interconnect: qcom: Consolidate interconnect RPMh support
authorDavid Dai <daidavid1@codeaurora.org>
Fri, 28 Feb 2020 10:11:40 +0000 (12:11 +0200)
committerGeorgi Djakov <georgi.djakov@linaro.org>
Fri, 28 Feb 2020 10:11:40 +0000 (12:11 +0200)
Add bcm voter driver and add support for RPMh specific interconnect
providers which implements the set and aggregate functionalities that
translates bandwidth requests into RPMh messages. These modules provide
a common set of functionalities for all Qualcomm RPMh based interconnect
providers and should help reduce code duplication when adding new
providers.

Signed-off-by: David Dai <daidavid1@codeaurora.org>
Signed-off-by: Odelu Kukatla <okukatla@codeaurora.org>
Reviewed-by: Evan Green <evgreen@chromium.org>
Signed-off-by: Sibi Sankar <sibis@codeaurora.org>
Link: https://lore.kernel.org/r/20200228095951.15457-1-sibis@codeaurora.org
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
drivers/interconnect/qcom/Kconfig
drivers/interconnect/qcom/Makefile
drivers/interconnect/qcom/bcm-voter.c [new file with mode: 0644]
drivers/interconnect/qcom/bcm-voter.h [new file with mode: 0644]
drivers/interconnect/qcom/icc-rpmh.c [new file with mode: 0644]
drivers/interconnect/qcom/icc-rpmh.h [new file with mode: 0644]

index 76938ec..87c9268 100644 (file)
@@ -5,6 +5,9 @@ config INTERCONNECT_QCOM
        help
          Support for Qualcomm's Network-on-Chip interconnect hardware.
 
+config INTERCONNECT_QCOM_BCM_VOTER
+       tristate
+
 config INTERCONNECT_QCOM_MSM8916
        tristate "Qualcomm MSM8916 interconnect driver"
        depends on INTERCONNECT_QCOM
@@ -32,10 +35,15 @@ config INTERCONNECT_QCOM_QCS404
          This is a driver for the Qualcomm Network-on-Chip on qcs404-based
          platforms.
 
+config INTERCONNECT_QCOM_RPMH
+       tristate
+
 config INTERCONNECT_QCOM_SDM845
        tristate "Qualcomm SDM845 interconnect driver"
        depends on INTERCONNECT_QCOM
        depends on (QCOM_RPMH && QCOM_COMMAND_DB && OF) || COMPILE_TEST
+       select INTERCONNECT_QCOM_RPMH
+       select INTERCONNECT_QCOM_BCM_VOTER
        help
          This is a driver for the Qualcomm Network-on-Chip on sdm845-based
          platforms.
index e827157..d591bb5 100644 (file)
@@ -1,13 +1,17 @@
 # SPDX-License-Identifier: GPL-2.0
 
+icc-bcm-voter-objs                     := bcm-voter.o
 qnoc-msm8916-objs                      := msm8916.o
 qnoc-msm8974-objs                      := msm8974.o
 qnoc-qcs404-objs                       := qcs404.o
+icc-rpmh-obj                           := icc-rpmh.o
 qnoc-sdm845-objs                       := sdm845.o
 icc-smd-rpm-objs                       := smd-rpm.o
 
+obj-$(CONFIG_INTERCONNECT_QCOM_BCM_VOTER) += icc-bcm-voter.o
 obj-$(CONFIG_INTERCONNECT_QCOM_MSM8916) += qnoc-msm8916.o
 obj-$(CONFIG_INTERCONNECT_QCOM_MSM8974) += qnoc-msm8974.o
 obj-$(CONFIG_INTERCONNECT_QCOM_QCS404) += qnoc-qcs404.o
+obj-$(CONFIG_INTERCONNECT_QCOM_RPMH) += icc-rpmh.o
 obj-$(CONFIG_INTERCONNECT_QCOM_SDM845) += qnoc-sdm845.o
 obj-$(CONFIG_INTERCONNECT_QCOM_SMD_RPM) += icc-smd-rpm.o
diff --git a/drivers/interconnect/qcom/bcm-voter.c b/drivers/interconnect/qcom/bcm-voter.c
new file mode 100644 (file)
index 0000000..2adfde8
--- /dev/null
@@ -0,0 +1,366 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2020, The Linux Foundation. All rights reserved.
+ */
+
+#include <asm/div64.h>
+#include <linux/interconnect-provider.h>
+#include <linux/list_sort.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+
+#include <soc/qcom/rpmh.h>
+#include <soc/qcom/tcs.h>
+
+#include "bcm-voter.h"
+#include "icc-rpmh.h"
+
+static LIST_HEAD(bcm_voters);
+static DEFINE_MUTEX(bcm_voter_lock);
+
+/**
+ * struct bcm_voter - Bus Clock Manager voter
+ * @dev: reference to the device that communicates with the BCM
+ * @np: reference to the device node to match bcm voters
+ * @lock: mutex to protect commit and wake/sleep lists in the voter
+ * @commit_list: list containing bcms to be committed to hardware
+ * @ws_list: list containing bcms that have different wake/sleep votes
+ * @voter_node: list of bcm voters
+ */
+struct bcm_voter {
+       struct device *dev;
+       struct device_node *np;
+       struct mutex lock;
+       struct list_head commit_list;
+       struct list_head ws_list;
+       struct list_head voter_node;
+};
+
+static int cmp_vcd(void *priv, struct list_head *a, struct list_head *b)
+{
+       const struct qcom_icc_bcm *bcm_a =
+                       list_entry(a, struct qcom_icc_bcm, list);
+       const struct qcom_icc_bcm *bcm_b =
+                       list_entry(b, struct qcom_icc_bcm, list);
+
+       if (bcm_a->aux_data.vcd < bcm_b->aux_data.vcd)
+               return -1;
+       else if (bcm_a->aux_data.vcd == bcm_b->aux_data.vcd)
+               return 0;
+       else
+               return 1;
+}
+
+static void bcm_aggregate(struct qcom_icc_bcm *bcm)
+{
+       size_t i, bucket;
+       u64 agg_avg[QCOM_ICC_NUM_BUCKETS] = {0};
+       u64 agg_peak[QCOM_ICC_NUM_BUCKETS] = {0};
+       u64 temp;
+
+       for (bucket = 0; bucket < QCOM_ICC_NUM_BUCKETS; bucket++) {
+               for (i = 0; i < bcm->num_nodes; i++) {
+                       temp = bcm->nodes[i]->sum_avg[bucket] * bcm->aux_data.width;
+                       do_div(temp, bcm->nodes[i]->buswidth * bcm->nodes[i]->channels);
+                       agg_avg[bucket] = max(agg_avg[bucket], temp);
+
+                       temp = bcm->nodes[i]->max_peak[bucket] * bcm->aux_data.width;
+                       do_div(temp, bcm->nodes[i]->buswidth);
+                       agg_peak[bucket] = max(agg_peak[bucket], temp);
+               }
+
+               temp = agg_avg[bucket] * 1000ULL;
+               do_div(temp, bcm->aux_data.unit);
+               bcm->vote_x[bucket] = temp;
+
+               temp = agg_peak[bucket] * 1000ULL;
+               do_div(temp, bcm->aux_data.unit);
+               bcm->vote_y[bucket] = temp;
+       }
+
+       if (bcm->keepalive && bcm->vote_x[QCOM_ICC_BUCKET_AMC] == 0 &&
+           bcm->vote_y[QCOM_ICC_BUCKET_AMC] == 0) {
+               bcm->vote_x[QCOM_ICC_BUCKET_AMC] = 1;
+               bcm->vote_x[QCOM_ICC_BUCKET_WAKE] = 1;
+               bcm->vote_y[QCOM_ICC_BUCKET_AMC] = 1;
+               bcm->vote_y[QCOM_ICC_BUCKET_WAKE] = 1;
+       }
+}
+
+static inline void tcs_cmd_gen(struct tcs_cmd *cmd, u64 vote_x, u64 vote_y,
+                              u32 addr, bool commit)
+{
+       bool valid = true;
+
+       if (!cmd)
+               return;
+
+       if (vote_x == 0 && vote_y == 0)
+               valid = false;
+
+       if (vote_x > BCM_TCS_CMD_VOTE_MASK)
+               vote_x = BCM_TCS_CMD_VOTE_MASK;
+
+       if (vote_y > BCM_TCS_CMD_VOTE_MASK)
+               vote_y = BCM_TCS_CMD_VOTE_MASK;
+
+       cmd->addr = addr;
+       cmd->data = BCM_TCS_CMD(commit, valid, vote_x, vote_y);
+
+       /*
+        * Set the wait for completion flag on command that need to be completed
+        * before the next command.
+        */
+       if (commit)
+               cmd->wait = true;
+}
+
+static void tcs_list_gen(struct list_head *bcm_list, int bucket,
+                        struct tcs_cmd tcs_list[MAX_BCMS],
+                        int n[MAX_VCD + 1])
+{
+       struct qcom_icc_bcm *bcm;
+       bool commit;
+       size_t idx = 0, batch = 0, cur_vcd_size = 0;
+
+       memset(n, 0, sizeof(int) * (MAX_VCD + 1));
+
+       list_for_each_entry(bcm, bcm_list, list) {
+               commit = false;
+               cur_vcd_size++;
+               if ((list_is_last(&bcm->list, bcm_list)) ||
+                   bcm->aux_data.vcd != list_next_entry(bcm, list)->aux_data.vcd) {
+                       commit = true;
+                       cur_vcd_size = 0;
+               }
+               tcs_cmd_gen(&tcs_list[idx], bcm->vote_x[bucket],
+                           bcm->vote_y[bucket], bcm->addr, commit);
+               idx++;
+               n[batch]++;
+               /*
+                * Batch the BCMs in such a way that we do not split them in
+                * multiple payloads when they are under the same VCD. This is
+                * to ensure that every BCM is committed since we only set the
+                * commit bit on the last BCM request of every VCD.
+                */
+               if (n[batch] >= MAX_RPMH_PAYLOAD) {
+                       if (!commit) {
+                               n[batch] -= cur_vcd_size;
+                               n[batch + 1] = cur_vcd_size;
+                       }
+                       batch++;
+               }
+       }
+}
+
+/**
+ * of_bcm_voter_get - gets a bcm voter handle from DT node
+ * @dev: device pointer for the consumer device
+ * @name: name for the bcm voter device
+ *
+ * This function will match a device_node pointer for the phandle
+ * specified in the device DT and return a bcm_voter handle on success.
+ *
+ * Returns bcm_voter pointer or ERR_PTR() on error. EPROBE_DEFER is returned
+ * when matching bcm voter is yet to be found.
+ */
+struct bcm_voter *of_bcm_voter_get(struct device *dev, const char *name)
+{
+       struct bcm_voter *voter = ERR_PTR(-EPROBE_DEFER);
+       struct bcm_voter *temp;
+       struct device_node *np, *node;
+       int idx = 0;
+
+       if (!dev || !dev->of_node)
+               return ERR_PTR(-ENODEV);
+
+       np = dev->of_node;
+
+       if (name) {
+               idx = of_property_match_string(np, "qcom,bcm-voter-names", name);
+               if (idx < 0)
+                       return ERR_PTR(idx);
+       }
+
+       node = of_parse_phandle(np, "qcom,bcm-voters", idx);
+
+       mutex_lock(&bcm_voter_lock);
+       list_for_each_entry(temp, &bcm_voters, voter_node) {
+               if (temp->np == node) {
+                       voter = temp;
+                       break;
+               }
+       }
+       mutex_unlock(&bcm_voter_lock);
+
+       return voter;
+}
+EXPORT_SYMBOL_GPL(of_bcm_voter_get);
+
+/**
+ * qcom_icc_bcm_voter_add - queues up the bcm nodes that require updates
+ * @voter: voter that the bcms are being added to
+ * @bcm: bcm to add to the commit and wake sleep list
+ */
+void qcom_icc_bcm_voter_add(struct bcm_voter *voter, struct qcom_icc_bcm *bcm)
+{
+       if (!voter)
+               return;
+
+       mutex_lock(&voter->lock);
+       if (list_empty(&bcm->list))
+               list_add_tail(&bcm->list, &voter->commit_list);
+
+       if (list_empty(&bcm->ws_list))
+               list_add_tail(&bcm->ws_list, &voter->ws_list);
+
+       mutex_unlock(&voter->lock);
+}
+EXPORT_SYMBOL_GPL(qcom_icc_bcm_voter_add);
+
+/**
+ * qcom_icc_bcm_voter_commit - generates and commits tcs cmds based on bcms
+ * @voter: voter that needs flushing
+ *
+ * This function generates a set of AMC commands and flushes to the BCM device
+ * associated with the voter. It conditionally generate WAKE and SLEEP commands
+ * based on deltas between WAKE/SLEEP requirements. The ws_list persists
+ * through multiple commit requests and bcm nodes are removed only when the
+ * requirements for WAKE matches SLEEP.
+ *
+ * Returns 0 on success, or an appropriate error code otherwise.
+ */
+int qcom_icc_bcm_voter_commit(struct bcm_voter *voter)
+{
+       struct qcom_icc_bcm *bcm;
+       struct qcom_icc_bcm *bcm_tmp;
+       int commit_idx[MAX_VCD + 1];
+       struct tcs_cmd cmds[MAX_BCMS];
+       int ret = 0;
+
+       if (!voter)
+               return 0;
+
+       mutex_lock(&voter->lock);
+       list_for_each_entry(bcm, &voter->commit_list, list)
+               bcm_aggregate(bcm);
+
+       /*
+        * Pre sort the BCMs based on VCD for ease of generating a command list
+        * that groups the BCMs with the same VCD together. VCDs are numbered
+        * with lowest being the most expensive time wise, ensuring that
+        * those commands are being sent the earliest in the queue. This needs
+        * to be sorted every commit since we can't guarantee the order in which
+        * the BCMs are added to the list.
+        */
+       list_sort(NULL, &voter->commit_list, cmp_vcd);
+
+       /*
+        * Construct the command list based on a pre ordered list of BCMs
+        * based on VCD.
+        */
+       tcs_list_gen(&voter->commit_list, QCOM_ICC_BUCKET_AMC, cmds, commit_idx);
+
+       if (!commit_idx[0])
+               goto out;
+
+       ret = rpmh_invalidate(voter->dev);
+       if (ret) {
+               pr_err("Error invalidating RPMH client (%d)\n", ret);
+               goto out;
+       }
+
+       ret = rpmh_write_batch(voter->dev, RPMH_ACTIVE_ONLY_STATE,
+                              cmds, commit_idx);
+       if (ret) {
+               pr_err("Error sending AMC RPMH requests (%d)\n", ret);
+               goto out;
+       }
+
+       list_for_each_entry_safe(bcm, bcm_tmp, &voter->commit_list, list)
+               list_del_init(&bcm->list);
+
+       list_for_each_entry_safe(bcm, bcm_tmp, &voter->ws_list, ws_list) {
+               /*
+                * Only generate WAKE and SLEEP commands if a resource's
+                * requirements change as the execution environment transitions
+                * between different power states.
+                */
+               if (bcm->vote_x[QCOM_ICC_BUCKET_WAKE] !=
+                   bcm->vote_x[QCOM_ICC_BUCKET_SLEEP] ||
+                   bcm->vote_y[QCOM_ICC_BUCKET_WAKE] !=
+                   bcm->vote_y[QCOM_ICC_BUCKET_SLEEP])
+                       list_add_tail(&bcm->list, &voter->commit_list);
+               else
+                       list_del_init(&bcm->ws_list);
+       }
+
+       if (list_empty(&voter->commit_list))
+               goto out;
+
+       list_sort(NULL, &voter->commit_list, cmp_vcd);
+
+       tcs_list_gen(&voter->commit_list, QCOM_ICC_BUCKET_WAKE, cmds, commit_idx);
+
+       ret = rpmh_write_batch(voter->dev, RPMH_WAKE_ONLY_STATE, cmds, commit_idx);
+       if (ret) {
+               pr_err("Error sending WAKE RPMH requests (%d)\n", ret);
+               goto out;
+       }
+
+       tcs_list_gen(&voter->commit_list, QCOM_ICC_BUCKET_SLEEP, cmds, commit_idx);
+
+       ret = rpmh_write_batch(voter->dev, RPMH_SLEEP_STATE, cmds, commit_idx);
+       if (ret) {
+               pr_err("Error sending SLEEP RPMH requests (%d)\n", ret);
+               goto out;
+       }
+
+out:
+       list_for_each_entry_safe(bcm, bcm_tmp, &voter->commit_list, list)
+               list_del_init(&bcm->list);
+
+       mutex_unlock(&voter->lock);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(qcom_icc_bcm_voter_commit);
+
+static int qcom_icc_bcm_voter_probe(struct platform_device *pdev)
+{
+       struct bcm_voter *voter;
+
+       voter = devm_kzalloc(&pdev->dev, sizeof(*voter), GFP_KERNEL);
+       if (!voter)
+               return -ENOMEM;
+
+       voter->dev = &pdev->dev;
+       voter->np = pdev->dev.of_node;
+       mutex_init(&voter->lock);
+       INIT_LIST_HEAD(&voter->commit_list);
+       INIT_LIST_HEAD(&voter->ws_list);
+
+       mutex_lock(&bcm_voter_lock);
+       list_add_tail(&voter->voter_node, &bcm_voters);
+       mutex_unlock(&bcm_voter_lock);
+
+       return 0;
+}
+
+static const struct of_device_id bcm_voter_of_match[] = {
+       { .compatible = "qcom,bcm-voter" },
+       { }
+};
+
+static struct platform_driver qcom_icc_bcm_voter_driver = {
+       .probe = qcom_icc_bcm_voter_probe,
+       .driver = {
+               .name           = "bcm_voter",
+               .of_match_table = bcm_voter_of_match,
+       },
+};
+module_platform_driver(qcom_icc_bcm_voter_driver);
+
+MODULE_AUTHOR("David Dai <daidavid1@codeaurora.org>");
+MODULE_DESCRIPTION("Qualcomm BCM Voter interconnect driver");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/interconnect/qcom/bcm-voter.h b/drivers/interconnect/qcom/bcm-voter.h
new file mode 100644 (file)
index 0000000..0f64c0b
--- /dev/null
@@ -0,0 +1,27 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (c) 2020, The Linux Foundation. All rights reserved.
+ */
+
+#ifndef __DRIVERS_INTERCONNECT_QCOM_BCM_VOTER_H__
+#define __DRIVERS_INTERCONNECT_QCOM_BCM_VOTER_H__
+
+#include <soc/qcom/cmd-db.h>
+#include <soc/qcom/rpmh.h>
+#include <soc/qcom/tcs.h>
+
+#include "icc-rpmh.h"
+
+#define DEFINE_QBCM(_name, _bcmname, _keepalive, ...)                         \
+static struct qcom_icc_bcm _name = {                                          \
+       .name = _bcmname,                                                      \
+       .keepalive = _keepalive,                                               \
+       .num_nodes = ARRAY_SIZE(((struct qcom_icc_node *[]){ __VA_ARGS__ })),  \
+       .nodes = { __VA_ARGS__ },                                              \
+}
+
+struct bcm_voter *of_bcm_voter_get(struct device *dev, const char *name);
+void qcom_icc_bcm_voter_add(struct bcm_voter *voter, struct qcom_icc_bcm *bcm);
+int qcom_icc_bcm_voter_commit(struct bcm_voter *voter);
+
+#endif
diff --git a/drivers/interconnect/qcom/icc-rpmh.c b/drivers/interconnect/qcom/icc-rpmh.c
new file mode 100644 (file)
index 0000000..3ac5182
--- /dev/null
@@ -0,0 +1,150 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2020, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/interconnect.h>
+#include <linux/interconnect-provider.h>
+#include <linux/module.h>
+
+#include "bcm-voter.h"
+#include "icc-rpmh.h"
+
+/**
+ * qcom_icc_pre_aggregate - cleans up stale values from prior icc_set
+ * @node: icc node to operate on
+ */
+void qcom_icc_pre_aggregate(struct icc_node *node)
+{
+       size_t i;
+       struct qcom_icc_node *qn;
+
+       qn = node->data;
+
+       for (i = 0; i < QCOM_ICC_NUM_BUCKETS; i++) {
+               qn->sum_avg[i] = 0;
+               qn->max_peak[i] = 0;
+       }
+}
+EXPORT_SYMBOL_GPL(qcom_icc_pre_aggregate);
+
+/**
+ * qcom_icc_aggregate - aggregate bw for buckets indicated by tag
+ * @node: node to aggregate
+ * @tag: tag to indicate which buckets to aggregate
+ * @avg_bw: new bw to sum aggregate
+ * @peak_bw: new bw to max aggregate
+ * @agg_avg: existing aggregate avg bw val
+ * @agg_peak: existing aggregate peak bw val
+ */
+int qcom_icc_aggregate(struct icc_node *node, u32 tag, u32 avg_bw,
+                      u32 peak_bw, u32 *agg_avg, u32 *agg_peak)
+{
+       size_t i;
+       struct qcom_icc_node *qn;
+       struct qcom_icc_provider *qp;
+
+       qn = node->data;
+       qp = to_qcom_provider(node->provider);
+
+       if (!tag)
+               tag = QCOM_ICC_TAG_ALWAYS;
+
+       for (i = 0; i < QCOM_ICC_NUM_BUCKETS; i++) {
+               if (tag & BIT(i)) {
+                       qn->sum_avg[i] += avg_bw;
+                       qn->max_peak[i] = max_t(u32, qn->max_peak[i], peak_bw);
+               }
+       }
+
+       *agg_avg += avg_bw;
+       *agg_peak = max_t(u32, *agg_peak, peak_bw);
+
+       for (i = 0; i < qn->num_bcms; i++)
+               qcom_icc_bcm_voter_add(qp->voter, qn->bcms[i]);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(qcom_icc_aggregate);
+
+/**
+ * qcom_icc_set - set the constraints based on path
+ * @src: source node for the path to set constraints on
+ * @dst: destination node for the path to set constraints on
+ *
+ * Return: 0 on success, or an error code otherwise
+ */
+int qcom_icc_set(struct icc_node *src, struct icc_node *dst)
+{
+       struct qcom_icc_provider *qp;
+       struct icc_node *node;
+
+       if (!src)
+               node = dst;
+       else
+               node = src;
+
+       qp = to_qcom_provider(node->provider);
+
+       qcom_icc_bcm_voter_commit(qp->voter);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(qcom_icc_set);
+
+/**
+ * qcom_icc_bcm_init - populates bcm aux data and connect qnodes
+ * @bcm: bcm to be initialized
+ * @dev: associated provider device
+ *
+ * Return: 0 on success, or an error code otherwise
+ */
+int qcom_icc_bcm_init(struct qcom_icc_bcm *bcm, struct device *dev)
+{
+       struct qcom_icc_node *qn;
+       const struct bcm_db *data;
+       size_t data_count;
+       int i;
+
+       /* BCM is already initialised*/
+       if (bcm->addr)
+               return 0;
+
+       bcm->addr = cmd_db_read_addr(bcm->name);
+       if (!bcm->addr) {
+               dev_err(dev, "%s could not find RPMh address\n",
+                       bcm->name);
+               return -EINVAL;
+       }
+
+       data = cmd_db_read_aux_data(bcm->name, &data_count);
+       if (IS_ERR(data)) {
+               dev_err(dev, "%s command db read error (%ld)\n",
+                       bcm->name, PTR_ERR(data));
+               return PTR_ERR(data);
+       }
+       if (!data_count) {
+               dev_err(dev, "%s command db missing or partial aux data\n",
+                       bcm->name);
+               return -EINVAL;
+       }
+
+       bcm->aux_data.unit = le32_to_cpu(data->unit);
+       bcm->aux_data.width = le16_to_cpu(data->width);
+       bcm->aux_data.vcd = data->vcd;
+       bcm->aux_data.reserved = data->reserved;
+       INIT_LIST_HEAD(&bcm->list);
+       INIT_LIST_HEAD(&bcm->ws_list);
+
+       /* Link Qnodes to their respective BCMs */
+       for (i = 0; i < bcm->num_nodes; i++) {
+               qn = bcm->nodes[i];
+               qn->bcms[qn->num_bcms] = bcm;
+               qn->num_bcms++;
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(qcom_icc_bcm_init);
+
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/interconnect/qcom/icc-rpmh.h b/drivers/interconnect/qcom/icc-rpmh.h
new file mode 100644 (file)
index 0000000..903d25e
--- /dev/null
@@ -0,0 +1,149 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (c) 2020, The Linux Foundation. All rights reserved.
+ */
+
+#ifndef __DRIVERS_INTERCONNECT_QCOM_ICC_RPMH_H__
+#define __DRIVERS_INTERCONNECT_QCOM_ICC_RPMH_H__
+
+#define to_qcom_provider(_provider) \
+       container_of(_provider, struct qcom_icc_provider, provider)
+
+/**
+ * struct qcom_icc_provider - Qualcomm specific interconnect provider
+ * @provider: generic interconnect provider
+ * @dev: reference to the NoC device
+ * @bcms: list of bcms that maps to the provider
+ * @num_bcms: number of @bcms
+ * @voter: bcm voter targeted by this provider
+ */
+struct qcom_icc_provider {
+       struct icc_provider provider;
+       struct device *dev;
+       struct qcom_icc_bcm **bcms;
+       size_t num_bcms;
+       struct bcm_voter *voter;
+};
+
+/**
+ * struct bcm_db - Auxiliary data pertaining to each Bus Clock Manager (BCM)
+ * @unit: divisor used to convert bytes/sec bw value to an RPMh msg
+ * @width: multiplier used to convert bytes/sec bw value to an RPMh msg
+ * @vcd: virtual clock domain that this bcm belongs to
+ * @reserved: reserved field
+ */
+struct bcm_db {
+       __le32 unit;
+       __le16 width;
+       u8 vcd;
+       u8 reserved;
+};
+
+#define MAX_LINKS              128
+#define MAX_BCMS               64
+#define MAX_BCM_PER_NODE       3
+#define MAX_VCD                        10
+
+/*
+ * The AMC bucket denotes constraints that are applied to hardware when
+ * icc_set_bw() completes, whereas the WAKE and SLEEP constraints are applied
+ * when the execution environment transitions between active and low power mode.
+ */
+#define QCOM_ICC_BUCKET_AMC            0
+#define QCOM_ICC_BUCKET_WAKE           1
+#define QCOM_ICC_BUCKET_SLEEP          2
+#define QCOM_ICC_NUM_BUCKETS           3
+#define QCOM_ICC_TAG_AMC               BIT(QCOM_ICC_BUCKET_AMC)
+#define QCOM_ICC_TAG_WAKE              BIT(QCOM_ICC_BUCKET_WAKE)
+#define QCOM_ICC_TAG_SLEEP             BIT(QCOM_ICC_BUCKET_SLEEP)
+#define QCOM_ICC_TAG_ACTIVE_ONLY       (QCOM_ICC_TAG_AMC | QCOM_ICC_TAG_WAKE)
+#define QCOM_ICC_TAG_ALWAYS            (QCOM_ICC_TAG_AMC | QCOM_ICC_TAG_WAKE |\
+                                        QCOM_ICC_TAG_SLEEP)
+
+/**
+ * struct qcom_icc_node - Qualcomm specific interconnect nodes
+ * @name: the node name used in debugfs
+ * @links: an array of nodes where we can go next while traversing
+ * @id: a unique node identifier
+ * @num_links: the total number of @links
+ * @channels: num of channels at this node
+ * @buswidth: width of the interconnect between a node and the bus
+ * @sum_avg: current sum aggregate value of all avg bw requests
+ * @max_peak: current max aggregate value of all peak bw requests
+ * @bcms: list of bcms associated with this logical node
+ * @num_bcms: num of @bcms
+ */
+struct qcom_icc_node {
+       const char *name;
+       u16 links[MAX_LINKS];
+       u16 id;
+       u16 num_links;
+       u16 channels;
+       u16 buswidth;
+       u64 sum_avg[QCOM_ICC_NUM_BUCKETS];
+       u64 max_peak[QCOM_ICC_NUM_BUCKETS];
+       struct qcom_icc_bcm *bcms[MAX_BCM_PER_NODE];
+       size_t num_bcms;
+};
+
+/**
+ * struct qcom_icc_bcm - Qualcomm specific hardware accelerator nodes
+ * known as Bus Clock Manager (BCM)
+ * @name: the bcm node name used to fetch BCM data from command db
+ * @type: latency or bandwidth bcm
+ * @addr: address offsets used when voting to RPMH
+ * @vote_x: aggregated threshold values, represents sum_bw when @type is bw bcm
+ * @vote_y: aggregated threshold values, represents peak_bw when @type is bw bcm
+ * @dirty: flag used to indicate whether the bcm needs to be committed
+ * @keepalive: flag used to indicate whether a keepalive is required
+ * @aux_data: auxiliary data used when calculating threshold values and
+ * communicating with RPMh
+ * @list: used to link to other bcms when compiling lists for commit
+ * @ws_list: used to keep track of bcms that may transition between wake/sleep
+ * @num_nodes: total number of @num_nodes
+ * @nodes: list of qcom_icc_nodes that this BCM encapsulates
+ */
+struct qcom_icc_bcm {
+       const char *name;
+       u32 type;
+       u32 addr;
+       u64 vote_x[QCOM_ICC_NUM_BUCKETS];
+       u64 vote_y[QCOM_ICC_NUM_BUCKETS];
+       bool dirty;
+       bool keepalive;
+       struct bcm_db aux_data;
+       struct list_head list;
+       struct list_head ws_list;
+       size_t num_nodes;
+       struct qcom_icc_node *nodes[];
+};
+
+struct qcom_icc_fabric {
+       struct qcom_icc_node **nodes;
+       size_t num_nodes;
+};
+
+struct qcom_icc_desc {
+       struct qcom_icc_node **nodes;
+       size_t num_nodes;
+       struct qcom_icc_bcm **bcms;
+       size_t num_bcms;
+};
+
+#define DEFINE_QNODE(_name, _id, _channels, _buswidth, ...)            \
+               static struct qcom_icc_node _name = {                   \
+               .id = _id,                                              \
+               .name = #_name,                                         \
+               .channels = _channels,                                  \
+               .buswidth = _buswidth,                                  \
+               .num_links = ARRAY_SIZE(((int[]){ __VA_ARGS__ })),      \
+               .links = { __VA_ARGS__ },                               \
+       }
+
+int qcom_icc_aggregate(struct icc_node *node, u32 tag, u32 avg_bw,
+                      u32 peak_bw, u32 *agg_avg, u32 *agg_peak);
+int qcom_icc_set(struct icc_node *src, struct icc_node *dst);
+int qcom_icc_bcm_init(struct qcom_icc_bcm *bcm, struct device *dev);
+void qcom_icc_pre_aggregate(struct icc_node *node);
+
+#endif