Skip to contentSkip to conversation

Per-user clock constraints

PATCH In review Tomeu Vizoso 13 participants 70 messages Jan 23, 2015 – Mar 13, 2015 1 message not in the archive

Subsystems (22)
lore
Share & tools
lore
#0 lore raw

v13 cover Per-user clock constraints

Hi,

in this v13 I have:

* added the latest R-b tags from Stephen,

* split some lines that went over the 80-char limit,

* made explicit in the docs that the range limits are inclusive, and

* removed an ifdef in clkdev.c by adding a dev_id param to __of_clk_get_by_name.

The first patch just removes some unneeded NULL checks.

The second removes __clk_register.

The third patch actually moves the per-clock data that was stored in struct clk to a new struct clk_core and adds references to it from both struct clk and struct clk_hw. struct clk is now ready to contain information that is specific to a given clk consumer.

The fourth patch adds API for setting floor and ceiling constraints and stores that information on the per-user struct clk, which is iterable from struct clk_core. The constraints are made available to clock implementations in the determine_rate callback.

The fifth patch exports clk_register_clkdev so that clk-test.ko can be built as a module.

The sixth patch adds a module that can be used to unit-test the CCF. It's very rough right now, but I hope to have time soon to polish it a bit. It has been quite useful during development of this series though.

http://cgit.collabora.com/git/user/tomeu/linux.git/log/?h=per-user-clk-constraints-v13

Thanks,

Tomeu

Tomeu Vizoso (6):
clk: Remove unneeded NULL checks
clk: Remove __clk_register
clk: Make clk API return per-user struct clk instances
clk: Add rate constraints to clocks
clkdev: Export clk_register_clkdev
clk: Add module for unit tests

 Documentation/clk.txt                   |   2 +
 arch/arm/mach-omap2/cclock3xxx_data.c   | 111 +++--
 arch/arm/mach-omap2/clock.h             |  11 +-
 arch/arm/mach-omap2/clock_common_data.c |   5 +-
 arch/arm/mach-omap2/dpll3xxx.c          |   2 +
 arch/arm/mach-omap2/dpll44xx.c          |   2 +
 arch/mips/alchemy/common/clock.c        |   8 +
 drivers/clk/Kconfig                     |   1 +
 drivers/clk/Kconfig.debug               |   6 +
 drivers/clk/Makefile                    |   1 +
 drivers/clk/at91/clk-programmable.c     |   2 +
 drivers/clk/bcm/clk-kona.c              |   2 +
 drivers/clk/clk-composite.c             |   9 +-
 drivers/clk/clk-test.c                  | 326 ++++++++++++
 drivers/clk/clk.c                       | 851 +++++++++++++++++++++-----------
 drivers/clk/clk.h                       |   5 +
 drivers/clk/clkdev.c                    |  85 +++-
 drivers/clk/hisilicon/clk-hi3620.c      |   2 +
 drivers/clk/mmp/clk-mix.c               |   2 +
 drivers/clk/qcom/clk-pll.c              |   1 +
 drivers/clk/qcom/clk-rcg.c              |  10 +-
 drivers/clk/qcom/clk-rcg2.c             |   6 +
 drivers/clk/sunxi/clk-factors.c         |   2 +
 drivers/clk/sunxi/clk-sun6i-ar100.c     |   2 +
 include/linux/clk-private.h             |  43 +-
 include/linux/clk-provider.h            |  27 +-
 include/linux/clk.h                     |  28 ++
 include/linux/clk/ti.h                  |   4 +
 28 files changed, 1187 insertions(+), 369 deletions(-)
 create mode 100644 drivers/clk/Kconfig.debug
 create mode 100644 drivers/clk/clk-test.c
Signature · 1 line
1.9.3
#1 lore raw

PATCH v13 1/6 clk: Remove unneeded NULL checks

As clk_unprepare_unused_subtree and clk_disable_unused_subtree are always called with a valid struct clk.

---
 drivers/clk/clk.c | 9 ---------
 1 file changed, 9 deletions(-)
drivers/clk/clk.c +0 -9
index f4963b7..97f3425 100644
@@ -432,9 +432,6 @@ static void clk_unprepare_unused_subtree(struct clk *clk)
 {
 	struct clk *child;
 
-	if (!clk)
-		return;
-
 	hlist_for_each_entry(child, &clk->children, child_node)
 		clk_unprepare_unused_subtree(child);
 
@@ -458,9 +455,6 @@ static void clk_disable_unused_subtree(struct clk *clk)
 	struct clk *child;
 	unsigned long flags;
 
-	if (!clk)
-		goto out;
-
 	hlist_for_each_entry(child, &clk->children, child_node)
 		clk_disable_unused_subtree(child);
 
@@ -486,9 +480,6 @@ static void clk_disable_unused_subtree(struct clk *clk)
 
 unlock_out:
 	clk_enable_unlock(flags);
-
-out:
-	return;
 }
 
 static bool clk_ignore_unused;
Signature · 1 line
1.9.3
#2 lore raw

PATCH v13 2/6 clk: Remove __clk_register

In reply to Tomeu Vizoso

As it has never been used.

---
 drivers/clk/clk.c           | 42 ------------------------------------------
 include/linux/clk-private.h |  2 --
 2 files changed, 44 deletions(-)
drivers/clk/clk.c +0 -42
index 97f3425..fd3d671 100644
@@ -1975,48 +1975,6 @@ out:
 }
 
 /**
- * __clk_register - register a clock and return a cookie.
- *
- * Same as clk_register, except that the .clk field inside hw shall point to a
- * preallocated (generally statically allocated) struct clk. None of the fields
- * of the struct clk need to be initialized.
- *
- * The data pointed to by .init and .clk field shall NOT be marked as init
- * data.
- *
- * __clk_register is only exposed via clk-private.h and is intended for use with
- * very large numbers of clocks that need to be statically initialized.  It is
- * a layering violation to include clk-private.h from any code which implements
- * a clock's .ops; as such any statically initialized clock data MUST be in a
- * separate C file from the logic that implements its operations.  Returns 0
- * on success, otherwise an error code.
- */
-struct clk *__clk_register(struct device *dev, struct clk_hw *hw)
-{
-	int ret;
-	struct clk *clk;
-
-	clk = hw->clk;
-	clk->name = hw->init->name;
-	clk->ops = hw->init->ops;
-	clk->hw = hw;
-	clk->flags = hw->init->flags;
-	clk->parent_names = hw->init->parent_names;
-	clk->num_parents = hw->init->num_parents;
-	if (dev && dev->driver)
-		clk->owner = dev->driver->owner;
-	else
-		clk->owner = NULL;
-
-	ret = __clk_init(dev, clk);
-	if (ret)
-		return ERR_PTR(ret);
-
-	return clk;
-}
-EXPORT_SYMBOL_GPL(__clk_register);
-
-/**
  * clk_register - allocate a new clock, register it and return an opaque cookie
  * @dev: device that is registering this clock
  * @hw: link to hardware-specific clock data
include/linux/clk-private.h +0 -2
index 0ca5f60..c5f40d0 100644
@@ -214,7 +214,5 @@ struct clk {
  */
 int __clk_init(struct device *dev, struct clk *clk);
 
-struct clk *__clk_register(struct device *dev, struct clk_hw *hw);
-
 #endif /* CONFIG_COMMON_CLK */
 #endif /* CLK_PRIVATE_H */
Signature · 1 line
1.9.3
#3 lore raw

PATCH v13 3/6 clk: Make clk API return per-user struct clk instances

In reply to Tomeu Vizoso

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

---
v13: * Split some lines to 80 columns
* Add dev_id parameter to __of_clk_get_by_name so it can be called by clk_get without losing information

v12: * Move __clk_create_clk so it's later next to __clk_free_clk
* NULLify a just-free'd pointer
* Check the value returned by __clk_create_clk in clk_register

v11: * Allow for clk_set_parent to be called with a NULL parent

v10: * Add more missing NULL checks
* Remove __clk_reparent as it's now unused
* Remove clk_core_round_rate as it's now unused
* Call nolock variants from __clk_mux_determine_rate

v9: * Add missing NULL checks
* Remove __clk_prepare and __clk_unprepare as they are unused now

v7: * Add stub for __of_clk_get_by_name to fix builds without OF

v6: * Guard against NULL pointer

v4: * Remove unused function __clk_core_to_clk
* Use "core" more often as the name for struct clk_core* variables
* Make sure we don't lose information about the caller in of_clk_get_*

v3: * Rebase on top of linux-next 20141009

v2: * Remove exported functions that aren't really used outside clk.c
* Rename new internal functions to clk_core_ prefix
* Remove redundant checks for error pointers in *_get_parent
* Change __clk_create_clk to take a struct clk_hw instead
* Match the original error behavior in clk_get_sys

---
 arch/arm/mach-omap2/cclock3xxx_data.c   | 111 ++++--
 arch/arm/mach-omap2/clock.h             |  11 +-
 arch/arm/mach-omap2/clock_common_data.c |   5 +-
 drivers/clk/clk.c                       | 621 ++++++++++++++++++++------------
 drivers/clk/clk.h                       |   5 +
 drivers/clk/clkdev.c                    |  84 ++++-
 include/linux/clk-private.h             |  35 +-
 include/linux/clk-provider.h            |  12 +-
 8 files changed, 589 insertions(+), 295 deletions(-)
arch/arm/mach-omap2/cclock3xxx_data.c +73 -38
index 644ff32..adb4e64 100644
@@ -82,7 +82,7 @@ DEFINE_CLK_MUX(osc_sys_ck, osc_sys_ck_parent_names, NULL, 0x0,
 	       OMAP3430_PRM_CLKSEL, OMAP3430_SYS_CLKIN_SEL_SHIFT,
 	       OMAP3430_SYS_CLKIN_SEL_WIDTH, 0x0, NULL);
 
-DEFINE_CLK_DIVIDER(sys_ck, "osc_sys_ck", &osc_sys_ck, 0x0,
+DEFINE_CLK_DIVIDER(sys_ck, "osc_sys_ck", &osc_sys_ck_core, 0x0,
 		   OMAP3430_PRM_CLKSRC_CTRL, OMAP_SYSCLKDIV_SHIFT,
 		   OMAP_SYSCLKDIV_WIDTH, CLK_DIVIDER_ONE_BASED, NULL);
 
@@ -132,7 +132,7 @@ static struct clk_hw_omap dpll3_ck_hw = {
 
 DEFINE_STRUCT_CLK(dpll3_ck, dpll3_ck_parent_names, dpll3_ck_ops);
 
-DEFINE_CLK_DIVIDER(dpll3_m2_ck, "dpll3_ck", &dpll3_ck, 0x0,
+DEFINE_CLK_DIVIDER(dpll3_m2_ck, "dpll3_ck", &dpll3_ck_core, 0x0,
 		   OMAP_CM_REGADDR(PLL_MOD, CM_CLKSEL1),
 		   OMAP3430_CORE_DPLL_CLKOUT_DIV_SHIFT,
 		   OMAP3430_CORE_DPLL_CLKOUT_DIV_WIDTH,
@@ -149,12 +149,12 @@ static const struct clk_ops core_ck_ops = {};
 DEFINE_STRUCT_CLK_HW_OMAP(core_ck, NULL);
 DEFINE_STRUCT_CLK(core_ck, core_ck_parent_names, core_ck_ops);
 
-DEFINE_CLK_DIVIDER(l3_ick, "core_ck", &core_ck, 0x0,
+DEFINE_CLK_DIVIDER(l3_ick, "core_ck", &core_ck_core, 0x0,
 		   OMAP_CM_REGADDR(CORE_MOD, CM_CLKSEL),
 		   OMAP3430_CLKSEL_L3_SHIFT, OMAP3430_CLKSEL_L3_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
 
-DEFINE_CLK_DIVIDER(l4_ick, "l3_ick", &l3_ick, 0x0,
+DEFINE_CLK_DIVIDER(l4_ick, "l3_ick", &l3_ick_core, 0x0,
 		   OMAP_CM_REGADDR(CORE_MOD, CM_CLKSEL),
 		   OMAP3430_CLKSEL_L4_SHIFT, OMAP3430_CLKSEL_L4_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -275,9 +275,9 @@ static struct clk_hw_omap dpll1_ck_hw = {
 
 DEFINE_STRUCT_CLK(dpll1_ck, dpll3_ck_parent_names, dpll1_ck_ops);
 
-DEFINE_CLK_FIXED_FACTOR(dpll1_x2_ck, "dpll1_ck", &dpll1_ck, 0x0, 2, 1);
+DEFINE_CLK_FIXED_FACTOR(dpll1_x2_ck, "dpll1_ck", &dpll1_ck_core, 0x0, 2, 1);
 
-DEFINE_CLK_DIVIDER(dpll1_x2m2_ck, "dpll1_x2_ck", &dpll1_x2_ck, 0x0,
+DEFINE_CLK_DIVIDER(dpll1_x2m2_ck, "dpll1_x2_ck", &dpll1_x2_ck_core, 0x0,
 		   OMAP_CM_REGADDR(MPU_MOD, OMAP3430_CM_CLKSEL2_PLL),
 		   OMAP3430_MPU_DPLL_CLKOUT_DIV_SHIFT,
 		   OMAP3430_MPU_DPLL_CLKOUT_DIV_WIDTH,
@@ -292,7 +292,7 @@ static const char *mpu_ck_parent_names[] = {
 DEFINE_STRUCT_CLK_HW_OMAP(mpu_ck, "mpu_clkdm");
 DEFINE_STRUCT_CLK(mpu_ck, mpu_ck_parent_names, core_l4_ick_ops);
 
-DEFINE_CLK_DIVIDER(arm_fck, "mpu_ck", &mpu_ck, 0x0,
+DEFINE_CLK_DIVIDER(arm_fck, "mpu_ck", &mpu_ck_core, 0x0,
 		   OMAP_CM_REGADDR(MPU_MOD, OMAP3430_CM_IDLEST_PLL),
 		   OMAP3430_ST_MPU_CLK_SHIFT, OMAP3430_ST_MPU_CLK_WIDTH,
 		   0x0, NULL);
@@ -424,7 +424,7 @@ static const struct clk_div_table dpll4_mx_ck_div_table[] = {
 	{ .div = 0 },
 };
 
-DEFINE_CLK_DIVIDER(dpll4_m5_ck, "dpll4_ck", &dpll4_ck, 0x0,
+DEFINE_CLK_DIVIDER(dpll4_m5_ck, "dpll4_ck", &dpll4_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_CAM_MOD, CM_CLKSEL),
 		   OMAP3430_CLKSEL_CAM_SHIFT, OMAP3630_CLKSEL_CAM_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -466,7 +466,7 @@ static struct clk_hw_omap dpll4_m5x2_ck_hw = {
 DEFINE_STRUCT_CLK_FLAGS(dpll4_m5x2_ck, dpll4_m5x2_ck_parent_names,
 			dpll4_m5x2_ck_ops, CLK_SET_RATE_PARENT);
 
-static struct clk dpll4_m5x2_ck_3630 = {
+static struct clk_core dpll4_m5x2_ck_3630_core = {
 	.name		= "dpll4_m5x2_ck",
 	.hw		= &dpll4_m5x2_ck_hw.hw,
 	.parent_names	= dpll4_m5x2_ck_parent_names,
@@ -475,6 +475,10 @@ static struct clk dpll4_m5x2_ck_3630 = {
 	.flags		= CLK_SET_RATE_PARENT,
 };
 
+static struct clk dpll4_m5x2_ck_3630 = {
+	.core = &dpll4_m5x2_ck_3630_core,
+};
+
 static struct clk cam_mclk;
 
 static const char *cam_mclk_parent_names[] = {
@@ -490,7 +494,7 @@ static struct clk_hw_omap cam_mclk_hw = {
 	.clkdm_name	= "cam_clkdm",
 };
 
-static struct clk cam_mclk = {
+static struct clk_core cam_mclk_core = {
 	.name		= "cam_mclk",
 	.hw		= &cam_mclk_hw.hw,
 	.parent_names	= cam_mclk_parent_names,
@@ -499,6 +503,10 @@ static struct clk cam_mclk = {
 	.flags		= CLK_SET_RATE_PARENT,
 };
 
+static struct clk cam_mclk = {
+	.core = &cam_mclk_core,
+};
+
 static const struct clksel_rate clkout2_src_core_rates[] = {
 	{ .div = 1, .val = 0, .flags = RATE_IN_3XXX },
 	{ .div = 0 }
@@ -514,7 +522,7 @@ static const struct clksel_rate clkout2_src_96m_rates[] = {
 	{ .div = 0 }
 };
 
-DEFINE_CLK_DIVIDER(dpll4_m2_ck, "dpll4_ck", &dpll4_ck, 0x0,
+DEFINE_CLK_DIVIDER(dpll4_m2_ck, "dpll4_ck", &dpll4_ck_core, 0x0,
 		   OMAP_CM_REGADDR(PLL_MOD, OMAP3430_CM_CLKSEL3),
 		   OMAP3430_DIV_96M_SHIFT, OMAP3630_DIV_96M_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -538,7 +546,7 @@ static struct clk_hw_omap dpll4_m2x2_ck_hw = {
 
 DEFINE_STRUCT_CLK(dpll4_m2x2_ck, dpll4_m2x2_ck_parent_names, dpll4_m5x2_ck_ops);
 
-static struct clk dpll4_m2x2_ck_3630 = {
+static struct clk_core dpll4_m2x2_ck_3630_core = {
 	.name		= "dpll4_m2x2_ck",
 	.hw		= &dpll4_m2x2_ck_hw.hw,
 	.parent_names	= dpll4_m2x2_ck_parent_names,
@@ -546,6 +554,10 @@ static struct clk dpll4_m2x2_ck_3630 = {
 	.ops		= &dpll4_m5x2_ck_3630_ops,
 };
 
+static struct clk dpll4_m2x2_ck_3630 = {
+	.core = &dpll4_m2x2_ck_3630_core,
+};
+
 static struct clk omap_96m_alwon_fck;
 
 static const char *omap_96m_alwon_fck_parent_names[] = {
@@ -570,7 +582,7 @@ static const struct clksel_rate clkout2_src_54m_rates[] = {
 	{ .div = 0 }
 };
 
-DEFINE_CLK_DIVIDER_TABLE(dpll4_m3_ck, "dpll4_ck", &dpll4_ck, 0x0,
+DEFINE_CLK_DIVIDER_TABLE(dpll4_m3_ck, "dpll4_ck", &dpll4_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_DSS_MOD, CM_CLKSEL),
 		   OMAP3430_CLKSEL_TV_SHIFT, OMAP3630_CLKSEL_TV_WIDTH,
 		   0, dpll4_mx_ck_div_table, NULL);
@@ -594,7 +606,7 @@ static struct clk_hw_omap dpll4_m3x2_ck_hw = {
 
 DEFINE_STRUCT_CLK(dpll4_m3x2_ck, dpll4_m3x2_ck_parent_names, dpll4_m5x2_ck_ops);
 
-static struct clk dpll4_m3x2_ck_3630 = {
+static struct clk_core dpll4_m3x2_ck_3630_core = {
 	.name		= "dpll4_m3x2_ck",
 	.hw		= &dpll4_m3x2_ck_hw.hw,
 	.parent_names	= dpll4_m3x2_ck_parent_names,
@@ -602,6 +614,10 @@ static struct clk dpll4_m3x2_ck_3630 = {
 	.ops		= &dpll4_m5x2_ck_3630_ops,
 };
 
+static struct clk dpll4_m3x2_ck_3630 = {
+	.core = &dpll4_m3x2_ck_3630_core,
+};
+
 static const char *omap_54m_fck_parent_names[] = {
 	"dpll4_m3x2_ck", "sys_altclk",
 };
@@ -677,7 +693,8 @@ static struct clk_hw_omap omap_48m_fck_hw = {
 
 DEFINE_STRUCT_CLK(omap_48m_fck, omap_48m_fck_parent_names, omap_48m_fck_ops);
 
-DEFINE_CLK_FIXED_FACTOR(omap_12m_fck, "omap_48m_fck", &omap_48m_fck, 0x0, 1, 4);
+DEFINE_CLK_FIXED_FACTOR(omap_12m_fck, "omap_48m_fck", &omap_48m_fck_core, 0x0,
+			1, 4);
 
 static struct clk core_12m_fck;
 
@@ -723,7 +740,8 @@ static const char *core_l3_ick_parent_names[] = {
 DEFINE_STRUCT_CLK_HW_OMAP(core_l3_ick, "core_l3_clkdm");
 DEFINE_STRUCT_CLK(core_l3_ick, core_l3_ick_parent_names, core_l4_ick_ops);
 
-DEFINE_CLK_FIXED_FACTOR(dpll3_m2x2_ck, "dpll3_m2_ck", &dpll3_m2_ck, 0x0, 2, 1);
+DEFINE_CLK_FIXED_FACTOR(dpll3_m2x2_ck, "dpll3_m2_ck", &dpll3_m2_ck_core, 0x0,
+			2, 1);
 
 static struct clk corex2_fck;
 
@@ -809,7 +827,7 @@ static struct clk_hw_omap des2_ick_hw = {
 
 DEFINE_STRUCT_CLK(des2_ick, aes2_ick_parent_names, aes2_ick_ops);
 
-DEFINE_CLK_DIVIDER(dpll1_fck, "core_ck", &core_ck, 0x0,
+DEFINE_CLK_DIVIDER(dpll1_fck, "core_ck", &core_ck_core, 0x0,
 		   OMAP_CM_REGADDR(MPU_MOD, OMAP3430_CM_CLKSEL1_PLL),
 		   OMAP3430_MPU_CLK_SRC_SHIFT, OMAP3430_MPU_CLK_SRC_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -852,18 +870,18 @@ static struct clk_hw_omap dpll2_ck_hw = {
 
 DEFINE_STRUCT_CLK(dpll2_ck, dpll3_ck_parent_names, dpll1_ck_ops);
 
-DEFINE_CLK_DIVIDER(dpll2_fck, "core_ck", &core_ck, 0x0,
+DEFINE_CLK_DIVIDER(dpll2_fck, "core_ck", &core_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_IVA2_MOD, OMAP3430_CM_CLKSEL1_PLL),
 		   OMAP3430_IVA2_CLK_SRC_SHIFT, OMAP3430_IVA2_CLK_SRC_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
 
-DEFINE_CLK_DIVIDER(dpll2_m2_ck, "dpll2_ck", &dpll2_ck, 0x0,
+DEFINE_CLK_DIVIDER(dpll2_m2_ck, "dpll2_ck", &dpll2_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_IVA2_MOD, OMAP3430_CM_CLKSEL2_PLL),
 		   OMAP3430_IVA2_DPLL_CLKOUT_DIV_SHIFT,
 		   OMAP3430_IVA2_DPLL_CLKOUT_DIV_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
 
-DEFINE_CLK_DIVIDER(dpll3_m3_ck, "dpll3_ck", &dpll3_ck, 0x0,
+DEFINE_CLK_DIVIDER(dpll3_m3_ck, "dpll3_ck", &dpll3_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_EMU_MOD, CM_CLKSEL1),
 		   OMAP3430_DIV_DPLL3_SHIFT, OMAP3430_DIV_DPLL3_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -887,7 +905,7 @@ static struct clk_hw_omap dpll3_m3x2_ck_hw = {
 
 DEFINE_STRUCT_CLK(dpll3_m3x2_ck, dpll3_m3x2_ck_parent_names, dpll4_m5x2_ck_ops);
 
-static struct clk dpll3_m3x2_ck_3630 = {
+static struct clk_core dpll3_m3x2_ck_3630_core = {
 	.name		= "dpll3_m3x2_ck",
 	.hw		= &dpll3_m3x2_ck_hw.hw,
 	.parent_names	= dpll3_m3x2_ck_parent_names,
@@ -895,9 +913,13 @@ static struct clk dpll3_m3x2_ck_3630 = {
 	.ops		= &dpll4_m5x2_ck_3630_ops,
 };
 
-DEFINE_CLK_FIXED_FACTOR(dpll3_x2_ck, "dpll3_ck", &dpll3_ck, 0x0, 2, 1);
+static struct clk dpll3_m3x2_ck_3630 = {
+	.core = &dpll3_m3x2_ck_3630_core,
+};
+
+DEFINE_CLK_FIXED_FACTOR(dpll3_x2_ck, "dpll3_ck", &dpll3_ck_core, 0x0, 2, 1);
 
-DEFINE_CLK_DIVIDER_TABLE(dpll4_m4_ck, "dpll4_ck", &dpll4_ck, 0x0,
+DEFINE_CLK_DIVIDER_TABLE(dpll4_m4_ck, "dpll4_ck", &dpll4_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_DSS_MOD, CM_CLKSEL),
 		   OMAP3430_CLKSEL_DSS1_SHIFT, OMAP3630_CLKSEL_DSS1_WIDTH,
 		   0, dpll4_mx_ck_div_table, NULL);
@@ -922,7 +944,7 @@ static struct clk_hw_omap dpll4_m4x2_ck_hw = {
 DEFINE_STRUCT_CLK_FLAGS(dpll4_m4x2_ck, dpll4_m4x2_ck_parent_names,
 		dpll4_m5x2_ck_ops, CLK_SET_RATE_PARENT);
 
-static struct clk dpll4_m4x2_ck_3630 = {
+static struct clk_core dpll4_m4x2_ck_3630_core = {
 	.name		= "dpll4_m4x2_ck",
 	.hw		= &dpll4_m4x2_ck_hw.hw,
 	.parent_names	= dpll4_m4x2_ck_parent_names,
@@ -931,7 +953,11 @@ static struct clk dpll4_m4x2_ck_3630 = {
 	.flags		= CLK_SET_RATE_PARENT,
 };
 
-DEFINE_CLK_DIVIDER(dpll4_m6_ck, "dpll4_ck", &dpll4_ck, 0x0,
+static struct clk dpll4_m4x2_ck_3630 = {
+	.core = &dpll4_m4x2_ck_3630_core,
+};
+
+DEFINE_CLK_DIVIDER(dpll4_m6_ck, "dpll4_ck", &dpll4_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_EMU_MOD, CM_CLKSEL1),
 		   OMAP3430_DIV_DPLL4_SHIFT, OMAP3630_DIV_DPLL4_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -955,7 +981,7 @@ static struct clk_hw_omap dpll4_m6x2_ck_hw = {
 
 DEFINE_STRUCT_CLK(dpll4_m6x2_ck, dpll4_m6x2_ck_parent_names, dpll4_m5x2_ck_ops);
 
-static struct clk dpll4_m6x2_ck_3630 = {
+static struct clk_core dpll4_m6x2_ck_3630_core = {
 	.name		= "dpll4_m6x2_ck",
 	.hw		= &dpll4_m6x2_ck_hw.hw,
 	.parent_names	= dpll4_m6x2_ck_parent_names,
@@ -963,7 +989,11 @@ static struct clk dpll4_m6x2_ck_3630 = {
 	.ops		= &dpll4_m5x2_ck_3630_ops,
 };
 
-DEFINE_CLK_FIXED_FACTOR(dpll4_x2_ck, "dpll4_ck", &dpll4_ck, 0x0, 2, 1);
+static struct clk dpll4_m6x2_ck_3630 = {
+	.core = &dpll4_m6x2_ck_3630_core,
+};
+
+DEFINE_CLK_FIXED_FACTOR(dpll4_x2_ck, "dpll4_ck", &dpll4_ck_core, 0x0, 2, 1);
 
 static struct dpll_data dpll5_dd = {
 	.mult_div1_reg	= OMAP_CM_REGADDR(PLL_MOD, OMAP3430ES2_CM_CLKSEL4),
@@ -1000,7 +1030,7 @@ static struct clk_hw_omap dpll5_ck_hw = {
 
 DEFINE_STRUCT_CLK(dpll5_ck, dpll3_ck_parent_names, dpll1_ck_ops);
 
-DEFINE_CLK_DIVIDER(dpll5_m2_ck, "dpll5_ck", &dpll5_ck, 0x0,
+DEFINE_CLK_DIVIDER(dpll5_m2_ck, "dpll5_ck", &dpll5_ck_core, 0x0,
 		   OMAP_CM_REGADDR(PLL_MOD, OMAP3430ES2_CM_CLKSEL5),
 		   OMAP3430ES2_DIV_120M_SHIFT, OMAP3430ES2_DIV_120M_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -1247,7 +1277,7 @@ static struct clk_hw_omap emu_src_ck_hw = {
 
 DEFINE_STRUCT_CLK(emu_src_ck, emu_src_ck_parent_names, emu_src_ck_ops);
 
-DEFINE_CLK_DIVIDER(atclk_fck, "emu_src_ck", &emu_src_ck, 0x0,
+DEFINE_CLK_DIVIDER(atclk_fck, "emu_src_ck", &emu_src_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_EMU_MOD, CM_CLKSEL1),
 		   OMAP3430_CLKSEL_ATCLK_SHIFT, OMAP3430_CLKSEL_ATCLK_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -1298,7 +1328,7 @@ static struct clk_hw_omap gfx_l3_ck_hw = {
 
 DEFINE_STRUCT_CLK(gfx_l3_ck, core_l3_ick_parent_names, aes1_ick_ops);
 
-DEFINE_CLK_DIVIDER(gfx_l3_fck, "l3_ick", &l3_ick, 0x0,
+DEFINE_CLK_DIVIDER(gfx_l3_fck, "l3_ick", &l3_ick_core, 0x0,
 		   OMAP_CM_REGADDR(GFX_MOD, CM_CLKSEL),
 		   OMAP_CLKSEL_GFX_SHIFT, OMAP_CLKSEL_GFX_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -2498,7 +2528,7 @@ static struct clk_hw_omap omap_96m_alwon_fck_3630_hw = {
 	.clksel_mask	= OMAP3630_CLKSEL_96M_MASK,
 };
 
-static struct clk omap_96m_alwon_fck_3630 = {
+static struct clk_core omap_96m_alwon_fck_3630_core = {
 	.name	= "omap_96m_alwon_fck",
 	.hw	= &omap_96m_alwon_fck_3630_hw.hw,
 	.parent_names	= omap_96m_alwon_fck_3630_parent_names,
@@ -2506,6 +2536,10 @@ static struct clk omap_96m_alwon_fck_3630 = {
 	.ops	= &omap_96m_alwon_fck_3630_ops,
 };
 
+static struct clk omap_96m_alwon_fck_3630 = {
+	.core = &omap_96m_alwon_fck_3630_core,
+};
+
 static struct clk omapctrl_ick;
 
 static struct clk_hw_omap omapctrl_ick_hw = {
@@ -2521,12 +2555,12 @@ static struct clk_hw_omap omapctrl_ick_hw = {
 
 DEFINE_STRUCT_CLK(omapctrl_ick, aes2_ick_parent_names, aes2_ick_ops);
 
-DEFINE_CLK_DIVIDER(pclk_fck, "emu_src_ck", &emu_src_ck, 0x0,
+DEFINE_CLK_DIVIDER(pclk_fck, "emu_src_ck", &emu_src_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_EMU_MOD, CM_CLKSEL1),
 		   OMAP3430_CLKSEL_PCLK_SHIFT, OMAP3430_CLKSEL_PCLK_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
 
-DEFINE_CLK_DIVIDER(pclkx2_fck, "emu_src_ck", &emu_src_ck, 0x0,
+DEFINE_CLK_DIVIDER(pclkx2_fck, "emu_src_ck", &emu_src_ck_core, 0x0,
 		   OMAP_CM_REGADDR(OMAP3430_EMU_MOD, CM_CLKSEL1),
 		   OMAP3430_CLKSEL_PCLKX2_SHIFT, OMAP3430_CLKSEL_PCLKX2_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -2558,7 +2592,7 @@ static struct clk_hw_omap pka_ick_hw = {
 
 DEFINE_STRUCT_CLK(pka_ick, pka_ick_parent_names, aes1_ick_ops);
 
-DEFINE_CLK_DIVIDER(rm_ick, "l4_ick", &l4_ick, 0x0,
+DEFINE_CLK_DIVIDER(rm_ick, "l4_ick", &l4_ick_core, 0x0,
 		   OMAP_CM_REGADDR(WKUP_MOD, CM_CLKSEL),
 		   OMAP3430_CLKSEL_RM_SHIFT, OMAP3430_CLKSEL_RM_WIDTH,
 		   CLK_DIVIDER_ONE_BASED, NULL);
@@ -2819,10 +2853,10 @@ DEFINE_CLK_OMAP_MUX_GATE(ssi_ssr_fck_3430es2, "core_l4_clkdm",
 			 ssi_ssr_fck_3430es1_ops);
 
 DEFINE_CLK_FIXED_FACTOR(ssi_sst_fck_3430es1, "ssi_ssr_fck_3430es1",
-			&ssi_ssr_fck_3430es1, 0x0, 1, 2);
+			&ssi_ssr_fck_3430es1_core, 0x0, 1, 2);
 
 DEFINE_CLK_FIXED_FACTOR(ssi_sst_fck_3430es2, "ssi_ssr_fck_3430es2",
-			&ssi_ssr_fck_3430es2, 0x0, 1, 2);
+			&ssi_ssr_fck_3430es2_core, 0x0, 1, 2);
 
 static struct clk sys_clkout1;
 
@@ -2840,7 +2874,7 @@ static struct clk_hw_omap sys_clkout1_hw = {
 
 DEFINE_STRUCT_CLK(sys_clkout1, sys_clkout1_parent_names, aes1_ick_ops);
 
-DEFINE_CLK_DIVIDER(sys_clkout2, "clkout2_src_ck", &clkout2_src_ck, 0x0,
+DEFINE_CLK_DIVIDER(sys_clkout2, "clkout2_src_ck", &clkout2_src_ck_core, 0x0,
 		   OMAP3430_CM_CLKOUT_CTRL, OMAP3430_CLKOUT2_DIV_SHIFT,
 		   OMAP3430_CLKOUT2_DIV_WIDTH, CLK_DIVIDER_POWER_OF_TWO, NULL);
 
@@ -2849,7 +2883,8 @@ DEFINE_CLK_MUX(traceclk_src_fck, emu_src_ck_parent_names, NULL, 0x0,
 	       OMAP3430_TRACE_MUX_CTRL_SHIFT, OMAP3430_TRACE_MUX_CTRL_WIDTH,
 	       0x0, NULL);
 
-DEFINE_CLK_DIVIDER(traceclk_fck, "traceclk_src_fck", &traceclk_src_fck, 0x0,
+DEFINE_CLK_DIVIDER(traceclk_fck, "traceclk_src_fck", &traceclk_src_fck_core,
+		   0x0,
 		   OMAP_CM_REGADDR(OMAP3430_EMU_MOD, CM_CLKSEL1),
 		   OMAP3430_CLKSEL_TRACECLK_SHIFT,
 		   OMAP3430_CLKSEL_TRACECLK_WIDTH, CLK_DIVIDER_ONE_BASED, NULL);
arch/arm/mach-omap2/clock.h +9 -2
index a4282e7..c5b3a7f 100644
@@ -40,23 +40,29 @@ struct omap_clk {
 struct clockdomain;
 
 #define DEFINE_STRUCT_CLK(_name, _parent_array_name, _clkops_name)	\
-	static struct clk _name = {				\
+	static struct clk_core _name##_core = {			\
 		.name = #_name,					\
 		.hw = &_name##_hw.hw,				\
 		.parent_names = _parent_array_name,		\
 		.num_parents = ARRAY_SIZE(_parent_array_name),	\
 		.ops = &_clkops_name,				\
+	};							\
+	static struct clk _name = {				\
+		.core = &_name##_core,				\
 	};
 
 #define DEFINE_STRUCT_CLK_FLAGS(_name, _parent_array_name,	\
 				_clkops_name, _flags)		\
-	static struct clk _name = {				\
+	static struct clk_core _name##_core = {			\
 		.name = #_name,					\
 		.hw = &_name##_hw.hw,				\
 		.parent_names = _parent_array_name,		\
 		.num_parents = ARRAY_SIZE(_parent_array_name),	\
 		.ops = &_clkops_name,				\
 		.flags = _flags,				\
+	};							\
+	static struct clk _name = {				\
+		.core = &_name##_core,				\
 	};
 
 #define DEFINE_STRUCT_CLK_HW_OMAP(_name, _clkdm_name)		\
@@ -248,6 +254,7 @@ extern const struct clksel_rate gpt_32k_rates[];
 extern const struct clksel_rate gpt_sys_rates[];
 extern const struct clksel_rate gfx_l3_rates[];
 extern const struct clksel_rate dsp_ick_rates[];
+extern struct clk_core dummy_ck_core;
 extern struct clk dummy_ck;
 
 extern const struct clk_hw_omap_ops clkhwops_iclk_wait;
arch/arm/mach-omap2/clock_common_data.c +4 -1
index ef4d21b..febd0a2 100644
@@ -119,8 +119,11 @@ const struct clksel_rate div31_1to31_rates[] = {
 
 static struct clk_ops dummy_ck_ops = {};
 
-struct clk dummy_ck = {
+struct clk_core dummy_ck_core = {
 	.name = "dummy_clk",
 	.ops = &dummy_ck_ops,
 	.flags = CLK_IS_BASIC,
 };
+struct clk dummy_ck = {
+	.core = &dummy_ck_core,
+};
drivers/clk/clk.c +399 -222

1,329 lines not on this page · raw message

drivers/clk/clk.h +5 -0
index c798138..23c44e5 100644
@@ -9,9 +9,14 @@
  * published by the Free Software Foundation.
  */
 
+struct clk_hw;
+
 #if defined(CONFIG_OF) && defined(CONFIG_COMMON_CLK)
 struct clk *of_clk_get_by_clkspec(struct of_phandle_args *clkspec);
 struct clk *__of_clk_get_from_provider(struct of_phandle_args *clkspec);
 void of_clk_lock(void);
 void of_clk_unlock(void);
 #endif
+
+struct clk *__clk_create_clk(struct clk_hw *hw, const char *dev_id,
+			     const char *con_id);
drivers/clk/clkdev.c +65 -19

138 lines not on this page · raw message

include/linux/clk-private.h +27 -8
index c5f40d0..ae55d99 100644
@@ -28,20 +28,20 @@
 
 struct module;
 
-struct clk {
+struct clk_core {
 	const char		*name;
 	const struct clk_ops	*ops;
 	struct clk_hw		*hw;
 	struct module		*owner;
-	struct clk		*parent;
+	struct clk_core		*parent;
 	const char		**parent_names;
-	struct clk		**parents;
+	struct clk_core		**parents;
 	u8			num_parents;
 	u8			new_parent_index;
 	unsigned long		rate;
 	unsigned long		new_rate;
-	struct clk		*new_parent;
-	struct clk		*new_child;
+	struct clk_core		*new_parent;
+	struct clk_core		*new_child;
 	unsigned long		flags;
 	unsigned int		enable_count;
 	unsigned int		prepare_count;
@@ -57,6 +57,12 @@ struct clk {
 	struct kref		ref;
 };
 
+struct clk {
+	struct clk_core	*core;
+	const char *dev_id;
+	const char *con_id;
+};
+
 /*
  * DOC: Basic clock implementations common to many platforms
  *
@@ -69,6 +75,9 @@ struct clk {
 #define DEFINE_CLK(_name, _ops, _flags, _parent_names,		\
 		_parents)					\
 	static struct clk _name = {				\
+		.core = &_name##_core				\
+	};							\
+	static struct clk_core _name##_core = {			\
 		.name = #_name,					\
 		.ops = &_ops,					\
 		.hw = &_name##_hw.hw,				\
@@ -81,9 +90,11 @@ struct clk {
 #define DEFINE_CLK_FIXED_RATE(_name, _flags, _rate,		\
 				_fixed_rate_flags)		\
 	static struct clk _name;				\
+	static struct clk_core _name##_core;			\
 	static const char *_name##_parent_names[] = {};		\
 	static struct clk_fixed_rate _name##_hw = {		\
 		.hw = {						\
+			.core = &_name##_core,			\
 			.clk = &_name,				\
 		},						\
 		.fixed_rate = _rate,				\
@@ -96,14 +107,16 @@ struct clk {
 				_flags, _reg, _bit_idx,		\
 				_gate_flags, _lock)		\
 	static struct clk _name;				\
+	static struct clk_core _name##_core;			\
 	static const char *_name##_parent_names[] = {		\
 		_parent_name,					\
 	};							\
-	static struct clk *_name##_parents[] = {		\
+	static struct clk_core *_name##_parents[] = {		\
 		_parent_ptr,					\
 	};							\
 	static struct clk_gate _name##_hw = {			\
 		.hw = {						\
+			.core = &_name##_core,			\
 			.clk = &_name,				\
 		},						\
 		.reg = _reg,					\
@@ -118,14 +131,16 @@ struct clk {
 				_flags, _reg, _shift, _width,	\
 				_divider_flags, _table, _lock)	\
 	static struct clk _name;				\
+	static struct clk_core _name##_core;			\
 	static const char *_name##_parent_names[] = {		\
 		_parent_name,					\
 	};							\
-	static struct clk *_name##_parents[] = {		\
+	static struct clk_core *_name##_parents[] = {		\
 		_parent_ptr,					\
 	};							\
 	static struct clk_divider _name##_hw = {		\
 		.hw = {						\
+			.core = &_name##_core,			\
 			.clk = &_name,				\
 		},						\
 		.reg = _reg,					\
@@ -157,8 +172,10 @@ struct clk {
 				_reg, _shift, _width,		\
 				_mux_flags, _lock)		\
 	static struct clk _name;				\
+	static struct clk_core _name##_core;			\
 	static struct clk_mux _name##_hw = {			\
 		.hw = {						\
+			.core = &_name##_core,			\
 			.clk = &_name,				\
 		},						\
 		.reg = _reg,					\
@@ -174,14 +191,16 @@ struct clk {
 				_parent_ptr, _flags,		\
 				_mult, _div)			\
 	static struct clk _name;				\
+	static struct clk_core _name##_core;			\
 	static const char *_name##_parent_names[] = {		\
 		_parent_name,					\
 	};							\
-	static struct clk *_name##_parents[] = {		\
+	static struct clk_core *_name##_parents[] = {		\
 		_parent_ptr,					\
 	};							\
 	static struct clk_fixed_factor _name##_hw = {		\
 		.hw = {						\
+			.core = &_name##_core,			\
 			.clk = &_name,				\
 		},						\
 		.mult = _mult,					\
include/linux/clk-provider.h +7 -5
index d936409..9afd438 100644
@@ -33,6 +33,7 @@
 #define CLK_GET_ACCURACY_NOCACHE BIT(8) /* do not use the cached clk accuracy */
 
 struct clk_hw;
+struct clk_core;
 struct dentry;
 
 /**
@@ -216,13 +217,17 @@ struct clk_init_data {
  * clk_foo and then referenced by the struct clk instance that uses struct
  * clk_foo's clk_ops
  *
- * @clk: pointer to the struct clk instance that points back to this struct
- * clk_hw instance
+ * @core: pointer to the struct clk_core instance that points back to this
+ * struct clk_hw instance
+ *
+ * @clk: pointer to the per-user struct clk instance that can be used to call
+ * into the clk API
  *
  * @init: pointer to struct clk_init_data that contains the init data shared
  * with the common clock framework.
  */
 struct clk_hw {
+	struct clk_core *core;
 	struct clk *clk;
 	const struct clk_init_data *init;
 };
@@ -556,9 +561,6 @@ long __clk_mux_determine_rate(struct clk_hw *hw, unsigned long rate,
 /*
  * FIXME clock api without lock protection
  */
-int __clk_prepare(struct clk *clk);
-void __clk_unprepare(struct clk *clk);
-void __clk_reparent(struct clk *clk, struct clk *new_parent);
 unsigned long __clk_round_rate(struct clk *clk, unsigned long rate);
 
 struct of_device_id;
Signature · 1 line
1.9.3
#4 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

Quoted 6 lines

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

Quoted 6 lines

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and struct dpll_data, namely this snippet from arch/arm/mach-omap2/dpll3xxx.c:

        parent = __clk_get_parent(hw->clk);

        if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                WARN(parent != dd->clk_bypass,
                                "here0, parent name is %s, bypass name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                r = _omap3_noncore_dpll_bypass(clk);
        } else {
                WARN(parent != dd->clk_ref,
                                "here1, parent name is %s, ref name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                r = _omap3_noncore_dpll_lock(clk);
        }

struct dpll_data has members clk_ref and clk_bypass which are struct clk pointers. This was always a bit of a violation of the clk.h contract since drivers are not supposed to deref struct clk pointers. Now that we generate unique pointers for each call to clk_get (clk_ref & clk_bypass are populated by of_clk_get in ti_clk_register_dpll) then the pointer comparisons above will never be equal (even if they resolve down to the same struct clk_core). I added the verbose traces to the WARNs above to illustrate the point: the names are always the same but the pointers differ.

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

        dd->clk_ref = of_clk_get(node, 0);
        dd->clk_bypass = of_clk_get(node, 1);

Tony, the same problem affects the FAPLL code which copy/pastes some of the DPLL code.

Thoughts?

Regards,
Mike

maintainer
#5 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 44 lines · Mike Turquette

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

Quoted 6 lines

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

Quoted 6 lines

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and struct dpll_data, namely this snippet from arch/arm/mach-omap2/dpll3xxx.c:

        parent = __clk_get_parent(hw->clk);

        if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                WARN(parent != dd->clk_bypass,
                                "here0, parent name is %s, bypass name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                r = _omap3_noncore_dpll_bypass(clk);
        } else {
                WARN(parent != dd->clk_ref,
                                "here1, parent name is %s, ref name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                r = _omap3_noncore_dpll_lock(clk);
        }

struct dpll_data has members clk_ref and clk_bypass which are struct clk pointers. This was always a bit of a violation of the clk.h contract since drivers are not supposed to deref struct clk pointers. Now that we generate unique pointers for each call to clk_get (clk_ref & clk_bypass are populated by of_clk_get in ti_clk_register_dpll) then the pointer comparisons above will never be equal (even if they resolve down to the same struct clk_core). I added the verbose traces to the WARNs above to illustrate the point: the names are always the same but the pointers differ.

Quoted 2 lines

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

Also on at least omap4 like I posted. So sounds like the check for WARN is wrong but harmless. Paul & Tero, what do you want to do about that?

Quoted 5 lines

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

Quoted 7 lines
        dd->clk_ref = of_clk_get(node, 0);
        dd->clk_bypass = of_clk_get(node, 1);

Tony, the same problem affects the FAPLL code which copy/pastes some of the DPLL code.

Thoughts?

Not seeing these warnings with dm186x as fapll.c does not use dpll3xxx.c. This is because of the way the PLL's child synthesizers need to also access the PLL registers for power and bypass mode.

Not related to the $subject bug, but to me it seems that we could possibly have Linux generic PLL code if we add support for parent_in_bypass_mode in addition to the parent_rate. This is because the PLL can in theory generate the same rate both in bypass mode and regular mode so parent_rate is not enough to tell it to the child synthesizers. Not sure how the PLL registers enabling and disabling it's children should be handled, maybe regmap would work there.

Regards,

Tony

#6 lore raw

PATCH v13 3/6 clk: Make clk API return per-user struct clk instances

Quoting Tony Lindgren (2015-02-02 09:04:59)

Quoted 45 lines · Mike Turquette
Quoted 44 lines

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

Quoted 6 lines

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

Quoted 6 lines

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and struct dpll_data, namely this snippet from arch/arm/mach-omap2/dpll3xxx.c:

        parent = __clk_get_parent(hw->clk);

        if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                WARN(parent != dd->clk_bypass,
                                "here0, parent name is %s, bypass name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                r = _omap3_noncore_dpll_bypass(clk);
        } else {
                WARN(parent != dd->clk_ref,
                                "here1, parent name is %s, ref name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                r = _omap3_noncore_dpll_lock(clk);
        }

struct dpll_data has members clk_ref and clk_bypass which are struct clk pointers. This was always a bit of a violation of the clk.h contract since drivers are not supposed to deref struct clk pointers. Now that we generate unique pointers for each call to clk_get (clk_ref & clk_bypass are populated by of_clk_get in ti_clk_register_dpll) then the pointer comparisons above will never be equal (even if they resolve down to the same struct clk_core). I added the verbose traces to the WARNs above to illustrate the point: the names are always the same but the pointers differ.

Quoted 4 lines
Quoted 2 lines

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

Also on at least omap4 like I posted.

Right, hence the + after OMAP3 ;-)

Quoted 3 lines

So sounds like the check for
WARN is wrong but harmless. Paul & Tero, what do you want to do about that?

I would be fine to simply silence the WARNs since we're so closed to the merge window and then revisit it with a proper fix. Of course the ideal solution would be to convert the pointer comparison scheme to one using parent_index.

Regards,
Mike

Quoted 29 lines
Quoted 6 lines

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

Quoted 7 lines
        dd->clk_ref = of_clk_get(node, 0);
        dd->clk_bypass = of_clk_get(node, 1);

Tony, the same problem affects the FAPLL code which copy/pastes some of the DPLL code.

Thoughts?

Not seeing these warnings with dm186x as fapll.c does not use dpll3xxx.c. This is because of the way the PLL's child synthesizers need to also access the PLL registers for power and bypass mode.

Not related to the $subject bug, but to me it seems that we could possibly have Linux generic PLL code if we add support for parent_in_bypass_mode in addition to the parent_rate. This is because the PLL can in theory generate the same rate both in bypass mode and regular mode so parent_rate is not enough to tell it to the child synthesizers. Not sure how the PLL registers enabling and disabling it's children should be handled, maybe regmap would work there.

Regards,

Tony

#7 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Mike Turquette
Quoted 44 lines · Mike Turquette

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

Quoted 6 lines

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

Quoted 6 lines

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and struct dpll_data, namely this snippet from arch/arm/mach-omap2/dpll3xxx.c:

         parent = __clk_get_parent(hw->clk);

         if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                 WARN(parent != dd->clk_bypass,
                                 "here0, parent name is %s, bypass name is %s\n",
                                 __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                 r = _omap3_noncore_dpll_bypass(clk);
         } else {
                 WARN(parent != dd->clk_ref,
                                 "here1, parent name is %s, ref name is %s\n",
                                 __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                 r = _omap3_noncore_dpll_lock(clk);
         }

struct dpll_data has members clk_ref and clk_bypass which are struct clk pointers. This was always a bit of a violation of the clk.h contract since drivers are not supposed to deref struct clk pointers. Now that we generate unique pointers for each call to clk_get (clk_ref & clk_bypass are populated by of_clk_get in ti_clk_register_dpll) then the pointer comparisons above will never be equal (even if they resolve down to the same struct clk_core). I added the verbose traces to the WARNs above to illustrate the point: the names are always the same but the pointers differ.

Quoted 6 lines

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

The problem is we still need to be able to get runtime parent clock rates (the parent rate may change also), so simple index value is not sufficient. We need a handle of some sort to the bypass/ref clocks. The DPLL code generally requires knowledge of the bypass + reference clock rates to work properly, as it calculates the M/N values based on these.

Shall I change the DPLL code to check against clk_hw pointers or what is the preferred approach here? The patch at the end does this and fixes the dpll related warnings.

Btw, the rate constraints patch broke boot for me completely, but sounds like you reverted it already.

-Tero

--------------------

Author: Tero Kristo <[email protected]>
Date:   Mon Feb 2 17:19:17 2015 +0200

ARM: OMAP3+: clock: dpll: fix logic for comparing parent clocks

DPLL code uses reference and bypass clock pointers for determining
runtime
properties for these clocks, like parent clock rates.

As clock API now returns per-user clock structs, using a global handle in the clock driver code does not work properly anymore. Fix this by using the clk_hw instead, and comparing this against the parents.

Fixes 59cf3fcf9baf ("clk: Make clk API return per-user struct clk

instances")

arch/arm/mach-omap2/dpll3xxx.c +4 -4
index c2da2a0..49752d7 100644
@@ -410,7 +410,7 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
  	struct clk_hw_omap *clk = to_clk_hw_omap(hw);
  	int r;
  	struct dpll_data *dd;
-	struct clk *parent;
+	struct clk_hw *parent;
 
  	dd = clk->dpll_data;
  	if (!dd)
@@ -427,13 +427,13 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
  		}
  	}
 
-	parent = __clk_get_parent(hw->clk);
+	parent = __clk_get_hw(__clk_get_parent(hw->clk));
 
  	if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
-		WARN_ON(parent != dd->clk_bypass);
+		WARN_ON(parent != __clk_get_hw(dd->clk_bypass));
  		r = _omap3_noncore_dpll_bypass(clk);
  	} else {
-		WARN_ON(parent != dd->clk_ref);
+		WARN_ON(parent != __clk_get_hw(dd->clk_ref));
  		r = _omap3_noncore_dpll_lock(clk);
  	}
 
@@ -549,7 +549,8 @@ int omap3_noncore_dpll_set_rate(struct clk_hw *hw, 

unsigned long rate,
if (!dd)
return -EINVAL;

- if (__clk_get_parent(hw->clk) != dd->clk_ref)
+ if (__clk_get_hw(__clk_get_parent(hw->clk)) !=
+ __clk_get_hw(dd->clk_ref))
return -EINVAL;

if (dd->last_rounded_rate == 0)

maintainer Tested-by
#8 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 16 lines · Tero Kristo
Quoted 8 lines · Mike Turquette

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

The problem is we still need to be able to get runtime parent clock rates (the parent rate may change also), so simple index value is not sufficient. We need a handle of some sort to the bypass/ref clocks. The DPLL code generally requires knowledge of the bypass + reference clock rates to work properly, as it calculates the M/N values based on these.

Quoted 6 lines

Shall I change the DPLL code to check against clk_hw pointers or what is the preferred approach here? The patch at the end does this and fixes the dpll related warnings.

Btw, the rate constraints patch broke boot for me completely, but sounds like you reverted it already.

Thanks Tero, looks like your fix fixes all the issues I'm seeing with commit 59cf3fcf9baf. That is noisy dmesg, dpll_abe_ck not locking on 4430sdp, and off-idle not working for omap3.

I could not get the patch to apply, below is what I applied manually.

Mike, If possible, maybe fold this into 59cf3fcf9baf? It applies with some fuzz on that too. And inn that case, please feel also to add the following for Tomeu's patch:

8<------------
From: Tero Kristo <[email protected]>
Date: Mon, 2 Feb 2015 12:17:00 -0800
Subject: [PATCH] ARM: OMAP3+: clock: dpll: fix logic for comparing parent clocks

DPLL code uses reference and bypass clock pointers for determining runtime properties for these clocks, like parent clock rates.
As clock API now returns per-user clock structs, using a global handle in the clock driver code does not work properly anymore. Fix this by using the clk_hw instead, and comparing this against the parents.

Fixes 59cf3fcf9baf ("clk: Make clk API return per-user struct clk instances")
[tony at atomide.com: updated to apply]
arch/arm/mach-omap2/dpll3xxx.c +6 -5
@@ -410,7 +410,7 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
 	struct clk_hw_omap *clk = to_clk_hw_omap(hw);
 	int r;
 	struct dpll_data *dd;
-	struct clk *parent;
+	struct clk_hw *parent;
 
 	dd = clk->dpll_data;
 	if (!dd)
@@ -427,13 +427,13 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
 		}
 	}
 
-	parent = __clk_get_parent(hw->clk);
+	parent = __clk_get_hw(__clk_get_parent(hw->clk));
 
 	if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
-		WARN_ON(parent != dd->clk_bypass);
+		WARN_ON(parent != __clk_get_hw(dd->clk_bypass));
 		r = _omap3_noncore_dpll_bypass(clk);
 	} else {
-		WARN_ON(parent != dd->clk_ref);
+		WARN_ON(parent != __clk_get_hw(dd->clk_ref));
 		r = _omap3_noncore_dpll_lock(clk);
 	}
 
@@ -549,7 +549,8 @@ int omap3_noncore_dpll_set_rate(struct clk_hw *hw, unsigned long rate,
 	if (!dd)
 		return -EINVAL;
 
-	if (__clk_get_parent(hw->clk) != dd->clk_ref)
+	if (__clk_get_hw(__clk_get_parent(hw->clk)) !=
+	    __clk_get_hw(dd->clk_ref))
 		return -EINVAL;
 
 	if (dd->last_rounded_rate == 0)
#9 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoting Tony Lindgren (2015-02-02 12:44:02)

Quoted 28 lines · Tero Kristo
Quoted 16 lines · Mike Turquette
Quoted 8 lines

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

The problem is we still need to be able to get runtime parent clock rates (the parent rate may change also), so simple index value is not sufficient. We need a handle of some sort to the bypass/ref clocks. The DPLL code generally requires knowledge of the bypass + reference clock rates to work properly, as it calculates the M/N values based on these.

Quoted 6 lines

Shall I change the DPLL code to check against clk_hw pointers or what is the preferred approach here? The patch at the end does this and fixes the dpll related warnings.

Btw, the rate constraints patch broke boot for me completely, but sounds like you reverted it already.

Thanks Tero, looks like your fix fixes all the issues I'm seeing with commit 59cf3fcf9baf. That is noisy dmesg, dpll_abe_ck not locking on 4430sdp, and off-idle not working for omap3.

Quoted 7 lines

I could not get the patch to apply, below is what I applied manually.

Mike, If possible, maybe fold this into 59cf3fcf9baf? It applies with some fuzz on that too. And inn that case, please feel also to add the following for Tomeu's patch:

Done and done. Things look good in my testing. I've pushed another branch out to the mirrors and hopefully the autobuild/autoboot testing will give us the green light.

This implementation can be revisited probably after 3.19 comes out if Tero doesn't like using clk_hw directly, or if we provide a better interface.

Thanks,
Mike

Quoted 55 lines

8<------------
From: Tero Kristo <[email protected]>
Date: Mon, 2 Feb 2015 12:17:00 -0800
Subject: [PATCH] ARM: OMAP3+: clock: dpll: fix logic for comparing parent clocks

DPLL code uses reference and bypass clock pointers for determining runtime properties for these clocks, like parent clock rates.
As clock API now returns per-user clock structs, using a global handle in the clock driver code does not work properly anymore. Fix this by using the clk_hw instead, and comparing this against the parents.

Fixes 59cf3fcf9baf ("clk: Make clk API return per-user struct clk instances")
[tony at atomide.com: updated to apply]
arch/arm/mach-omap2/dpll3xxx.c +6 -5
@@ -410,7 +410,7 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
        struct clk_hw_omap *clk = to_clk_hw_omap(hw);
        int r;
        struct dpll_data *dd;
-       struct clk *parent;
+       struct clk_hw *parent;
 
        dd = clk->dpll_data;
        if (!dd)
@@ -427,13 +427,13 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
                }
        }
 
-       parent = __clk_get_parent(hw->clk);
+       parent = __clk_get_hw(__clk_get_parent(hw->clk));
 
        if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
-               WARN_ON(parent != dd->clk_bypass);
+               WARN_ON(parent != __clk_get_hw(dd->clk_bypass));
                r = _omap3_noncore_dpll_bypass(clk);
        } else {
-               WARN_ON(parent != dd->clk_ref);
+               WARN_ON(parent != __clk_get_hw(dd->clk_ref));
                r = _omap3_noncore_dpll_lock(clk);
        }
 
@@ -549,7 +549,8 @@ int omap3_noncore_dpll_set_rate(struct clk_hw *hw, unsigned long rate,
        if (!dd)
                return -EINVAL;
 
-       if (__clk_get_parent(hw->clk) != dd->clk_ref)
+       if (__clk_get_hw(__clk_get_parent(hw->clk)) !=
+           __clk_get_hw(dd->clk_ref))
                return -EINVAL;
 
        if (dd->last_rounded_rate == 0)
maintainer
#10 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 12 lines · Mike Turquette

Quoting Tony Lindgren (2015-02-02 12:44:02)

Quoted 11 lines

Thanks Tero, looks like your fix fixes all the issues I'm seeing with commit 59cf3fcf9baf. That is noisy dmesg, dpll_abe_ck not locking on 4430sdp, and off-idle not working for omap3.

I could not get the patch to apply, below is what I applied manually.

Mike, If possible, maybe fold this into 59cf3fcf9baf? It applies with some fuzz on that too. And inn that case, please feel also to add the following for Tomeu's patch:

Quoted 5 lines
Quoted 1 line

Done and done. Things look good in my testing. I've pushed another branch out to the mirrors and hopefully the autobuild/autoboot testing will give us the green light.

Thanks I just checked that your updated branch works for me now.

Quoted 3 lines

This implementation can be revisited probably after 3.19 comes out if Tero doesn't like using clk_hw directly, or if we provide a better interface.

Sounds like what Tero is saying also relates to knowing if the parent clock is in bypass mode or not in addition to the parent rate.

Regards,

Tony

#11 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Tero Kristo

Quoting Tero Kristo (2015-02-02 11:32:01)

Quoted 52 lines · Mike Turquette
Quoted 44 lines

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

Quoted 6 lines

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

Quoted 6 lines

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and struct dpll_data, namely this snippet from arch/arm/mach-omap2/dpll3xxx.c:

         parent = __clk_get_parent(hw->clk);

         if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                 WARN(parent != dd->clk_bypass,
                                 "here0, parent name is %s, bypass name is %s\n",
                                 __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                 r = _omap3_noncore_dpll_bypass(clk);
         } else {
                 WARN(parent != dd->clk_ref,
                                 "here1, parent name is %s, ref name is %s\n",
                                 __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                 r = _omap3_noncore_dpll_lock(clk);
         }

struct dpll_data has members clk_ref and clk_bypass which are struct clk pointers. This was always a bit of a violation of the clk.h contract since drivers are not supposed to deref struct clk pointers. Now that we generate unique pointers for each call to clk_get (clk_ref & clk_bypass are populated by of_clk_get in ti_clk_register_dpll) then the pointer comparisons above will never be equal (even if they resolve down to the same struct clk_core). I added the verbose traces to the WARNs above to illustrate the point: the names are always the same but the pointers differ.

Quoted 6 lines

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

Quoted 5 lines

The problem is we still need to be able to get runtime parent clock rates (the parent rate may change also), so simple index value is not sufficient. We need a handle of some sort to the bypass/ref clocks. The DPLL code generally requires knowledge of the bypass + reference clock rates to work properly, as it calculates the M/N values based on these.

We can maybe introduce something like of_clk_get_parent_rate, as we have analogous stuff for getting parent names and indexes. Without introducing a new helper you could probably just do:

	clk_ref = clk_get_parent_by_index(dpll_clk, 0);
	ref_rate = clk_get_rate(clk_ref);

	clk_bypass = clk_get_parent_by_index(dpll_clk, 1);
	bypass_rate = clk_get_rate(clk_bypass);

Currently the semantics around this call are weird. It seems like it would create a new struct clk pointer but it does not. So don't call clk_put on clk_ref and clk_bypass yet. That might change in the future as we iron out this brave new world that we all live in. Probably best to leave a FIXME in there.

Stephen & Tomeu, let me know if I got any of that wrong.

Quoted 4 lines

Shall I change the DPLL code to check against clk_hw pointers or what is the preferred approach here? The patch at the end does this and fixes the dpll related warnings.

Yes, for now that is fine, but feels a bit hacky to me. I don't know honestly, let me sleep on it. Anyways for 3.20 that is perfectly fine but we might want to switch to something like the scheme above.

Quoted 3 lines

Btw, the rate constraints patch broke boot for me completely, but sounds like you reverted it already.

Fixed with Stephen's patch from last week. Thanks for dealing with all the breakage so promptly. It has helped a lot!

Regards,
Mike

Quoted 65 lines

-Tero

--------------------

Author: Tero Kristo <[email protected]>
Date:   Mon Feb 2 17:19:17 2015 +0200

ARM: OMAP3+: clock: dpll: fix logic for comparing parent clocks

DPLL code uses reference and bypass clock pointers for determining
runtime
properties for these clocks, like parent clock rates.

As clock API now returns per-user clock structs, using a global handle in the clock driver code does not work properly anymore. Fix this by using the clk_hw instead, and comparing this against the parents.

Fixes 59cf3fcf9baf ("clk: Make clk API return per-user struct clk

instances")

arch/arm/mach-omap2/dpll3xxx.c +6 -5
index c2da2a0..49752d7 100644
@@ -410,7 +410,7 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
        struct clk_hw_omap *clk = to_clk_hw_omap(hw);
        int r;
        struct dpll_data *dd;
-       struct clk *parent;
+       struct clk_hw *parent;
 
        dd = clk->dpll_data;
        if (!dd)
@@ -427,13 +427,13 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
                }
        }
 
-       parent = __clk_get_parent(hw->clk);
+       parent = __clk_get_hw(__clk_get_parent(hw->clk));
 
        if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
-               WARN_ON(parent != dd->clk_bypass);
+               WARN_ON(parent != __clk_get_hw(dd->clk_bypass));
                r = _omap3_noncore_dpll_bypass(clk);
        } else {
-               WARN_ON(parent != dd->clk_ref);
+               WARN_ON(parent != __clk_get_hw(dd->clk_ref));
                r = _omap3_noncore_dpll_lock(clk);
        }
 
@@ -549,7 +549,8 @@ int omap3_noncore_dpll_set_rate(struct clk_hw *hw, 
 unsigned long rate,
        if (!dd)
                return -EINVAL;
 
-       if (__clk_get_parent(hw->clk) != dd->clk_ref)
+       if (__clk_get_hw(__clk_get_parent(hw->clk)) !=
+           __clk_get_hw(dd->clk_ref))
                return -EINVAL;
 
        if (dd->last_rounded_rate == 0)
#12 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 24 lines · Mike Turquette

Quoting Tero Kristo (2015-02-02 11:32:01)

Quoted 5 lines · Mike Turquette
Quoted 4 lines

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

Quoted 8 lines
Quoted 3 lines

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

The problem is we still need to be able to get runtime parent clock rates (the parent rate may change also), so simple index value is not sufficient. We need a handle of some sort to the bypass/ref clocks. The DPLL code generally requires knowledge of the bypass + reference clock rates to work properly, as it calculates the M/N values based on these.

We can maybe introduce something like of_clk_get_parent_rate, as we have analogous stuff for getting parent names and indexes. Without introducing a new helper you could probably just do:

	clk_ref = clk_get_parent_by_index(dpll_clk, 0);
	ref_rate = clk_get_rate(clk_ref);

	clk_bypass = clk_get_parent_by_index(dpll_clk, 1);
	bypass_rate = clk_get_rate(clk_bypass);
Quoted 7 lines

Currently the semantics around this call are weird. It seems like it would create a new struct clk pointer but it does not. So don't call clk_put on clk_ref and clk_bypass yet. That might change in the future as we iron out this brave new world that we all live in. Probably best to leave a FIXME in there.

Stephen & Tomeu, let me know if I got any of that wrong.

The plan is to make clk_get_parent_by_index() return a clk_hw pointer instead of a clk pointer (probably with a new clk_get_parent_hw_by_index() API). Then drivers that are clk providers can deal in struct clk_hw and clk consumers can deal in struct clk, nicely splitting the API between consumers and providers on the structures they use to interact with the framework.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#13 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Mike Turquette
Quoted 69 lines · Mike Turquette

Quoting Tero Kristo (2015-02-02 11:32:01)

Quoted 52 lines · Mike Turquette
Quoted 44 lines

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

Quoted 6 lines

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

Quoted 6 lines

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and struct dpll_data, namely this snippet from arch/arm/mach-omap2/dpll3xxx.c:

         parent = __clk_get_parent(hw->clk);

         if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                 WARN(parent != dd->clk_bypass,
                                 "here0, parent name is %s, bypass name is %s\n",
                                 __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                 r = _omap3_noncore_dpll_bypass(clk);
         } else {
                 WARN(parent != dd->clk_ref,
                                 "here1, parent name is %s, ref name is %s\n",
                                 __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                 r = _omap3_noncore_dpll_lock(clk);
         }

struct dpll_data has members clk_ref and clk_bypass which are struct clk pointers. This was always a bit of a violation of the clk.h contract since drivers are not supposed to deref struct clk pointers. Now that we generate unique pointers for each call to clk_get (clk_ref & clk_bypass are populated by of_clk_get in ti_clk_register_dpll) then the pointer comparisons above will never be equal (even if they resolve down to the same struct clk_core). I added the verbose traces to the WARNs above to illustrate the point: the names are always the same but the pointers differ.

Quoted 6 lines

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

Quoted 5 lines

The problem is we still need to be able to get runtime parent clock rates (the parent rate may change also), so simple index value is not sufficient. We need a handle of some sort to the bypass/ref clocks. The DPLL code generally requires knowledge of the bypass + reference clock rates to work properly, as it calculates the M/N values based on these.

We can maybe introduce something like of_clk_get_parent_rate, as we have analogous stuff for getting parent names and indexes. Without introducing a new helper you could probably just do:

	clk_ref = clk_get_parent_by_index(dpll_clk, 0);
	ref_rate = clk_get_rate(clk_ref);

	clk_bypass = clk_get_parent_by_index(dpll_clk, 1);
	bypass_rate = clk_get_rate(clk_bypass);
Quoted 7 lines

Currently the semantics around this call are weird. It seems like it would create a new struct clk pointer but it does not. So don't call clk_put on clk_ref and clk_bypass yet. That might change in the future as we iron out this brave new world that we all live in. Probably best to leave a FIXME in there.

Stephen & Tomeu, let me know if I got any of that wrong.

I think you got it right, just wanted to mention that we can and probably should make the clk_get_parent_* calls in the consumer API to return per-user clk instances but that we need to make sure first that callers call clk_put afterwards.

This should also allow us to remove the reference to struct clk from clk_hw, which is at best awkward.

Regards,

Tomeu

Quoted 84 lines
Quoted 4 lines

Shall I change the DPLL code to check against clk_hw pointers or what is the preferred approach here? The patch at the end does this and fixes the dpll related warnings.

Yes, for now that is fine, but feels a bit hacky to me. I don't know honestly, let me sleep on it. Anyways for 3.20 that is perfectly fine but we might want to switch to something like the scheme above.

Quoted 3 lines

Btw, the rate constraints patch broke boot for me completely, but sounds like you reverted it already.

Fixed with Stephen's patch from last week. Thanks for dealing with all the breakage so promptly. It has helped a lot!

Regards,
Mike

Quoted 65 lines

-Tero

--------------------

Author: Tero Kristo <[email protected]>
Date:   Mon Feb 2 17:19:17 2015 +0200

ARM: OMAP3+: clock: dpll: fix logic for comparing parent clocks

DPLL code uses reference and bypass clock pointers for determining
runtime
properties for these clocks, like parent clock rates.

As clock API now returns per-user clock structs, using a global handle in the clock driver code does not work properly anymore. Fix this by using the clk_hw instead, and comparing this against the parents.

Fixes 59cf3fcf9baf ("clk: Make clk API return per-user struct clk

instances")

arch/arm/mach-omap2/dpll3xxx.c +6 -5
index c2da2a0..49752d7 100644
@@ -410,7 +410,7 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
        struct clk_hw_omap *clk = to_clk_hw_omap(hw);
        int r;
        struct dpll_data *dd;
-       struct clk *parent;
+       struct clk_hw *parent;
 
        dd = clk->dpll_data;
        if (!dd)
@@ -427,13 +427,13 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
                }
        }
 
-       parent = __clk_get_parent(hw->clk);
+       parent = __clk_get_hw(__clk_get_parent(hw->clk));
 
        if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
-               WARN_ON(parent != dd->clk_bypass);
+               WARN_ON(parent != __clk_get_hw(dd->clk_bypass));
                r = _omap3_noncore_dpll_bypass(clk);
        } else {
-               WARN_ON(parent != dd->clk_ref);
+               WARN_ON(parent != __clk_get_hw(dd->clk_ref));
                r = _omap3_noncore_dpll_lock(clk);
        }
 
@@ -549,7 +549,8 @@ int omap3_noncore_dpll_set_rate(struct clk_hw *hw, 
 unsigned long rate,
        if (!dd)
                return -EINVAL;
 
-       if (__clk_get_parent(hw->clk) != dd->clk_ref)
+       if (__clk_get_hw(__clk_get_parent(hw->clk)) !=
+           __clk_get_hw(dd->clk_ref))
                return -EINVAL;
 
        if (dd->last_rounded_rate == 0)
#14 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 83 lines · Tomeu Vizoso
Quoted 69 lines · Mike Turquette

Quoting Tero Kristo (2015-02-02 11:32:01)

Quoted 52 lines · Mike Turquette
Quoted 44 lines

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

Quoted 6 lines

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

Quoted 6 lines

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and struct dpll_data, namely this snippet from arch/arm/mach-omap2/dpll3xxx.c:

          parent = __clk_get_parent(hw->clk);

          if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                  WARN(parent != dd->clk_bypass,
                                  "here0, parent name is %s, bypass name is %s\n",
                                  __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                  r = _omap3_noncore_dpll_bypass(clk);
          } else {
                  WARN(parent != dd->clk_ref,
                                  "here1, parent name is %s, ref name is %s\n",
                                  __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                  r = _omap3_noncore_dpll_lock(clk);
          }

struct dpll_data has members clk_ref and clk_bypass which are struct clk pointers. This was always a bit of a violation of the clk.h contract since drivers are not supposed to deref struct clk pointers. Now that we generate unique pointers for each call to clk_get (clk_ref & clk_bypass are populated by of_clk_get in ti_clk_register_dpll) then the pointer comparisons above will never be equal (even if they resolve down to the same struct clk_core). I added the verbose traces to the WARNs above to illustrate the point: the names are always the same but the pointers differ.

Quoted 6 lines

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

Quoted 5 lines

The problem is we still need to be able to get runtime parent clock rates (the parent rate may change also), so simple index value is not sufficient. We need a handle of some sort to the bypass/ref clocks. The DPLL code generally requires knowledge of the bypass + reference clock rates to work properly, as it calculates the M/N values based on these.

We can maybe introduce something like of_clk_get_parent_rate, as we have analogous stuff for getting parent names and indexes. Without introducing a new helper you could probably just do:

	clk_ref = clk_get_parent_by_index(dpll_clk, 0);
	ref_rate = clk_get_rate(clk_ref);

	clk_bypass = clk_get_parent_by_index(dpll_clk, 1);
	bypass_rate = clk_get_rate(clk_bypass);
Quoted 7 lines

Currently the semantics around this call are weird. It seems like it would create a new struct clk pointer but it does not. So don't call clk_put on clk_ref and clk_bypass yet. That might change in the future as we iron out this brave new world that we all live in. Probably best to leave a FIXME in there.

Stephen & Tomeu, let me know if I got any of that wrong.

I think you got it right, just wanted to mention that we can and probably should make the clk_get_parent_* calls in the consumer API to return per-user clk instances but that we need to make sure first that callers call clk_put afterwards.

Quoted 4 lines

This should also allow us to remove the reference to struct clk from clk_hw, which is at best awkward.

Regards,

For the DPLL code it should just be fine to be able to get the current parent index (not parent clock handle), and read a parent clock rate based on an arbitrary index (not just the current one.) I don't think there is any other need for having the clk_ref / clk_bypass clock handles around.

-Tero

Quoted 88 lines

Tomeu

Quoted 84 lines
Quoted 4 lines

Shall I change the DPLL code to check against clk_hw pointers or what is the preferred approach here? The patch at the end does this and fixes the dpll related warnings.

Yes, for now that is fine, but feels a bit hacky to me. I don't know honestly, let me sleep on it. Anyways for 3.20 that is perfectly fine but we might want to switch to something like the scheme above.

Quoted 3 lines

Btw, the rate constraints patch broke boot for me completely, but sounds like you reverted it already.

Fixed with Stephen's patch from last week. Thanks for dealing with all the breakage so promptly. It has helped a lot!

Regards,
Mike

Quoted 65 lines

-Tero

--------------------

Author: Tero Kristo <[email protected]>
Date:   Mon Feb 2 17:19:17 2015 +0200

ARM: OMAP3+: clock: dpll: fix logic for comparing parent clocks

DPLL code uses reference and bypass clock pointers for determining
runtime
properties for these clocks, like parent clock rates.

As clock API now returns per-user clock structs, using a global handle in the clock driver code does not work properly anymore. Fix this by using the clk_hw instead, and comparing this against the parents.

Fixes 59cf3fcf9baf ("clk: Make clk API return per-user struct clk

instances")

arch/arm/mach-omap2/dpll3xxx.c +6 -5
index c2da2a0..49752d7 100644
@@ -410,7 +410,7 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
         struct clk_hw_omap *clk = to_clk_hw_omap(hw);
         int r;
         struct dpll_data *dd;
-       struct clk *parent;
+       struct clk_hw *parent;
 
         dd = clk->dpll_data;
         if (!dd)
@@ -427,13 +427,13 @@ int omap3_noncore_dpll_enable(struct clk_hw *hw)
                 }
         }
 
-       parent = __clk_get_parent(hw->clk);
+       parent = __clk_get_hw(__clk_get_parent(hw->clk));
 
         if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
-               WARN_ON(parent != dd->clk_bypass);
+               WARN_ON(parent != __clk_get_hw(dd->clk_bypass));
                 r = _omap3_noncore_dpll_bypass(clk);
         } else {
-               WARN_ON(parent != dd->clk_ref);
+               WARN_ON(parent != __clk_get_hw(dd->clk_ref));
                 r = _omap3_noncore_dpll_lock(clk);
         }
 
@@ -549,7 +549,8 @@ int omap3_noncore_dpll_set_rate(struct clk_hw *hw,
 unsigned long rate,
         if (!dd)
                 return -EINVAL;
 
-       if (__clk_get_parent(hw->clk) != dd->clk_ref)
+       if (__clk_get_hw(__clk_get_parent(hw->clk)) !=
+           __clk_get_hw(dd->clk_ref))
                 return -EINVAL;
 
         if (dd->last_rounded_rate == 0)
maintainer
#15 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 7 lines · Tero Kristo
Quoted 6 lines · Tomeu Vizoso

I think you got it right, just wanted to mention that we can and probably should make the clk_get_parent_* calls in the consumer API to return per-user clk instances but that we need to make sure first that callers call clk_put afterwards.

Quoted 7 lines
Quoted 2 lines

This should also allow us to remove the reference to struct clk from clk_hw, which is at best awkward.

For the DPLL code it should just be fine to be able to get the current parent index (not parent clock handle), and read a parent clock rate based on an arbitrary index (not just the current one.) I don't think there is any other need for having the clk_ref / clk_bypass clock handles around.

I'd like to avoid the situation where the children have know that certain parent index and parent rate means bypass mode for both parent and children.

Maybe we can hide that knowledge into some Linux generic PLL code so the children can get the PLL output as a mux clock. For a PLL, there can be three mux clock outputs: refclk*multi/div, bypass clock, or no output.

Regards,

Tony

#16 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Mike Turquette
Quoted 33 lines · Mike Turquette

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

Quoted 6 lines

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

Quoted 6 lines

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and struct dpll_data, namely this snippet from arch/arm/mach-omap2/dpll3xxx.c:

        parent = __clk_get_parent(hw->clk);

        if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                WARN(parent != dd->clk_bypass,
                                "here0, parent name is %s, bypass name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                r = _omap3_noncore_dpll_bypass(clk);
        } else {
                WARN(parent != dd->clk_ref,
                                "here1, parent name is %s, ref name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                r = _omap3_noncore_dpll_lock(clk);
        }
Quoted 3 lines

struct dpll_data has members clk_ref and clk_bypass which are struct clk pointers. This was always a bit of a violation of the clk.h contract since drivers are not supposed to deref struct clk pointers.

Julia,

Is there a way we can write a coccinelle script to check for this? The goal being to find all drivers that are comparing struct clk pointers or attempting to dereference them. There are probably other frameworks that could use the same type of check (regulator, gpiod, reset, pwm, etc.). Probably anything that has a get/put API.

-Stephen

Quoted 24 lines

Now that we
generate unique pointers for each call to clk_get (clk_ref & clk_bypass are populated by of_clk_get in ti_clk_register_dpll) then the pointer comparisons above will never be equal (even if they resolve down to the same struct clk_core). I added the verbose traces to the WARNs above to illustrate the point: the names are always the same but the pointers differ.

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

        dd->clk_ref = of_clk_get(node, 0);
        dd->clk_bypass = of_clk_get(node, 1);

Tony, the same problem affects the FAPLL code which copy/pastes some of the DPLL code.

Thoughts?

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#17 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 38 lines · Stephen Boyd
Quoted 33 lines · Mike Turquette

Quoting Tomeu Vizoso (2015-01-23 03:03:30)

Quoted 6 lines

Moves clock state to struct clk_core, but takes care to change as little API as possible.

struct clk_hw still has a pointer to a struct clk, which is the implementation's per-user clk instance, for backwards compatibility.

Quoted 6 lines

The struct clk that clk_get_parent() returns isn't owned by the caller, but by the clock implementation, so the former shouldn't call clk_put() on it.

Because some boards in mach-omap2 still register clocks statically, their clock registration had to be updated to take into account that the clock information is stored in struct clk_core now.

Tero, Paul & Tony,

Tomeu's patch unveils a problem with omap3_noncore_dpll_enable and struct dpll_data, namely this snippet from arch/arm/mach-omap2/dpll3xxx.c:

        parent = __clk_get_parent(hw->clk);

        if (__clk_get_rate(hw->clk) == __clk_get_rate(dd->clk_bypass)) {
                WARN(parent != dd->clk_bypass,
                                "here0, parent name is %s, bypass name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_bypass));
                r = _omap3_noncore_dpll_bypass(clk);
        } else {
                WARN(parent != dd->clk_ref,
                                "here1, parent name is %s, ref name is %s\n",
                                __clk_get_name(parent), __clk_get_name(dd->clk_ref));
                r = _omap3_noncore_dpll_lock(clk);
        }
Quoted 3 lines

struct dpll_data has members clk_ref and clk_bypass which are struct clk pointers. This was always a bit of a violation of the clk.h contract since drivers are not supposed to deref struct clk pointers.

Quoted 7 lines

Julia,

Is there a way we can write a coccinelle script to check for this? The goal being to find all drivers that are comparing struct clk pointers or attempting to dereference them. There are probably other frameworks that could use the same type of check (regulator, gpiod, reset, pwm, etc.). Probably anything that has a get/put API.

Comparing or dereferencing pointers of a particular type should be straightforward to check for. Is there an example of how to use the parent_index value to fix the problem?

julia

Quoted 32 lines

-Stephen

Quoted 24 lines

Now that we
generate unique pointers for each call to clk_get (clk_ref & clk_bypass are populated by of_clk_get in ti_clk_register_dpll) then the pointer comparisons above will never be equal (even if they resolve down to the same struct clk_core). I added the verbose traces to the WARNs above to illustrate the point: the names are always the same but the pointers differ.

AFAICT this doesn't break anything, but booting on OMAP3+ results in noisy WARNs.

I think the correct fix is to replace clk_bypass and clk_ref pointers with a simple integer parent_index. In fact we already have this index. See how the pointers are populated in ti_clk_register_dpll:

        dd->clk_ref = of_clk_get(node, 0);
        dd->clk_bypass = of_clk_get(node, 1);

Tony, the same problem affects the FAPLL code which copy/pastes some of the DPLL code.

Thoughts?

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#18 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 5 lines · Julia Lawall
Quoted 2 lines · Stephen Boyd

Julia,

Quoted 9 lines
Quoted 5 lines

Is there a way we can write a coccinelle script to check for this? The goal being to find all drivers that are comparing struct clk pointers or attempting to dereference them. There are probably other frameworks that could use the same type of check (regulator, gpiod, reset, pwm, etc.). Probably anything that has a get/put API.

Comparing or dereferencing pointers of a particular type should be straightforward to check for. Is there an example of how to use the parent_index value to fix the problem?

I'm not sure how to fix this case with parent_index values generically. I imagine it would be highly specific to the surrounding code to the point where we couldn't fix it automatically. For example, if it's a clk consumer (not provider like in this case) using an index wouldn't be the right fix. We would need some sort of clk API that we don't currently have and I would wonder why clock consumers even care to compare such pointers in the first place.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#19 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoting Stephen Boyd (2015-02-02 14:35:59)

Quoted 16 lines · Julia Lawall
Quoted 14 lines · Stephen Boyd
Quoted 8 lines

Julia,

Is there a way we can write a coccinelle script to check for this? The goal being to find all drivers that are comparing struct clk pointers or attempting to dereference them. There are probably other frameworks that could use the same type of check (regulator, gpiod, reset, pwm, etc.). Probably anything that has a get/put API.

Comparing or dereferencing pointers of a particular type should be straightforward to check for. Is there an example of how to use the parent_index value to fix the problem?

Quoted 7 lines

I'm not sure how to fix this case with parent_index values generically. I imagine it would be highly specific to the surrounding code to the point where we couldn't fix it automatically. For example, if it's a clk consumer (not provider like in this case) using an index wouldn't be the right fix. We would need some sort of clk API that we don't currently have and I would wonder why clock consumers even care to compare such pointers in the first place.

Ack. Is there precedent for a "Don't do that" kind of response?

Regards,
Mike

Quoted 5 lines
Signature · 3 lines
-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#20 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Hello,
Julia asked me to have a look and see if I can help.

Quoted 27 lines · Mike Turquette

Quoting Stephen Boyd (2015-02-02 14:35:59)

Quoted 14 lines · Julia Lawall
Quoted 13 lines · Stephen Boyd
Quoted 8 lines

Julia,

Is there a way we can write a coccinelle script to check for this? The goal being to find all drivers that are comparing struct clk pointers or attempting to dereference them. There are probably other frameworks that could use the same type of check (regulator, gpiod, reset, pwm, etc.). Probably anything that has a get/put API.

Comparing or dereferencing pointers of a particular type should be straightforward to check for. Is there an example of how to use the parent_index value to fix the problem?

Quoted 7 lines

I'm not sure how to fix this case with parent_index values generically. I imagine it would be highly specific to the surrounding code to the point where we couldn't fix it automatically. For example, if it's a clk consumer (not provider like in this case) using an index wouldn't be the right fix. We would need some sort of clk API that we don't currently have and I would wonder why clock consumers even care to compare such pointers in the first place.

Ack. Is there precedent for a "Don't do that" kind of response?

Regards,
Mike

Quoted 8 lines
Quoted 4 lines
Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

_______________________________________________
Cocci mailing list
[email protected]
https://systeme.lip6.fr/mailman/listinfo/cocci

I have found these three cases using Coccinnelle in the mach-omap2 directory.

./arch/arm/mach-omap2/clkt_clksel.c
@@ -67,7 +67,6 @@ static const struct clksel *_get_clksel_

                return NULL;

        for (clks = clk->clksel; clks->parent; clks++)
                if (clks->parent == src_clk)
                        break; /* Found the requested parent */

        if (!clks->parent) {
./arch/arm/mach-omap2/dpll3xxx.c
@@ -551,7 +549,6 @@ int omap3_noncore_dpll_set_rate(struct c
        if (!dd)
                return -EINVAL;

        if (__clk_get_parent(hw->clk) != dd->clk_ref)
                return -EINVAL;

        if (dd->last_rounded_rate == 0)
./arch/arm/mach-omap2/timer.c
@@ -298,7 +298,6 @@ static int __init omap_dm_timer_init_one
        if (IS_ERR(src))
                return PTR_ERR(src);

        if (clk_get_parent(timer->fclk) != src) {
                r = clk_set_parent(timer->fclk, src);
                if (r < 0) {
                        pr_warn("%s: %s cannot set source\n", __func__,

Are they instances of your issue?
Do you have any suggestion on how to fix them?
If you want me to I can enlarge the search to other directories.

Quentin

#21 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 10 lines · Quentin Lambert

Hello,
Julia asked me to have a look and see if I can help.

I have found these three cases using Coccinnelle in the mach-omap2 directory.

./arch/arm/mach-omap2/clkt_clksel.c

Review static const struct clksel *_get_clksel_ · line 70

Quoted 4 lines
(quoted diff) +0 -0
@@ -67,7 +67,6 @@ static const struct clksel *_get_clksel_
                return NULL;
 
        for (clks = clk->clksel; clks->parent; clks++)
                if (clks->parent == src_clk)

This probably needs to compare hw pointers again given that it's in the clk-provider code. It would be nice if we could use indices instead though.

Review int omap3_noncore_dpll_set_rate(struct c · line 552

Quoted 9 lines
(quoted diff) +0 -0
                        break; /* Found the requested parent */
 
        if (!clks->parent) {

./arch/arm/mach-omap2/dpll3xxx.c

(quoted diff) +0 -0
@@ -551,7 +549,6 @@ int omap3_noncore_dpll_set_rate(struct c
        if (!dd)
                return -EINVAL;
 
        if (__clk_get_parent(hw->clk) != dd->clk_ref)

This one is what this thread has started on about. Comparing hw pointers is ok for now...

Review static int __init omap_dm_timer_init_one · line 301

Quoted 9 lines
(quoted diff) +0 -0
                return -EINVAL;
 
        if (dd->last_rounded_rate == 0)

./arch/arm/mach-omap2/timer.c

(quoted diff) +0 -0
@@ -298,7 +298,6 @@ static int __init omap_dm_timer_init_one
        if (IS_ERR(src))
                return PTR_ERR(src);
 
        if (clk_get_parent(timer->fclk) != src) {

This one looks like an optimization. We can just always try to set the parent and if it's already the current parent then the core should bail out early without an error. So the fix would be to just remove the if condition.

Quoted 6 lines
(quoted diff) +0 -0
                r = clk_set_parent(timer->fclk, src);
                if (r < 0) {
                        pr_warn("%s: %s cannot set source\n", __func__,

Are they instances of your issue?

Yes these all look like instances of the problem.

Quoted 1 line

If you want me to I can enlarge the search to other directories.

Yes please do. And if you could share the coccinelle patch that would be great. Thanks.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#22 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 4 lines · Stephen Boyd
Quoted 1 line

If you want me to I can enlarge the search to other directories.

Yes please do. And if you could share the coccinelle patch that would be great. Thanks.

structclk.cocci is the coccinelle patch
structclk-arm.patch is the result I got when applying it to the arch/arm directory

Is there anything else I can do to help?

tmp/nothing/mach-imx/mach-imx6q.c renamed from arch/arm/mach-imx/mach-imx6q.c +0 -1
diff -u -p ./arch/arm/mach-imx/mach-imx6q.c /tmp/nothing/mach-imx/mach-imx6q.c
@@ -211,7 +211,6 @@ static void __init imx6q_1588_init(void)
 	 * set bit IOMUXC_GPR1[21].  Or the PTP clock must be from pad
 	 * (external OSC), and we need to clear the bit.
 	 */
-	clksel = ptp_clk == enet_ref ? IMX6Q_GPR1_ENET_CLK_SEL_ANATOP :
 				       IMX6Q_GPR1_ENET_CLK_SEL_PAD;
 	gpr = syscon_regmap_lookup_by_compatible("fsl,imx6q-iomuxc-gpr");
 	if (!IS_ERR(gpr))
tmp/nothing/mach-shmobile/clock-r8a73a4.c renamed from arch/arm/mach-shmobile/clock-r8a73a4.c +0 -1
diff -u -p ./arch/arm/mach-shmobile/clock-r8a73a4.c /tmp/nothing/mach-shmobile/clock-r8a73a4.c
@@ -139,7 +139,6 @@ static int pll_set_parent(struct clk *cl
 
 	/* Search the parent */
 	for (i = 0; i < clk->parent_num; i++)
-		if (clk->parent_table[i] == parent)
 			break;
 
 	if (i == clk->parent_num)
tmp/nothing/mach-shmobile/clock-sh7372.c renamed from arch/arm/mach-shmobile/clock-sh7372.c +0 -1
diff -u -p ./arch/arm/mach-shmobile/clock-sh7372.c /tmp/nothing/mach-shmobile/clock-sh7372.c
@@ -223,7 +223,6 @@ static int pllc2_set_parent(struct clk *
 
 	/* Search the parent */
 	for (i = 0; i < clk->parent_num; i++)
-		if (clk->parent_table[i] == parent)
 			break;
 
 	if (i == clk->parent_num)
tmp/nothing/mach-shmobile/clock-r8a7740.c renamed from arch/arm/mach-shmobile/clock-r8a7740.c +0 -1
diff -u -p ./arch/arm/mach-shmobile/clock-r8a7740.c /tmp/nothing/mach-shmobile/clock-r8a7740.c
@@ -195,7 +195,6 @@ static int usb24s_set_parent(struct clk
 
 	/* Search the parent */
 	for (i = 0; i < clk->parent_num; i++)
-		if (clk->parent_table[i] == parent)
 			break;
 
 	if (i == clk->parent_num)
tmp/nothing/mach-omap2/clkt_clksel.c renamed from arch/arm/mach-omap2/clkt_clksel.c +0 -1
diff -u -p ./arch/arm/mach-omap2/clkt_clksel.c /tmp/nothing/mach-omap2/clkt_clksel.c
@@ -67,7 +67,6 @@ static const struct clksel *_get_clksel_
 		return NULL;
 
 	for (clks = clk->clksel; clks->parent; clks++)
-		if (clks->parent == src_clk)
 			break; /* Found the requested parent */
 
 	if (!clks->parent) {
tmp/nothing/mach-omap2/timer.c renamed from arch/arm/mach-omap2/timer.c +0 -1
diff -u -p ./arch/arm/mach-omap2/timer.c /tmp/nothing/mach-omap2/timer.c
@@ -298,7 +298,6 @@ static int __init omap_dm_timer_init_one
 	if (IS_ERR(src))
 		return PTR_ERR(src);
 
-	if (clk_get_parent(timer->fclk) != src) {
 		r = clk_set_parent(timer->fclk, src);
 		if (r < 0) {
 			pr_warn("%s: %s cannot set source\n", __func__,

/// Find any attempt to compare or dereference struct clk pointers. ///
// Confidence: High
// Copyright: (C) 2015 Quentin Lambert, INRIA/LiP6. GPLv2 // URL: http://coccinelle.lip6.fr/
// Options: --recursive-includes --relax-include-path
// Options: --include-headers-for-types

virtual context
virtual org
virtual report

// ----------------------------------------------------------------------------

@comparison_dereference depends on context || org || report@ struct clk *x1, x2;
position j0;
@@

(
* x1@j0 == x2
|
* x1@j0 != x2
|
* *x1@j0
)

// ----------------------------------------------------------------------------

@script:python comparison_dereference_org depends on org@ j0 << comparison_dereference.j0;
@@

msg = "WARNING trying to compare or dereference struct clk pointers." coccilib.org.print_todo(j0[0], msg)

// ----------------------------------------------------------------------------

@script:python comparison_dereference_report depends on report@ j0 << comparison_dereference.j0;
@@

msg = "WARNING trying to compare or dereference struct clk pointers." coccilib.report.print_report(j0[0], msg)

#23 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Sorry let me do that properly.

Quoted 12 lines · Quentin Lambert
Quoted 4 lines · Stephen Boyd
Quoted 1 line

If you want me to I can enlarge the search to other directories.

Yes please do. And if you could share the coccinelle patch that would be great. Thanks.

structclk.cocci is the coccinelle patch
structclk-arm.patch is the result I got when applying it to the arch/arm directory

Is there anything else I can do to help?

These are the new instances I found by applying the patch to arch/arm directory:

./arch/arm/mach-imx/mach-imx6q.c
@@ -211,7 +211,6 @@ static void __init imx6q_1588_init(void)

      * set bit IOMUXC_GPR1[21].  Or the PTP clock must be from pad
      * (external OSC), and we need to clear the bit.
      */
     clksel = ptp_clk == enet_ref ? IMX6Q_GPR1_ENET_CLK_SEL_ANATOP :
                        IMX6Q_GPR1_ENET_CLK_SEL_PAD;
     gpr = syscon_regmap_lookup_by_compatible("fsl,imx6q-iomuxc-gpr");
     if (!IS_ERR(gpr))

./arch/arm/mach-shmobile/clock-r8a73a4.c
@@ -139,7 +139,6 @@ static int pll_set_parent(struct clk *cl

     /* Search the parent */
     for (i = 0; i < clk->parent_num; i++)
         if (clk->parent_table[i] == parent)
             break;

     if (i == clk->parent_num)

./arch/arm/mach-shmobile/clock-sh7372.c
@@ -223,7 +223,6 @@ static int pllc2_set_parent(struct clk *

     /* Search the parent */
     for (i = 0; i < clk->parent_num; i++)
         if (clk->parent_table[i] == parent)
             break;

     if (i == clk->parent_num)

./arch/arm/mach-shmobile/clock-r8a7740.c
@@ -195,7 +195,6 @@ static int usb24s_set_parent(struct clk

     /* Search the parent */
     for (i = 0; i < clk->parent_num; i++)
         if (clk->parent_table[i] == parent)
             break;

     if (i == clk->parent_num)
#24 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 23 lines · Quentin Lambert

Sorry let me do that properly.

Quoted 7 lines · Quentin Lambert
Quoted 5 lines · Stephen Boyd
Quoted 1 line

If you want me to I can enlarge the search to other directories.

Yes please do. And if you could share the coccinelle patch that would be
great. Thanks.

Quoted 7 lines

structclk.cocci is the coccinelle patch structclk-arm.patch is the result I got when applying it to the arch/arm directory

Is there anything else I can do to help?

These are the new instances I found by applying the patch to arch/arm directory:

./arch/arm/mach-imx/mach-imx6q.c

tmp/nothing/mach-imx/mach-imx6q.c static void __init imx6q_1588_init(void) · line 216

Quoted 7 lines
tmp/nothing/mach-imx/mach-imx6q.c renamed from arch/arm/mach-imx/mach-imx6q.c +0 -0
@@ -211,7 +211,6 @@ static void __init imx6q_1588_init(void)
      * set bit IOMUXC_GPR1[21].  Or the PTP clock must be from pad
      * (external OSC), and we need to clear the bit.
      */
     clksel = ptp_clk == enet_ref ? IMX6Q_GPR1_ENET_CLK_SEL_ANATOP :
                        IMX6Q_GPR1_ENET_CLK_SEL_PAD;
     gpr = syscon_regmap_lookup_by_compatible("fsl,imx6q-iomuxc-gpr");
     if (!IS_ERR(gpr))

This one looks like a real problem and it could probably use a clk_equal(struct clk *a, struct clk *b) function.

Quoted 26 lines

./arch/arm/mach-shmobile/clock-r8a73a4.c

tmp/nothing/mach-shmobile/clock-r8a73a4.c renamed from arch/arm/mach-shmobile/clock-r8a73a4.c +0 -0
@@ -139,7 +139,6 @@ static int pll_set_parent(struct clk *cl
 
     /* Search the parent */
     for (i = 0; i < clk->parent_num; i++)
         if (clk->parent_table[i] == parent)
             break;
 
     if (i == clk->parent_num)

./arch/arm/mach-shmobile/clock-sh7372.c

tmp/nothing/mach-shmobile/clock-r8a73a4.c renamed from arch/arm/mach-shmobile/clock-r8a73a4.c +0 -0
@@ -223,7 +223,6 @@ static int pllc2_set_parent(struct clk *
 
     /* Search the parent */
     for (i = 0; i < clk->parent_num; i++)
         if (clk->parent_table[i] == parent)
             break;
 
     if (i == clk->parent_num)

./arch/arm/mach-shmobile/clock-r8a7740.c

tmp/nothing/mach-shmobile/clock-r8a73a4.c static int usb24s_set_parent(struct clk · line 144

Quoted 7 lines
tmp/nothing/mach-shmobile/clock-r8a73a4.c renamed from arch/arm/mach-shmobile/clock-r8a73a4.c +0 -0
@@ -195,7 +195,6 @@ static int usb24s_set_parent(struct clk
 
     /* Search the parent */
     for (i = 0; i < clk->parent_num; i++)
         if (clk->parent_table[i] == parent)
             break;
 
     if (i == clk->parent_num)

I don't think shmobile uses the CCF so this should be safe, but we should probably fix them up to also use a clk_equal() function that just does pointer comparisons.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#25 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Quentin Lambert
Quoted 13 lines · Quentin Lambert
Quoted 4 lines · Stephen Boyd
Quoted 1 line

If you want me to I can enlarge the search to other directories.

Yes please do. And if you could share the coccinelle patch that would be great. Thanks.

structclk.cocci is the coccinelle patch structclk-arm.patch is the result I got when applying it to the arch/arm directory

Is there anything else I can do to help?

Thanks for the coccinelle patch. Thinking more about it, I don't think we care if the pointer is dereferenced because that would require a definition of struct clk and that is most likely not the case outside of the clock framework. Did you scan the entire kernel? I'm running it now but it seems to be taking a while.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#26 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 4 lines · Stephen Boyd

Thanks for the coccinelle patch. Thinking more about it, I don't think we care if the pointer is dereferenced because that would require a definition of struct clk and that is most likely not the case outside of the clock framework. Did you scan the entire kernel?

No I haven't.

Quoted 3 lines

I'm running it now
but it seems to be taking a while.

Yes, that's why, as a first step, I chose to limit the scan to the arm directory.

#27 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 6 lines · Quentin Lambert
Quoted 4 lines · Stephen Boyd

Thanks for the coccinelle patch. Thinking more about it, I don't think we care if the pointer is dereferenced because that would require a definition of struct clk and that is most likely not the case outside of the clock framework. Did you scan the entire kernel?

Quoted 6 lines

No I haven't.

Quoted 3 lines

I'm running it now
but it seems to be taking a while.

Yes, that's why, as a first step, I chose to limit the scan to the arm directory.

Are you sure to be using all of the options provided:

// Options: --recursive-includes --relax-include-path // Options: --include-headers-for-types

And are you using 1.0.0-rc23 or 1.0.0-rc24? Those should save parsed header files so that they don't have to be parsed over and over.

If you are using rc24, then you can use the -j option for parallelism. But then you should also use an option like --chunksize 10 (I don't know what number would be good), because the default is chunksize 1, and in that case the saved parsed header files are not reused, because the fies are all processed individually. In general, it is only the files within a chunk that will share parsed header files.

julia

#28 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 22 lines · Julia Lawall
Quoted 11 lines · Quentin Lambert
Quoted 4 lines · Stephen Boyd

Thanks for the coccinelle patch. Thinking more about it, I don't think we care if the pointer is dereferenced because that would require a definition of struct clk and that is most likely not the case outside of the clock framework. Did you scan the entire kernel?

No I haven't.

Quoted 3 lines

I'm running it now
but it seems to be taking a while.

Yes, that's why, as a first step, I chose to limit the scan to the arm directory.

Are you sure to be using all of the options provided:

// Options: --recursive-includes --relax-include-path // Options: --include-headers-for-types

And are you using 1.0.0-rc23 or 1.0.0-rc24? Those should save parsed header files so that they don't have to be parsed over and over.

Quoted 6 lines

If you are using rc24, then you can use the -j option for parallelism. But then you should also use an option like --chunksize 10 (I don't know what number would be good), because the default is chunksize 1, and in that case the saved parsed header files are not reused, because the fies are all processed individually. In general, it is only the files within a chunk that will share parsed header files.

Thanks for the info.

$ spatch --version
spatch version 1.0.0-rc22 with Python support and with PCRE support

so I guess I need to update. I tried throwing it into scripts/coccinelle/misc and then did

make O=../obj/ coccicheck COCCI=../kernel/scripts/coccinelle/misc/structclk.cocci

at the toplevel but it didn't find anything.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#29 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Stephen Boyd
Quoted 10 lines · Stephen Boyd
Quoted 9 lines · Quentin Lambert
Quoted 4 lines · Stephen Boyd
Quoted 1 line

If you want me to I can enlarge the search to other directories.

Yes please do. And if you could share the coccinelle patch that would be great. Thanks.

structclk.cocci is the coccinelle patch structclk-arm.patch is the result I got when applying it to the arch/arm directory

Quoted 9 lines
Quoted 3 lines

Is there anything else I can do to help?

Thanks for the coccinelle patch. Thinking more about it, I don't think we care if the pointer is dereferenced because that would require a definition of struct clk and that is most likely not the case outside of the clock framework. Did you scan the entire kernel? I'm running it now but it seems to be taking a while.

I ran the script on all files that include <linux/clk.h>. I've also trimmed out mips and unicore32 because they're not using the common clock framework.

tmp/cocci-output-11792-b62223-mach-imx6q.c renamed from arm/mach-imx/mach-imx6q.c +0 -1
diff = 
@@ -211,7 +211,6 @@ static void __init imx6q_1588_init(void)
 	 * set bit IOMUXC_GPR1[21].  Or the PTP clock must be from pad
 	 * (external OSC), and we need to clear the bit.
 	 */
-	clksel = ptp_clk == enet_ref ? IMX6Q_GPR1_ENET_CLK_SEL_ANATOP :
 				       IMX6Q_GPR1_ENET_CLK_SEL_PAD;
 	gpr = syscon_regmap_lookup_by_compatible("fsl,imx6q-iomuxc-gpr");
 	if (!IS_ERR(gpr))
tmp/cocci-output-12321-a5f298-armada_510.c renamed from gpu/drm/armada/armada_510.c +0 -1
diff = 
@@ -53,7 +53,6 @@ static int armada510_crtc_compute_clock(
 	if (IS_ERR(clk))
 		return PTR_ERR(clk);
 
-	if (dcrtc->clk != clk) {
 		ret = clk_prepare_enable(clk);
 		if (ret)
 			return ret;

drivers/gpu/drm/armada/armada_510.c:56:5-15: WARNING trying to compare or dereference struct clk pointers.

tmp/cocci-output-12679-3c5195-pwm-atmel-hlcdc.c renamed from pwm/pwm-atmel-hlcdc.c +0 -1
diff = 
@@ -91,7 +91,6 @@ static int atmel_hlcdc_pwm_config(struct
 
 	pwmcfg = ATMEL_HLCDC_PWMPS(pres);
 
-	if (new_clk != chip->cur_clk) {
 		u32 gencfg = 0;
 		int ret;
 

drivers/pwm/pwm-atmel-hlcdc.c:94:5-12: WARNING trying to compare or dereference struct clk pointers.

tmp/cocci-output-12827-715e72-samsung.c renamed from tty/serial/samsung.c +0 -1
diff = 
@@ -750,7 +750,6 @@ static void s3c24xx_serial_set_termios(s
 
 	/* check to see if we need  to change clock source */
 
-	if (ourport->baudclk != clk) {
 		s3c24xx_serial_setsource(port, clk_sel);
 
 		if (!IS_ERR(ourport->baudclk)) {

drivers/tty/serial/samsung.c:753:5-21: WARNING trying to compare or dereference struct clk pointers.

tmp/cocci-output-13020-d518c3-fsl_esai.c renamed from soc/fsl/fsl_esai.c +0 -1
diff = 
@@ -269,7 +269,6 @@ static int fsl_esai_set_dai_sysclk(struc
 	}
 
 	/* Only EXTAL source can be output directly without using PSR and PM */
-	if (ratio == 1 && clksrc == esai_priv->extalclk) {
 		/* Bypass all the dividers if not being needed */
 		ecr |= tx ? ESAI_ECR_ETO : ESAI_ECR_ERO;
 		goto out;

sound/soc/fsl/fsl_esai.c:272:19-25: WARNING trying to compare or dereference struct clk pointers.

tmp/cocci-output-13024-7acb1d-fsl_spdif.c renamed from soc/fsl/fsl_spdif.c +0 -2
diff = 
@@ -1054,7 +1054,6 @@ static u32 fsl_spdif_txclk_caldiv(struct
 				enum spdif_txrate index, bool round)
 {
 	const u32 rate[] = { 32000, 44100, 48000, 96000, 192000 };
-	bool is_sysclk = clk == spdif_priv->sysclk;
 	u64 rate_ideal, rate_actual, sub;
 	u32 sysclk_dfmin, sysclk_dfmax;
 	u32 txclk_df, sysclk_df, arate;
@@ -1148,7 +1147,6 @@ static int fsl_spdif_probe_txclk(struct
 			spdif_priv->txclk_src[index], rate[index]);
 	dev_dbg(&pdev->dev, "use txclk df %d for %dHz sample rate\n",
 			spdif_priv->txclk_df[index], rate[index]);
-	if (spdif_priv->txclk[index] == spdif_priv->sysclk)
 		dev_dbg(&pdev->dev, "use sysclk df %d for %dHz sample rate\n",
 				spdif_priv->sysclk_df[index], rate[index]);
 	dev_dbg(&pdev->dev, "the best rate for %dHz sample rate is %dHz\n",

sound/soc/fsl/fsl_spdif.c:1151:5-29: WARNING trying to compare or dereference struct clk pointers. sound/soc/fsl/fsl_spdif.c:1057:18-21: WARNING trying to compare or dereference struct clk pointers.

tmp/cocci-output-13041-3200a6-kirkwood-i2s.c renamed from soc/kirkwood/kirkwood-i2s.c +0 -1
diff = 
@@ -579,7 +579,6 @@ static int kirkwood_i2s_dev_probe(struct
 		if (PTR_ERR(priv->extclk) == -EPROBE_DEFER)
 			return -EPROBE_DEFER;
 	} else {
-		if (priv->extclk == priv->clk) {
 			devm_clk_put(&pdev->dev, priv->extclk);
 			priv->extclk = ERR_PTR(-EINVAL);
 		} else {

sound/soc/kirkwood/kirkwood-i2s.c:582:6-18: WARNING trying to compare or dereference struct clk pointers.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#30 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

tmp/cocci-output-11792-b62223-mach-imx6q.c static void __init imx6q_1588_init(void) · line 216

Quoted 11 lines · Stephen Boyd

diff =

tmp/cocci-output-11792-b62223-mach-imx6q.c renamed from arm/mach-imx/mach-imx6q.c +0 -1
@@ -211,7 +211,6 @@ static void __init imx6q_1588_init(void)
 	 * set bit IOMUXC_GPR1[21].  Or the PTP clock must be from pad
 	 * (external OSC), and we need to clear the bit.
 	 */
-	clksel = ptp_clk == enet_ref ? IMX6Q_GPR1_ENET_CLK_SEL_ANATOP :
 				       IMX6Q_GPR1_ENET_CLK_SEL_PAD;
 	gpr = syscon_regmap_lookup_by_compatible("fsl,imx6q-iomuxc-gpr");
 	if (!IS_ERR(gpr))

Any idea how to do the comparison here? Or should we rely that the bootloader sets this properly (it managed already to select a frequency)? The phy has no clock node in current DT's so we can check this.

Sebastian

#31 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 16 lines · Sebastian Andrzej Siewior

tmp/cocci-output-11792-b62223-mach-imx6q.c static void __init imx6q_1588_init(void) · line 216

Quoted 11 lines · Stephen Boyd

diff =

tmp/cocci-output-11792-b62223-mach-imx6q.c renamed from arm/mach-imx/mach-imx6q.c +0 -1
@@ -211,7 +211,6 @@ static void __init imx6q_1588_init(void)
 	 * set bit IOMUXC_GPR1[21].  Or the PTP clock must be from pad
 	 * (external OSC), and we need to clear the bit.
 	 */
-	clksel = ptp_clk == enet_ref ? IMX6Q_GPR1_ENET_CLK_SEL_ANATOP :
 				       IMX6Q_GPR1_ENET_CLK_SEL_PAD;
 	gpr = syscon_regmap_lookup_by_compatible("fsl,imx6q-iomuxc-gpr");
 	if (!IS_ERR(gpr))

Any idea how to do the comparison here? Or should we rely that the bootloader sets this properly (it managed already to select a frequency)? The phy has no clock node in current DT's so we can check this.

This has been fixed by adding a clk_is_match() helper and using that to compare instead of comparing raw pointers. It would be nice if we could replace the patch with something else that doesn't require this helper though. It looks like this is static board configuration, so I wonder why we didn't just have a DT property that indicates how the gpr should be configured for this particular board.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#32 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 18 lines · Stephen Boyd
Quoted 16 lines · Sebastian Andrzej Siewior

tmp/cocci-output-11792-b62223-mach-imx6q.c static void __init imx6q_1588_init(void) · line 216

Quoted 11 lines · Stephen Boyd

diff =

tmp/cocci-output-11792-b62223-mach-imx6q.c renamed from arm/mach-imx/mach-imx6q.c +0 -1
@@ -211,7 +211,6 @@ static void __init imx6q_1588_init(void)
 	 * set bit IOMUXC_GPR1[21].  Or the PTP clock must be from pad
 	 * (external OSC), and we need to clear the bit.
 	 */
-	clksel = ptp_clk == enet_ref ? IMX6Q_GPR1_ENET_CLK_SEL_ANATOP :
 				       IMX6Q_GPR1_ENET_CLK_SEL_PAD;
 	gpr = syscon_regmap_lookup_by_compatible("fsl,imx6q-iomuxc-gpr");
 	if (!IS_ERR(gpr))

Any idea how to do the comparison here? Or should we rely that the bootloader sets this properly (it managed already to select a frequency)? The phy has no clock node in current DT's so we can check this.

Quoted 6 lines

This has been fixed by adding a clk_is_match() helper and using that to compare instead of comparing raw pointers. It would be nice if we could replace the patch with something else that doesn't require this helper though. It looks like this is static board configuration, so I wonder why we didn't just have a DT property that indicates how the gpr should be configured for this particular board.

We did not add a DT property for it, because there was already enough info (clock configuration) in DT for kernel to figure it out.

Shawn

#33 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Hi Shawn,

Quoted 2 lines · Shawn Guo

We did not add a DT property for it, because there was already enough info (clock configuration) in DT for kernel to figure it out.

Correct. My understanding is whatever can be figured out without DT should be done that way.

Is there a way to get this clock-select bit set without enable_fec_anatop_clock() in u-boot? Because this one selects the clock and frequency and also sets the proper bit in gpr1. My question is simply why do we need to do this here as well.

Quoted 1 line

Shawn

Sebastian

#34 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 6 lines · Sebastian Andrzej Siewior

Hi Shawn,

Quoted 2 lines · Shawn Guo

We did not add a DT property for it, because there was already enough info (clock configuration) in DT for kernel to figure it out.

Quoted 7 lines

Correct. My understanding is whatever can be figured out without DT should be done that way.

Is there a way to get this clock-select bit set without enable_fec_anatop_clock() in u-boot? Because this one selects the clock and frequency and also sets the proper bit in gpr1. My question is simply why do we need to do this here as well.

I'm not sure I follow your question. But we had been doing this setup in kernel function imx6q_1588_init() for a while. The problem we're having now is that the clk core change broke it since v4.0-rc1. And the fix is already queued there [1].

Shawn

[1] https://git.kernel.org/cgit/linux/kernel/git/clk/linux.git/log/?h=clk-fixes

#35 lore raw

Re: [Cocci] [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Shawn Guo
Quoted 18 lines · Shawn Guo
Quoted 17 lines · Stephen Boyd
Quoted 16 lines · Sebastian Andrzej Siewior

tmp/cocci-output-11792-b62223-mach-imx6q.c static void __init imx6q_1588_init(void) · line 216

Quoted 11 lines · Stephen Boyd

diff =

tmp/cocci-output-11792-b62223-mach-imx6q.c renamed from arm/mach-imx/mach-imx6q.c +0 -1
@@ -211,7 +211,6 @@ static void __init imx6q_1588_init(void)
 	 * set bit IOMUXC_GPR1[21].  Or the PTP clock must be from pad
 	 * (external OSC), and we need to clear the bit.
 	 */
-	clksel = ptp_clk == enet_ref ? IMX6Q_GPR1_ENET_CLK_SEL_ANATOP :
 				       IMX6Q_GPR1_ENET_CLK_SEL_PAD;
 	gpr = syscon_regmap_lookup_by_compatible("fsl,imx6q-iomuxc-gpr");
 	if (!IS_ERR(gpr))

Any idea how to do the comparison here? Or should we rely that the bootloader sets this properly (it managed already to select a frequency)? The phy has no clock node in current DT's so we can check this.

Quoted 8 lines
Quoted 6 lines

This has been fixed by adding a clk_is_match() helper and using that to compare instead of comparing raw pointers. It would be nice if we could replace the patch with something else that doesn't require this helper though. It looks like this is static board configuration, so I wonder why we didn't just have a DT property that indicates how the gpr should be configured for this particular board.

We did not add a DT property for it, because there was already enough info (clock configuration) in DT for kernel to figure it out.

Ok well then if everything is in DT we don't need to be comparing clock pointers at all, right? We should be able to write up some DT parsing logic to figure it out?

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#36 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Tomeu Vizoso

Hi Tomeu,

Quoted 31 lines · Tomeu Vizoso

int __clk_get(struct clk *clk)
{

drivers/clk/clk.c +8 -8
-	if (clk) {
-		if (!try_module_get(clk->owner))
+	struct clk_core *core = !clk ? NULL : clk->core;
+
+	if (core) {
+		if (!try_module_get(core->owner))
 			return 0;
 
-		kref_get(&clk->ref);
+		kref_get(&core->ref);
 	}
 	return 1;
 }
 
-void __clk_put(struct clk *clk)
+static void clk_core_put(struct clk_core *core)
 {
 	struct module *owner;
 
-	if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
-		return;
+	owner = core->owner;
 
 	clk_prepare_lock();
-	owner = clk->owner;
-	kref_put(&clk->ref, __clk_release);
+	kref_put(&core->ref, __clk_release);
 	clk_prepare_unlock();
 

drivers/clk/clk.c EXPORT_SYMBOL_GPL(devm_clk_unregister); · line 2413

Quoted 10 lines
drivers/clk/clk.c +7 -0
 	module_put(owner);
 }
 
+void __clk_put(struct clk *clk)
+{
+	if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+		return;
+
+	clk_core_put(clk->core);
+	kfree(clk);

Why do we have kfree() here? clk_get() doesn't allocate the data structure being freed here. What happens if we do clk_get(), clk_put(), clk_get() on same clock?

I suspect __clk_free_clk() should be called in __clk_release() callback instead, but then there is an issue of safely getting reference to struct clk from struct clk_core pointer.

I tested linux-next on Odroid U3 and booting fails with oopses as below. There is no problems when the above kfree() is commented out.

Quoted 1 line

+}

[    1.345850] Unable to handle kernel paging request at virtual address 00200200
[    1.349319] pgd = c0004000
[    1.352072] [00200200] *pgd=00000000
[    1.355574] Internal error: Oops: 805 [#1] PREEMPT SMP ARM
[    1.361035] Modules linked in:
[    1.364078] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.19.0-rc1-00104-ga251361a-dirty #992
[    1.372405] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
[    1.378483] task: ee00b000 ti: ee088000 task.ti: ee088000
[    1.383879] PC is at __clk_put+0x24/0xd0
[    1.387773] LR is at clk_prepare_lock+0xc/0xec
[    1.392198] pc : [<c03eef38>]    lr : [<c03ec1f4>]    psr: 20000153
[    1.392198] sp : ee089de8  ip : 00000000  fp : 00000000
[    1.403653] r10: ee02f480  r9 : 00000001  r8 : 00000000
[    1.408862] r7 : ee031cc0  r6 : ee089e08  r5 : 00000000  r4 : ee02f480
[    1.415373] r3 : 00100100  r2 : 00200200  r1 : 0000091e  r0 : 00000001
[    1.421884] Flags: nzCv  IRQs on  FIQs off  Mode SVC_32  ISA ARM  Segment kernel
[    1.429261] Control: 10c5387d  Table: 4000404a  DAC: 00000015
[    1.434989] Process swapper/0 (pid: 1, stack limit = 0xee088238)
[    1.440978] Stack: (0xee089de8 to 0xee08a000)
[    1.445321] 9de0:                   ee7c8f14 c03f0ec8 ee089e08 00000000 c0718dc8 00000001
[    1.453480] 9e00: 00000000 c04ee0f0 ee7e0844 00000001 00000181 c04edb58 ee2bd320 00000000
[    1.461639] 9e20: 00000000 c011dc5c ee16a1e0 00000000 00000000 c0718dc8 ee16a1e0 ee2bd1e0
[    1.469798] 9e40: c0641740 ee16a1e0 00000000 ee2bd320 c0718dc8 ee1d3e10 ee1d3e10 00000000
[    1.477957] 9e60: c0769a88 00000000 c0718dc8 00000000 00000000 c02c3124 c02c310c ee1d3e10
[    1.486117] 9e80: c07b4eec 00000000 c0769a88 c02c1d0c ee1d3e10 c0769a88 ee1d3e44 00000000
[    1.494276] 9ea0: c07091dc c02c1eb8 00000000 c0769a88 c02c1e2c c02c0544 ee005478 ee1676c0
[    1.502435] 9ec0: c0769a88 ee3a4e80 c0760ce8 c02c150c c0669b90 c0769a88 c0746cd8 c0769a88
[    1.510594] 9ee0: c0746cd8 ee2bc4c0 c0778c00 c02c24e0 00000000 c0746cd8 c0746cd8 c07091f0
[    1.518753] 9f00: 00000000 c0008944 c04f405c 00000025 ee00b000 60000153 c074ab00 00000000
[    1.526913] 9f20: 00000000 c074ab90 60000153 00000000 ef7fca5d c050860c 000000b6 c0036b88
[    1.535071] 9f40: c065ecc4 c06bc728 00000006 00000006 c074ab30 ef7fca40 c0739bdc 00000006
[    1.543231] 9f60: c0718dbc c0778c00 000000b6 c0718dc8 c06ed598 c06edd64 00000006 00000006
[    1.551390] 9f80: c06ed598 c003b438 00000000 c04e64f4 00000000 00000000 00000000 00000000
[    1.559549] 9fa0: 00000000 c04e64fc 00000000 c000e838 00000000 00000000 00000000 00000000
[    1.567708] 9fc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
[    1.575867] 9fe0: 00000000 00000000 00000000 00000000 00000013 00000000 c0c0c0c0 c0c0c0c0
[    1.584045] [<c03eef38>] (__clk_put) from [<c03f0ec8>] (of_clk_set_defaults+0xe0/0x2c0)
[    1.592024] [<c03f0ec8>] (of_clk_set_defaults) from [<c02c3124>] (platform_drv_probe+0x18/0xa4)
[    1.600698] [<c02c3124>] (platform_drv_probe) from [<c02c1d0c>] (driver_probe_device+0x10c/0x22c)
[    1.609549] [<c02c1d0c>] (driver_probe_device) from [<c02c1eb8>] (__driver_attach+0x8c/0x90)
[    1.617968] [<c02c1eb8>] (__driver_attach) from [<c02c0544>] (bus_for_each_dev+0x54/0x88)
[    1.626128] [<c02c0544>] (bus_for_each_dev) from [<c02c150c>] (bus_add_driver+0xd4/0x1d0)
[    1.634286] [<c02c150c>] (bus_add_driver) from [<c02c24e0>] (driver_register+0x78/0xf4)
[    1.642275] [<c02c24e0>] (driver_register) from [<c07091f0>] (fimc_md_init+0x14/0x30)
[    1.650089] [<c07091f0>] (fimc_md_init) from [<c0008944>] (do_one_initcall+0x80/0x1d0)
[    1.657989] [<c0008944>] (do_one_initcall) from [<c06edd64>] (kernel_init_freeable+0x108/0x1d4)
[    1.666675] [<c06edd64>] (kernel_init_freeable) from [<c04e64fc>] (kernel_init+0x8/0xec)
[    1.674743] [<c04e64fc>] (kernel_init) from [<c000e838>] (ret_from_fork+0x14/0x3c)
[    1.682287] Code: ebfff4ae e5943014 e5942018 e3530000 (e5823000) 
[    1.688606] ---[ end trace bb367704ce3168e1 ]---
Signature · 2 lines
Regards,
Sylwester
#37 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 12 lines · Sylwester Nawrocki
Quoted 7 lines

+void __clk_put(struct clk *clk)

drivers/clk/clk.c EXPORT_SYMBOL_GPL(devm_clk_unregister); · line 2413

Quoted 6 lines
drivers/clk/clk.c +6 -0
+{
+	if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+		return;
+
+	clk_core_put(clk->core);
+	kfree(clk);

Why do we have kfree() here? clk_get() doesn't allocate the data structure being freed here. What happens if we do clk_get(), clk_put(), clk_get() on same clock?

Quoted 3 lines

I suspect __clk_free_clk() should be called in __clk_release() callback instead, but then there is an issue of safely getting reference to struct clk from struct clk_core pointer.

Please ignore this comment, I missed __clk_create_clk() calls in clkdev.c Anyway, in current -next I'm seeing random pointer dereferences while booting Odroid U3, I'll get back to debugging this tomorrow morning.

Signature · 2 lines
Regards,
Sylwester
#38 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Sylwester Nawrocki
Quoted 48 lines · Sylwester Nawrocki

Hi Tomeu,

Quoted 31 lines · Tomeu Vizoso

int __clk_get(struct clk *clk)
{

(quoted diff) +8 -8
-	if (clk) {
-		if (!try_module_get(clk->owner))
+	struct clk_core *core = !clk ? NULL : clk->core;
+
+	if (core) {
+		if (!try_module_get(core->owner))
 			return 0;
 
-		kref_get(&clk->ref);
+		kref_get(&core->ref);
 	}
 	return 1;
 }
 
-void __clk_put(struct clk *clk)
+static void clk_core_put(struct clk_core *core)
 {
 	struct module *owner;
 
-	if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
-		return;
+	owner = core->owner;
 
 	clk_prepare_lock();
-	owner = clk->owner;
-	kref_put(&clk->ref, __clk_release);
+	kref_put(&core->ref, __clk_release);
 	clk_prepare_unlock();
 

Review

Quoted 10 lines
(quoted diff) +7 -0
 	module_put(owner);
 }
 
+void __clk_put(struct clk *clk)
+{
+	if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+		return;
+
+	clk_core_put(clk->core);
+	kfree(clk);

Why do we have kfree() here? clk_get() doesn't allocate the data structure being freed here. What happens if we do clk_get(), clk_put(), clk_get() on same clock?

Quoted 8 lines

I suspect __clk_free_clk() should be called in __clk_release() callback instead, but then there is an issue of safely getting reference to struct clk from struct clk_core pointer.

I tested linux-next on Odroid U3 and booting fails with oopses as below. There is no problems when the above kfree() is commented out.

Ah now I get it. You meant to say that of_clk_get_by_clkspec() doesn't return an allocated clk pointer. Let's fix that.

----8<----

From: Stephen Boyd <[email protected]>
Subject: [PATCH] clkdev: Always allocate a struct clk in OF functions

of_clk_get_by_clkspec() returns a struct clk pointer but it doesn't create a new handle for the consumers. Instead it just returns whatever the OF clk provider hands out. Let's create a per-user handle here so that clk_put() can properly unlink it and free it when the consumer is done.

Fixes 035a61c314eb "clk: Make clk API return per-user struct clk instances"
---
 drivers/clk/clkdev.c | 44 +++++++++++++++++++++++---------------------
 1 file changed, 23 insertions(+), 21 deletions(-)
drivers/clk/clkdev.c +23 -21
index 29a1ab7af4b8..00d747d09b2a 100644
@@ -29,15 +29,8 @@ static DEFINE_MUTEX(clocks_mutex);
 
 #if defined(CONFIG_OF) && defined(CONFIG_COMMON_CLK)
 
-/**
- * of_clk_get_by_clkspec() - Lookup a clock form a clock provider
- * @clkspec: pointer to a clock specifier data structure
- *
- * This function looks up a struct clk from the registered list of clock
- * providers, an input is a clock specifier data structure as returned
- * from the of_parse_phandle_with_args() function call.
- */
-struct clk *of_clk_get_by_clkspec(struct of_phandle_args *clkspec)
+static struct clk *__of_clk_get_by_clkspec(struct of_phandle_args *clkspec,
+					 const char *dev_id, const char *con_id)
 {
 	struct clk *clk;
 
@@ -47,6 +40,8 @@ struct clk *of_clk_get_by_clkspec(struct of_phandle_args *clkspec)
 	of_clk_lock();
 	clk = __of_clk_get_from_provider(clkspec);
 
+	if (!IS_ERR(clk))
+		clk = __clk_create_clk(__clk_get_hw(clk), dev_id, con_id);
 	if (!IS_ERR(clk) && !__clk_get(clk))
 		clk = ERR_PTR(-ENOENT);
 
@@ -54,7 +49,21 @@ struct clk *of_clk_get_by_clkspec(struct of_phandle_args *clkspec)
 	return clk;
 }
 
-static struct clk *__of_clk_get(struct device_node *np, int index)
+/**
+ * of_clk_get_by_clkspec() - Lookup a clock form a clock provider
+ * @clkspec: pointer to a clock specifier data structure
+ *
+ * This function looks up a struct clk from the registered list of clock
+ * providers, an input is a clock specifier data structure as returned
+ * from the of_parse_phandle_with_args() function call.
+ */
+struct clk *of_clk_get_by_clkspec(struct of_phandle_args *clkspec)
+{
+	return __of_clk_get_by_clkspec(clkspec, NULL, __func__);
+}
+
+static struct clk *__of_clk_get(struct device_node *np, int index,
+			       const char *dev_id, const char *con_id)
 {
 	struct of_phandle_args clkspec;
 	struct clk *clk;
@@ -68,7 +77,7 @@ static struct clk *__of_clk_get(struct device_node *np, int index)
 	if (rc)
 		return ERR_PTR(rc);
 
-	clk = of_clk_get_by_clkspec(&clkspec);
+	clk = __of_clk_get_by_clkspec(&clkspec, dev_id, con_id);
 	of_node_put(clkspec.np);
 
 	return clk;
@@ -76,12 +85,7 @@ static struct clk *__of_clk_get(struct device_node *np, int index)
 
 struct clk *of_clk_get(struct device_node *np, int index)
 {
-	struct clk *clk = __of_clk_get(np, index);
-
-	if (!IS_ERR(clk))
-		clk = __clk_create_clk(__clk_get_hw(clk), np->full_name, NULL);
-
-	return clk;
+	return __of_clk_get(np, index, np->full_name, NULL);
 }
 EXPORT_SYMBOL(of_clk_get);
 
@@ -102,12 +106,10 @@ static struct clk *__of_clk_get_by_name(struct device_node *np,
 		 */
 		if (name)
 			index = of_property_match_string(np, "clock-names", name);
-		clk = __of_clk_get(np, index);
+		clk = __of_clk_get(np, index, dev_id, name);
 		if (!IS_ERR(clk)) {
-			clk = __clk_create_clk(__clk_get_hw(clk), dev_id, name);
 			break;
-		}
-		else if (name && index >= 0) {
+		} else if (name && index >= 0) {
 			if (PTR_ERR(clk) != -EPROBE_DEFER)
 				pr_err("ERROR: could not get clock %s:%s(%i)\n",
 					np->full_name, name ? name : "", index);
Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#39 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 101 lines · Stephen Boyd
Quoted 48 lines · Sylwester Nawrocki

Hi Tomeu,

Quoted 31 lines · Tomeu Vizoso

int __clk_get(struct clk *clk)
{

(quoted diff) +8 -8
-	if (clk) {
-		if (!try_module_get(clk->owner))
+	struct clk_core *core = !clk ? NULL : clk->core;
+
+	if (core) {
+		if (!try_module_get(core->owner))
 			return 0;
 
-		kref_get(&clk->ref);
+		kref_get(&core->ref);
 	}
 	return 1;
 }
 
-void __clk_put(struct clk *clk)
+static void clk_core_put(struct clk_core *core)
 {
 	struct module *owner;
 
-	if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
-		return;
+	owner = core->owner;
 
 	clk_prepare_lock();
-	owner = clk->owner;
-	kref_put(&clk->ref, __clk_release);
+	kref_put(&core->ref, __clk_release);
 	clk_prepare_unlock();
 

Review

Quoted 10 lines
(quoted diff) +7 -0
 	module_put(owner);
 }
 
+void __clk_put(struct clk *clk)
+{
+	if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+		return;
+
+	clk_core_put(clk->core);
+	kfree(clk);

Why do we have kfree() here? clk_get() doesn't allocate the data structure being freed here. What happens if we do clk_get(), clk_put(), clk_get() on same clock?

Quoted 8 lines

I suspect __clk_free_clk() should be called in __clk_release() callback instead, but then there is an issue of safely getting reference to struct clk from struct clk_core pointer.

I tested linux-next on Odroid U3 and booting fails with oopses as below. There is no problems when the above kfree() is commented out.

Ah now I get it. You meant to say that of_clk_get_by_clkspec() doesn't return an allocated clk pointer. Let's fix that.

----8<----

From: Stephen Boyd <[email protected]>
Subject: [PATCH] clkdev: Always allocate a struct clk in OF functions

of_clk_get_by_clkspec() returns a struct clk pointer but it doesn't create a new handle for the consumers. Instead it just returns whatever the OF clk provider hands out. Let's create a per-user handle here so that clk_put() can properly unlink it and free it when the consumer is done.

Fixes 035a61c314eb "clk: Make clk API return per-user struct clk instances"
---
 drivers/clk/clkdev.c | 44 +++++++++++++++++++++++---------------------
 1 file changed, 23 insertions(+), 21 deletions(-)
drivers/clk/clkdev.c +2 -9
index 29a1ab7af4b8..00d747d09b2a 100644
@@ -29,15 +29,8 @@ static DEFINE_MUTEX(clocks_mutex);
 
 #if defined(CONFIG_OF) && defined(CONFIG_COMMON_CLK)
 
-/**
- * of_clk_get_by_clkspec() - Lookup a clock form a clock provider
- * @clkspec: pointer to a clock specifier data structure
- *
- * This function looks up a struct clk from the registered list of clock
- * providers, an input is a clock specifier data structure as returned
- * from the of_parse_phandle_with_args() function call.
- */
-struct clk *of_clk_get_by_clkspec(struct of_phandle_args *clkspec)
+static struct clk *__of_clk_get_by_clkspec(struct of_phandle_args *clkspec,
+					 const char *dev_id, const char *con_id)
 {
 	struct clk *clk;
 

drivers/clk/clkdev.c struct clk *of_clk_get_by_clkspec(struct of_phandle_args *clkspec) · line 47

Quoted 8 lines
drivers/clk/clkdev.c +2 -0
@@ -47,6 +40,8 @@ struct clk *of_clk_get_by_clkspec(struct of_phandle_args *clkspec)
 	of_clk_lock();
 	clk = __of_clk_get_from_provider(clkspec);
 
+	if (!IS_ERR(clk))
+		clk = __clk_create_clk(__clk_get_hw(clk), dev_id, con_id);
 	if (!IS_ERR(clk) && !__clk_get(clk))
 		clk = ERR_PTR(-ENOENT);
 

Actually we can bury the __clk_create_clk() inside __of_clk_get_from_provider(). We should also move __clk_get() into there because right now we have a hole where whoever calls of_clk_get_from_provider() never calls __clk_get() on the clk, leading to possible badness. v2 coming soon.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#40 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 5 lines · Stephen Boyd

Actually we can bury the __clk_create_clk() inside __of_clk_get_from_provider(). We should also move __clk_get() into there because right now we have a hole where whoever calls of_clk_get_from_provider() never calls __clk_get() on the clk, leading to possible badness. v2 coming soon.

There's some other issues here too...

sound/soc/kirkwood/kirkwood-i2s.c:

        priv->clk = devm_clk_get(&pdev->dev, np ? "internal" : NULL);
...
        priv->extclk = devm_clk_get(&pdev->dev, "extclk");
        if (IS_ERR(priv->extclk)) {
...
	} else {
                if (priv->extclk == priv->clk) {
                        devm_clk_put(&pdev->dev, priv->extclk);
                        priv->extclk = ERR_PTR(-EINVAL);
                } else {
                        dev_info(&pdev->dev, "found external clock\n");
                        clk_prepare_enable(priv->extclk);
                        soc_dai = kirkwood_i2s_dai_extclk;
                }

It should be fine provided your "trick" is only done for DT clocks, but not for legacy - with legacy, a NULL in the clkdev tables will match both these requests, hence the need to compare the clk_get() return value to tell whether we get the same clock.

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
#41 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 25 lines · Russell King - ARM Linux
Quoted 5 lines · Stephen Boyd

Actually we can bury the __clk_create_clk() inside __of_clk_get_from_provider(). We should also move __clk_get() into there because right now we have a hole where whoever calls of_clk_get_from_provider() never calls __clk_get() on the clk, leading to possible badness. v2 coming soon.

There's some other issues here too...

sound/soc/kirkwood/kirkwood-i2s.c:

        priv->clk = devm_clk_get(&pdev->dev, np ? "internal" : NULL);
...
        priv->extclk = devm_clk_get(&pdev->dev, "extclk");
        if (IS_ERR(priv->extclk)) {
...
	} else {
                if (priv->extclk == priv->clk) {
                        devm_clk_put(&pdev->dev, priv->extclk);
                        priv->extclk = ERR_PTR(-EINVAL);
                } else {
                        dev_info(&pdev->dev, "found external clock\n");
                        clk_prepare_enable(priv->extclk);
                        soc_dai = kirkwood_i2s_dai_extclk;
                }
Quoted 5 lines

It should be fine provided your "trick" is only done for DT clocks, but not for legacy - with legacy, a NULL in the clkdev tables will match both these requests, hence the need to compare the clk_get() return value to tell whether we get the same clock.

Are we still talking about of_clk_get_from_provider()? Or are we talking about comparing struct clk pointers? From what I can tell this code is now broken because we made all clk getting functions (there's quite a few...) return unique pointers every time they're called. It seems that the driver wants to know if extclk and clk are the same so it can do something differently in kirkwood_set_rate(). Do we need some sort of clk_equal(struct clk *a, struct clk *b) function for drivers like this?

Also, even on DT this could fail if the DT author made internal and extclk map to the same clock provider and cell.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#42 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 26 lines · Stephen Boyd
Quoted 25 lines · Russell King - ARM Linux
Quoted 5 lines · Stephen Boyd

Actually we can bury the __clk_create_clk() inside __of_clk_get_from_provider(). We should also move __clk_get() into there because right now we have a hole where whoever calls of_clk_get_from_provider() never calls __clk_get() on the clk, leading to possible badness. v2 coming soon.

There's some other issues here too...

sound/soc/kirkwood/kirkwood-i2s.c:

        priv->clk = devm_clk_get(&pdev->dev, np ? "internal" : NULL);
...
        priv->extclk = devm_clk_get(&pdev->dev, "extclk");
        if (IS_ERR(priv->extclk)) {
...
	} else {
                if (priv->extclk == priv->clk) {
                        devm_clk_put(&pdev->dev, priv->extclk);
                        priv->extclk = ERR_PTR(-EINVAL);
                } else {
                        dev_info(&pdev->dev, "found external clock\n");
                        clk_prepare_enable(priv->extclk);
                        soc_dai = kirkwood_i2s_dai_extclk;
                }
Quoted 8 lines
Quoted 5 lines

It should be fine provided your "trick" is only done for DT clocks, but not for legacy - with legacy, a NULL in the clkdev tables will match both these requests, hence the need to compare the clk_get() return value to tell whether we get the same clock.

Are we still talking about of_clk_get_from_provider()? Or are we talking about comparing struct clk pointers?

Comparing struct clk pointers, and the implications of the patch changing the clk_get() et.al. to be unique struct clk pointers.

Quoted 6 lines

From what I can tell this code is
now broken because we made all clk getting functions (there's quite a few...) return unique pointers every time they're called. It seems that the driver wants to know if extclk and clk are the same so it can do something differently in kirkwood_set_rate(). Do we need some sort of clk_equal(struct clk *a, struct clk *b) function for drivers like this?

Well, the clocks in question are the SoC internal clock (which is more or less fixed, but has a programmable divider) and an externally supplied clock, and the IP has a multiplexer on its input which allows us to select between those two sources.

If it were possible to bind both to the same clock, it wouldn't be a useful configuration - nothing would be gained from doing so in terms of available rates.

What the comparison is there for is to catch the case with legacy lookups where a clkdev lookup entry with a NULL connection ID results in matching any connection ID passed to clk_get(). If the patch changes this, then we will have a regression - and this is something which needs fixing _before_ we do this "return unique clocks".

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
#43 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 17 lines · Russell King - ARM Linux
Quoted 6 lines · Stephen Boyd

From what I can tell this code is
now broken because we made all clk getting functions (there's quite a few...) return unique pointers every time they're called. It seems that the driver wants to know if extclk and clk are the same so it can do something differently in kirkwood_set_rate(). Do we need some sort of clk_equal(struct clk *a, struct clk *b) function for drivers like this?

Well, the clocks in question are the SoC internal clock (which is more or less fixed, but has a programmable divider) and an externally supplied clock, and the IP has a multiplexer on its input which allows us to select between those two sources.

If it were possible to bind both to the same clock, it wouldn't be a useful configuration - nothing would be gained from doing so in terms of available rates.

Quoted 6 lines

What the comparison is there for is to catch the case with legacy lookups where a clkdev lookup entry with a NULL connection ID results in matching any connection ID passed to clk_get(). If the patch changes this, then we will have a regression - and this is something which needs fixing _before_ we do this "return unique clocks".

Ok. It seems that we might need a clk_equal() or similar API after all. My understanding is that this driver is calling clk_get() twice with NULL for the con_id and then "extclk" in attempts to get the SoC internal clock and the externally supplied clock. If we're using legacy lookups then both clk_get() calls may map to the same clk_lookup entry and before Tomeu's patch that returns unique clocks the driver could detect this case and know that there isn't an external clock. Looking at arch/arm/mach-dove/common.c it seems that there is only one lookup per device and it has a wildcard NULL for con_id so both clk_get() calls here are going to find the same lookup and get a unique struct clk pointer.

Why don't we make the legacy lookup more specific and actually indicate "internal" for the con_id? Then the external clock would fail to be found, but we can detect that case and figure out that it's not due to probe defer, but instead due to the fact that there really isn't any mapping. It looks like the code is already prepared for this anyway.

----8<----

From: Stephen Boyd <[email protected]>
Subject: [PATCH] ARM: dove: Remove wildcard from mvebu-audio device clk lookup

This i2s driver is using the wildcard nature of clkdev lookups to figure out if there's an external clock or not. It does this by calling clk_get() twice with NULL for the con_id first and then "external" for the con_id the second time around and then compares the two pointers. With DT the wildcard feature of clk_get() is gone and so the driver has to handle an error from the second clk_get() call as meaning "no external clock specified". Let's use that logic even with clk lookups to simplify the code and remove the struct clk pointer comparisons which may not work in the future when clk_get() returns unique pointers.

---
 arch/arm/mach-dove/common.c       |  4 ++--
 sound/soc/kirkwood/kirkwood-i2s.c | 13 ++++---------
 2 files changed, 6 insertions(+), 11 deletions(-)
arch/arm/mach-dove/common.c +2 -2
index 0d1a89298ece..f290fc944cc1 100644
@@ -124,8 +124,8 @@ static void __init dove_clk_init(void)
 	orion_clkdev_add(NULL, "sdhci-dove.1", sdio1);
 	orion_clkdev_add(NULL, "orion_nand", nand);
 	orion_clkdev_add(NULL, "cafe1000-ccic.0", camera);
-	orion_clkdev_add(NULL, "mvebu-audio.0", i2s0);
-	orion_clkdev_add(NULL, "mvebu-audio.1", i2s1);
+	orion_clkdev_add("internal", "mvebu-audio.0", i2s0);
+	orion_clkdev_add("internal", "mvebu-audio.1", i2s1);
 	orion_clkdev_add(NULL, "mv_crypto", crypto);
 	orion_clkdev_add(NULL, "dove-ac97", ac97);
 	orion_clkdev_add(NULL, "dove-pdma", pdma);
sound/soc/kirkwood/kirkwood-i2s.c +4 -9
index def7d8260c4e..0bfeb712a997 100644
@@ -564,7 +564,7 @@ static int kirkwood_i2s_dev_probe(struct platform_device *pdev)
 		return -EINVAL;
 	}
 
-	priv->clk = devm_clk_get(&pdev->dev, np ? "internal" : NULL);
+	priv->clk = devm_clk_get(&pdev->dev, "internal");
 	if (IS_ERR(priv->clk)) {
 		dev_err(&pdev->dev, "no clock\n");
 		return PTR_ERR(priv->clk);
@@ -579,14 +579,9 @@ static int kirkwood_i2s_dev_probe(struct platform_device *pdev)
 		if (PTR_ERR(priv->extclk) == -EPROBE_DEFER)
 			return -EPROBE_DEFER;
 	} else {
-		if (priv->extclk == priv->clk) {
-			devm_clk_put(&pdev->dev, priv->extclk);
-			priv->extclk = ERR_PTR(-EINVAL);
-		} else {
-			dev_info(&pdev->dev, "found external clock\n");
-			clk_prepare_enable(priv->extclk);
-			soc_dai = kirkwood_i2s_dai_extclk;
-		}
+		dev_info(&pdev->dev, "found external clock\n");
+		clk_prepare_enable(priv->extclk);
+		soc_dai = kirkwood_i2s_dai_extclk;
 	}
 
 	/* Some sensible defaults - this reflects the powerup values */
Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#44 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 5 lines · Stephen Boyd

Why don't we make the legacy lookup more specific and actually indicate "internal" for the con_id? Then the external clock would fail to be found, but we can detect that case and figure out that it's not due to probe defer, but instead due to the fact that there really isn't any mapping. It looks like the code is already prepared for this anyway.

We _could_, and that would solve this specific issue, but I'd suggest coccinelle is used to locate any other similar instances of this.

As I'm allergic to coccinelle (or coccinelle is allergic to me since I never seem to be able to get it to do what I want...)

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
#45 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 6 lines · Russell King - ARM Linux
Quoted 5 lines · Stephen Boyd

Why don't we make the legacy lookup more specific and actually indicate "internal" for the con_id? Then the external clock would fail to be found, but we can detect that case and figure out that it's not due to probe defer, but instead due to the fact that there really isn't any mapping. It looks like the code is already prepared for this anyway.

Quoted 6 lines

We _could_, and that would solve this specific issue, but I'd suggest coccinelle is used to locate any other similar instances of this.

As I'm allergic to coccinelle (or coccinelle is allergic to me since I never seem to be able to get it to do what I want...)

Great. I'd like to avoid adding clk_equal() until we actually need it, and I hope we don't ever need it. We've already got coccinelle in the works to find similar issues and it seems you and I have the same allergies because I can't get it to work for me right now.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#46 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

In reply to Stephen Boyd

Quoting Stephen Boyd (2015-02-06 11:30:18)

Quoted 59 lines · Russell King - ARM Linux
Quoted 17 lines · Stephen Boyd
Quoted 7 lines

From what I can tell this code is
now broken because we made all clk getting functions (there's quite a few...) return unique pointers every time they're called. It seems that the driver wants to know if extclk and clk are the same so it can do something differently in kirkwood_set_rate(). Do we need some sort of clk_equal(struct clk *a, struct clk *b) function for drivers like this?

Well, the clocks in question are the SoC internal clock (which is more or less fixed, but has a programmable divider) and an externally supplied clock, and the IP has a multiplexer on its input which allows us to select between those two sources.

If it were possible to bind both to the same clock, it wouldn't be a useful configuration - nothing would be gained from doing so in terms of available rates.

Quoted 6 lines

What the comparison is there for is to catch the case with legacy lookups where a clkdev lookup entry with a NULL connection ID results in matching any connection ID passed to clk_get(). If the patch changes this, then we will have a regression - and this is something which needs fixing _before_ we do this "return unique clocks".

Ok. It seems that we might need a clk_equal() or similar API after all. My understanding is that this driver is calling clk_get() twice with NULL for the con_id and then "extclk" in attempts to get the SoC internal clock and the externally supplied clock. If we're using legacy lookups then both clk_get() calls may map to the same clk_lookup entry and before Tomeu's patch that returns unique clocks the driver could detect this case and know that there isn't an external clock. Looking at arch/arm/mach-dove/common.c it seems that there is only one lookup per device and it has a wildcard NULL for con_id so both clk_get() calls here are going to find the same lookup and get a unique struct clk pointer.

Why don't we make the legacy lookup more specific and actually indicate "internal" for the con_id? Then the external clock would fail to be found, but we can detect that case and figure out that it's not due to probe defer, but instead due to the fact that there really isn't any mapping. It looks like the code is already prepared for this anyway.

----8<----

From: Stephen Boyd <[email protected]>
Subject: [PATCH] ARM: dove: Remove wildcard from mvebu-audio device clk lookup

This i2s driver is using the wildcard nature of clkdev lookups to figure out if there's an external clock or not. It does this by calling clk_get() twice with NULL for the con_id first and then "external" for the con_id the second time around and then compares the two pointers. With DT the wildcard feature of clk_get() is gone and so the driver has to handle an error from the second clk_get() call as meaning "no external clock specified". Let's use that logic even with clk lookups to simplify the code and remove the struct clk pointer comparisons which may not work in the future when clk_get() returns unique pointers.

Quoted 1 line

Russell et al,

I'm happy to take this patch through the clock tree (where the problem shows up) with an ack.

Regards,
Mike

Quoted 56 lines
---
 arch/arm/mach-dove/common.c       |  4 ++--
 sound/soc/kirkwood/kirkwood-i2s.c | 13 ++++---------
 2 files changed, 6 insertions(+), 11 deletions(-)
arch/arm/mach-dove/common.c +2 -2
index 0d1a89298ece..f290fc944cc1 100644
@@ -124,8 +124,8 @@ static void __init dove_clk_init(void)
        orion_clkdev_add(NULL, "sdhci-dove.1", sdio1);
        orion_clkdev_add(NULL, "orion_nand", nand);
        orion_clkdev_add(NULL, "cafe1000-ccic.0", camera);
-       orion_clkdev_add(NULL, "mvebu-audio.0", i2s0);
-       orion_clkdev_add(NULL, "mvebu-audio.1", i2s1);
+       orion_clkdev_add("internal", "mvebu-audio.0", i2s0);
+       orion_clkdev_add("internal", "mvebu-audio.1", i2s1);
        orion_clkdev_add(NULL, "mv_crypto", crypto);
        orion_clkdev_add(NULL, "dove-ac97", ac97);
        orion_clkdev_add(NULL, "dove-pdma", pdma);
sound/soc/kirkwood/kirkwood-i2s.c +4 -9
index def7d8260c4e..0bfeb712a997 100644
@@ -564,7 +564,7 @@ static int kirkwood_i2s_dev_probe(struct platform_device *pdev)
                return -EINVAL;
        }
 
-       priv->clk = devm_clk_get(&pdev->dev, np ? "internal" : NULL);
+       priv->clk = devm_clk_get(&pdev->dev, "internal");
        if (IS_ERR(priv->clk)) {
                dev_err(&pdev->dev, "no clock\n");
                return PTR_ERR(priv->clk);
@@ -579,14 +579,9 @@ static int kirkwood_i2s_dev_probe(struct platform_device *pdev)
                if (PTR_ERR(priv->extclk) == -EPROBE_DEFER)
                        return -EPROBE_DEFER;
        } else {
-               if (priv->extclk == priv->clk) {
-                       devm_clk_put(&pdev->dev, priv->extclk);
-                       priv->extclk = ERR_PTR(-EINVAL);
-               } else {
-                       dev_info(&pdev->dev, "found external clock\n");
-                       clk_prepare_enable(priv->extclk);
-                       soc_dai = kirkwood_i2s_dai_extclk;
-               }
+               dev_info(&pdev->dev, "found external clock\n");
+               clk_prepare_enable(priv->extclk);
+               soc_dai = kirkwood_i2s_dai_extclk;
        }
 
        /* Some sensible defaults - this reflects the powerup values */
Signature · 3 lines
-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#47 lore raw

Re: [PATCH v13 3/6] clk: Make clk API return per-user struct clk instances

Quoted 60 lines · Mike Turquette

Quoting Stephen Boyd (2015-02-06 11:30:18)

Quoted 59 lines · Russell King - ARM Linux
Quoted 17 lines · Stephen Boyd
Quoted 7 lines

From what I can tell this code is
now broken because we made all clk getting functions (there's quite a few...) return unique pointers every time they're called. It seems that the driver wants to know if extclk and clk are the same so it can do something differently in kirkwood_set_rate(). Do we need some sort of clk_equal(struct clk *a, struct clk *b) function for drivers like this?

Well, the clocks in question are the SoC internal clock (which is more or less fixed, but has a programmable divider) and an externally supplied clock, and the IP has a multiplexer on its input which allows us to select between those two sources.

If it were possible to bind both to the same clock, it wouldn't be a useful configuration - nothing would be gained from doing so in terms of available rates.

Quoted 6 lines

What the comparison is there for is to catch the case with legacy lookups where a clkdev lookup entry with a NULL connection ID results in matching any connection ID passed to clk_get(). If the patch changes this, then we will have a regression - and this is something which needs fixing _before_ we do this "return unique clocks".

Ok. It seems that we might need a clk_equal() or similar API after all. My understanding is that this driver is calling clk_get() twice with NULL for the con_id and then "extclk" in attempts to get the SoC internal clock and the externally supplied clock. If we're using legacy lookups then both clk_get() calls may map to the same clk_lookup entry and before Tomeu's patch that returns unique clocks the driver could detect this case and know that there isn't an external clock. Looking at arch/arm/mach-dove/common.c it seems that there is only one lookup per device and it has a wildcard NULL for con_id so both clk_get() calls here are going to find the same lookup and get a unique struct clk pointer.

Why don't we make the legacy lookup more specific and actually indicate "internal" for the con_id? Then the external clock would fail to be found, but we can detect that case and figure out that it's not due to probe defer, but instead due to the fact that there really isn't any mapping. It looks like the code is already prepared for this anyway.

----8<----

From: Stephen Boyd <[email protected]>
Subject: [PATCH] ARM: dove: Remove wildcard from mvebu-audio device clk lookup

This i2s driver is using the wildcard nature of clkdev lookups to figure out if there's an external clock or not. It does this by calling clk_get() twice with NULL for the con_id first and then "external" for the con_id the second time around and then compares the two pointers. With DT the wildcard feature of clk_get() is gone and so the driver has to handle an error from the second clk_get() call as meaning "no external clock specified". Let's use that logic even with clk lookups to simplify the code and remove the struct clk pointer comparisons which may not work in the future when clk_get() returns unique pointers.

Quoted 6 lines
Quoted 1 line

Russell et al,

I'm happy to take this patch through the clock tree (where the problem shows up) with an ack.

It's not up to me - I don't maintain this driver. I'm just an interested party.

Note that much more than just this has now broken. The iMX6 code has broken as well, and it's not going to take such a simple fix there to fix it either.

Please either revert the patches creating this breakage (and have another attempt at the next merge window) or supply fixes for these places.

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
#48 lore raw

PATCH v13 3/6 clk: Make clk API return per-user struct clk instances

Quoting Russell King - ARM Linux (2015-02-24 06:08:08)

Quoted 63 lines · Mike Turquette
Quoted 62 lines

Quoting Stephen Boyd (2015-02-06 11:30:18)

Quoted 60 lines · Russell King - ARM Linux
Quoted 17 lines · Stephen Boyd
Quoted 7 lines

From what I can tell this code is
now broken because we made all clk getting functions (there's quite a few...) return unique pointers every time they're called. It seems that the driver wants to know if extclk and clk are the same so it can do something differently in kirkwood_set_rate(). Do we need some sort of clk_equal(struct clk *a, struct clk *b) function for drivers like this?

Well, the clocks in question are the SoC internal clock (which is more or less fixed, but has a programmable divider) and an externally supplied clock, and the IP has a multiplexer on its input which allows us to select between those two sources.

If it were possible to bind both to the same clock, it wouldn't be a useful configuration - nothing would be gained from doing so in terms of available rates.

Quoted 6 lines

What the comparison is there for is to catch the case with legacy lookups where a clkdev lookup entry with a NULL connection ID results in matching any connection ID passed to clk_get(). If the patch changes this, then we will have a regression - and this is something which needs fixing _before_ we do this "return unique clocks".

Ok. It seems that we might need a clk_equal() or similar API after all. My understanding is that this driver is calling clk_get() twice with NULL for the con_id and then "extclk" in attempts to get the SoC internal clock and the externally supplied clock. If we're using legacy lookups then both clk_get() calls may map to the same clk_lookup entry and before Tomeu's patch that returns unique clocks the driver could detect this case and know that there isn't an external clock. Looking at arch/arm/mach-dove/common.c it seems that there is only one lookup per device and it has a wildcard NULL for con_id so both clk_get() calls here are going to find the same lookup and get a unique struct clk pointer.

Why don't we make the legacy lookup more specific and actually indicate "internal" for the con_id? Then the external clock would fail to be found, but we can detect that case and figure out that it's not due to probe defer, but instead due to the fact that there really isn't any mapping. It looks like the code is already prepared for this anyway.

----8<----

From: Stephen Boyd <[email protected]>
Subject: [PATCH] ARM: dove: Remove wildcard from mvebu-audio device clk lookup

This i2s driver is using the wildcard nature of clkdev lookups to figure out if there's an external clock or not. It does this by calling clk_get() twice with NULL for the con_id first and then "external" for the con_id the second time around and then compares the two pointers. With DT the wildcard feature of clk_get() is gone and so the driver has to handle an error from the second clk_get() call as meaning "no external clock specified". Let's use that logic even with clk lookups to simplify the code and remove the struct clk pointer comparisons which may not work in the future when clk_get() returns unique pointers.

Quoted 7 lines
Quoted 4 lines

Russell et al,

I'm happy to take this patch through the clock tree (where the problem shows up) with an ack.

It's not up to me - I don't maintain this driver. I'm just an interested party.

Sure.

Quoted 7 lines

Note that much more than just this has now broken. The iMX6 code has broken as well, and it's not going to take such a simple fix there to fix it either.

Please either revert the patches creating this breakage (and have another attempt at the next merge window) or supply fixes for these places.

Let's try the latter. Stephen used coccinelle to find similar instances of clk pointer comparisons[0]. As a stop-gap solution we can introduce a clk_is_match(struct clk *a, struct clk *b) function and sub it for the handful of drivers that do this behavior and use CCF.

[0] http://lkml.kernel.org/r/<[email protected]>

Regards,
Mike

Quoted 4 lines
Signature · 3 lines
-- 
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
#49 lore raw

PATCH v13 4/6 clk: Add rate constraints to clocks

In reply to Tomeu Vizoso

Adds a way for clock consumers to set maximum and minimum rates. This can be used for thermal drivers to set minimum rates, or by misc. drivers to set maximum rates to assure a minimum performance level.

Changes the signature of the determine_rate callback by adding the parameters min_rate and max_rate.

---
v13: * Cut one line to the 80-column limit
* Make clear in the docs that the ranges are inclusive

v12: * Refactor locking so that __clk_put takes the lock only once

v11: * Recalculate the rate before putting the reference to clk_core
* Don't recalculate the rate when freeing the per-user clock in the initialization error paths
* Move __clk_create_clk to be next to __clk_free_clk for more comfortable reading

v10: * Refactor __clk_determine_rate to share code with
clk_round_rate.
* Remove clk_core_round_rate_nolock as it's unused now

v9: * s/floor/min and s/ceiling/max
* Add a bunch of NULL checks
* Propagate our rate range when querying our parent for the rate
* Take constraints into account in clk_round_rate
* Add __clk_determine_rate() for clk providers to ask their parents for a rate within their range
* Make sure that what ops->round_rate returns when changing rates is within the range

v7: * Update a few more instances in new code

v6: * Take the prepare lock before removing a per-user clk
* Init per-user clks list before adding the first clk
* Pass the constraints to determine_rate and let clk implementations deal with constraints
* Add clk_set_rate_range

v5: * Initialize clk.ceiling_constraint to ULONG_MAX
* Warn about inconsistent constraints

v4: * Copy function docs from header
* Move WARN out of critical section
* Refresh rate after removing a per-user clk
* Rename clk_core.per_user_clks to clk_core.clks
* Store requested rate and re-apply it when constraints are updated

---
 Documentation/clk.txt               |   2 +
 arch/arm/mach-omap2/dpll3xxx.c      |   2 +
 arch/arm/mach-omap2/dpll44xx.c      |   2 +
 arch/mips/alchemy/common/clock.c    |   8 ++
 drivers/clk/at91/clk-programmable.c |   2 +
 drivers/clk/bcm/clk-kona.c          |   2 +
 drivers/clk/clk-composite.c         |   9 +-
 drivers/clk/clk.c                   | 249 +++++++++++++++++++++++++++++-------
 drivers/clk/hisilicon/clk-hi3620.c  |   2 +
 drivers/clk/mmp/clk-mix.c           |   2 +
 drivers/clk/qcom/clk-pll.c          |   1 +
 drivers/clk/qcom/clk-rcg.c          |  10 +-
 drivers/clk/qcom/clk-rcg2.c         |   6 +
 drivers/clk/sunxi/clk-factors.c     |   2 +
 drivers/clk/sunxi/clk-sun6i-ar100.c |   2 +
 include/linux/clk-private.h         |   6 +
 include/linux/clk-provider.h        |  15 ++-
 include/linux/clk.h                 |  28 ++++
 include/linux/clk/ti.h              |   4 +
 19 files changed, 298 insertions(+), 56 deletions(-)
Documentation/clk.txt +2 -0
index 4ff8462..0e4f90a 100644
@@ -73,6 +73,8 @@ the operations defined in clk.h:
 						unsigned long *parent_rate);
 		long		(*determine_rate)(struct clk_hw *hw,
 						unsigned long rate,
+						unsigned long min_rate,
+						unsigned long max_rate,
 						unsigned long *best_parent_rate,
 						struct clk_hw **best_parent_clk);
 		int		(*set_parent)(struct clk_hw *hw, u8 index);
arch/arm/mach-omap2/dpll3xxx.c +2 -0
index c2da2a0..ac3fb11 100644
@@ -473,6 +473,8 @@ void omap3_noncore_dpll_disable(struct clk_hw *hw)
  * in failure.
  */
 long omap3_noncore_dpll_determine_rate(struct clk_hw *hw, unsigned long rate,
+				       unsigned long min_rate,
+				       unsigned long max_rate,
 				       unsigned long *best_parent_rate,
 				       struct clk_hw **best_parent_clk)
 {
arch/arm/mach-omap2/dpll44xx.c +2 -0
index 0e58e5a..acacb90 100644
@@ -222,6 +222,8 @@ out:
  * in failure.
  */
 long omap4_dpll_regm4xen_determine_rate(struct clk_hw *hw, unsigned long rate,
+					unsigned long min_rate,
+					unsigned long max_rate,
 					unsigned long *best_parent_rate,
 					struct clk_hw **best_parent_clk)
 {
arch/mips/alchemy/common/clock.c +8 -0
index 48a9dfc..4e65404 100644
@@ -373,6 +373,8 @@ static long alchemy_calc_div(unsigned long rate, unsigned long prate,
 }
 
 static long alchemy_clk_fgcs_detr(struct clk_hw *hw, unsigned long rate,
+					unsigned long min_rate,
+					unsigned long max_rate,
 					unsigned long *best_parent_rate,
 					struct clk_hw **best_parent_clk,
 					int scale, int maxdiv)
@@ -546,6 +548,8 @@ static unsigned long alchemy_clk_fgv1_recalc(struct clk_hw *hw,
 }
 
 static long alchemy_clk_fgv1_detr(struct clk_hw *hw, unsigned long rate,
+					unsigned long min_rate,
+					unsigned long max_rate,
 					unsigned long *best_parent_rate,
 					struct clk_hw **best_parent_clk)
 {
@@ -678,6 +682,8 @@ static unsigned long alchemy_clk_fgv2_recalc(struct clk_hw *hw,
 }
 
 static long alchemy_clk_fgv2_detr(struct clk_hw *hw, unsigned long rate,
+					unsigned long min_rate,
+					unsigned long max_rate,
 					unsigned long *best_parent_rate,
 					struct clk_hw **best_parent_clk)
 {
@@ -897,6 +903,8 @@ static int alchemy_clk_csrc_setr(struct clk_hw *hw, unsigned long rate,
 }
 
 static long alchemy_clk_csrc_detr(struct clk_hw *hw, unsigned long rate,
+					unsigned long min_rate,
+					unsigned long max_rate,
 					unsigned long *best_parent_rate,
 					struct clk_hw **best_parent_clk)
 {
drivers/clk/at91/clk-programmable.c +2 -0
index bbdb1b9..86c8a07 100644
@@ -56,6 +56,8 @@ static unsigned long clk_programmable_recalc_rate(struct clk_hw *hw,
 
 static long clk_programmable_determine_rate(struct clk_hw *hw,
 					    unsigned long rate,
+					    unsigned long min_rate,
+					    unsigned long max_rate,
 					    unsigned long *best_parent_rate,
 					    struct clk_hw **best_parent_hw)
 {
drivers/clk/bcm/clk-kona.c +2 -0
index 1c06f6f..05abae8 100644
@@ -1032,6 +1032,8 @@ static long kona_peri_clk_round_rate(struct clk_hw *hw, unsigned long rate,
 }
 
 static long kona_peri_clk_determine_rate(struct clk_hw *hw, unsigned long rate,
+		unsigned long min_rate,
+		unsigned long max_rate,
 		unsigned long *best_parent_rate, struct clk_hw **best_parent)
 {
 	struct kona_clk *bcm_clk = to_kona_clk(hw);
drivers/clk/clk-composite.c +7 -2
index 4386697..dee81b8 100644
@@ -56,6 +56,8 @@ static unsigned long clk_composite_recalc_rate(struct clk_hw *hw,
 }
 
 static long clk_composite_determine_rate(struct clk_hw *hw, unsigned long rate,
+					unsigned long min_rate,
+					unsigned long max_rate,
 					unsigned long *best_parent_rate,
 					struct clk_hw **best_parent_p)
 {
@@ -73,7 +75,9 @@ static long clk_composite_determine_rate(struct clk_hw *hw, unsigned long rate,
 
 	if (rate_hw && rate_ops && rate_ops->determine_rate) {
 		rate_hw->clk = hw->clk;
-		return rate_ops->determine_rate(rate_hw, rate, best_parent_rate,
+		return rate_ops->determine_rate(rate_hw, rate, min_rate,
+						max_rate,
+						best_parent_rate,
 						best_parent_p);
 	} else if (rate_hw && rate_ops && rate_ops->round_rate &&
 		   mux_hw && mux_ops && mux_ops->set_parent) {
@@ -117,7 +121,8 @@ static long clk_composite_determine_rate(struct clk_hw *hw, unsigned long rate,
 		return best_rate;
 	} else if (mux_hw && mux_ops && mux_ops->determine_rate) {
 		mux_hw->clk = hw->clk;
-		return mux_ops->determine_rate(mux_hw, rate, best_parent_rate,
+		return mux_ops->determine_rate(mux_hw, rate, min_rate,
+					       max_rate, best_parent_rate,
 					       best_parent_p);
 	} else {
 		pr_err("clk: clk_composite_determine_rate function called, but no mux or rate callback set!\n");
drivers/clk/clk.c +200 -49

402 lines not on this page · raw message

drivers/clk/hisilicon/clk-hi3620.c +2 -0
index 007144f..2e4f6d4 100644
@@ -295,6 +295,8 @@ static unsigned long mmc_clk_recalc_rate(struct clk_hw *hw,
 }
 
 static long mmc_clk_determine_rate(struct clk_hw *hw, unsigned long rate,
+			      unsigned long min_rate,
+			      unsigned long max_rate,
 			      unsigned long *best_parent_rate,
 			      struct clk_hw **best_parent_p)
 {
drivers/clk/mmp/clk-mix.c +2 -0
index 48fa53c..de6a873 100644
@@ -202,6 +202,8 @@ error:
 }
 
 static long mmp_clk_mix_determine_rate(struct clk_hw *hw, unsigned long rate,
+					unsigned long min_rate,
+					unsigned long max_rate,
 					unsigned long *best_parent_rate,
 					struct clk_hw **best_parent_clk)
 {
drivers/clk/qcom/clk-pll.c +1 -0
index 60873a7..b4325f6 100644
@@ -141,6 +141,7 @@ struct pll_freq_tbl *find_freq(const struct pll_freq_tbl *f, unsigned long rate)
 
 static long
 clk_pll_determine_rate(struct clk_hw *hw, unsigned long rate,
+		       unsigned long min_rate, unsigned long max_rate,
 		       unsigned long *p_rate, struct clk_hw **p)
 {
 	struct clk_pll *pll = to_clk_pll(hw);
drivers/clk/qcom/clk-rcg.c +8 -2
index 0b93972..0039bd7 100644
@@ -368,6 +368,7 @@ clk_dyn_rcg_recalc_rate(struct clk_hw *hw, unsigned long parent_rate)
 
 static long _freq_tbl_determine_rate(struct clk_hw *hw,
 		const struct freq_tbl *f, unsigned long rate,
+		unsigned long min_rate, unsigned long max_rate,
 		unsigned long *p_rate, struct clk_hw **p_hw)
 {
 	unsigned long clk_flags;
@@ -397,22 +398,27 @@ static long _freq_tbl_determine_rate(struct clk_hw *hw,
 }
 
 static long clk_rcg_determine_rate(struct clk_hw *hw, unsigned long rate,
+		unsigned long min_rate, unsigned long max_rate,
 		unsigned long *p_rate, struct clk_hw **p)
 {
 	struct clk_rcg *rcg = to_clk_rcg(hw);
 
-	return _freq_tbl_determine_rate(hw, rcg->freq_tbl, rate, p_rate, p);
+	return _freq_tbl_determine_rate(hw, rcg->freq_tbl, rate, min_rate,
+			max_rate, p_rate, p);
 }
 
 static long clk_dyn_rcg_determine_rate(struct clk_hw *hw, unsigned long rate,
+		unsigned long min_rate, unsigned long max_rate,
 		unsigned long *p_rate, struct clk_hw **p)
 {
 	struct clk_dyn_rcg *rcg = to_clk_dyn_rcg(hw);
 
-	return _freq_tbl_determine_rate(hw, rcg->freq_tbl, rate, p_rate, p);
+	return _freq_tbl_determine_rate(hw, rcg->freq_tbl, rate, min_rate,
+			max_rate, p_rate, p);
 }
 
 static long clk_rcg_bypass_determine_rate(struct clk_hw *hw, unsigned long rate,
+		unsigned long min_rate, unsigned long max_rate,
 		unsigned long *p_rate, struct clk_hw **p_hw)
 {
 	struct clk_rcg *rcg = to_clk_rcg(hw);
drivers/clk/qcom/clk-rcg2.c +6 -0
index 08b8b37..742acfa 100644
@@ -208,6 +208,7 @@ static long _freq_tbl_determine_rate(struct clk_hw *hw,
 }
 
 static long clk_rcg2_determine_rate(struct clk_hw *hw, unsigned long rate,
+		unsigned long min_rate, unsigned long max_rate,
 		unsigned long *p_rate, struct clk_hw **p)
 {
 	struct clk_rcg2 *rcg = to_clk_rcg2(hw);
@@ -361,6 +362,8 @@ static int clk_edp_pixel_set_rate_and_parent(struct clk_hw *hw,
 }
 
 static long clk_edp_pixel_determine_rate(struct clk_hw *hw, unsigned long rate,
+				 unsigned long min_rate,
+				 unsigned long max_rate,
 				 unsigned long *p_rate, struct clk_hw **p)
 {
 	struct clk_rcg2 *rcg = to_clk_rcg2(hw);
@@ -412,6 +415,7 @@ const struct clk_ops clk_edp_pixel_ops = {
 EXPORT_SYMBOL_GPL(clk_edp_pixel_ops);
 
 static long clk_byte_determine_rate(struct clk_hw *hw, unsigned long rate,
+			 unsigned long min_rate, unsigned long max_rate,
 			 unsigned long *p_rate, struct clk_hw **p_hw)
 {
 	struct clk_rcg2 *rcg = to_clk_rcg2(hw);
@@ -476,6 +480,8 @@ static const struct frac_entry frac_table_pixel[] = {
 };
 
 static long clk_pixel_determine_rate(struct clk_hw *hw, unsigned long rate,
+				 unsigned long min_rate,
+				 unsigned long max_rate,
 				 unsigned long *p_rate, struct clk_hw **p)
 {
 	struct clk_rcg2 *rcg = to_clk_rcg2(hw);
drivers/clk/sunxi/clk-factors.c +2 -0
index 62e08fb..761029b 100644
@@ -80,6 +80,8 @@ static long clk_factors_round_rate(struct clk_hw *hw, unsigned long rate,
 }
 
 static long clk_factors_determine_rate(struct clk_hw *hw, unsigned long rate,
+				       unsigned long min_rate,
+				       unsigned long max_rate,
 				       unsigned long *best_parent_rate,
 				       struct clk_hw **best_parent_p)
 {
drivers/clk/sunxi/clk-sun6i-ar100.c +2 -0
index 3d282fb..63cf149 100644
@@ -45,6 +45,8 @@ static unsigned long ar100_recalc_rate(struct clk_hw *hw,
 }
 
 static long ar100_determine_rate(struct clk_hw *hw, unsigned long rate,
+				 unsigned long min_rate,
+				 unsigned long max_rate,
 				 unsigned long *best_parent_rate,
 				 struct clk_hw **best_parent_clk)
 {
include/linux/clk-private.h +6 -0
index ae55d99..5136b30 100644
@@ -39,6 +39,7 @@ struct clk_core {
 	u8			num_parents;
 	u8			new_parent_index;
 	unsigned long		rate;
+	unsigned long		req_rate;
 	unsigned long		new_rate;
 	struct clk_core		*new_parent;
 	struct clk_core		*new_child;
@@ -50,6 +51,7 @@ struct clk_core {
 	struct hlist_head	children;
 	struct hlist_node	child_node;
 	struct hlist_node	debug_node;
+	struct hlist_head	clks;
 	unsigned int		notifier_count;
 #ifdef CONFIG_DEBUG_FS
 	struct dentry		*dentry;
@@ -61,6 +63,10 @@ struct clk {
 	struct clk_core	*core;
 	const char *dev_id;
 	const char *con_id;
+
+	unsigned long min_rate;
+	unsigned long max_rate;
+	struct hlist_node child_node;
 };
 
 /*
include/linux/clk-provider.h +12 -3
index 9afd438..2416026 100644
@@ -175,9 +175,12 @@ struct clk_ops {
 					unsigned long parent_rate);
 	long		(*round_rate)(struct clk_hw *hw, unsigned long rate,
 					unsigned long *parent_rate);
-	long		(*determine_rate)(struct clk_hw *hw, unsigned long rate,
-					unsigned long *best_parent_rate,
-					struct clk_hw **best_parent_hw);
+	long		(*determine_rate)(struct clk_hw *hw,
+					  unsigned long rate,
+					  unsigned long min_rate,
+					  unsigned long max_rate,
+					  unsigned long *best_parent_rate,
+					  struct clk_hw **best_parent_hw);
 	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 rate,
@@ -555,8 +558,14 @@ 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 min_rate,
+			      unsigned long max_rate,
 			      unsigned long *best_parent_rate,
 			      struct clk_hw **best_parent_p);
+unsigned long __clk_determine_rate(struct clk_hw *core,
+				   unsigned long rate,
+				   unsigned long min_rate,
+				   unsigned long max_rate);
 
 /*
  * FIXME clock api without lock protection
include/linux/clk.h +28 -0
index c7f258a..a891e21 100644
@@ -302,6 +302,34 @@ long clk_round_rate(struct clk *clk, unsigned long rate);
 int clk_set_rate(struct clk *clk, unsigned long rate);
 
 /**
+ * clk_set_rate_range - set a rate range for a clock source
+ * @clk: clock source
+ * @min: desired minimum clock rate in Hz, inclusive
+ * @max: desired maximum clock rate in Hz, inclusive
+ *
+ * Returns success (0) or negative errno.
+ */
+int clk_set_rate_range(struct clk *clk, unsigned long min, unsigned long max);
+
+/**
+ * clk_set_min_rate - set a minimum clock rate for a clock source
+ * @clk: clock source
+ * @rate: desired minimum clock rate in Hz, inclusive
+ *
+ * Returns success (0) or negative errno.
+ */
+int clk_set_min_rate(struct clk *clk, unsigned long rate);
+
+/**
+ * clk_set_max_rate - set a maximum clock rate for a clock source
+ * @clk: clock source
+ * @rate: desired maximum clock rate in Hz, inclusive
+ *
+ * Returns success (0) or negative errno.
+ */
+int clk_set_max_rate(struct clk *clk, unsigned long rate);
+
+/**
  * clk_set_parent - set the parent clock source for this clock
  * @clk: clock source
  * @parent: parent clock source
include/linux/clk/ti.h +4 -0
index 55ef529..cfb9e55 100644
@@ -263,6 +263,8 @@ int omap3_noncore_dpll_set_rate_and_parent(struct clk_hw *hw,
 					   u8 index);
 long omap3_noncore_dpll_determine_rate(struct clk_hw *hw,
 				       unsigned long rate,
+				       unsigned long min_rate,
+				       unsigned long max_rate,
 				       unsigned long *best_parent_rate,
 				       struct clk_hw **best_parent_clk);
 unsigned long omap4_dpll_regm4xen_recalc(struct clk_hw *hw,
@@ -272,6 +274,8 @@ long omap4_dpll_regm4xen_round_rate(struct clk_hw *hw,
 				    unsigned long *parent_rate);
 long omap4_dpll_regm4xen_determine_rate(struct clk_hw *hw,
 					unsigned long rate,
+					unsigned long min_rate,
+					unsigned long max_rate,
 					unsigned long *best_parent_rate,
 					struct clk_hw **best_parent_clk);
 u8 omap2_init_dpll_parent(struct clk_hw *hw);
Signature · 1 line
1.9.3
#50 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Hi Tomeu, Mike,

Quoted 9 lines · Tomeu Vizoso

Adds a way for clock consumers to set maximum and minimum rates. This can be used for thermal drivers to set minimum rates, or by misc. drivers to set maximum rates to assure a minimum performance level.

Changes the signature of the determine_rate callback by adding the parameters min_rate and max_rate.

This is now in clk-next, and causes division by zeroes on all shmobile platforms that use renesas,cpg-div6-clock (verified on r8a73a4, r8a7740, r8a7791, sh73a0):

Division by zero in kernel.

CPU: 0 PID: 0 Comm: swapper/0 Not tainted
3.19.0-rc6-koelsch-04360-g48d797d57c5932c8-dirty #792
Hardware name: Generic R8A7791 (Flattened Device Tree)
Backtrace:
[<c001216c>] (dump_backtrace) from [<c001238c>] (show_stack+0x18/0x1c)
 r6:c051b124 r5:00000000 r4:00000000 r3:00200000
[<c0012374>] (show_stack) from [<c03955d0>] (dump_stack+0x78/0x94)
[<c0395558>] (dump_stack) from [<c00122f4>] (__div0+0x18/0x20)
 r4:2e7ddb00 r3:c05093c8
[<c00122dc>] (__div0) from [<c01bdc9c>] (Ldiv0+0x8/0x10)

[<c02d8efc>] (cpg_div6_clock_round_rate) from [<c02d56a0>] (clk_calc_new_rates+0xc8/0x1d4)
r4:eec14e00 r3:c03cb52c
[<c02d55d8>] (clk_calc_new_rates) from [<c02d57f4>] (clk_core_set_rate_nolock+0x48/0x90)
r9:eec02f40 r8:00000001 r7:c051b0b8 r6:c051b124 r5:00000000 r4:eec14e00
[<c02d57ac>] (clk_core_set_rate_nolock) from [<c02d6848>] (__clk_put+0x78/0xdc)
r7:c051b0b8 r6:c051b124 r5:eec08100 r4:eec029c0
[<c02d67d0>] (__clk_put) from [<c02d3238>] (clk_put+0x10/0x14)
r5:eec08100 r4:00000000
[<c02d3228>] (clk_put) from [<c04db4d0>] (of_clk_init+0x144/0x178) [<c04db38c>] (of_clk_init) from [<c04dbd34>] (rcar_gen2_clocks_init+0x1c/0x24)
r10:c04ed098 r9:c05023c0 r8:ffffffff r7:19432124 r6:c0502404 r5:00989680 r4:f0006000
[<c04dbd18>] (rcar_gen2_clocks_init) from [<c04c6a3c>] (rcar_gen2_timer_init+0x130/0x14c)
[<c04c690c>] (rcar_gen2_timer_init) from [<c04c15b4>] (time_init+0x24/0x38)
r7:00000000 r6:c0520c80 r5:00000000 r4:ef7fcbc0
[<c04c1590>] (time_init) from [<c04bdb84>] (start_kernel+0x248/0x3bc) [<c04bd93c>] (start_kernel) from [<40008084>] (0x40008084)
r10:00000000 r9:413fc0f2 r8:40007000 r7:c0505870 r6:c04ed094 r5:c0502440 r4:c0521054

Quoted 11 lines
drivers/clk/clk.c +1 -1
@@ -2391,25 +2543,24 @@ int __clk_get(struct clk *clk)
        return 1;
 }
 
-static void clk_core_put(struct clk_core *core)
+void __clk_put(struct clk *clk)
 {
        struct module *owner;

drivers/clk/clk.c int __clk_get(struct clk *clk) · line 2556

Quoted 10 lines
drivers/clk/clk.c +5 -2
 
-       owner = core->owner;
+       if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+               return;
 
        clk_prepare_lock();
-       kref_put(&core->ref, __clk_release);
+
+       hlist_del(&clk->child_node);
+       clk_core_set_rate_nolock(clk->core, clk->core->req_rate);

At this point, clk->core->req_rate is still zero, causing cpg_div6_clock_round_rate() to be called with a zero "rate" parameter, e.g. on r8a7791:

cpg_div6_clock_round_rate: clock sd2 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd1 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock mmc0 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd1 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd1 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd2 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd2 rate 0 parent_rate 780000000 div 1

and cpg_div6_clock_calc_div() is called to calculate

        div = DIV_ROUND_CLOSEST(parent_rate, rate);

Why was this call to clk_core_set_rate_nolock() in __clk_put() added? Before, there was no rate setting done at this point, and cpg_div6_clock_round_rate() was not called.

Have the semantics changed? Should .round_rate() be ready to accept a "zero" rate, and use e.g. the current rate instead?

Quoted 6 lines
drivers/clk/clk.c +3 -0
+       owner = clk->core->owner;
+       kref_put(&clk->core->ref, __clk_release);
+
        clk_prepare_unlock();
 
        module_put(owner);

drivers/clk/clk.c int __clk_get(struct clk *clk) · line 2565

Quoted 10 lines
drivers/clk/clk.c +0 -7
-}
-
-void __clk_put(struct clk *clk)
-{
-       if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
-               return;
 
-       clk_core_put(clk->core);
        kfree(clk);
 }

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

#51 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Quoted 19 lines · Geert Uytterhoeven

Hi Tomeu, Mike,

Quoted 15 lines · Tomeu Vizoso
drivers/clk/clk.c +3 -2
@@ -2391,25 +2543,24 @@ int __clk_get(struct clk *clk)
        return 1;
 }
 
-static void clk_core_put(struct clk_core *core)
+void __clk_put(struct clk *clk)
 {
        struct module *owner;
 
-       owner = core->owner;
+       if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+               return;
 
Quoted 8 lines

drivers/clk/clk.c int __clk_get(struct clk *clk) · line 2556

Quoted 5 lines
drivers/clk/clk.c +3 -1
        clk_prepare_lock();
-       kref_put(&core->ref, __clk_release);
+
+       hlist_del(&clk->child_node);
+       clk_core_set_rate_nolock(clk->core, clk->core->req_rate);

At this point, clk->core->req_rate is still zero, causing cpg_div6_clock_round_rate() to be called with a zero "rate" parameter, e.g. on r8a7791:

Hmm.. I wonder if we should assign core->req_rate to be the same as core->rate during __clk_init()? That would make this call to clk_core_set_rate_nolock() a nop in this case.

Quoted 9 lines

cpg_div6_clock_round_rate: clock sd2 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd1 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock mmc0 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd1 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd1 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd2 rate 0 parent_rate 780000000 div 1
cpg_div6_clock_round_rate: clock sd2 rate 0 parent_rate 780000000 div 1

Quoted 7 lines

and cpg_div6_clock_calc_div() is called to calculate

        div = DIV_ROUND_CLOSEST(parent_rate, rate);

Why was this call to clk_core_set_rate_nolock() in __clk_put() added? Before, there was no rate setting done at this point, and cpg_div6_clock_round_rate() was not called.

We need to call clk_core_set_rate_nolock() here to drop any min/max rate request that this consumer has.

Quoted 3 lines

Have the semantics changed? Should .round_rate() be ready to accept a "zero" rate, and use e.g. the current rate instead?

It seems like we've also exposed a bug in cpg_div6_clock_calc_div(). Technically any driver could have called clk_round_rate() with a zero rate before this change and that would have caused the same division by zero.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#52 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Quoted 29 lines · Stephen Boyd
Quoted 19 lines · Geert Uytterhoeven

Hi Tomeu, Mike,

Quoted 15 lines · Tomeu Vizoso
drivers/clk/clk.c +3 -2
@@ -2391,25 +2543,24 @@ int __clk_get(struct clk *clk)
        return 1;
 }
 
-static void clk_core_put(struct clk_core *core)
+void __clk_put(struct clk *clk)
 {
        struct module *owner;
 
-       owner = core->owner;
+       if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+               return;
 
Quoted 8 lines

drivers/clk/clk.c int __clk_get(struct clk *clk) · line 2556

Quoted 5 lines
drivers/clk/clk.c +3 -1
        clk_prepare_lock();
-       kref_put(&core->ref, __clk_release);
+
+       hlist_del(&clk->child_node);
+       clk_core_set_rate_nolock(clk->core, clk->core->req_rate);

At this point, clk->core->req_rate is still zero, causing cpg_div6_clock_round_rate() to be called with a zero "rate" parameter, e.g. on r8a7791:

Quoted 4 lines

Hmm.. I wonder if we should assign core->req_rate to be the same as core->rate during __clk_init()? That would make this call to clk_core_set_rate_nolock() a nop in this case.

Here's a patch to do this

---8<----
From: Stephen Boyd <[email protected]>
Subject: [PATCH] clk: Assign a requested rate by default

We need to assign a requested rate here so that we avoid requesting a rate of 0 on clocks when we remove clock consumers.

---
 drivers/clk/clk.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/clk/clk.c +5 -3
index a29daf9edea4..8416ed1c40be 100644
@@ -2142,6 +2142,7 @@ int __clk_init(struct device *dev, struct clk *clk_user)
 	struct clk_core *orphan;
 	struct hlist_node *tmp2;
 	struct clk_core *clk;
+	unsigned long rate;
 
 	if (!clk_user)
 		return -EINVAL;
@@ -2266,12 +2267,13 @@ int __clk_init(struct device *dev, struct clk *clk_user)
 	 * then rate is set to zero.
 	 */
 	if (clk->ops->recalc_rate)
-		clk->rate = clk->ops->recalc_rate(clk->hw,
+		rate = clk->ops->recalc_rate(clk->hw,
 				clk_core_get_rate_nolock(clk->parent));
 	else if (clk->parent)
-		clk->rate = clk->parent->rate;
+		rate = clk->parent->rate;
 	else
-		clk->rate = 0;
+		rate = 0;
+	clk->rate = clk->req_rate = rate;
 
 	/*
 	 * walk the list of orphan clocks and reparent any that are children of
Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
author Reviewed-by
#53 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Quoted 35 lines · Stephen Boyd
Quoted 29 lines · Stephen Boyd
Quoted 19 lines · Geert Uytterhoeven

Hi Tomeu, Mike,

Quoted 15 lines · Tomeu Vizoso
drivers/clk/clk.c +3 -2
@@ -2391,25 +2543,24 @@ int __clk_get(struct clk *clk)
        return 1;
 }
 
-static void clk_core_put(struct clk_core *core)
+void __clk_put(struct clk *clk)
 {
        struct module *owner;
 
-       owner = core->owner;
+       if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+               return;
 
Quoted 8 lines

drivers/clk/clk.c int __clk_get(struct clk *clk) · line 2556

Quoted 5 lines
drivers/clk/clk.c +3 -1
        clk_prepare_lock();
-       kref_put(&core->ref, __clk_release);
+
+       hlist_del(&clk->child_node);
+       clk_core_set_rate_nolock(clk->core, clk->core->req_rate);

At this point, clk->core->req_rate is still zero, causing cpg_div6_clock_round_rate() to be called with a zero "rate" parameter, e.g. on r8a7791:

Quoted 4 lines

Hmm.. I wonder if we should assign core->req_rate to be the same as core->rate during __clk_init()? That would make this call to clk_core_set_rate_nolock() a nop in this case.

Quoted 8 lines

Here's a patch to do this

---8<----
From: Stephen Boyd <[email protected]>
Subject: [PATCH] clk: Assign a requested rate by default

We need to assign a requested rate here so that we avoid requesting a rate of 0 on clocks when we remove clock consumers.

Hi, this looks good to me.

Thanks,

Tomeu

Quoted 42 lines
---
 drivers/clk/clk.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/clk/clk.c +5 -4
index a29daf9edea4..8416ed1c40be 100644
@@ -2142,6 +2142,7 @@ int __clk_init(struct device *dev, struct clk *clk_user)
        struct clk_core *orphan;
        struct hlist_node *tmp2;
        struct clk_core *clk;
+       unsigned long rate;
 
        if (!clk_user)
                return -EINVAL;
@@ -2266,12 +2267,13 @@ int __clk_init(struct device *dev, struct clk *clk_user)
         * then rate is set to zero.
         */
        if (clk->ops->recalc_rate)
-               clk->rate = clk->ops->recalc_rate(clk->hw,
+               rate = clk->ops->recalc_rate(clk->hw,
                                clk_core_get_rate_nolock(clk->parent));
        else if (clk->parent)
-               clk->rate = clk->parent->rate;
+               rate = clk->parent->rate;
        else
-               clk->rate = 0;
+               rate = 0;
+       clk->rate = clk->req_rate = rate;
 
        /*
         * walk the list of orphan clocks and reparent any that are children of
--

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

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#54 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Quoting Tomeu Vizoso (2015-01-31 10:36:22)

Quoted 42 lines · Stephen Boyd
Quoted 41 lines · Stephen Boyd
Quoted 29 lines · Geert Uytterhoeven
Quoted 19 lines

Hi Tomeu, Mike,

Quoted 15 lines · Tomeu Vizoso
drivers/clk/clk.c +3 -2
@@ -2391,25 +2543,24 @@ int __clk_get(struct clk *clk)
        return 1;
 }
 
-static void clk_core_put(struct clk_core *core)
+void __clk_put(struct clk *clk)
 {
        struct module *owner;
 
-       owner = core->owner;
+       if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+               return;
 
Quoted 8 lines

drivers/clk/clk.c int __clk_get(struct clk *clk) · line 2556

Quoted 5 lines
drivers/clk/clk.c +3 -1
        clk_prepare_lock();
-       kref_put(&core->ref, __clk_release);
+
+       hlist_del(&clk->child_node);
+       clk_core_set_rate_nolock(clk->core, clk->core->req_rate);

At this point, clk->core->req_rate is still zero, causing cpg_div6_clock_round_rate() to be called with a zero "rate" parameter, e.g. on r8a7791:

Quoted 4 lines

Hmm.. I wonder if we should assign core->req_rate to be the same as core->rate during __clk_init()? That would make this call to clk_core_set_rate_nolock() a nop in this case.

Here's a patch to do this

---8<----
From: Stephen Boyd <[email protected]>
Subject: [PATCH] clk: Assign a requested rate by default

Quoted 6 lines
Quoted 2 lines

We need to assign a requested rate here so that we avoid requesting a rate of 0 on clocks when we remove clock consumers.

Hi, this looks good to me.

It seems to fix the total boot failure on OMAPs, and hopefully does the same for SH Mobile and others. I've squashed this into Tomeu's rate constraints patch to maintain bisect.

Regards,
Mike

Quoted 47 lines

Thanks,

Tomeu

Quoted 42 lines
---
 drivers/clk/clk.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/clk/clk.c +5 -4
index a29daf9edea4..8416ed1c40be 100644
@@ -2142,6 +2142,7 @@ int __clk_init(struct device *dev, struct clk *clk_user)
        struct clk_core *orphan;
        struct hlist_node *tmp2;
        struct clk_core *clk;
+       unsigned long rate;
 
        if (!clk_user)
                return -EINVAL;
@@ -2266,12 +2267,13 @@ int __clk_init(struct device *dev, struct clk *clk_user)
         * then rate is set to zero.
         */
        if (clk->ops->recalc_rate)
-               clk->rate = clk->ops->recalc_rate(clk->hw,
+               rate = clk->ops->recalc_rate(clk->hw,
                                clk_core_get_rate_nolock(clk->parent));
        else if (clk->parent)
-               clk->rate = clk->parent->rate;
+               rate = clk->parent->rate;
        else
-               clk->rate = 0;
+               rate = 0;
+       clk->rate = clk->req_rate = rate;
 
        /*
         * walk the list of orphan clocks and reparent any that are children of
--

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

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#55 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Quoted 46 lines · Mike Turquette

Quoting Tomeu Vizoso (2015-01-31 10:36:22)

Quoted 45 lines · Stephen Boyd
Quoted 43 lines · Stephen Boyd
Quoted 29 lines · Geert Uytterhoeven
Quoted 19 lines

Hi Tomeu, Mike,

Quoted 15 lines · Tomeu Vizoso
drivers/clk/clk.c +3 -2
@@ -2391,25 +2543,24 @@ int __clk_get(struct clk *clk)
        return 1;
 }
 
-static void clk_core_put(struct clk_core *core)
+void __clk_put(struct clk *clk)
 {
        struct module *owner;
 
-       owner = core->owner;
+       if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+               return;
 
Quoted 8 lines

drivers/clk/clk.c int __clk_get(struct clk *clk) · line 2556

Quoted 5 lines
drivers/clk/clk.c +3 -1
        clk_prepare_lock();
-       kref_put(&core->ref, __clk_release);
+
+       hlist_del(&clk->child_node);
+       clk_core_set_rate_nolock(clk->core, clk->core->req_rate);

At this point, clk->core->req_rate is still zero, causing cpg_div6_clock_round_rate() to be called with a zero "rate" parameter, e.g. on r8a7791:

Quoted 4 lines

Hmm.. I wonder if we should assign core->req_rate to be the same as core->rate during __clk_init()? That would make this call to clk_core_set_rate_nolock() a nop in this case.

Here's a patch to do this

---8<----
From: Stephen Boyd <[email protected]>
Subject: [PATCH] clk: Assign a requested rate by default

We need to assign a requested rate here so that we avoid requesting a rate of 0 on clocks when we remove clock consumers.

Quoted 7 lines
Quoted 3 lines

Hi, this looks good to me.

It seems to fix the total boot failure on OMAPs, and hopefully does the same for SH Mobile and others. I've squashed this into Tomeu's rate constraints patch to maintain bisect.

Yes, it fixes shmobile. .round_rate() is now called with a sane value of rate.

Thanks!

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

maintainer
#56 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Quoted 49 lines · Geert Uytterhoeven
Quoted 48 lines · Mike Turquette

Quoting Tomeu Vizoso (2015-01-31 10:36:22)

Quoted 47 lines · Stephen Boyd
Quoted 35 lines · Stephen Boyd
Quoted 29 lines · Geert Uytterhoeven
Quoted 19 lines

Hi Tomeu, Mike,

Quoted 15 lines · Tomeu Vizoso
drivers/clk/clk.c +3 -2
@@ -2391,25 +2543,24 @@ int __clk_get(struct clk *clk)
        return 1;
 }
 
-static void clk_core_put(struct clk_core *core)
+void __clk_put(struct clk *clk)
 {
        struct module *owner;
 
-       owner = core->owner;
+       if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+               return;
 
Quoted 8 lines

drivers/clk/clk.c int __clk_get(struct clk *clk) · line 2556

Quoted 5 lines
drivers/clk/clk.c +3 -1
        clk_prepare_lock();
-       kref_put(&core->ref, __clk_release);
+
+       hlist_del(&clk->child_node);
+       clk_core_set_rate_nolock(clk->core, clk->core->req_rate);

At this point, clk->core->req_rate is still zero, causing cpg_div6_clock_round_rate() to be called with a zero "rate" parameter, e.g. on r8a7791:

Quoted 4 lines

Hmm.. I wonder if we should assign core->req_rate to be the same as core->rate during __clk_init()? That would make this call to clk_core_set_rate_nolock() a nop in this case.

Quoted 8 lines

Here's a patch to do this

---8<----
From: Stephen Boyd <[email protected]>
Subject: [PATCH] clk: Assign a requested rate by default

We need to assign a requested rate here so that we avoid requesting a rate of 0 on clocks when we remove clock consumers.

Hi, this looks good to me.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

It seems to fix the total boot failure on OMAPs, and hopefully does the same for SH Mobile and others. I've squashed this into Tomeu's rate constraints patch to maintain bisect.

Yes, it fixes shmobile. .round_rate() is now called with a sane value of rate.

Looks like next-20150202 now produces tons of the following errors, these from omap4:

[   10.568206] ------------[ cut here ]------------
[   10.568206] WARNING: CPU: 0 PID: 1 at drivers/clk/clk.c:925 clk_disable+0x28/0x34()
[   10.568237] Modules linked in:
[   10.568237] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W       3.19.0-rc6-next-20150202 #2037
[   10.568237] Hardware name: Generic OMAP4 (Flattened Device Tree)
[   10.568267] [<c0015bdc>] (unwind_backtrace) from [<c001222c>] (show_stack+0x10/0x14)
[   10.568267] [<c001222c>] (show_stack) from [<c05d2014>] (dump_stack+0x84/0x9c)
[   10.568267] [<c05d2014>] (dump_stack) from [<c003ea90>] (warn_slowpath_common+0x7c/0xb8)
[   10.568298] [<c003ea90>] (warn_slowpath_common) from [<c003eb68>] (warn_slowpath_null+0x1c/0x24)
[   10.568298] [<c003eb68>] (warn_slowpath_null) from [<c04c1ffc>] (clk_disable+0x28/0x34)
[   10.568328] [<c04c1ffc>] (clk_disable) from [<c0025b3c>] (_disable_clocks+0x18/0x68)
[   10.568328] [<c0025b3c>] (_disable_clocks) from [<c0026f14>] (_idle+0x10c/0x214)
[   10.568328] [<c0026f14>] (_idle) from [<c0855fac>] (_setup+0x338/0x410)
[   10.568359] [<c0855fac>] (_setup) from [<c0027360>] (omap_hwmod_for_each+0x34/0x60)
[   10.568359] [<c0027360>] (omap_hwmod_for_each) from [<c08563c4>] (__omap_hwmod_setup_all+0x30/0x40)
[   10.568389] [<c08563c4>] (__omap_hwmod_setup_all) from [<c0008a04>] (do_one_initcall+0x80/0x1dc)
[   10.568389] [<c0008a04>] (do_one_initcall) from [<c0848ea0>] (kernel_init_freeable+0x204/0x2d0)
[   10.568420] [<c0848ea0>] (kernel_init_freeable) from [<c05cdab8>] (kernel_init+0x8/0xec)
[   10.568420] [<c05cdab8>] (kernel_init) from [<c000e790>] (ret_from_fork+0x14/0x24)
[   10.568420] ---[ end trace cb88537fdc8fa211 ]---
[   10.568450] ------------[ cut here ]------------
[   10.568450] WARNING: CPU: 0 PID: 1 at arch/arm/mach-omap2/dpll3xxx.c:436 omap3_noncore_dpll_enable+0xdc/0
x10c()
[   10.568450] Modules linked in:
[   10.568481] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W       3.19.0-rc6-next-20150202 #2037
[   10.568481] Hardware name: Generic OMAP4 (Flattened Device Tree)
[   10.568481] [<c0015bdc>] (unwind_backtrace) from [<c001222c>] (show_stack+0x10/0x14)
[   10.568511] [<c001222c>] (show_stack) from [<c05d2014>] (dump_stack+0x84/0x9c)
[   10.568511] [<c05d2014>] (dump_stack) from [<c003ea90>] (warn_slowpath_common+0x7c/0xb8)
[   10.568511] [<c003ea90>] (warn_slowpath_common) from [<c003eb68>] (warn_slowpath_null+0x1c/0x24)
[   10.568542] [<c003eb68>] (warn_slowpath_null) from [<c0035800>] (omap3_noncore_dpll_enable+0xdc/0x10c)
[   10.568542] [<c0035800>] (omap3_noncore_dpll_enable) from [<c04c0a10>] (clk_core_enable+0x60/0x9c)
[   10.568572] [<c04c0a10>] (clk_core_enable) from [<c04c09f0>] (clk_core_enable+0x40/0x9c)
[   10.568572] ---[ end trace cb88537fdc8fa212 ]---

...

Regards,

Tony

#57 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Quoting Tony Lindgren (2015-02-02 08:12:37)

Quoted 52 lines · Geert Uytterhoeven
Quoted 51 lines · Mike Turquette
Quoted 50 lines

Quoting Tomeu Vizoso (2015-01-31 10:36:22)

Quoted 48 lines · Stephen Boyd
Quoted 35 lines · Stephen Boyd
Quoted 29 lines · Geert Uytterhoeven
Quoted 19 lines

Hi Tomeu, Mike,

Quoted 15 lines · Tomeu Vizoso
drivers/clk/clk.c +3 -2
@@ -2391,25 +2543,24 @@ int __clk_get(struct clk *clk)
        return 1;
 }
 
-static void clk_core_put(struct clk_core *core)
+void __clk_put(struct clk *clk)
 {
        struct module *owner;
 
-       owner = core->owner;
+       if (!clk || WARN_ON_ONCE(IS_ERR(clk)))
+               return;
 
Quoted 8 lines

drivers/clk/clk.c int __clk_get(struct clk *clk) · line 2556

Quoted 5 lines
drivers/clk/clk.c +3 -1
        clk_prepare_lock();
-       kref_put(&core->ref, __clk_release);
+
+       hlist_del(&clk->child_node);
+       clk_core_set_rate_nolock(clk->core, clk->core->req_rate);

At this point, clk->core->req_rate is still zero, causing cpg_div6_clock_round_rate() to be called with a zero "rate" parameter, e.g. on r8a7791:

Quoted 4 lines

Hmm.. I wonder if we should assign core->req_rate to be the same as core->rate during __clk_init()? That would make this call to clk_core_set_rate_nolock() a nop in this case.

Quoted 8 lines

Here's a patch to do this

---8<----
From: Stephen Boyd <[email protected]>
Subject: [PATCH] clk: Assign a requested rate by default

We need to assign a requested rate here so that we avoid requesting a rate of 0 on clocks when we remove clock consumers.

Hi, this looks good to me.

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

It seems to fix the total boot failure on OMAPs, and hopefully does the same for SH Mobile and others. I've squashed this into Tomeu's rate constraints patch to maintain bisect.

Yes, it fixes shmobile. .round_rate() is now called with a sane value of rate.

Looks like next-20150202 now produces tons of the following errors, these from omap4:

next-20150202 is the rolled-back changes from last Friday. I removed the clock constraints patch and in doing so also rolled back the TI clock driver migration and clk-private.h removal patches. Those are all back in clk-next as of last night and it looks as though they missed being pulled into todays linux-next by a matter of minutes :-/

Quoted 13 lines
[   10.568206] ------------[ cut here ]------------
[   10.568206] WARNING: CPU: 0 PID: 1 at drivers/clk/clk.c:925 clk_disable+0x28/0x34()
[   10.568237] Modules linked in:
[   10.568237] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W       3.19.0-rc6-next-20150202 #2037
[   10.568237] Hardware name: Generic OMAP4 (Flattened Device Tree)
[   10.568267] [<c0015bdc>] (unwind_backtrace) from [<c001222c>] (show_stack+0x10/0x14)
[   10.568267] [<c001222c>] (show_stack) from [<c05d2014>] (dump_stack+0x84/0x9c)
[   10.568267] [<c05d2014>] (dump_stack) from [<c003ea90>] (warn_slowpath_common+0x7c/0xb8)
[   10.568298] [<c003ea90>] (warn_slowpath_common) from [<c003eb68>] (warn_slowpath_null+0x1c/0x24)
[   10.568298] [<c003eb68>] (warn_slowpath_null) from [<c04c1ffc>] (clk_disable+0x28/0x34)
[   10.568328] [<c04c1ffc>] (clk_disable) from [<c0025b3c>] (_disable_clocks+0x18/0x68)
[   10.568328] [<c0025b3c>] (_disable_clocks) from [<c0026f14>] (_idle+0x10c/0x214)
Quoted 8 lines
[   10.568328] [<c0026f14>] (_idle) from [<c0855fac>] (_setup+0x338/0x410)
[   10.568359] [<c0855fac>] (_setup) from [<c0027360>] (omap_hwmod_for_each+0x34/0x60)
[   10.568359] [<c0027360>] (omap_hwmod_for_each) from [<c08563c4>] (__omap_hwmod_setup_all+0x30/0x40)
[   10.568389] [<c08563c4>] (__omap_hwmod_setup_all) from [<c0008a04>] (do_one_initcall+0x80/0x1dc)
[   10.568389] [<c0008a04>] (do_one_initcall) from [<c0848ea0>] (kernel_init_freeable+0x204/0x2d0)
[   10.568420] [<c0848ea0>] (kernel_init_freeable) from [<c05cdab8>] (kernel_init+0x8/0xec)
[   10.568420] [<c05cdab8>] (kernel_init) from [<c000e790>] (ret_from_fork+0x14/0x24)
[   10.568420] ---[ end trace cb88537fdc8fa211 ]---

This looks like mis-matched enable/disable calls. We now have unique struct clk pointers for every call to clk_get. I haven't yet looked through the hwmod code but I have a feeling that we're doing something like this:

	/* enable clock */
	my_clk = clk_get(...);
	clk_prepare_enable(my_clk);
	clk_put(my_clk);

	/* do some work */
	do_work();

	/* disable clock */
	my_clk = clk_get(...);
	clk_disable_unprepare(my_clk);
	clk_put(my_clk);

The above pattern no longer works since my_clk will be two different unique pointers, but it really should be one stable pointer across the whole usage of the clk. E.g:

	/* enable clock */
	my_clk = clk_get(...);
	clk_prepare_enable(my_clk);

	/* do some work */
	do_work();

	/* disable clock */
	clk_disable_unprepare(my_clk);
	clk_put(my_clk);

Again, I haven't looked through the code, so the above is just an educated guess.

Anyways I am testing with an OMAP4460 Panda ES and I didn't see the above. Is there a test you are running to get this?

Quoted 16 lines

[   10.568450] ------------[ cut here ]------------
[   10.568450] WARNING: CPU: 0 PID: 1 at arch/arm/mach-omap2/dpll3xxx.c:436 omap3_noncore_dpll_enable+0xdc/0
x10c()
[   10.568450] Modules linked in:
[   10.568481] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W       3.19.0-rc6-next-20150202 #2037
[   10.568481] Hardware name: Generic OMAP4 (Flattened Device Tree)
[   10.568481] [<c0015bdc>] (unwind_backtrace) from [<c001222c>] (show_stack+0x10/0x14)
[   10.568511] [<c001222c>] (show_stack) from [<c05d2014>] (dump_stack+0x84/0x9c)
[   10.568511] [<c05d2014>] (dump_stack) from [<c003ea90>] (warn_slowpath_common+0x7c/0xb8)
[   10.568511] [<c003ea90>] (warn_slowpath_common) from [<c003eb68>] (warn_slowpath_null+0x1c/0x24)
[   10.568542] [<c003eb68>] (warn_slowpath_null) from [<c0035800>] (omap3_noncore_dpll_enable+0xdc/0x10c)
[   10.568542] [<c0035800>] (omap3_noncore_dpll_enable) from [<c04c0a10>] (clk_core_enable+0x60/0x9c)
[   10.568572] [<c04c0a10>] (clk_core_enable) from [<c04c09f0>] (clk_core_enable+0x40/0x9c)
[   10.568572] ---[ end trace cb88537fdc8fa212 ]---
Quoted 1 line

...

This is the same issue discussed already in this thread[0]. Feedback from Tero & Paul on how to handle it would be nice.

Please let me know if anything else breaks for you.

Regards,
Mike

Quoted 4 lines

Regards,

Tony

#58 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Quoted 18 lines · Mike Turquette

This looks like mis-matched enable/disable calls. We now have unique struct clk pointers for every call to clk_get. I haven't yet looked through the hwmod code but I have a feeling that we're doing something like this:

	/* enable clock */
	my_clk = clk_get(...);
	clk_prepare_enable(my_clk);
	clk_put(my_clk);

	/* do some work */
	do_work();

	/* disable clock */
	my_clk = clk_get(...);
	clk_disable_unprepare(my_clk);
	clk_put(my_clk);
Quoted 3 lines

The above pattern no longer works since my_clk will be two different unique pointers, but it really should be one stable pointer across the whole usage of the clk. E.g:

Yes, it has always been documented that shall be the case. Anyone doing the above is basically broken.

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
maintainer
#59 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

In reply to Mike Turquette
Quoted 22 lines · Mike Turquette

Quoting Tony Lindgren (2015-02-02 08:12:37)

Quoted 21 lines
[   10.568206] ------------[ cut here ]------------
[   10.568206] WARNING: CPU: 0 PID: 1 at drivers/clk/clk.c:925 clk_disable+0x28/0x34()
[   10.568237] Modules linked in:
[   10.568237] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W       3.19.0-rc6-next-20150202 #2037
[   10.568237] Hardware name: Generic OMAP4 (Flattened Device Tree)
[   10.568267] [<c0015bdc>] (unwind_backtrace) from [<c001222c>] (show_stack+0x10/0x14)
[   10.568267] [<c001222c>] (show_stack) from [<c05d2014>] (dump_stack+0x84/0x9c)
[   10.568267] [<c05d2014>] (dump_stack) from [<c003ea90>] (warn_slowpath_common+0x7c/0xb8)
[   10.568298] [<c003ea90>] (warn_slowpath_common) from [<c003eb68>] (warn_slowpath_null+0x1c/0x24)
[   10.568298] [<c003eb68>] (warn_slowpath_null) from [<c04c1ffc>] (clk_disable+0x28/0x34)
[   10.568328] [<c04c1ffc>] (clk_disable) from [<c0025b3c>] (_disable_clocks+0x18/0x68)
[   10.568328] [<c0025b3c>] (_disable_clocks) from [<c0026f14>] (_idle+0x10c/0x214)
[   10.568328] [<c0026f14>] (_idle) from [<c0855fac>] (_setup+0x338/0x410)
[   10.568359] [<c0855fac>] (_setup) from [<c0027360>] (omap_hwmod_for_each+0x34/0x60)
[   10.568359] [<c0027360>] (omap_hwmod_for_each) from [<c08563c4>] (__omap_hwmod_setup_all+0x30/0x40)
[   10.568389] [<c08563c4>] (__omap_hwmod_setup_all) from [<c0008a04>] (do_one_initcall+0x80/0x1dc)
[   10.568389] [<c0008a04>] (do_one_initcall) from [<c0848ea0>] (kernel_init_freeable+0x204/0x2d0)
[   10.568420] [<c0848ea0>] (kernel_init_freeable) from [<c05cdab8>] (kernel_init+0x8/0xec)
[   10.568420] [<c05cdab8>] (kernel_init) from [<c000e790>] (ret_from_fork+0x14/0x24)
[   10.568420] ---[ end trace cb88537fdc8fa211 ]---

For reference, the above is line 992 in clk-next.

Quoted 33 lines

This looks like mis-matched enable/disable calls. We now have unique struct clk pointers for every call to clk_get. I haven't yet looked through the hwmod code but I have a feeling that we're doing something like this:

	/* enable clock */
	my_clk = clk_get(...);
	clk_prepare_enable(my_clk);
	clk_put(my_clk);

	/* do some work */
	do_work();

	/* disable clock */
	my_clk = clk_get(...);
	clk_disable_unprepare(my_clk);
	clk_put(my_clk);

The above pattern no longer works since my_clk will be two different unique pointers, but it really should be one stable pointer across the whole usage of the clk. E.g:

	/* enable clock */
	my_clk = clk_get(...);
	clk_prepare_enable(my_clk);

	/* do some work */
	do_work();

	/* disable clock */
	clk_disable_unprepare(my_clk);
	clk_put(my_clk);
Quoted 5 lines

Again, I haven't looked through the code, so the above is just an educated guess.

Anyways I am testing with an OMAP4460 Panda ES and I didn't see the above. Is there a test you are running to get this?

Just booting 4430-sdp with omap2plus_defconfig. And git bisect points to 59cf3fcf9baf ("clk: Make clk API return per-user struct clk instances") as you already guessed.

Quoted 16 lines
Quoted 14 lines
[   10.568450] ------------[ cut here ]------------
[   10.568450] WARNING: CPU: 0 PID: 1 at arch/arm/mach-omap2/dpll3xxx.c:436 omap3_noncore_dpll_enable+0xdc/0
x10c()
[   10.568450] Modules linked in:
[   10.568481] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W       3.19.0-rc6-next-20150202 #2037
[   10.568481] Hardware name: Generic OMAP4 (Flattened Device Tree)
[   10.568481] [<c0015bdc>] (unwind_backtrace) from [<c001222c>] (show_stack+0x10/0x14)
[   10.568511] [<c001222c>] (show_stack) from [<c05d2014>] (dump_stack+0x84/0x9c)
[   10.568511] [<c05d2014>] (dump_stack) from [<c003ea90>] (warn_slowpath_common+0x7c/0xb8)
[   10.568511] [<c003ea90>] (warn_slowpath_common) from [<c003eb68>] (warn_slowpath_null+0x1c/0x24)
[   10.568542] [<c003eb68>] (warn_slowpath_null) from [<c0035800>] (omap3_noncore_dpll_enable+0xdc/0x10c)
[   10.568542] [<c0035800>] (omap3_noncore_dpll_enable) from [<c04c0a10>] (clk_core_enable+0x60/0x9c)
[   10.568572] [<c04c0a10>] (clk_core_enable) from [<c04c09f0>] (clk_core_enable+0x40/0x9c)
[   10.568572] ---[ end trace cb88537fdc8fa212 ]---
Quoted 1 line

...

Quoted 2 lines

This is the same issue discussed already in this thread[0]. Feedback from Tero & Paul on how to handle it would be nice.

Yes that one is a separate issue.

Quoted 1 line

Please let me know if anything else breaks for you.

Also off-idle on omap3 seems to be broken by commit 59cf3fcf9baf.

Regards,

Tony

maintainer
#60 lore raw

Re: [PATCH v13 4/6] clk: Add rate constraints to clocks

Quoted 25 lines · Tony Lindgren
Quoted 22 lines · Mike Turquette

Quoting Tony Lindgren (2015-02-02 08:12:37)

Quoted 21 lines
[   10.568206] ------------[ cut here ]------------
[   10.568206] WARNING: CPU: 0 PID: 1 at drivers/clk/clk.c:925 clk_disable+0x28/0x34()
[   10.568237] Modules linked in:
[   10.568237] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W       3.19.0-rc6-next-20150202 #2037
[   10.568237] Hardware name: Generic OMAP4 (Flattened Device Tree)
[   10.568267] [<c0015bdc>] (unwind_backtrace) from [<c001222c>] (show_stack+0x10/0x14)
[   10.568267] [<c001222c>] (show_stack) from [<c05d2014>] (dump_stack+0x84/0x9c)
[   10.568267] [<c05d2014>] (dump_stack) from [<c003ea90>] (warn_slowpath_common+0x7c/0xb8)
[   10.568298] [<c003ea90>] (warn_slowpath_common) from [<c003eb68>] (warn_slowpath_null+0x1c/0x24)
[   10.568298] [<c003eb68>] (warn_slowpath_null) from [<c04c1ffc>] (clk_disable+0x28/0x34)
[   10.568328] [<c04c1ffc>] (clk_disable) from [<c0025b3c>] (_disable_clocks+0x18/0x68)
[   10.568328] [<c0025b3c>] (_disable_clocks) from [<c0026f14>] (_idle+0x10c/0x214)
[   10.568328] [<c0026f14>] (_idle) from [<c0855fac>] (_setup+0x338/0x410)
[   10.568359] [<c0855fac>] (_setup) from [<c0027360>] (omap_hwmod_for_each+0x34/0x60)
[   10.568359] [<c0027360>] (omap_hwmod_for_each) from [<c08563c4>] (__omap_hwmod_setup_all+0x30/0x40)
[   10.568389] [<c08563c4>] (__omap_hwmod_setup_all) from [<c0008a04>] (do_one_initcall+0x80/0x1dc)
[   10.568389] [<c0008a04>] (do_one_initcall) from [<c0848ea0>] (kernel_init_freeable+0x204/0x2d0)
[   10.568420] [<c0848ea0>] (kernel_init_freeable) from [<c05cdab8>] (kernel_init+0x8/0xec)
[   10.568420] [<c05cdab8>] (kernel_init) from [<c000e790>] (ret_from_fork+0x14/0x24)
[   10.568420] ---[ end trace cb88537fdc8fa211 ]---

For reference, the above is line 992 in clk-next.

...

Quoted 20 lines

Just booting 4430-sdp with omap2plus_defconfig. And git bisect points to 59cf3fcf9baf ("clk: Make clk API return per-user struct clk instances") as you already guessed.

Quoted 16 lines
Quoted 15 lines
[   10.568450] ------------[ cut here ]------------
[   10.568450] WARNING: CPU: 0 PID: 1 at arch/arm/mach-omap2/dpll3xxx.c:436 omap3_noncore_dpll_enable+0xdc/0
x10c()
[   10.568450] Modules linked in:
[   10.568481] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W       3.19.0-rc6-next-20150202 #2037
[   10.568481] Hardware name: Generic OMAP4 (Flattened Device Tree)
[   10.568481] [<c0015bdc>] (unwind_backtrace) from [<c001222c>] (show_stack+0x10/0x14)
[   10.568511] [<c001222c>] (show_stack) from [<c05d2014>] (dump_stack+0x84/0x9c)
[   10.568511] [<c05d2014>] (dump_stack) from [<c003ea90>] (warn_slowpath_common+0x7c/0xb8)
[   10.568511] [<c003ea90>] (warn_slowpath_common) from [<c003eb68>] (warn_slowpath_null+0x1c/0x24)
[   10.568542] [<c003eb68>] (warn_slowpath_null) from [<c0035800>] (omap3_noncore_dpll_enable+0xdc/0x10c)
[   10.568542] [<c0035800>] (omap3_noncore_dpll_enable) from [<c04c0a10>] (clk_core_enable+0x60/0x9c)
[   10.568572] [<c04c0a10>] (clk_core_enable) from [<c04c09f0>] (clk_core_enable+0x40/0x9c)
[   10.568572] ---[ end trace cb88537fdc8fa212 ]---

...

Quoted 8 lines
Quoted 2 lines

This is the same issue discussed already in this thread[0]. Feedback from Tero & Paul on how to handle it would be nice.

Yes that one is a separate issue.

Quoted 1 line

Please let me know if anything else breaks for you.

Also off-idle on omap3 seems to be broken by commit 59cf3fcf9baf.

Actually the two warnigns above are all caused by the same issue. And the patch Tero just posted fixes both of the issue. It's the thread "[PATCH v13 3/6] clk: Make clk API return per-user struct clk instances" for reference.

Regards,

Tony

#61 lore raw

PATCH v13 5/6 clkdev: Export clk_register_clkdev

In reply to Tomeu Vizoso

So it can be used from modules such as clk-test.ko.

---
 drivers/clk/clkdev.c | 1 +
 1 file changed, 1 insertion(+)
drivers/clk/clkdev.c +1 -0
index 901d242..29a1ab7 100644
@@ -377,6 +377,7 @@ int clk_register_clkdev(struct clk *clk, const char *con_id,
 
 	return 0;
 }
+EXPORT_SYMBOL(clk_register_clkdev);
 
 /**
  * clk_register_clkdevs - register a set of clk_lookup for a struct clk
Signature · 1 line
1.9.3
#62 lore raw

Re: [PATCH v13 5/6] clkdev: Export clk_register_clkdev

Quoted 4 lines · Tomeu Vizoso

So it can be used from modules such as clk-test.ko.

Sorry for late answer, but this is useless.

We have to provide the way how to get the allocated clk_lookup. Otherwise you have no way but memory leak at the ->remove() stage.

(By the way I used to have the exact patch locally, that's why discovered the change in upsteam)

Quoted 24 lines
---
 drivers/clk/clkdev.c | 1 +
 1 file changed, 1 insertion(+)
drivers/clk/clkdev.c +1 -1
index 901d242..29a1ab7 100644
@@ -377,6 +377,7 @@ int clk_register_clkdev(struct clk *clk, const char *con_id,
 
        return 0;
 }
+EXPORT_SYMBOL(clk_register_clkdev);
 
 /**
  * clk_register_clkdevs - register a set of clk_lookup for a struct clk
--

1.9.3

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#63 lore raw

Re: [PATCH v13 5/6] clkdev: Export clk_register_clkdev

Quoted 7 lines · Andy Shevchenko
Quoted 4 lines · Tomeu Vizoso

So it can be used from modules such as clk-test.ko.

Quoted 8 lines

Sorry for late answer, but this is useless.

We have to provide the way how to get the allocated clk_lookup. Otherwise you have no way but memory leak at the ->remove() stage.

(By the way I used to have the exact patch locally, that's why discovered the change in upsteam)

Will send a patch soon to fix this.

Quoted 30 lines
Quoted 19 lines
---
 drivers/clk/clkdev.c | 1 +
 1 file changed, 1 insertion(+)
drivers/clk/clkdev.c +1 -1
index 901d242..29a1ab7 100644
@@ -377,6 +377,7 @@ int clk_register_clkdev(struct clk *clk, const char *con_id,
 
        return 0;
 }
+EXPORT_SYMBOL(clk_register_clkdev);
 
 /**
  * clk_register_clkdevs - register a set of clk_lookup for a struct clk
--

1.9.3

Quoted 5 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

--
With Best Regards,
Andy Shevchenko

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#64 lore raw

PATCH v13 6/6 clk: Add module for unit tests

In reply to Tomeu Vizoso

This module registers a clock hierarchy and performs several operations against them checking that the result is expected.

---
 drivers/clk/Kconfig       |   1 +
 drivers/clk/Kconfig.debug |   6 +
 drivers/clk/Makefile      |   1 +
 drivers/clk/clk-test.c    | 326 ++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 334 insertions(+)
 create mode 100644 drivers/clk/Kconfig.debug
 create mode 100644 drivers/clk/clk-test.c
drivers/clk/Kconfig +1 -0
index 3f44f29..03c500f 100644
@@ -135,6 +135,7 @@ config COMMON_CLK_PXA
 	  Sypport for the Marvell PXA SoC.
 
 source "drivers/clk/qcom/Kconfig"
+source "drivers/clk/Kconfig.debug"
 
 endmenu
 
drivers/clk/Kconfig.debug new file +6 -0
new file mode 100644
index 0000000..840b790
@@ -0,0 +1,6 @@
+config COMMON_CLK_TEST
+	tristate "Unit tests for the Common Clock Framework"
+	default n
+	---help---
+	  This driver runs several tests on the Common Clock Framework.
+
drivers/clk/Makefile +1 -0
index d5fba5b..4ee1475 100644
@@ -68,3 +68,4 @@ obj-$(CONFIG_ARCH_U8500)		+= ux500/
 obj-$(CONFIG_COMMON_CLK_VERSATILE)	+= versatile/
 obj-$(CONFIG_X86)			+= x86/
 obj-$(CONFIG_ARCH_ZYNQ)			+= zynq/
+obj-$(CONFIG_COMMON_CLK_TEST)		+= clk-test.o
drivers/clk/clk-test.c new file +326 -0

326 lines not on this page · raw message

Signature · 1 line
1.9.3
#65 lore raw
In reply to Tomeu Vizoso
Quoted 4 lines · Tomeu Vizoso

Hi,

in this v13 I have:

Quoted 8 lines

* added the latest R-b tags from Stephen,

* split some lines that went over the 80-char limit,

* made explicit in the docs that the range limits are inclusive, and

* removed an ifdef in clkdev.c by adding a dev_id param to __of_clk_get_by_name.

I think this patch series is ready to go except for the last one? Do you plan on polishing that one some more?

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#66 lore raw
Quoted 9 lines · Stephen Boyd
Quoted 8 lines · Tomeu Vizoso

Hi,

in this v13 I have:

* added the latest R-b tags from Stephen,

* split some lines that went over the 80-char limit,

Quoted 7 lines
Quoted 4 lines

* made explicit in the docs that the range limits are inclusive, and

* removed an ifdef in clkdev.c by adding a dev_id param to __of_clk_get_by_name.

I think this patch series is ready to go except for the last one? Do you plan on polishing that one some more?

Not in time for 3.20, I'm afraid. Want to spin first each of my other series, then I will go back to it.

Regards,

Tomeu

#67 lore raw
In reply to Stephen Boyd
Quoted 9 lines · Stephen Boyd
Quoted 8 lines · Tomeu Vizoso

Hi,

in this v13 I have:

* added the latest R-b tags from Stephen,

* split some lines that went over the 80-char limit,

Quoted 7 lines
Quoted 4 lines

* made explicit in the docs that the range limits are inclusive, and

* removed an ifdef in clkdev.c by adding a dev_id param to __of_clk_get_by_name.

I think this patch series is ready to go except for the last one? Do you plan on polishing that one some more?

Hi Mike,

do you plan to merge this for 3.20?

Thanks,

Tomeu

A message that is not in the archive (sent off-list or not yet received): 20150129022633.22722.78592@quantum

#69 lore raw
Quoted 29 lines · Mike Turquette

Quoting Tomeu Vizoso (2015-01-27 22:59:59)

Quoted 15 lines · Stephen Boyd
Quoted 14 lines · Tomeu Vizoso
Quoted 12 lines

Hi,

in this v13 I have:

* added the latest R-b tags from Stephen,

* split some lines that went over the 80-char limit,

* made explicit in the docs that the range limits are inclusive, and

* removed an ifdef in clkdev.c by adding a dev_id param to __of_clk_get_by_name.

Quoted 6 lines
Quoted 2 lines

I think this patch series is ready to go except for the last one? Do you plan on polishing that one some more?

Hi Mike,

do you plan to merge this for 3.20?

Tomeu,

Yes. I have taken patches 1-5.

I edited patch #3 to adapt clk_has_parent for struct clk_core.

Quoted 8 lines

https://git.kernel.org/cgit/linux/kernel/git/clk/linux.git/commit/?h=clk-next&id=59cf3fcf9bafe8fdb6183f13dafb81fc8213a8f9

Patch #4 need an update to convert sun6i_ahb1_clk_determine_rate to have min/max rates.

https://git.kernel.org/cgit/linux/kernel/git/clk/linux.git/commit/?h=clk-next&id=cb75a8fcd14e71c115decc416dd8d223e7c8b2e0

Can you look it over and make sure I didn't screw anything up?

Yup, they look good to me. I have also done some testing here.

Thanks,

Tomeu

#70 lore raw
Quoted 32 lines · Tomeu Vizoso
Quoted 31 lines · Mike Turquette

Quoting Tomeu Vizoso (2015-01-27 22:59:59)

Quoted 15 lines · Stephen Boyd
Quoted 14 lines · Tomeu Vizoso
Quoted 12 lines

Hi,

in this v13 I have:

* added the latest R-b tags from Stephen,

* split some lines that went over the 80-char limit,

* made explicit in the docs that the range limits are inclusive, and

* removed an ifdef in clkdev.c by adding a dev_id param to __of_clk_get_by_name.

Quoted 6 lines
Quoted 2 lines

I think this patch series is ready to go except for the last one? Do you plan on polishing that one some more?

Hi Mike,

do you plan to merge this for 3.20?

Tomeu,

Yes. I have taken patches 1-5.

I edited patch #3 to adapt clk_has_parent for struct clk_core.

https://git.kernel.org/cgit/linux/kernel/git/clk/linux.git/commit/?h=clk-next&id=59cf3fcf9bafe8fdb6183f13dafb81fc8213a8f9

Quoted 8 lines
Quoted 6 lines

Patch #4 need an update to convert sun6i_ahb1_clk_determine_rate to have min/max rates.

https://git.kernel.org/cgit/linux/kernel/git/clk/linux.git/commit/?h=clk-next&id=cb75a8fcd14e71c115decc416dd8d223e7c8b2e0

Can you look it over and make sure I didn't screw anything up?

Yup, they look good to me. I have also done some testing here.

Great! Thanks again for your hard work on this series.

Regards,
Mike

Quoted 5 lines

Thanks,

Tomeu