2023-01-17 14:36:14

by Christian Marangi

[permalink] [raw]
Subject: [PATCH v3 0/6] clk: qcom: clk-rcg2: introduce support for multiple conf for same freq

This small series fix a current problem with ipq8074 where the 2 uniphy
port doesn't work in some corner case with some clk configuration. The
port to correctly work require a specific frequency, using the wrong one
results in the port not transmitting data.

With the current code with a requested freq of 125MHz, the frequency is
set to 105MHz. This is caused by the fact that there are 2 different
configuration to set 125MHz and it's always selected the first one that
results in 105MHz.

In the original QSDK code, the frequency configuration selection is
different and the CEIL FLOOR logic is not present. Instead it's used a
BEST approach where the frequency table is checked and then it's checked
if there are duplicate entry.

This proposed implementation is more specific and keep the CEIL FLOOR
logic while maitaining the possibility to provide multiple
configuration.

The first 2 patch drop redundant F entry redefinition.

The 3rd and 4th one try to align rcg2 set rate to what clock core suggest
by using just what determine_clock pass to clk core.

The 5th one implement the change with also some macro description on how
this new implementation works. On determine the old v2 logic is used but
for set the requested parent is used instead of researching.

The 6th one migrate the gcc-ipq8074 driver to this new implementation.

Changes v3:
- Add qcom_find_freq_exact
- Drop re-search on rcg2_set_rate
- Rework multiple conf patch to follow new implementation
Changes v2:
- Out of RFC
- Fix compile warning from buildbot related to F redefinition

Christian Marangi (6):
clk: qcom: gcc-ipq6018: drop redundant F define
clk: qcom: gcc-sdm660: drop redundant F define
clk: qcom: common: add qcom_find_freq_exact
clk: qcom: clk-rcg2: don't re-search config on rcg2_set_rate
clk: qcom: clk-rcg2: introduce support for multiple conf for same freq
clk: qcom: gcc-ipq8074: rework nss_port5/6 clock to multiple conf

drivers/clk/qcom/clk-rcg.h | 14 +++-
drivers/clk/qcom/clk-rcg2.c | 140 +++++++++++++++++++++++++++------
drivers/clk/qcom/common.c | 17 ++++
drivers/clk/qcom/common.h | 2 +
drivers/clk/qcom/gcc-ipq6018.c | 2 -
drivers/clk/qcom/gcc-ipq8074.c | 64 +++++++++++----
drivers/clk/qcom/gcc-sdm660.c | 2 -
7 files changed, 196 insertions(+), 45 deletions(-)

--
2.38.1


2023-01-17 14:45:41

by Christian Marangi

[permalink] [raw]
Subject: [PATCH v3 2/6] clk: qcom: gcc-sdm660: drop redundant F define

The same exact F frequency table entry is defined in clk-rcg.h
Drop the redundant define to cleanup code.

Signed-off-by: Christian Marangi <[email protected]>
---
drivers/clk/qcom/gcc-sdm660.c | 2 --
1 file changed, 2 deletions(-)

diff --git a/drivers/clk/qcom/gcc-sdm660.c b/drivers/clk/qcom/gcc-sdm660.c
index db918c92a522..6afce8e42ede 100644
--- a/drivers/clk/qcom/gcc-sdm660.c
+++ b/drivers/clk/qcom/gcc-sdm660.c
@@ -25,8 +25,6 @@
#include "reset.h"
#include "gdsc.h"

-#define F(f, s, h, m, n) { (f), (s), (2 * (h) - 1), (m), (n) }
-
enum {
P_XO,
P_SLEEP_CLK,
--
2.38.1

2023-01-17 14:46:19

by Christian Marangi

[permalink] [raw]
Subject: [PATCH v3 4/6] clk: qcom: clk-rcg2: don't re-search config on rcg2_set_rate

Currently the rcg2 driver search the rate to apply 2 times.
- In _freq_tbl_determine_rate for the determine_rate function used by
core clk to understand the best rate to set with set_rate
- In rcg2_set_rate where the suggested rate is not trusted and searched
another time using a CEIL or FLOOR policy.

This is fundamentally wrong as we are ignoring what core clock is
deciding and just setting whatever clock configuration we want for the
suggested clock. The problem is in the fact that the correct clock
should have already be searched and selected with the determine_rate
function and set_rate should just apply whatever clock was provided.

Rework the function using the find_req_exact and by trusting whatever
core clk is asking to set.

Signed-off-by: Christian Marangi <[email protected]>
---
drivers/clk/qcom/clk-rcg2.c | 23 ++++++-----------------
1 file changed, 6 insertions(+), 17 deletions(-)

diff --git a/drivers/clk/qcom/clk-rcg2.c b/drivers/clk/qcom/clk-rcg2.c
index 76551534f10d..3f15e993dc04 100644
--- a/drivers/clk/qcom/clk-rcg2.c
+++ b/drivers/clk/qcom/clk-rcg2.c
@@ -352,23 +352,12 @@ static int clk_rcg2_configure(struct clk_rcg2 *rcg, const struct freq_tbl *f)
return update_config(rcg);
}

-static int __clk_rcg2_set_rate(struct clk_hw *hw, unsigned long rate,
- enum freq_policy policy)
+static int __clk_rcg2_set_rate(struct clk_hw *hw, unsigned long rate)
{
struct clk_rcg2 *rcg = to_clk_rcg2(hw);
const struct freq_tbl *f;

- switch (policy) {
- case FLOOR:
- f = qcom_find_freq_floor(rcg->freq_tbl, rate);
- break;
- case CEIL:
- f = qcom_find_freq(rcg->freq_tbl, rate);
- break;
- default:
- return -EINVAL;
- }
-
+ f = qcom_find_freq_exact(rcg->freq_tbl, rate);
if (!f)
return -EINVAL;

@@ -378,25 +367,25 @@ static int __clk_rcg2_set_rate(struct clk_hw *hw, unsigned long rate,
static int clk_rcg2_set_rate(struct clk_hw *hw, unsigned long rate,
unsigned long parent_rate)
{
- return __clk_rcg2_set_rate(hw, rate, CEIL);
+ return __clk_rcg2_set_rate(hw, rate);
}

static int clk_rcg2_set_floor_rate(struct clk_hw *hw, unsigned long rate,
unsigned long parent_rate)
{
- return __clk_rcg2_set_rate(hw, rate, FLOOR);
+ return __clk_rcg2_set_rate(hw, rate);
}

static int clk_rcg2_set_rate_and_parent(struct clk_hw *hw,
unsigned long rate, unsigned long parent_rate, u8 index)
{
- return __clk_rcg2_set_rate(hw, rate, CEIL);
+ return __clk_rcg2_set_rate(hw, rate);
}

static int clk_rcg2_set_floor_rate_and_parent(struct clk_hw *hw,
unsigned long rate, unsigned long parent_rate, u8 index)
{
- return __clk_rcg2_set_rate(hw, rate, FLOOR);
+ return __clk_rcg2_set_rate(hw, rate);
}

static int clk_rcg2_get_duty_cycle(struct clk_hw *hw, struct clk_duty *duty)
--
2.38.1

2023-01-17 14:47:48

by Christian Marangi

[permalink] [raw]
Subject: [PATCH v3 5/6] clk: qcom: clk-rcg2: introduce support for multiple conf for same freq

Some RCG frequency can be reached by multiple configuration.

We currently declare multiple configuration for the same frequency but
that is not supported and always the first configuration will be taken.

These multiple configuration are needed as based on the current parent
configuration, it may be needed to use a different configuration to
reach the same frequency.

To handle this introduce 2 new macro, FM and C.

- FM is used to declare an empty freq_tbl with just the frequency and an
array of confs to insert all the config for the provided frequency.

- C is used to declare a freq_conf where src, pre_div, m and n are
provided.

The driver is changed to handle this special freq_tbl and select the
correct config by calculating the final rate and deciding based on the
one that is less different than the requested one.
Function set_rate(_and_parent) will then select the config based on the
current or the requested parent for the set_rate_and_parent variant.

Tested-by: Robert Marko <[email protected]>
Signed-off-by: Christian Marangi <[email protected]>
---
drivers/clk/qcom/clk-rcg.h | 14 +++-
drivers/clk/qcom/clk-rcg2.c | 129 ++++++++++++++++++++++++++++++++----
2 files changed, 129 insertions(+), 14 deletions(-)

diff --git a/drivers/clk/qcom/clk-rcg.h b/drivers/clk/qcom/clk-rcg.h
index 01581f4d2c39..18f4f7b59f36 100644
--- a/drivers/clk/qcom/clk-rcg.h
+++ b/drivers/clk/qcom/clk-rcg.h
@@ -7,7 +7,17 @@
#include <linux/clk-provider.h>
#include "clk-regmap.h"

-#define F(f, s, h, m, n) { (f), (s), (2 * (h) - 1), (m), (n) }
+#define F(f, s, h, m, n) { (f), (s), (2 * (h) - 1), (m), (n), 0, NULL }
+
+#define FM(_f, _confs) { .freq = (_f), .confs_num = ARRAY_SIZE(_confs), .confs = (_confs) }
+#define C(s, h, m, n) { (s), (2 * (h) - 1), (m), (n) }
+
+struct freq_conf {
+ u8 src;
+ u8 pre_div;
+ u16 m;
+ u16 n;
+};

struct freq_tbl {
unsigned long freq;
@@ -15,6 +25,8 @@ struct freq_tbl {
u8 pre_div;
u16 m;
u16 n;
+ int confs_num;
+ const struct freq_conf *confs;
};

/**
diff --git a/drivers/clk/qcom/clk-rcg2.c b/drivers/clk/qcom/clk-rcg2.c
index 3f15e993dc04..dfc2ab1cd59a 100644
--- a/drivers/clk/qcom/clk-rcg2.c
+++ b/drivers/clk/qcom/clk-rcg2.c
@@ -209,11 +209,60 @@ clk_rcg2_recalc_rate(struct clk_hw *hw, unsigned long parent_rate)
return __clk_rcg2_recalc_rate(hw, parent_rate, cfg);
}

+static void
+_freq_tbl_determine_parent(struct clk_hw *hw, struct freq_tbl *f_tbl,
+ const struct freq_tbl *f, unsigned long req_rate)
+{
+ unsigned long best_rate = 0, parent_rate, rate;
+ const struct freq_conf *conf, *best_conf;
+ struct clk_rcg2 *rcg = to_clk_rcg2(hw);
+ struct clk_hw *p;
+ int index, i;
+
+ /* Search in each provided config the one that is near the wanted rate */
+ for (i = 0, conf = f->confs; i < f->confs_num; i++, conf++) {
+ index = qcom_find_src_index(hw, rcg->parent_map, conf->src);
+ if (index < 0)
+ continue;
+
+ p = clk_hw_get_parent_by_index(hw, index);
+ if (!p)
+ continue;
+
+ parent_rate = clk_hw_get_rate(p);
+ rate = calc_rate(parent_rate, conf->n, conf->m, conf->n, conf->pre_div);
+
+ if (rate == req_rate) {
+ best_conf = conf;
+ break;
+ }
+
+ if (abs(req_rate - rate) < abs(best_rate - rate)) {
+ best_rate = rate;
+ best_conf = conf;
+ }
+ }
+
+ /*
+ * Very unlikely.
+ * Force the first conf if we can't find a correct config.
+ */
+ if (unlikely(i == f->confs_num))
+ best_conf = f->confs;
+
+ /* Apply the config */
+ f_tbl->src = best_conf->src;
+ f_tbl->pre_div = best_conf->pre_div;
+ f_tbl->m = best_conf->m;
+ f_tbl->n = best_conf->n;
+}
+
static int _freq_tbl_determine_rate(struct clk_hw *hw, const struct freq_tbl *f,
struct clk_rate_request *req,
enum freq_policy policy)
{
unsigned long clk_flags, rate = req->rate;
+ struct freq_tbl f_tbl;
struct clk_hw *p;
struct clk_rcg2 *rcg = to_clk_rcg2(hw);
int index;
@@ -232,7 +281,15 @@ static int _freq_tbl_determine_rate(struct clk_hw *hw, const struct freq_tbl *f,
if (!f)
return -EINVAL;

- index = qcom_find_src_index(hw, rcg->parent_map, f->src);
+ f_tbl = *f;
+ /*
+ * A single freq may be reached by multiple configuration.
+ * Try to find the bast one if we have this kind of freq_table.
+ */
+ if (f->confs)
+ _freq_tbl_determine_parent(hw, &f_tbl, f, rate);
+
+ index = qcom_find_src_index(hw, rcg->parent_map, f_tbl.src);
if (index < 0)
return index;

@@ -242,18 +299,18 @@ static int _freq_tbl_determine_rate(struct clk_hw *hw, const struct freq_tbl *f,
return -EINVAL;

if (clk_flags & CLK_SET_RATE_PARENT) {
- rate = f->freq;
- if (f->pre_div) {
+ rate = f_tbl.freq;
+ if (f_tbl.pre_div) {
if (!rate)
rate = req->rate;
rate /= 2;
- rate *= f->pre_div + 1;
+ rate *= f_tbl.pre_div + 1;
}

if (f->n) {
u64 tmp = rate;
- tmp = tmp * f->n;
- do_div(tmp, f->m);
+ tmp = tmp * f_tbl.n;
+ do_div(tmp, f_tbl.m);
rate = tmp;
}
} else {
@@ -261,7 +318,7 @@ static int _freq_tbl_determine_rate(struct clk_hw *hw, const struct freq_tbl *f,
}
req->best_parent_hw = p;
req->best_parent_rate = rate;
- req->rate = f->freq;
+ req->rate = f_tbl.freq;

return 0;
}
@@ -352,40 +409,86 @@ static int clk_rcg2_configure(struct clk_rcg2 *rcg, const struct freq_tbl *f)
return update_config(rcg);
}

-static int __clk_rcg2_set_rate(struct clk_hw *hw, unsigned long rate)
+static void
+__clk_rcg2_select_conf(struct clk_hw *hw, struct freq_tbl *f_tbl,
+ const struct freq_tbl *f, u8 parent_index)
+{
+ const struct freq_conf *conf, *best_conf;
+ struct clk_rcg2 *rcg = to_clk_rcg2(hw);
+ int index, i;
+
+ /* Search in each provided config the one that is near the wanted rate */
+ for (i = 0, conf = f->confs; i < f->confs_num; i++, conf++) {
+ index = qcom_find_src_index(hw, rcg->parent_map, conf->src);
+ if (index < 0)
+ continue;
+
+ if (index == parent_index)
+ break;
+ }
+
+ /*
+ * Very unlikely.
+ * Force the first conf if we can't find a correct config.
+ */
+ if (unlikely(i == f->confs_num))
+ best_conf = f->confs;
+
+ /* Apply the config */
+ f_tbl->src = best_conf->src;
+ f_tbl->pre_div = best_conf->pre_div;
+ f_tbl->m = best_conf->m;
+ f_tbl->n = best_conf->n;
+}
+
+static int __clk_rcg2_set_rate(struct clk_hw *hw, unsigned long rate, u8 index)
{
struct clk_rcg2 *rcg = to_clk_rcg2(hw);
const struct freq_tbl *f;
+ struct freq_tbl f_tbl;

f = qcom_find_freq_exact(rcg->freq_tbl, rate);
if (!f)
return -EINVAL;

- return clk_rcg2_configure(rcg, f);
+ f_tbl = *f;
+ /*
+ * A single freq may be reached by multiple configuration.
+ * Select the conf for the requested parent.
+ */
+ if (f->confs) {
+ /* Use current parent if not set */
+ if (index < 0)
+ index = clk_hw_get_parent_index(hw);
+
+ __clk_rcg2_select_conf(hw, &f_tbl, f, index);
+ }
+
+ return clk_rcg2_configure(rcg, &f_tbl);
}

static int clk_rcg2_set_rate(struct clk_hw *hw, unsigned long rate,
unsigned long parent_rate)
{
- return __clk_rcg2_set_rate(hw, rate);
+ return __clk_rcg2_set_rate(hw, rate, -1);
}

static int clk_rcg2_set_floor_rate(struct clk_hw *hw, unsigned long rate,
unsigned long parent_rate)
{
- return __clk_rcg2_set_rate(hw, rate);
+ return __clk_rcg2_set_rate(hw, rate, -1);
}

static int clk_rcg2_set_rate_and_parent(struct clk_hw *hw,
unsigned long rate, unsigned long parent_rate, u8 index)
{
- return __clk_rcg2_set_rate(hw, rate);
+ return __clk_rcg2_set_rate(hw, rate, index);
}

static int clk_rcg2_set_floor_rate_and_parent(struct clk_hw *hw,
unsigned long rate, unsigned long parent_rate, u8 index)
{
- return __clk_rcg2_set_rate(hw, rate);
+ return __clk_rcg2_set_rate(hw, rate, index);
}

static int clk_rcg2_get_duty_cycle(struct clk_hw *hw, struct clk_duty *duty)
--
2.38.1

2023-01-17 14:57:08

by Christian Marangi

[permalink] [raw]
Subject: [PATCH v3 6/6] clk: qcom: gcc-ipq8074: rework nss_port5/6 clock to multiple conf

Rework nss_port5/6 to use the new multiple configuration implementation
and correctly fix the clocks for these port under some corner case.

This is particularly relevant for device that have 2.5G or 10G port
connected to port5 or port 6 on ipq8074. As the parent are shared
across multiple port it may be required to select the correct
configuration to accomplish the desired clock. Without this patch such
port doesn't work in some specific ethernet speed as the clock will be
set to the wrong frequency as we just select the first configuration for
the related frequency instead of selecting the best one.

Tested-by: Robert Marko <[email protected]> # ipq8074 Qnap QHora-301W
Signed-off-by: Christian Marangi <[email protected]>
---
drivers/clk/qcom/gcc-ipq8074.c | 64 +++++++++++++++++++++++++---------
1 file changed, 48 insertions(+), 16 deletions(-)

diff --git a/drivers/clk/qcom/gcc-ipq8074.c b/drivers/clk/qcom/gcc-ipq8074.c
index 42d185fe19c8..02d04a552b78 100644
--- a/drivers/clk/qcom/gcc-ipq8074.c
+++ b/drivers/clk/qcom/gcc-ipq8074.c
@@ -1787,13 +1787,21 @@ static struct clk_regmap_div nss_port4_tx_div_clk_src = {
},
};

+static const struct freq_conf ftbl_nss_port5_rx_clk_src_25[] = {
+ C(P_UNIPHY1_RX, 12.5, 0, 0),
+ C(P_UNIPHY0_RX, 5, 0, 0),
+};
+
+static const struct freq_conf ftbl_nss_port5_rx_clk_src_125[] = {
+ C(P_UNIPHY1_RX, 2.5, 0, 0),
+ C(P_UNIPHY0_RX, 1, 0, 0),
+};
+
static const struct freq_tbl ftbl_nss_port5_rx_clk_src[] = {
F(19200000, P_XO, 1, 0, 0),
- F(25000000, P_UNIPHY1_RX, 12.5, 0, 0),
- F(25000000, P_UNIPHY0_RX, 5, 0, 0),
+ FM(25000000, ftbl_nss_port5_rx_clk_src_25),
F(78125000, P_UNIPHY1_RX, 4, 0, 0),
- F(125000000, P_UNIPHY1_RX, 2.5, 0, 0),
- F(125000000, P_UNIPHY0_RX, 1, 0, 0),
+ FM(125000000, ftbl_nss_port5_rx_clk_src_125),
F(156250000, P_UNIPHY1_RX, 2, 0, 0),
F(312500000, P_UNIPHY1_RX, 1, 0, 0),
{ }
@@ -1829,13 +1837,21 @@ static struct clk_regmap_div nss_port5_rx_div_clk_src = {
},
};

+static struct freq_conf ftbl_nss_port5_tx_clk_src_25[] = {
+ C(P_UNIPHY1_TX, 12.5, 0, 0),
+ C(P_UNIPHY0_TX, 5, 0, 0),
+};
+
+static struct freq_conf ftbl_nss_port5_tx_clk_src_125[] = {
+ C(P_UNIPHY1_TX, 2.5, 0, 0),
+ C(P_UNIPHY0_TX, 1, 0, 0),
+};
+
static const struct freq_tbl ftbl_nss_port5_tx_clk_src[] = {
F(19200000, P_XO, 1, 0, 0),
- F(25000000, P_UNIPHY1_TX, 12.5, 0, 0),
- F(25000000, P_UNIPHY0_TX, 5, 0, 0),
+ FM(25000000, ftbl_nss_port5_tx_clk_src_25),
F(78125000, P_UNIPHY1_TX, 4, 0, 0),
- F(125000000, P_UNIPHY1_TX, 2.5, 0, 0),
- F(125000000, P_UNIPHY0_TX, 1, 0, 0),
+ FM(125000000, ftbl_nss_port5_tx_clk_src_125),
F(156250000, P_UNIPHY1_TX, 2, 0, 0),
F(312500000, P_UNIPHY1_TX, 1, 0, 0),
{ }
@@ -1871,13 +1887,21 @@ static struct clk_regmap_div nss_port5_tx_div_clk_src = {
},
};

+static struct freq_conf ftbl_nss_port6_rx_clk_src_25[] = {
+ C(P_UNIPHY2_RX, 5, 0, 0),
+ C(P_UNIPHY2_RX, 12.5, 0, 0),
+};
+
+static struct freq_conf ftbl_nss_port6_rx_clk_src_125[] = {
+ C(P_UNIPHY2_RX, 1, 0, 0),
+ C(P_UNIPHY2_RX, 2.5, 0, 0),
+};
+
static const struct freq_tbl ftbl_nss_port6_rx_clk_src[] = {
F(19200000, P_XO, 1, 0, 0),
- F(25000000, P_UNIPHY2_RX, 5, 0, 0),
- F(25000000, P_UNIPHY2_RX, 12.5, 0, 0),
+ FM(25000000, ftbl_nss_port6_rx_clk_src_25),
F(78125000, P_UNIPHY2_RX, 4, 0, 0),
- F(125000000, P_UNIPHY2_RX, 1, 0, 0),
- F(125000000, P_UNIPHY2_RX, 2.5, 0, 0),
+ FM(125000000, ftbl_nss_port6_rx_clk_src_125),
F(156250000, P_UNIPHY2_RX, 2, 0, 0),
F(312500000, P_UNIPHY2_RX, 1, 0, 0),
{ }
@@ -1913,13 +1937,21 @@ static struct clk_regmap_div nss_port6_rx_div_clk_src = {
},
};

+static struct freq_conf ftbl_nss_port6_tx_clk_src_25[] = {
+ C(P_UNIPHY2_TX, 5, 0, 0),
+ C(P_UNIPHY2_TX, 12.5, 0, 0),
+};
+
+static struct freq_conf ftbl_nss_port6_tx_clk_src_125[] = {
+ C(P_UNIPHY2_TX, 1, 0, 0),
+ C(P_UNIPHY2_TX, 2.5, 0, 0),
+};
+
static const struct freq_tbl ftbl_nss_port6_tx_clk_src[] = {
F(19200000, P_XO, 1, 0, 0),
- F(25000000, P_UNIPHY2_TX, 5, 0, 0),
- F(25000000, P_UNIPHY2_TX, 12.5, 0, 0),
+ FM(25000000, ftbl_nss_port6_tx_clk_src_25),
F(78125000, P_UNIPHY2_TX, 4, 0, 0),
- F(125000000, P_UNIPHY2_TX, 1, 0, 0),
- F(125000000, P_UNIPHY2_TX, 2.5, 0, 0),
+ FM(125000000, ftbl_nss_port6_tx_clk_src_125),
F(156250000, P_UNIPHY2_TX, 2, 0, 0),
F(312500000, P_UNIPHY2_TX, 1, 0, 0),
{ }
--
2.38.1

2023-03-29 20:03:14

by Stephen Boyd

[permalink] [raw]
Subject: Re: [PATCH v3 4/6] clk: qcom: clk-rcg2: don't re-search config on rcg2_set_rate

Quoting Christian Marangi (2023-01-17 05:54:57)
> Currently the rcg2 driver search the rate to apply 2 times.
> - In _freq_tbl_determine_rate for the determine_rate function used by
> core clk to understand the best rate to set with set_rate
> - In rcg2_set_rate where the suggested rate is not trusted and searched
> another time using a CEIL or FLOOR policy.
>
> This is fundamentally wrong as we are ignoring what core clock is
> deciding and just setting whatever clock configuration we want for the
> suggested clock. The problem is in the fact that the correct clock
> should have already be searched and selected with the determine_rate
> function and set_rate should just apply whatever clock was provided.

It sounds like you're assuming the rate coming into the set_rate clk_op
is rounded? Don't make that assumption. The set_rate clk_op should round
the rate again. The parent rate could have changed.

2023-03-29 20:18:04

by Stephen Boyd

[permalink] [raw]
Subject: Re: [PATCH v3 5/6] clk: qcom: clk-rcg2: introduce support for multiple conf for same freq

Quoting Christian Marangi (2023-01-17 05:54:58)
> diff --git a/drivers/clk/qcom/clk-rcg.h b/drivers/clk/qcom/clk-rcg.h
> index 01581f4d2c39..18f4f7b59f36 100644
> --- a/drivers/clk/qcom/clk-rcg.h
> +++ b/drivers/clk/qcom/clk-rcg.h
> @@ -15,6 +25,8 @@ struct freq_tbl {
> u8 pre_div;
> u16 m;
> u16 n;
> + int confs_num;
> + const struct freq_conf *confs;

This design is uncommon. I suggest you have a different clk_ops
structure and/or entire clk_rcg (clk_rcg_fm?) that does things
differently. Then we don't have to worry that all the other rcg2 clks
(of which there are many) could be broken by this patch.

2023-03-29 22:08:48

by Dmitry Baryshkov

[permalink] [raw]
Subject: Re: [PATCH v3 4/6] clk: qcom: clk-rcg2: don't re-search config on rcg2_set_rate

On 29/03/2023 22:52, Stephen Boyd wrote:
> Quoting Christian Marangi (2023-01-17 05:54:57)
>> Currently the rcg2 driver search the rate to apply 2 times.
>> - In _freq_tbl_determine_rate for the determine_rate function used by
>> core clk to understand the best rate to set with set_rate
>> - In rcg2_set_rate where the suggested rate is not trusted and searched
>> another time using a CEIL or FLOOR policy.
>>
>> This is fundamentally wrong as we are ignoring what core clock is
>> deciding and just setting whatever clock configuration we want for the
>> suggested clock. The problem is in the fact that the correct clock
>> should have already be searched and selected with the determine_rate
>> function and set_rate should just apply whatever clock was provided.
>
> It sounds like you're assuming the rate coming into the set_rate clk_op
> is rounded? Don't make that assumption. The set_rate clk_op should round
> the rate again. The parent rate could have changed.

My fault, I suggested this some time ago on the basis that... "CCF
switching the parent rate during the clk_set_rate() call. Then the
second lookup might end up selecting different parent/mnd configuration."

--
With best wishes
Dmitry