                              cpc_obj->type);
 +              goto out_free;
 +      }
 + 
 +      /* Only support CPPCv2. Bail otherwise. */
 +      if (num_ent != CPPC_NUM_ENT) {
 +              pr_debug("Firmware exports %d entries. Expected: %d\n",
 +                              num_ent, CPPC_NUM_ENT);
 +              goto out_free;
 +      }
 + 
 +      /* Second entry should be revision. */
 +      cpc_obj = &out_obj->package.elements[1];
 +      if (cpc_obj->type == ACPI_TYPE_INTEGER) {
 +              cpc_rev = cpc_obj->integer.value;
 +      } else {
 +              pr_debug("Unexpected entry type(%d) for Revision\n",
 +                              cpc_obj->type);
 +              goto out_free;
 +      }
 + 
 +      if (cpc_rev != CPPC_REV) {
 +              pr_debug("Firmware exports revision:%d. Expected:%d\n",
 +                              cpc_rev, CPPC_REV);
 +              goto out_free;
 +      }
 + 
 +      /* Iterate through remaining entries in _CPC */
 +      for (i = 2; i < num_ent; i++) {
 +              cpc_obj = &out_obj->package.elements[i];
 + 
 +              if (cpc_obj->type == ACPI_TYPE_INTEGER) {
 +                      cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_INTEGER;
 +                      cpc_ptr->cpc_regs[i-2].cpc_entry.int_value = cpc_obj->integer.value;
 +              } else if (cpc_obj->type == ACPI_TYPE_BUFFER) {
 +                      gas_t = (struct cpc_reg *)
 +                              cpc_obj->buffer.pointer;
 + 
 +                      /*
 +                       * The PCC Subspace index is encoded inside
 +                       * the CPC table entries. The same PCC index
 +                       * will be used for all the PCC entries,
 +                       * so extract it only once.
 +                       */
 +                      if (gas_t->space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
 +                              if (pcc_subspace_idx < 0)
 +                                      pcc_subspace_idx = gas_t->access_width;
 +                              else if (pcc_subspace_idx != gas_t->access_width) {
 +                                      pr_debug("Mismatched PCC ids.\n");
 +                                      goto out_free;
 +                              }
 +                      } else if (gas_t->space_id != ACPI_ADR_SPACE_SYSTEM_MEMORY) {
 +                              /* Support only PCC and SYS MEM type regs */
 +                              pr_debug("Unsupported register type: %d\n", gas_t->space_id);
 +                              goto out_free;
 +                      }
 + 
 +                      cpc_ptr->cpc_regs[i-2].type = ACPI_TYPE_BUFFER;
 +                      memcpy(&cpc_ptr->cpc_regs[i-2].cpc_entry.reg, gas_t, sizeof(*gas_t));
 +              } else {
 +                      pr_debug("Err in entry:%d in CPC table of CPU:%d \n", i, pr->id);
 +                      goto out_free;
 +              }
 +      }
 +      /* Store CPU Logical ID */
 +      cpc_ptr->cpu_id = pr->id;
 + 
 +      /* Plug it into this CPUs CPC descriptor. */
 +      per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
 + 
 +      /* Parse PSD data for this CPU */
 +      ret = acpi_get_psd(cpc_ptr, handle);
 +      if (ret)
 +              goto out_free;
 + 
 +      /* Register PCC channel once for all CPUs. */
 +      if (!pcc_channel_acquired) {
 +              ret = register_pcc_channel(pcc_subspace_idx);
 +              if (ret)
 +                      goto out_free;
 +      }
 + 
 +      /* Everything looks okay */
 +      pr_debug("Parsed CPC struct for CPU: %d\n", pr->id);
 + 
 +      kfree(output.pointer);
 +      return 0;
 + 
 + out_free:
 +      kfree(cpc_ptr);
 + 
 + out_buf_free:
 +      kfree(output.pointer);
 +      return ret;
 + }
 + EXPORT_SYMBOL_GPL(acpi_cppc_processor_probe);
 + 
 + /**
 +  * acpi_cppc_processor_exit - Cleanup CPC structs.
 +  * @pr: Ptr to acpi_processor containing this CPUs logical Id.
 +  *
 +  * Return: Void
 +  */
 + void acpi_cppc_processor_exit(struct acpi_processor *pr)
 + {
 +      struct cpc_desc *cpc_ptr;
 +      cpc_ptr = per_cpu(cpc_desc_ptr, pr->id);
 +      kfree(cpc_ptr);
 + }
 + EXPORT_SYMBOL_GPL(acpi_cppc_processor_exit);
 + 
 + static u64 get_phys_addr(struct cpc_reg *reg)
 + {
 +      /* PCC communication addr space begins at byte offset 0x8. */
 +      if (reg->space_id == ACPI_ADR_SPACE_PLATFORM_COMM)
 +              return (u64)comm_base_addr + 0x8 + reg->address;
 +      else
 +              return reg->address;
 + }
 + 
 + static void cpc_read(struct cpc_reg *reg, u64 *val)
 + {
 +      u64 addr = get_phys_addr(reg);
 + 
 +      acpi_os_read_memory((acpi_physical_address)addr,
 +                      val, reg->bit_width);
 + }
 + 
 + static void cpc_write(struct cpc_reg *reg, u64 val)
 + {
 +      u64 addr = get_phys_addr(reg);
 + 
 +      acpi_os_write_memory((acpi_physical_address)addr,
 +                      val, reg->bit_width);
 + }
 + 
 + /**
 +  * cppc_get_perf_caps - Get a CPUs performance capabilities.
 +  * @cpunum: CPU from which to get capabilities info.
 +  * @perf_caps: ptr to cppc_perf_caps. See cppc_acpi.h
 +  *
 +  * Return: 0 for success with perf_caps populated else -ERRNO.
 +  */
 + int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
 + {
 +      struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
 +      struct cpc_register_resource *highest_reg, *lowest_reg, *ref_perf,
 +                                                               *nom_perf;
 +      u64 high, low, ref, nom;
 +      int ret = 0;
 + 
 +      if (!cpc_desc) {
 +              pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
 +              return -ENODEV;
 +      }
 + 
 +      highest_reg = &cpc_desc->cpc_regs[HIGHEST_PERF];
 +      lowest_reg = &cpc_desc->cpc_regs[LOWEST_PERF];
 +      ref_perf = &cpc_desc->cpc_regs[REFERENCE_PERF];
 +      nom_perf = &cpc_desc->cpc_regs[NOMINAL_PERF];
 + 
 +      spin_lock(&pcc_lock);
 + 
 +      /* Are any of the regs PCC ?*/
 +      if ((highest_reg->cpc_entry.reg.space_id == ACPI_ADR_SPACE_PLATFORM_COMM) ||
 +                      (lowest_reg->cpc_entry.reg.space_id == ACPI_ADR_SPACE_PLATFORM_COMM) ||
 +                      (ref_perf->cpc_entry.reg.space_id == ACPI_ADR_SPACE_PLATFORM_COMM) ||
 +                      (nom_perf->cpc_entry.reg.space_id == ACPI_ADR_SPACE_PLATFORM_COMM)) {
 +              /* Ring doorbell once to update PCC subspace */
 +              if (send_pcc_cmd(CMD_READ)) {
 +                      ret = -EIO;
 +                      goto out_err;
 +              }
 +      }
 + 
 +      cpc_read(&highest_reg->cpc_entry.reg, &high);
 +      perf_caps->highest_perf = high;
 + 
 +      cpc_read(&lowest_reg->cpc_entry.reg, &low);
 +      perf_caps->lowest_perf = low;
 + 
 +      cpc_read(&ref_perf->cpc_entry.reg, &ref);
 +      perf_caps->reference_perf = ref;
 + 
 +      cpc_read(&nom_perf->cpc_entry.reg, &nom);
 +      perf_caps->nominal_perf = nom;
 + 
 +      if (!ref)
 +              perf_caps->reference_perf = perf_caps->nominal_perf;
 + 
 +      if (!high || !low || !nom)
 +              ret = -EFAULT;
 + 
 + out_err:
 +      spin_unlock(&pcc_lock);
 +      return ret;
 + }
 + EXPORT_SYMBOL_GPL(cppc_get_perf_caps);
 + 
 + /**
 +  * cppc_get_perf_ctrs - Read a CPUs performance feedback counters.
 +  * @cpunum: CPU from which to read counters.
 +  * @perf_fb_ctrs: ptr to cppc_perf_fb_ctrs. See cppc_acpi.h
 +  *
 +  * Return: 0 for success with perf_fb_ctrs populated else -ERRNO.
 +  */
 + int cppc_get_perf_ctrs(int cpunum, struct cppc_perf_fb_ctrs *perf_fb_ctrs)
 + {
 +      struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
 +      struct cpc_register_resource *delivered_reg, *reference_reg;
 +      u64 delivered, reference;
 +      int ret = 0;
 + 
 +      if (!cpc_desc) {
 +              pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
 +              return -ENODEV;
 +      }
 + 
 +      delivered_reg = &cpc_desc->cpc_regs[DELIVERED_CTR];
 +      reference_reg = &cpc_desc->cpc_regs[REFERENCE_CTR];
 + 
 +      spin_lock(&pcc_lock);
 + 
 +      /* Are any of the regs PCC ?*/
 +      if ((delivered_reg->cpc_entry.reg.space_id == ACPI_ADR_SPACE_PLATFORM_COMM) ||
 +                      (reference_reg->cpc_entry.reg.space_id == ACPI_ADR_SPACE_PLATFORM_COMM)) {
 +              /* Ring doorbell once to update PCC subspace */
 +              if (send_pcc_cmd(CMD_READ)) {
 +                      ret = -EIO;
 +                      goto out_err;
 +              }
 +      }
 + 
 +      cpc_read(&delivered_reg->cpc_entry.reg, &delivered);
 +      cpc_read(&reference_reg->cpc_entry.reg, &reference);
 + 
 +      if (!delivered || !reference) {
 +              ret = -EFAULT;
 +              goto out_err;
 +      }
 + 
 +      perf_fb_ctrs->delivered = delivered;
 +      perf_fb_ctrs->reference = reference;
 + 
 +      perf_fb_ctrs->delivered -= perf_fb_ctrs->prev_delivered;
 +      perf_fb_ctrs->reference -= perf_fb_ctrs->prev_reference;
 + 
 +      perf_fb_ctrs->prev_delivered = delivered;
 +      perf_fb_ctrs->prev_reference = reference;
 + 
 + out_err:
 +      spin_unlock(&pcc_lock);
 +      return ret;
 + }
 + EXPORT_SYMBOL_GPL(cppc_get_perf_ctrs);
 + 
 + /**
 +  * cppc_set_perf - Set a CPUs performance controls.
 +  * @cpu: CPU for which to set performance controls.
 +  * @perf_ctrls: ptr to cppc_perf_ctrls. See cppc_acpi.h
 +  *
 +  * Return: 0 for success, -ERRNO otherwise.
 +  */
 + int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls)
 + {
 +      struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpu);
 +      struct cpc_register_resource *desired_reg;
 +      int ret = 0;
 + 
 +      if (!cpc_desc) {
 +              pr_debug("No CPC descriptor for CPU:%d\n", cpu);
 +              return -ENODEV;
 +      }
 + 
 +      desired_reg = &cpc_desc->cpc_regs[DESIRED_PERF];
 + 
 +      spin_lock(&pcc_lock);
 + 
 +      /*
 +       * Skip writing MIN/MAX until Linux knows how to come up with
 +       * useful values.
 +       */
 +      cpc_write(&desired_reg->cpc_entry.reg, perf_ctrls->desired_perf);
 + 
 +      /* Is this a PCC reg ?*/
 +      if (desired_reg->cpc_entry.reg.space_id == ACPI_ADR_SPACE_PLATFORM_COMM) {
 +              /* Ring doorbell so Remote can get our perf request. */
 +              if (send_pcc_cmd(CMD_WRITE))
 +                      ret = -EIO;
 +      }
 + 
 +      spin_unlock(&pcc_lock);
 + 
 +      return ret;
 + }
 + EXPORT_SYMBOL_GPL(cppc_set_perf);
Simple merge