[PATCH v6 0/5] clk: implement remuxing during set_rate

[PATCH v6 0/5] clk: implement remuxing during set_rate

[James Hogan]

This patchset adds support for automatic selection of the best parent
for a clock mux, i.e. the one which can provide the closest clock rate
to that requested. It can be disabled by a new CLK_SET_RATE_NO_REPARENT
flag (which is set for all uses of clk_register_mux().

This works by way of adding a new op, determine_rate, similar to
round_rate but with an extra parameter to allow the clock driver to
optionally select a different parent clock. This is used in
clk_calc_new_rates to decide whether to initiate a set_parent operation.

Changes in v6:

* Rebased on v3.11-rc1.
* Added changes to samsung/clk-exynos-audss.c, zynq/clkc.c.

Changes in v5:

* Rebased on latest clk-next ("clk: tegra: Use common of_clk_init
  function")
* Updated clock hardware characteristics table in documentation.

Changes in v4:

* rebased on clk-next ("clk: sun5i: Add compatibles for Allwinner A13").
* replace __clk_set_parent_no_recalc with __clk_set_parent.
* never pass NULL to determine_rate's best_parent_clk parameter, and
  slight refactor of __clk_round_rate to use local copy of clk->parent.
* a few new comments around use of clk::new_child.
* new patch (patch 2) split out of patch 3 to avoid having to declare
  static __clk_set_parent() at the top of clk.c, and to ease readability
  of patch 3.

Changes in v3:

* rebased on v3.10-rc1.
* remove double underscore prefix from clk_get_parent_by_index()
* store new_parent_index in struct clk too (calculated from
  clk_fetch_parent_index, and passed through __clk_set_parent_no_recalc
  to __clk_set_parent).
* allow determine_rate to satisfy recalc_rate check in __clk_init.
* rename/invert CLK_SET_RATE_REMUX to CLK_SET_RATE_NO_REPARENT and move
  to patch 3.
* patch 3: add CLK_SET_RATE_NO_REPARENT flag to all callers of
  clk_register_mux. If you don't mind your clocks being reparented in
  response to set_rate please let me know and I'll drop the relevant
  portion of the patch.

Changes in v2:

I've moved the mux determine_rate implementation into a core helper, but
I haven't pushed it fully into the core, as I think it just wouldn't
work correctly for more complex clocks, e.g. if you (theoretically) had
a combined mux and divide, you'd want to intercept the determine_rate
and ask for a larger rate from the parent clocks, then return the
divided rate. This should be possible by wrapping the mux determine_rate
helper.

Patch 1 still exports the __clk_get_parent_by_index as it seems like it
might be a useful thing for clock implementations to have access to if
they ever wanted to do something more fancy with changing clock parents.

I haven't made any attempt to implement the atomic set_parent+set_rate
as I don't have hardware that could take proper advantage of it, but it
shouldn't be too difficult for others to implement if they wanted since
they're fairly close to one another (in clk_change_rate()).

* switched to using new determine_rate op rather than adding an argument
  to round_rate.
* moved mux implementation into a single helper which should be usable
  from more complex clocks which can mux.
* rewrite main implementation so that no changes are made until after
  the PRE notifications have been sent, and in a way that should ensure
  correct notifications without duplicates, and I think should be safe
  in the event of a notification failing.
* various tidy ups and fixes.

James Hogan (5):
  clk: abstract parent cache
  clk: move some parent related functions upwards
  clk: add support for clock reparent on set_rate
  clk: add CLK_SET_RATE_NO_REPARENT flag
  clk: clk-mux: implement remuxing on set_rate

 Documentation/clk.txt                  |  46 ++--
 arch/arm/mach-imx/clk.h                |   5 +-
 drivers/clk/clk-mux.c                  |   1 +
 drivers/clk/clk.c                      | 420 +++++++++++++++++++++------------
 drivers/clk/mmp/clk-mmp2.c             |  39 ++-
 drivers/clk/mmp/clk-pxa168.c           |  40 ++--
 drivers/clk/mmp/clk-pxa910.c           |  31 ++-
 drivers/clk/mxs/clk.h                  |   4 +-
 drivers/clk/samsung/clk-exynos-audss.c |   6 +-
 drivers/clk/samsung/clk.h              |   2 +-
 drivers/clk/spear/spear1310_clock.c    | 179 +++++++-------
 drivers/clk/spear/spear1340_clock.c    |  97 ++++----
 drivers/clk/spear/spear3xx_clock.c     |  57 +++--
 drivers/clk/spear/spear6xx_clock.c     |  35 +--
 drivers/clk/sunxi/clk-sunxi.c          |   3 +-
 drivers/clk/tegra/clk-tegra114.c       |  36 ++-
 drivers/clk/tegra/clk-tegra20.c        |   6 +-
 drivers/clk/tegra/clk-tegra30.c        |  33 ++-
 drivers/clk/versatile/clk-vexpress.c   |   4 +-
 drivers/clk/zynq/clkc.c                |  86 ++++---
 include/linux/clk-private.h            |   3 +
 include/linux/clk-provider.h           |  12 +
 22 files changed, 692 insertions(+), 453 deletions(-)

-- 
1.8.1.2

[PATCH v6 1/5] clk: abstract parent cache

[James Hogan]

Abstract access to the clock parent cache by defining
clk_get_parent_by_index(clk, index). This allows access to parent
clocks from clock drivers.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Cc: Mike Turquette <mturquette@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
---
Changes in v3:

* remove double underscore prefix from clk_get_parent_by_index()

 drivers/clk/clk.c            | 21 ++++++++++++++-------
 include/linux/clk-provider.h |  1 +
 2 files changed, 15 insertions(+), 7 deletions(-)

diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
index 54a191c..e06a9ac 100644
--- a/drivers/clk/clk.c
+++ b/drivers/clk/clk.c
@@ -559,6 +559,19 @@ struct clk *__clk_get_parent(struct clk *clk)
 	return !clk ? NULL : clk->parent;
 }
 
+struct clk *clk_get_parent_by_index(struct clk *clk, u8 index)
+{
+	if (!clk || index >= clk->num_parents)
+		return NULL;
+	else if (!clk->parents)
+		return __clk_lookup(clk->parent_names[index]);
+	else if (!clk->parents[index])
+		return clk->parents[index] =
+			__clk_lookup(clk->parent_names[index]);
+	else
+		return clk->parents[index];
+}
+
 unsigned int __clk_get_enable_count(struct clk *clk)
 {
 	return !clk ? 0 : clk->enable_count;
@@ -1315,13 +1328,7 @@ static struct clk *__clk_init_parent(struct clk *clk)
 			kzalloc((sizeof(struct clk*) * clk->num_parents),
 					GFP_KERNEL);
 
-	if (!clk->parents)
-		ret = __clk_lookup(clk->parent_names[index]);
-	else if (!clk->parents[index])
-		ret = clk->parents[index] =
-			__clk_lookup(clk->parent_names[index]);
-	else
-		ret = clk->parents[index];
+	ret = clk_get_parent_by_index(clk, index);
 
 out:
 	return ret;
diff --git a/include/linux/clk-provider.h b/include/linux/clk-provider.h
index 1ec14a7..e54dac5 100644
--- a/include/linux/clk-provider.h
+++ b/include/linux/clk-provider.h
@@ -418,6 +418,7 @@ const char *__clk_get_name(struct clk *clk);
 struct clk_hw *__clk_get_hw(struct clk *clk);
 u8 __clk_get_num_parents(struct clk *clk);
 struct clk *__clk_get_parent(struct clk *clk);
+struct clk *clk_get_parent_by_index(struct clk *clk, u8 index);
 unsigned int __clk_get_enable_count(struct clk *clk);
 unsigned int __clk_get_prepare_count(struct clk *clk);
 unsigned long __clk_get_rate(struct clk *clk);
-- 
1.8.1.2

[PATCH v6 2/5] clk: move some parent related functions upwards

[James Hogan]

Move some parent related functions up in clk.c so they can be used by
the modifications in the following patch which enables clock reparenting
during set_rate. No other changes are made so this patch makes no
functional difference in isolation. This is separate from the following
patch primarily to ease readability of that patch.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Cc: Mike Turquette <mturquette@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
---
Changes in v5:

* Rebased on latest clk-next ("clk: tegra: Use common of_clk_init
  function")

Changes in v4:

* new patch (patch 2) split out of patch 3 to avoid having to declare
  static __clk_set_parent() at the top of clk.c, and to ease readability
  of patch 3.

 drivers/clk/clk.c | 208 +++++++++++++++++++++++++++---------------------------
 1 file changed, 104 insertions(+), 104 deletions(-)

diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
index e06a9ac..b766db1 100644
--- a/drivers/clk/clk.c
+++ b/drivers/clk/clk.c
@@ -1027,6 +1027,110 @@ unsigned long clk_get_rate(struct clk *clk)
 }
 EXPORT_SYMBOL_GPL(clk_get_rate);
 
+static u8 clk_fetch_parent_index(struct clk *clk, struct clk *parent)
+{
+	u8 i;
+
+	if (!clk->parents)
+		clk->parents = kzalloc((sizeof(struct clk*) * clk->num_parents),
+								GFP_KERNEL);
+
+	/*
+	 * find index of new parent clock using cached parent ptrs,
+	 * or if not yet cached, use string name comparison and cache
+	 * them now to avoid future calls to __clk_lookup.
+	 */
+	for (i = 0; i < clk->num_parents; i++) {
+		if (clk->parents && clk->parents[i] == parent)
+			break;
+		else if (!strcmp(clk->parent_names[i], parent->name)) {
+			if (clk->parents)
+				clk->parents[i] = __clk_lookup(parent->name);
+			break;
+		}
+	}
+
+	return i;
+}
+
+static void clk_reparent(struct clk *clk, struct clk *new_parent)
+{
+	hlist_del(&clk->child_node);
+
+	if (new_parent)
+		hlist_add_head(&clk->child_node, &new_parent->children);
+	else
+		hlist_add_head(&clk->child_node, &clk_orphan_list);
+
+	clk->parent = new_parent;
+}
+
+static int __clk_set_parent(struct clk *clk, struct clk *parent, u8 p_index)
+{
+	unsigned long flags;
+	int ret = 0;
+	struct clk *old_parent = clk->parent;
+
+	/*
+	 * Migrate prepare state between parents and prevent race with
+	 * clk_enable().
+	 *
+	 * If the clock is not prepared, then a race with
+	 * clk_enable/disable() is impossible since we already have the
+	 * prepare lock (future calls to clk_enable() need to be preceded by
+	 * a clk_prepare()).
+	 *
+	 * If the clock is prepared, migrate the prepared state to the new
+	 * parent and also protect against a race with clk_enable() by
+	 * forcing the clock and the new parent on.  This ensures that all
+	 * future calls to clk_enable() are practically NOPs with respect to
+	 * hardware and software states.
+	 *
+	 * See also: Comment for clk_set_parent() below.
+	 */
+	if (clk->prepare_count) {
+		__clk_prepare(parent);
+		clk_enable(parent);
+		clk_enable(clk);
+	}
+
+	/* update the clk tree topology */
+	flags = clk_enable_lock();
+	clk_reparent(clk, parent);
+	clk_enable_unlock(flags);
+
+	/* change clock input source */
+	if (parent && clk->ops->set_parent)
+		ret = clk->ops->set_parent(clk->hw, p_index);
+
+	if (ret) {
+		flags = clk_enable_lock();
+		clk_reparent(clk, old_parent);
+		clk_enable_unlock(flags);
+
+		if (clk->prepare_count) {
+			clk_disable(clk);
+			clk_disable(parent);
+			__clk_unprepare(parent);
+		}
+		return ret;
+	}
+
+	/*
+	 * Finish the migration of prepare state and undo the changes done
+	 * for preventing a race with clk_enable().
+	 */
+	if (clk->prepare_count) {
+		clk_disable(clk);
+		clk_disable(old_parent);
+		__clk_unprepare(old_parent);
+	}
+
+	/* update debugfs with new clk tree topology */
+	clk_debug_reparent(clk, parent);
+	return 0;
+}
+
 /**
  * __clk_speculate_rates
  * @clk: first clk in the subtree
@@ -1334,18 +1438,6 @@ out:
 	return ret;
 }
 
-static void clk_reparent(struct clk *clk, struct clk *new_parent)
-{
-	hlist_del(&clk->child_node);
-
-	if (new_parent)
-		hlist_add_head(&clk->child_node, &new_parent->children);
-	else
-		hlist_add_head(&clk->child_node, &clk_orphan_list);
-
-	clk->parent = new_parent;
-}
-
 void __clk_reparent(struct clk *clk, struct clk *new_parent)
 {
 	clk_reparent(clk, new_parent);
@@ -1353,98 +1445,6 @@ void __clk_reparent(struct clk *clk, struct clk *new_parent)
 	__clk_recalc_rates(clk, POST_RATE_CHANGE);
 }
 
-static u8 clk_fetch_parent_index(struct clk *clk, struct clk *parent)
-{
-	u8 i;
-
-	if (!clk->parents)
-		clk->parents = kzalloc((sizeof(struct clk*) * clk->num_parents),
-								GFP_KERNEL);
-
-	/*
-	 * find index of new parent clock using cached parent ptrs,
-	 * or if not yet cached, use string name comparison and cache
-	 * them now to avoid future calls to __clk_lookup.
-	 */
-	for (i = 0; i < clk->num_parents; i++) {
-		if (clk->parents && clk->parents[i] == parent)
-			break;
-		else if (!strcmp(clk->parent_names[i], parent->name)) {
-			if (clk->parents)
-				clk->parents[i] = __clk_lookup(parent->name);
-			break;
-		}
-	}
-
-	return i;
-}
-
-static int __clk_set_parent(struct clk *clk, struct clk *parent, u8 p_index)
-{
-	unsigned long flags;
-	int ret = 0;
-	struct clk *old_parent = clk->parent;
-
-	/*
-	 * Migrate prepare state between parents and prevent race with
-	 * clk_enable().
-	 *
-	 * If the clock is not prepared, then a race with
-	 * clk_enable/disable() is impossible since we already have the
-	 * prepare lock (future calls to clk_enable() need to be preceded by
-	 * a clk_prepare()).
-	 *
-	 * If the clock is prepared, migrate the prepared state to the new
-	 * parent and also protect against a race with clk_enable() by
-	 * forcing the clock and the new parent on.  This ensures that all
-	 * future calls to clk_enable() are practically NOPs with respect to
-	 * hardware and software states.
-	 *
-	 * See also: Comment for clk_set_parent() below.
-	 */
-	if (clk->prepare_count) {
-		__clk_prepare(parent);
-		clk_enable(parent);
-		clk_enable(clk);
-	}
-
-	/* update the clk tree topology */
-	flags = clk_enable_lock();
-	clk_reparent(clk, parent);
-	clk_enable_unlock(flags);
-
-	/* change clock input source */
-	if (parent && clk->ops->set_parent)
-		ret = clk->ops->set_parent(clk->hw, p_index);
-
-	if (ret) {
-		flags = clk_enable_lock();
-		clk_reparent(clk, old_parent);
-		clk_enable_unlock(flags);
-
-		if (clk->prepare_count) {
-			clk_disable(clk);
-			clk_disable(parent);
-			__clk_unprepare(parent);
-		}
-		return ret;
-	}
-
-	/*
-	 * Finish the migration of prepare state and undo the changes done
-	 * for preventing a race with clk_enable().
-	 */
-	if (clk->prepare_count) {
-		clk_disable(clk);
-		clk_disable(old_parent);
-		__clk_unprepare(old_parent);
-	}
-
-	/* update debugfs with new clk tree topology */
-	clk_debug_reparent(clk, parent);
-	return 0;
-}
-
 /**
  * clk_set_parent - switch the parent of a mux clk
  * @clk: the mux clk whose input we are switching
-- 
1.8.1.2

[PATCH v6 3/5] clk: add support for clock reparent on set_rate

[James Hogan]

Add core support to allow clock implementations to select the best
parent clock when rounding a rate, e.g. the one which can provide the
closest clock rate to that requested. This is by way of adding a new
clock op, determine_rate(), which is like round_rate() but has an extra
parameter to allow the clock implementation to optionally select a
different parent clock. The core then takes care of reparenting the
clock when setting the rate.

The parent change takes place with the help of some new private data
members. struct clk::new_parent specifies a clock's new parent (NULL
indicates no change), and struct clk::new_child specifies a clock's new
child (whose new_parent member points back to it). The purpose of these
are to allow correct walking of the future tree for notifications prior
to actually reparenting any clocks, specifically to skip child clocks
who are being reparented to another clock (they will be notified via the
new parent), and to include any new child clock. These pointers are set
by clk_calc_subtree(), and the new_child pointer gets cleared when a
child is actually reparented to avoid duplicate POST_RATE_CHANGE
notifications.

Each place where round_rate() is called, determine_rate() is checked
first and called in preference. This restructures a few of the call
sites to simplify the logic into if/else blocks.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Cc: Mike Turquette <mturquette@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
---
Changes in v5:

* Updated clock hardware characteristics table in documentation.

Changes in v4:

* (not included Stephen Boyd's Reviewed-by due to changes)
* replace __clk_set_parent_no_recalc with __clk_set_parent.
* never pass NULL to determine_rate's best_parent_clk parameter, and
  slight refactor of __clk_round_rate to use local copy of clk->parent.
* a few new comments around use of clk::new_child.

Changes in v3:

* rebased on v3.10-rc1.
* store new_parent_index in struct clk too (calculated from
  clk_fetch_parent_index, and passed through __clk_set_parent_no_recalc
  to __clk_set_parent).
* allow determine_rate to satisfy recalc_rate check in __clk_init.

 Documentation/clk.txt        |  46 ++++++++------
 drivers/clk/clk.c            | 142 +++++++++++++++++++++++++++++--------------
 include/linux/clk-private.h  |   3 +
 include/linux/clk-provider.h |   7 +++
 4 files changed, 132 insertions(+), 66 deletions(-)

diff --git a/Documentation/clk.txt b/Documentation/clk.txt
index 6f68ba0..3aeb5c4 100644
--- a/Documentation/clk.txt
+++ b/Documentation/clk.txt
@@ -70,6 +70,10 @@ the operations defined in clk.h:
 						unsigned long parent_rate);
 		long		(*round_rate)(struct clk_hw *hw, unsigned long,
 						unsigned long *);
+		long		(*determine_rate)(struct clk_hw *hw,
+						unsigned long rate,
+						unsigned long *best_parent_rate,
+						struct clk **best_parent_clk);
 		int		(*set_parent)(struct clk_hw *hw, u8 index);
 		u8		(*get_parent)(struct clk_hw *hw);
 		int		(*set_rate)(struct clk_hw *hw, unsigned long);
@@ -179,26 +183,28 @@ mandatory, a cell marked as "n" implies that either including that
 callback is invalid or otherwise unnecessary.  Empty cells are either
 optional or must be evaluated on a case-by-case basis.
 
-                           clock hardware characteristics
-	     -----------------------------------------------------------
-             | gate | change rate | single parent | multiplexer | root |
-             |------|-------------|---------------|-------------|------|
-.prepare     |      |             |               |             |      |
-.unprepare   |      |             |               |             |      |
-             |      |             |               |             |      |
-.enable      | y    |             |               |             |      |
-.disable     | y    |             |               |             |      |
-.is_enabled  | y    |             |               |             |      |
-             |      |             |               |             |      |
-.recalc_rate |      | y           |               |             |      |
-.round_rate  |      | y           |               |             |      |
-.set_rate    |      | y           |               |             |      |
-             |      |             |               |             |      |
-.set_parent  |      |             | n             | y           | n    |
-.get_parent  |      |             | n             | y           | n    |
-             |      |             |               |             |      |
-.init        |      |             |               |             |      |
-	     -----------------------------------------------------------
+                              clock hardware characteristics
+                -----------------------------------------------------------
+                | gate | change rate | single parent | multiplexer | root |
+                |------|-------------|---------------|-------------|------|
+.prepare        |      |             |               |             |      |
+.unprepare      |      |             |               |             |      |
+                |      |             |               |             |      |
+.enable         | y    |             |               |             |      |
+.disable        | y    |             |               |             |      |
+.is_enabled     | y    |             |               |             |      |
+                |      |             |               |             |      |
+.recalc_rate    |      | y           |               |             |      |
+.round_rate     |      | y [1]       |               |             |      |
+.determine_rate |      | y [1]       |               |             |      |
+.set_rate       |      | y           |               |             |      |
+                |      |             |               |             |      |
+.set_parent     |      |             | n             | y           | n    |
+.get_parent     |      |             | n             | y           | n    |
+                |      |             |               |             |      |
+.init           |      |             |               |             |      |
+                -----------------------------------------------------------
+[1] either one of round_rate or determine_rate is required.
 
 Finally, register your clock at run-time with a hardware-specific
 registration function.  This function simply populates struct clk_foo's
diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
index b766db1..2a047a0 100644
--- a/drivers/clk/clk.c
+++ b/drivers/clk/clk.c
@@ -888,21 +888,24 @@ EXPORT_SYMBOL_GPL(clk_enable);
 unsigned long __clk_round_rate(struct clk *clk, unsigned long rate)
 {
 	unsigned long parent_rate = 0;
+	struct clk *parent;
 
 	if (!clk)
 		return 0;
 
-	if (!clk->ops->round_rate) {
-		if (clk->flags & CLK_SET_RATE_PARENT)
-			return __clk_round_rate(clk->parent, rate);
-		else
-			return clk->rate;
-	}
-
-	if (clk->parent)
-		parent_rate = clk->parent->rate;
-
-	return clk->ops->round_rate(clk->hw, rate, &parent_rate);
+	parent = clk->parent;
+	if (parent)
+		parent_rate = parent->rate;
+
+	if (clk->ops->determine_rate)
+		return clk->ops->determine_rate(clk->hw, rate, &parent_rate,
+						&parent);
+	else if (clk->ops->round_rate)
+		return clk->ops->round_rate(clk->hw, rate, &parent_rate);
+	else if (clk->flags & CLK_SET_RATE_PARENT)
+		return __clk_round_rate(clk->parent, rate);
+	else
+		return clk->rate;
 }
 
 /**
@@ -1055,6 +1058,10 @@ static u8 clk_fetch_parent_index(struct clk *clk, struct clk *parent)
 
 static void clk_reparent(struct clk *clk, struct clk *new_parent)
 {
+	/* avoid duplicate POST_RATE_CHANGE notifications */
+	if (new_parent->new_child == clk)
+		new_parent->new_child = NULL;
+
 	hlist_del(&clk->child_node);
 
 	if (new_parent)
@@ -1175,18 +1182,25 @@ out:
 	return ret;
 }
 
-static void clk_calc_subtree(struct clk *clk, unsigned long new_rate)
+static void clk_calc_subtree(struct clk *clk, unsigned long new_rate,
+			     struct clk *new_parent, u8 p_index)
 {
 	struct clk *child;
 
 	clk->new_rate = new_rate;
+	clk->new_parent = new_parent;
+	clk->new_parent_index = p_index;
+	/* include clk in new parent's PRE_RATE_CHANGE notifications */
+	clk->new_child = NULL;
+	if (new_parent && new_parent != clk->parent)
+		new_parent->new_child = clk;
 
 	hlist_for_each_entry(child, &clk->children, child_node) {
 		if (child->ops->recalc_rate)
 			child->new_rate = child->ops->recalc_rate(child->hw, new_rate);
 		else
 			child->new_rate = new_rate;
-		clk_calc_subtree(child, child->new_rate);
+		clk_calc_subtree(child, child->new_rate, NULL, 0);
 	}
 }
 
@@ -1197,50 +1211,63 @@ static void clk_calc_subtree(struct clk *clk, unsigned long new_rate)
 static struct clk *clk_calc_new_rates(struct clk *clk, unsigned long rate)
 {
 	struct clk *top = clk;
+	struct clk *old_parent, *parent;
 	unsigned long best_parent_rate = 0;
 	unsigned long new_rate;
+	u8 p_index = 0;
 
 	/* sanity */
 	if (IS_ERR_OR_NULL(clk))
 		return NULL;
 
 	/* save parent rate, if it exists */
-	if (clk->parent)
-		best_parent_rate = clk->parent->rate;
-
-	/* never propagate up to the parent */
-	if (!(clk->flags & CLK_SET_RATE_PARENT)) {
-		if (!clk->ops->round_rate) {
-			clk->new_rate = clk->rate;
-			return NULL;
-		}
-		new_rate = clk->ops->round_rate(clk->hw, rate, &best_parent_rate);
+	parent = old_parent = clk->parent;
+	if (parent)
+		best_parent_rate = parent->rate;
+
+	/* find the closest rate and parent clk/rate */
+	if (clk->ops->determine_rate) {
+		new_rate = clk->ops->determine_rate(clk->hw, rate,
+						    &best_parent_rate,
+						    &parent);
+	} else if (clk->ops->round_rate) {
+		new_rate = clk->ops->round_rate(clk->hw, rate,
+						&best_parent_rate);
+	} else if (!parent || !(clk->flags & CLK_SET_RATE_PARENT)) {
+		/* pass-through clock without adjustable parent */
+		clk->new_rate = clk->rate;
+		return NULL;
+	} else {
+		/* pass-through clock with adjustable parent */
+		top = clk_calc_new_rates(parent, rate);
+		new_rate = parent->new_rate;
 		goto out;
 	}
 
-	/* need clk->parent from here on out */
-	if (!clk->parent) {
-		pr_debug("%s: %s has NULL parent\n", __func__, clk->name);
+	/* some clocks must be gated to change parent */
+	if (parent != old_parent &&
+	    (clk->flags & CLK_SET_PARENT_GATE) && clk->prepare_count) {
+		pr_debug("%s: %s not gated but wants to reparent\n",
+			 __func__, clk->name);
 		return NULL;
 	}
 
-	if (!clk->ops->round_rate) {
-		top = clk_calc_new_rates(clk->parent, rate);
-		new_rate = clk->parent->new_rate;
-
-		goto out;
+	/* try finding the new parent index */
+	if (parent) {
+		p_index = clk_fetch_parent_index(clk, parent);
+		if (p_index == clk->num_parents) {
+			pr_debug("%s: clk %s can not be parent of clk %s\n",
+				 __func__, parent->name, clk->name);
+			return NULL;
+		}
 	}
 
-	new_rate = clk->ops->round_rate(clk->hw, rate, &best_parent_rate);
-
-	if (best_parent_rate != clk->parent->rate) {
-		top = clk_calc_new_rates(clk->parent, best_parent_rate);
-
-		goto out;
-	}
+	if ((clk->flags & CLK_SET_RATE_PARENT) && parent &&
+	    best_parent_rate != parent->rate)
+		top = clk_calc_new_rates(parent, best_parent_rate);
 
 out:
-	clk_calc_subtree(clk, new_rate);
+	clk_calc_subtree(clk, new_rate, parent, p_index);
 
 	return top;
 }
@@ -1252,7 +1279,7 @@ out:
  */
 static struct clk *clk_propagate_rate_change(struct clk *clk, unsigned long event)
 {
-	struct clk *child, *fail_clk = NULL;
+	struct clk *child, *tmp_clk, *fail_clk = NULL;
 	int ret = NOTIFY_DONE;
 
 	if (clk->rate == clk->new_rate)
@@ -1265,9 +1292,19 @@ static struct clk *clk_propagate_rate_change(struct clk *clk, unsigned long even
 	}
 
 	hlist_for_each_entry(child, &clk->children, child_node) {
-		clk = clk_propagate_rate_change(child, event);
-		if (clk)
-			fail_clk = clk;
+		/* Skip children who will be reparented to another clock */
+		if (child->new_parent && child->new_parent != clk)
+			continue;
+		tmp_clk = clk_propagate_rate_change(child, event);
+		if (tmp_clk)
+			fail_clk = tmp_clk;
+	}
+
+	/* handle the new child who might not be in clk->children yet */
+	if (clk->new_child) {
+		tmp_clk = clk_propagate_rate_change(clk->new_child, event);
+		if (tmp_clk)
+			fail_clk = tmp_clk;
 	}
 
 	return fail_clk;
@@ -1285,6 +1322,10 @@ static void clk_change_rate(struct clk *clk)
 
 	old_rate = clk->rate;
 
+	/* set parent */
+	if (clk->new_parent && clk->new_parent != clk->parent)
+		__clk_set_parent(clk, clk->new_parent, clk->new_parent_index);
+
 	if (clk->parent)
 		best_parent_rate = clk->parent->rate;
 
@@ -1299,8 +1340,16 @@ static void clk_change_rate(struct clk *clk)
 	if (clk->notifier_count && old_rate != clk->rate)
 		__clk_notify(clk, POST_RATE_CHANGE, old_rate, clk->rate);
 
-	hlist_for_each_entry(child, &clk->children, child_node)
+	hlist_for_each_entry(child, &clk->children, child_node) {
+		/* Skip children who will be reparented to another clock */
+		if (child->new_parent && child->new_parent != clk)
+			continue;
 		clk_change_rate(child);
+	}
+
+	/* handle the new child who might not be in clk->children yet */
+	if (clk->new_child)
+		clk_change_rate(clk->new_child);
 }
 
 /**
@@ -1551,8 +1600,9 @@ int __clk_init(struct device *dev, struct clk *clk)
 
 	/* check that clk_ops are sane.  See Documentation/clk.txt */
 	if (clk->ops->set_rate &&
-			!(clk->ops->round_rate && clk->ops->recalc_rate)) {
-		pr_warning("%s: %s must implement .round_rate & .recalc_rate\n",
+	    !((clk->ops->round_rate || clk->ops->determine_rate) &&
+	      clk->ops->recalc_rate)) {
+		pr_warning("%s: %s must implement .round_rate or .determine_rate in addition to .recalc_rate\n",
 				__func__, clk->name);
 		ret = -EINVAL;
 		goto out;
diff --git a/include/linux/clk-private.h b/include/linux/clk-private.h
index dd7adff..8138c94 100644
--- a/include/linux/clk-private.h
+++ b/include/linux/clk-private.h
@@ -33,8 +33,11 @@ struct clk {
 	const char		**parent_names;
 	struct clk		**parents;
 	u8			num_parents;
+	u8			new_parent_index;
 	unsigned long		rate;
 	unsigned long		new_rate;
+	struct clk		*new_parent;
+	struct clk		*new_child;
 	unsigned long		flags;
 	unsigned int		enable_count;
 	unsigned int		prepare_count;
diff --git a/include/linux/clk-provider.h b/include/linux/clk-provider.h
index e54dac5..d37e518 100644
--- a/include/linux/clk-provider.h
+++ b/include/linux/clk-provider.h
@@ -79,6 +79,10 @@ struct clk_hw;
  * @round_rate:	Given a target rate as input, returns the closest rate actually
  * 		supported by the clock.
  *
+ * @determine_rate: Given a target rate as input, returns the closest rate
+ *		actually supported by the clock, and optionally the parent clock
+ *		that should be used to provide the clock rate.
+ *
  * @get_parent:	Queries the hardware to determine the parent of a clock.  The
  * 		return value is a u8 which specifies the index corresponding to
  * 		the parent clock.  This index can be applied to either the
@@ -126,6 +130,9 @@ struct clk_ops {
 					unsigned long parent_rate);
 	long		(*round_rate)(struct clk_hw *hw, unsigned long,
 					unsigned long *);
+	long		(*determine_rate)(struct clk_hw *hw, unsigned long rate,
+					unsigned long *best_parent_rate,
+					struct clk **best_parent_clk);
 	int		(*set_parent)(struct clk_hw *hw, u8 index);
 	u8		(*get_parent)(struct clk_hw *hw);
 	int		(*set_rate)(struct clk_hw *hw, unsigned long,
-- 
1.8.1.2

[PATCH v6 5/5] clk: clk-mux: implement remuxing on set_rate

[James Hogan]

Implement clk-mux remuxing if the CLK_SET_RATE_NO_REPARENT flag isn't
set. This implements determine_rate for clk-mux to propagate to each
parent and to choose the best one (like clk-divider this chooses the
parent which provides the fastest rate <= the requested rate).

The determine_rate op is implemented as a core helper function so that
it can be easily used by more complex clocks which incorporate muxes.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Cc: Mike Turquette <mturquette@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
---
Changes in v4:

* never pass NULL to determine_rate's best_parent_clk parameter.

Changes in v3:

* rename/invert CLK_SET_RATE_REMUX to CLK_SET_RATE_NO_REPARENT and move
  to patch 3.

 drivers/clk/clk-mux.c        |  1 +
 drivers/clk/clk.c            | 49 ++++++++++++++++++++++++++++++++++++++++++++
 include/linux/clk-provider.h |  3 +++
 3 files changed, 53 insertions(+)

diff --git a/drivers/clk/clk-mux.c b/drivers/clk/clk-mux.c
index 614444c..b487b3c 100644
--- a/drivers/clk/clk-mux.c
+++ b/drivers/clk/clk-mux.c
@@ -104,6 +104,7 @@ static int clk_mux_set_parent(struct clk_hw *hw, u8 index)
 const struct clk_ops clk_mux_ops = {
 	.get_parent = clk_mux_get_parent,
 	.set_parent = clk_mux_set_parent,
+	.determine_rate = __clk_mux_determine_rate,
 };
 EXPORT_SYMBOL_GPL(clk_mux_ops);
 
diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
index 2a047a0..1ed9bdd 100644
--- a/drivers/clk/clk.c
+++ b/drivers/clk/clk.c
@@ -692,6 +692,55 @@ struct clk *__clk_lookup(const char *name)
 	return NULL;
 }
 
+/*
+ * Helper for finding best parent to provide a given frequency. This can be used
+ * directly as a determine_rate callback (e.g. for a mux), or from a more
+ * complex clock that may combine a mux with other operations.
+ */
+long __clk_mux_determine_rate(struct clk_hw *hw, unsigned long rate,
+			      unsigned long *best_parent_rate,
+			      struct clk **best_parent_p)
+{
+	struct clk *clk = hw->clk, *parent, *best_parent = NULL;
+	int i, num_parents;
+	unsigned long parent_rate, best = 0;
+
+	/* if NO_REPARENT flag set, pass through to current parent */
+	if (clk->flags & CLK_SET_RATE_NO_REPARENT) {
+		parent = clk->parent;
+		if (clk->flags & CLK_SET_RATE_PARENT)
+			best = __clk_round_rate(parent, rate);
+		else if (parent)
+			best = __clk_get_rate(parent);
+		else
+			best = __clk_get_rate(clk);
+		goto out;
+	}
+
+	/* find the parent that can provide the fastest rate <= rate */
+	num_parents = clk->num_parents;
+	for (i = 0; i < num_parents; i++) {
+		parent = clk_get_parent_by_index(clk, i);
+		if (!parent)
+			continue;
+		if (clk->flags & CLK_SET_RATE_PARENT)
+			parent_rate = __clk_round_rate(parent, rate);
+		else
+			parent_rate = __clk_get_rate(parent);
+		if (parent_rate <= rate && parent_rate > best) {
+			best_parent = parent;
+			best = parent_rate;
+		}
+	}
+
+out:
+	if (best_parent)
+		*best_parent_p = best_parent;
+	*best_parent_rate = best;
+
+	return best;
+}
+
 /***        clk api        ***/
 
 void __clk_unprepare(struct clk *clk)
diff --git a/include/linux/clk-provider.h b/include/linux/clk-provider.h
index 8ffe06c..9861cee 100644
--- a/include/linux/clk-provider.h
+++ b/include/linux/clk-provider.h
@@ -434,6 +434,9 @@ unsigned long __clk_get_flags(struct clk *clk);
 bool __clk_is_prepared(struct clk *clk);
 bool __clk_is_enabled(struct clk *clk);
 struct clk *__clk_lookup(const char *name);
+long __clk_mux_determine_rate(struct clk_hw *hw, unsigned long rate,
+			      unsigned long *best_parent_rate,
+			      struct clk **best_parent_p);
 
 /*
  * FIXME clock api without lock protection
-- 
1.8.1.2

Re: [PATCH v6 0/5] clk: implement remuxing during set_rate

[Mike Turquette]

Quoting James Hogan (2013-07-29 04:24:57)
> This patchset adds support for automatic selection of the best parent
> for a clock mux, i.e. the one which can provide the closest clock rate
> to that requested. It can be disabled by a new CLK_SET_RATE_NO_REPARENT
> flag (which is set for all uses of clk_register_mux().
> 
> This works by way of adding a new op, determine_rate, similar to
> round_rate but with an extra parameter to allow the clock driver to
> optionally select a different parent clock. This is used in
> clk_calc_new_rates to decide whether to initiate a set_parent operation.

Hi James,

I'm performing some testing on this series now. If I don't spot any
regressions than I'll take it into clk-next. Thanks a lot for the rework
you've done.

I think that work on re-muxing is not yet complete. There are still
unresolved issues about picking "best parents", and likely some other
loose ends. We can fix those up later.

Regards,
Mike

> 
> Changes in v6:
> 
> * Rebased on v3.11-rc1.
> * Added changes to samsung/clk-exynos-audss.c, zynq/clkc.c.
> 
> Changes in v5:
> 
> * Rebased on latest clk-next ("clk: tegra: Use common of_clk_init
>   function")
> * Updated clock hardware characteristics table in documentation.
> 
> Changes in v4:
> 
> * rebased on clk-next ("clk: sun5i: Add compatibles for Allwinner A13").
> * replace __clk_set_parent_no_recalc with __clk_set_parent.
> * never pass NULL to determine_rate's best_parent_clk parameter, and
>   slight refactor of __clk_round_rate to use local copy of clk->parent.
> * a few new comments around use of clk::new_child.
> * new patch (patch 2) split out of patch 3 to avoid having to declare
>   static __clk_set_parent() at the top of clk.c, and to ease readability
>   of patch 3.
> 
> Changes in v3:
> 
> * rebased on v3.10-rc1.
> * remove double underscore prefix from clk_get_parent_by_index()
> * store new_parent_index in struct clk too (calculated from
>   clk_fetch_parent_index, and passed through __clk_set_parent_no_recalc
>   to __clk_set_parent).
> * allow determine_rate to satisfy recalc_rate check in __clk_init.
> * rename/invert CLK_SET_RATE_REMUX to CLK_SET_RATE_NO_REPARENT and move
>   to patch 3.
> * patch 3: add CLK_SET_RATE_NO_REPARENT flag to all callers of
>   clk_register_mux. If you don't mind your clocks being reparented in
>   response to set_rate please let me know and I'll drop the relevant
>   portion of the patch.
> 
> Changes in v2:
> 
> I've moved the mux determine_rate implementation into a core helper, but
> I haven't pushed it fully into the core, as I think it just wouldn't
> work correctly for more complex clocks, e.g. if you (theoretically) had
> a combined mux and divide, you'd want to intercept the determine_rate
> and ask for a larger rate from the parent clocks, then return the
> divided rate. This should be possible by wrapping the mux determine_rate
> helper.
> 
> Patch 1 still exports the __clk_get_parent_by_index as it seems like it
> might be a useful thing for clock implementations to have access to if
> they ever wanted to do something more fancy with changing clock parents.
> 
> I haven't made any attempt to implement the atomic set_parent+set_rate
> as I don't have hardware that could take proper advantage of it, but it
> shouldn't be too difficult for others to implement if they wanted since
> they're fairly close to one another (in clk_change_rate()).
> 
> * switched to using new determine_rate op rather than adding an argument
>   to round_rate.
> * moved mux implementation into a single helper which should be usable
>   from more complex clocks which can mux.
> * rewrite main implementation so that no changes are made until after
>   the PRE notifications have been sent, and in a way that should ensure
>   correct notifications without duplicates, and I think should be safe
>   in the event of a notification failing.
> * various tidy ups and fixes.
> 
> James Hogan (5):
>   clk: abstract parent cache
>   clk: move some parent related functions upwards
>   clk: add support for clock reparent on set_rate
>   clk: add CLK_SET_RATE_NO_REPARENT flag
>   clk: clk-mux: implement remuxing on set_rate
> 
>  Documentation/clk.txt                  |  46 ++--
>  arch/arm/mach-imx/clk.h                |   5 +-
>  drivers/clk/clk-mux.c                  |   1 +
>  drivers/clk/clk.c                      | 420 +++++++++++++++++++++------------
>  drivers/clk/mmp/clk-mmp2.c             |  39 ++-
>  drivers/clk/mmp/clk-pxa168.c           |  40 ++--
>  drivers/clk/mmp/clk-pxa910.c           |  31 ++-
>  drivers/clk/mxs/clk.h                  |   4 +-
>  drivers/clk/samsung/clk-exynos-audss.c |   6 +-
>  drivers/clk/samsung/clk.h              |   2 +-
>  drivers/clk/spear/spear1310_clock.c    | 179 +++++++-------
>  drivers/clk/spear/spear1340_clock.c    |  97 ++++----
>  drivers/clk/spear/spear3xx_clock.c     |  57 +++--
>  drivers/clk/spear/spear6xx_clock.c     |  35 +--
>  drivers/clk/sunxi/clk-sunxi.c          |   3 +-
>  drivers/clk/tegra/clk-tegra114.c       |  36 ++-
>  drivers/clk/tegra/clk-tegra20.c        |   6 +-
>  drivers/clk/tegra/clk-tegra30.c        |  33 ++-
>  drivers/clk/versatile/clk-vexpress.c   |   4 +-
>  drivers/clk/zynq/clkc.c                |  86 ++++---
>  include/linux/clk-private.h            |   3 +
>  include/linux/clk-provider.h           |  12 +
>  22 files changed, 692 insertions(+), 453 deletions(-)
> 
> -- 
> 1.8.1.2