[PATCH v4 0/6] interconnect: qcom: Misc bcm-voter changes and fixes

[PATCH v4 0/6] interconnect: qcom: Misc bcm-voter changes and fixes

[Mike Tipton]

These changes are mostly unrelated, but there are some dependencies
between them.

v4:
- Reorder series such that the fix comes first
- Collect reviewed/acked-by from v3

v3:
- Improve qcom,tcs-wait property description
- Stop using #define in property `default` doc

v2:
- New patch for generic qcom,icc.h bindings
- New patch for documenting qcom,tcs-wait property
- Update bcm_div() 'base' parameter from u64 to u32

Mike Tipton (6):
  interconnect: qcom: Fix small BW votes being truncated to zero
  dt-bindings: interconnect: Add generic qcom bindings
  dt-bindings: interconnect: Add property to set BCM TCS wait behavior
  interconnect: qcom: Support bcm-voter-specific TCS wait behavior
  interconnect: qcom: Only wait for completion in AMC/WAKE by default
  interconnect: qcom: Add support for per-BCM scaling factors

 .../bindings/interconnect/qcom,bcm-voter.yaml | 20 ++++++
 drivers/interconnect/qcom/bcm-voter.c         | 63 ++++++++++++-------
 drivers/interconnect/qcom/icc-rpmh.c          |  3 +
 drivers/interconnect/qcom/icc-rpmh.h          | 20 ++----
 include/dt-bindings/interconnect/qcom,icc.h   | 26 ++++++++
 5 files changed, 95 insertions(+), 37 deletions(-)
 create mode 100644 include/dt-bindings/interconnect/qcom,icc.h

-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v4 1/6] interconnect: qcom: Fix small BW votes being truncated to zero

[Mike Tipton]

Small BW votes that translate to less than a single BCM unit are
currently truncated to zero. Ensure that non-zero BW requests always
result in at least a vote of 1 to BCM.

Fixes: 976daac4a1c5 ("interconnect: qcom: Consolidate interconnect RPMh support")
Signed-off-by: Mike Tipton <mdtipton@codeaurora.org>
---
 drivers/interconnect/qcom/bcm-voter.c | 27 +++++++++++++++++++--------
 1 file changed, 19 insertions(+), 8 deletions(-)

diff --git a/drivers/interconnect/qcom/bcm-voter.c b/drivers/interconnect/qcom/bcm-voter.c
index a3d2ef1d9903..609db9c95fd7 100644
--- a/drivers/interconnect/qcom/bcm-voter.c
+++ b/drivers/interconnect/qcom/bcm-voter.c
@@ -52,8 +52,20 @@ static int cmp_vcd(void *priv, struct list_head *a, struct list_head *b)
 		return 1;
 }
 
+static u64 bcm_div(u64 num, u32 base)
+{
+	/* Ensure that small votes aren't lost. */
+	if (num && num < base)
+		return 1;
+
+	do_div(num, base);
+
+	return num;
+}
+
 static void bcm_aggregate(struct qcom_icc_bcm *bcm)
 {
+	struct qcom_icc_node *node;
 	size_t i, bucket;
 	u64 agg_avg[QCOM_ICC_NUM_BUCKETS] = {0};
 	u64 agg_peak[QCOM_ICC_NUM_BUCKETS] = {0};
@@ -61,22 +73,21 @@ static void bcm_aggregate(struct qcom_icc_bcm *bcm)
 
 	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);
+			node = bcm->nodes[i];
+			temp = bcm_div(node->sum_avg[bucket] * bcm->aux_data.width,
+				       node->buswidth * node->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);
+			temp = bcm_div(node->max_peak[bucket] * bcm->aux_data.width,
+				       node->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;
+		bcm->vote_x[bucket] = bcm_div(temp, bcm->aux_data.unit);
 
 		temp = agg_peak[bucket] * 1000ULL;
-		do_div(temp, bcm->aux_data.unit);
-		bcm->vote_y[bucket] = temp;
+		bcm->vote_y[bucket] = bcm_div(temp, bcm->aux_data.unit);
 	}
 
 	if (bcm->keepalive && bcm->vote_x[QCOM_ICC_BUCKET_AMC] == 0 &&
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v4 2/6] dt-bindings: interconnect: Add generic qcom bindings

[Mike Tipton]

Add generic qcom interconnect bindings that are common across platforms. In
particular, these include QCOM_ICC_TAG_* macros that clients can use when
calling icc_set_tag().

Signed-off-by: Mike Tipton <mdtipton@codeaurora.org>
Acked-by: Rob Herring <robh@kernel.org>
---
 drivers/interconnect/qcom/icc-rpmh.h        | 18 ++------------
 include/dt-bindings/interconnect/qcom,icc.h | 26 +++++++++++++++++++++
 2 files changed, 28 insertions(+), 16 deletions(-)
 create mode 100644 include/dt-bindings/interconnect/qcom,icc.h

diff --git a/drivers/interconnect/qcom/icc-rpmh.h b/drivers/interconnect/qcom/icc-rpmh.h
index 903d25e61984..cb736b745e1a 100644
--- a/drivers/interconnect/qcom/icc-rpmh.h
+++ b/drivers/interconnect/qcom/icc-rpmh.h
@@ -6,6 +6,8 @@
 #ifndef __DRIVERS_INTERCONNECT_QCOM_ICC_RPMH_H__
 #define __DRIVERS_INTERCONNECT_QCOM_ICC_RPMH_H__
 
+#include <dt-bindings/interconnect/qcom,icc.h>
+
 #define to_qcom_provider(_provider) \
 	container_of(_provider, struct qcom_icc_provider, provider)
 
@@ -44,22 +46,6 @@ struct bcm_db {
 #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
diff --git a/include/dt-bindings/interconnect/qcom,icc.h b/include/dt-bindings/interconnect/qcom,icc.h
new file mode 100644
index 000000000000..cd34f36daaaa
--- /dev/null
+++ b/include/dt-bindings/interconnect/qcom,icc.h
@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2020, The Linux Foundation. All rights reserved.
+ */
+
+#ifndef __DT_BINDINGS_INTERCONNECT_QCOM_ICC_H
+#define __DT_BINDINGS_INTERCONNECT_QCOM_ICC_H
+
+/*
+ * 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		(1 << QCOM_ICC_BUCKET_AMC)
+#define QCOM_ICC_TAG_WAKE		(1 << QCOM_ICC_BUCKET_WAKE)
+#define QCOM_ICC_TAG_SLEEP		(1 << 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)
+
+#endif
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v4 3/6] dt-bindings: interconnect: Add property to set BCM TCS wait behavior

[Mike Tipton]

Add "qcom,tcs-wait" property to set which TCS should wait for completion
when triggering.

Signed-off-by: Mike Tipton <mdtipton@codeaurora.org>
Reviewed-by: Rob Herring <robh@kernel.org>
---
 .../bindings/interconnect/qcom,bcm-voter.yaml | 20 +++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/Documentation/devicetree/bindings/interconnect/qcom,bcm-voter.yaml b/Documentation/devicetree/bindings/interconnect/qcom,bcm-voter.yaml
index 5971fc1df08d..e23df4836c6f 100644
--- a/Documentation/devicetree/bindings/interconnect/qcom,bcm-voter.yaml
+++ b/Documentation/devicetree/bindings/interconnect/qcom,bcm-voter.yaml
@@ -21,6 +21,23 @@ properties:
     enum:
       - qcom,bcm-voter
 
+  qcom,tcs-wait:
+    description: |
+      Optional mask of which TCSs (Triggered Command Sets) wait for completion
+      upon triggering. If not specified, then the AMC and WAKE sets wait for
+      completion. The mask bits are available in the QCOM_ICC_TAG_* defines.
+
+      The AMC TCS is triggered immediately when icc_set_bw() is called. The
+      WAKE/SLEEP TCSs are triggered when the RSC transitions between active and
+      sleep modes.
+
+      In most cases, it's necessary to wait in both the AMC and WAKE sets to
+      ensure resources are available before use. If a specific RSC and its use
+      cases can ensure sufficient delay by other means, then this can be
+      overridden to reduce latencies.
+
+    $ref: /schemas/types.yaml#/definitions/uint32
+
 required:
   - compatible
 
@@ -39,7 +56,10 @@ examples:
   # as defined in Documentation/devicetree/bindings/soc/qcom/rpmh-rsc.txt
   - |
 
+    #include <dt-bindings/interconnect/qcom,icc.h>
+
     disp_bcm_voter: bcm_voter {
         compatible = "qcom,bcm-voter";
+        qcom,tcs-wait = <QCOM_ICC_TAG_AMC>;
     };
 ...
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v4 4/6] interconnect: qcom: Support bcm-voter-specific TCS wait behavior

[Mike Tipton]

Currently, all bcm-voters set tcs_cmd::wait=true for the last VCD
command in each TCS (AMC, WAKE, and SLEEP). However, some bcm-voters
don't need the completion and instead need to optimize for latency. For
instance, disabling wait-for-completion in the WAKE set can decrease
resume latency and allow for certain operations to occur in parallel
with the WAKE TCS triggering. This is only safe in very specific
situations. Keep the default behavior of always waiting, but allow it to
be overridden in devicetree.

Signed-off-by: Mike Tipton <mdtipton@codeaurora.org>
---
 drivers/interconnect/qcom/bcm-voter.c | 32 ++++++++++++++++++---------
 1 file changed, 21 insertions(+), 11 deletions(-)

diff --git a/drivers/interconnect/qcom/bcm-voter.c b/drivers/interconnect/qcom/bcm-voter.c
index 609db9c95fd7..f7db2b900599 100644
--- a/drivers/interconnect/qcom/bcm-voter.c
+++ b/drivers/interconnect/qcom/bcm-voter.c
@@ -27,6 +27,7 @@ static DEFINE_MUTEX(bcm_voter_lock);
  * @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
+ * @tcs_wait: mask for which buckets require TCS completion
  */
 struct bcm_voter {
 	struct device *dev;
@@ -35,6 +36,7 @@ struct bcm_voter {
 	struct list_head commit_list;
 	struct list_head ws_list;
 	struct list_head voter_node;
+	u32 tcs_wait;
 };
 
 static int cmp_vcd(void *priv, struct list_head *a, struct list_head *b)
@@ -100,7 +102,7 @@ static void bcm_aggregate(struct qcom_icc_bcm *bcm)
 }
 
 static inline void tcs_cmd_gen(struct tcs_cmd *cmd, u64 vote_x, u64 vote_y,
-			       u32 addr, bool commit)
+			       u32 addr, bool commit, bool wait)
 {
 	bool valid = true;
 
@@ -125,15 +127,16 @@ static inline void tcs_cmd_gen(struct tcs_cmd *cmd, u64 vote_x, u64 vote_y,
 	 * Set the wait for completion flag on command that need to be completed
 	 * before the next command.
 	 */
-	cmd->wait = commit;
+	cmd->wait = wait;
 }
 
-static void tcs_list_gen(struct list_head *bcm_list, int bucket,
-			 struct tcs_cmd tcs_list[MAX_BCMS],
+static void tcs_list_gen(struct bcm_voter *voter, int bucket,
+			 struct tcs_cmd tcs_list[MAX_VCD],
 			 int n[MAX_VCD + 1])
 {
+	struct list_head *bcm_list = &voter->commit_list;
 	struct qcom_icc_bcm *bcm;
-	bool commit;
+	bool commit, wait;
 	size_t idx = 0, batch = 0, cur_vcd_size = 0;
 
 	memset(n, 0, sizeof(int) * (MAX_VCD + 1));
@@ -146,8 +149,11 @@ static void tcs_list_gen(struct list_head *bcm_list, int bucket,
 			commit = true;
 			cur_vcd_size = 0;
 		}
+
+		wait = commit && (voter->tcs_wait & BIT(bucket));
+
 		tcs_cmd_gen(&tcs_list[idx], bcm->vote_x[bucket],
-			    bcm->vote_y[bucket], bcm->addr, commit);
+			    bcm->vote_y[bucket], bcm->addr, commit, wait);
 		idx++;
 		n[batch]++;
 		/*
@@ -272,8 +278,7 @@ int qcom_icc_bcm_voter_commit(struct bcm_voter *voter)
 	 * 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);
-
+	tcs_list_gen(voter, QCOM_ICC_BUCKET_AMC, cmds, commit_idx);
 	if (!commit_idx[0])
 		goto out;
 
@@ -309,7 +314,7 @@ int qcom_icc_bcm_voter_commit(struct bcm_voter *voter)
 
 	list_sort(NULL, &voter->commit_list, cmp_vcd);
 
-	tcs_list_gen(&voter->commit_list, QCOM_ICC_BUCKET_WAKE, cmds, commit_idx);
+	tcs_list_gen(voter, QCOM_ICC_BUCKET_WAKE, cmds, commit_idx);
 
 	ret = rpmh_write_batch(voter->dev, RPMH_WAKE_ONLY_STATE, cmds, commit_idx);
 	if (ret) {
@@ -317,7 +322,7 @@ int qcom_icc_bcm_voter_commit(struct bcm_voter *voter)
 		goto out;
 	}
 
-	tcs_list_gen(&voter->commit_list, QCOM_ICC_BUCKET_SLEEP, cmds, commit_idx);
+	tcs_list_gen(voter, QCOM_ICC_BUCKET_SLEEP, cmds, commit_idx);
 
 	ret = rpmh_write_batch(voter->dev, RPMH_SLEEP_STATE, cmds, commit_idx);
 	if (ret) {
@@ -336,6 +341,7 @@ EXPORT_SYMBOL_GPL(qcom_icc_bcm_voter_commit);
 
 static int qcom_icc_bcm_voter_probe(struct platform_device *pdev)
 {
+	struct device_node *np = pdev->dev.of_node;
 	struct bcm_voter *voter;
 
 	voter = devm_kzalloc(&pdev->dev, sizeof(*voter), GFP_KERNEL);
@@ -343,7 +349,11 @@ static int qcom_icc_bcm_voter_probe(struct platform_device *pdev)
 		return -ENOMEM;
 
 	voter->dev = &pdev->dev;
-	voter->np = pdev->dev.of_node;
+	voter->np = np;
+
+	if (of_property_read_u32(np, "qcom,tcs-wait", &voter->tcs_wait))
+		voter->tcs_wait = QCOM_ICC_TAG_ALWAYS;
+
 	mutex_init(&voter->lock);
 	INIT_LIST_HEAD(&voter->commit_list);
 	INIT_LIST_HEAD(&voter->ws_list);
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v4 5/6] interconnect: qcom: Only wait for completion in AMC/WAKE by default

[Mike Tipton]

Change the default TCS wait behavior to only wait for completion in AMC
and WAKE. Waiting isn't necessary in the SLEEP TCS, since votes are only
being removed in this case. Resources can be safely disabled
asynchronously in parallel with the rest of the power collapse sequence.
This reduces the sleep entry latency.

Signed-off-by: Mike Tipton <mdtipton@codeaurora.org>
---
 drivers/interconnect/qcom/bcm-voter.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/interconnect/qcom/bcm-voter.c b/drivers/interconnect/qcom/bcm-voter.c
index f7db2b900599..d58b08f1f495 100644
--- a/drivers/interconnect/qcom/bcm-voter.c
+++ b/drivers/interconnect/qcom/bcm-voter.c
@@ -352,7 +352,7 @@ static int qcom_icc_bcm_voter_probe(struct platform_device *pdev)
 	voter->np = np;
 
 	if (of_property_read_u32(np, "qcom,tcs-wait", &voter->tcs_wait))
-		voter->tcs_wait = QCOM_ICC_TAG_ALWAYS;
+		voter->tcs_wait = QCOM_ICC_TAG_ACTIVE_ONLY;
 
 	mutex_init(&voter->lock);
 	INIT_LIST_HEAD(&voter->commit_list);
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v4 6/6] interconnect: qcom: Add support for per-BCM scaling factors

[Mike Tipton]

Currently, bcm-voter always assumes requests are made in KBps and that
BCM HW always wants them in Bps, so it always scales the requests by
1000. However, certain use cases and BCMs may use different units.
Thus, add support for BCM-specific scaling factors.

Signed-off-by: Mike Tipton <mdtipton@codeaurora.org>
---
 drivers/interconnect/qcom/bcm-voter.c | 4 ++--
 drivers/interconnect/qcom/icc-rpmh.c  | 3 +++
 drivers/interconnect/qcom/icc-rpmh.h  | 2 ++
 3 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/interconnect/qcom/bcm-voter.c b/drivers/interconnect/qcom/bcm-voter.c
index d58b08f1f495..887d13721e52 100644
--- a/drivers/interconnect/qcom/bcm-voter.c
+++ b/drivers/interconnect/qcom/bcm-voter.c
@@ -85,10 +85,10 @@ static void bcm_aggregate(struct qcom_icc_bcm *bcm)
 			agg_peak[bucket] = max(agg_peak[bucket], temp);
 		}
 
-		temp = agg_avg[bucket] * 1000ULL;
+		temp = agg_avg[bucket] * bcm->vote_scale;
 		bcm->vote_x[bucket] = bcm_div(temp, bcm->aux_data.unit);
 
-		temp = agg_peak[bucket] * 1000ULL;
+		temp = agg_peak[bucket] * bcm->vote_scale;
 		bcm->vote_y[bucket] = bcm_div(temp, bcm->aux_data.unit);
 	}
 
diff --git a/drivers/interconnect/qcom/icc-rpmh.c b/drivers/interconnect/qcom/icc-rpmh.c
index 3ac5182c9ab2..008846c17bec 100644
--- a/drivers/interconnect/qcom/icc-rpmh.c
+++ b/drivers/interconnect/qcom/icc-rpmh.c
@@ -136,6 +136,9 @@ int qcom_icc_bcm_init(struct qcom_icc_bcm *bcm, struct device *dev)
 	INIT_LIST_HEAD(&bcm->list);
 	INIT_LIST_HEAD(&bcm->ws_list);
 
+	if (!bcm->vote_scale)
+		bcm->vote_scale = 1000;
+
 	/* Link Qnodes to their respective BCMs */
 	for (i = 0; i < bcm->num_nodes; i++) {
 		qn = bcm->nodes[i];
diff --git a/drivers/interconnect/qcom/icc-rpmh.h b/drivers/interconnect/qcom/icc-rpmh.h
index cb736b745e1a..10b1890345b1 100644
--- a/drivers/interconnect/qcom/icc-rpmh.h
+++ b/drivers/interconnect/qcom/icc-rpmh.h
@@ -80,6 +80,7 @@ struct qcom_icc_node {
  * @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
+ * @vote_scale: scaling factor for vote_x and vote_y
  * @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
@@ -95,6 +96,7 @@ struct qcom_icc_bcm {
 	u32 addr;
 	u64 vote_x[QCOM_ICC_NUM_BUCKETS];
 	u64 vote_y[QCOM_ICC_NUM_BUCKETS];
+	u64 vote_scale;
 	bool dirty;
 	bool keepalive;
 	struct bcm_db aux_data;
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project