Hi Michal,

On Fri, 18 Jan 2019 at 08:13, Michal Simek michal.simek@xilinx.com wrote:

The same functionality was added to Linux for i2c bus registration with this commit message:

" of: base: add function to get highest id of an alias stem

I2C supports adding adapters using either a dynamic or fixed id. The latter is provided by aliases in the DT case. To prevent id collisions of those two types, install this function which gives us the highest fixed id, so we can then let the dynamically created ones come after this highest number.

Signed-off-by: Wolfram Sang wsa+renesas@sang-engineering.com Acked-by: Rob Herring robh@kernel.org Signed-off-by: Wolfram Sang wsa@the-dreams.de "

Add it also to U-Boot for DM I2C support.

Signed-off-by: Michal Simek michal.simek@xilinx.com

drivers/core/of_access.c | 18 ++++++++++++++++++ include/dm/of_access.h | 9 +++++++++ 2 files changed, 27 insertions(+)

diff --git a/drivers/core/of_access.c b/drivers/core/of_access.c index 14c020a687b7..7c2df2354109 100644 --- a/drivers/core/of_access.c +++ b/drivers/core/of_access.c @@ -812,6 +812,24 @@ int of_alias_get_id(const struct device_node *np, const char *stem) return id; }

+int of_alias_get_highest_id(const char *stem) +{

•   struct alias_prop *app;
  

•   int id = -ENODEV;
  

•   mutex_lock(&of_mutex);
  

•   list_for_each_entry(app, &aliases_lookup, link) {
  

•           if (strcmp(app->stem, stem) != 0)
  

•                   continue;
  

•           if (app->id > id)
  

•                   id = app->id;
  

•   }
  

•   mutex_unlock(&of_mutex);
  

•   return id;
  

+}

struct device_node *of_get_stdout(void) { return of_stdout; diff --git a/include/dm/of_access.h b/include/dm/of_access.h index 5ed1a0cdb427..5cbfd220bfd4 100644 --- a/include/dm/of_access.h +++ b/include/dm/of_access.h @@ -425,6 +425,15 @@ int of_alias_scan(void); int of_alias_get_id(const struct device_node *np, const char *stem);

/**

• • of_alias_get_highest_id - Get highest alias id for the given stem
• • @stem: Alias stem to be examined
• • The function travels the lookup table to get the highest alias id for the
• • given alias stem. It returns the alias id if found.

@return

• */

+int of_alias_get_highest_id(const char *stem);

+/**

• of_get_stdout() - Get node to use for stdout
• @return node referred to by stdout-path alias, or NULL if none

-- 1.9.1

Can we place have a test that calls this for a few values?

Also see uclass_find_next_free_req_seq() and uclass_resolve_seq() which is what U-Boot does today. It is first in, first served. This is supposed to mean that devices without an alias don't get a req_seq value, meaning that the seq value is set to the next available value.

I suspect it is OK to change the behaviour though. It might affect some boards which rely on not having aliases but still getting bus numbers.

Regards, Simon

Hi Michal,

On Thu, 31 Jan 2019 at 03:19, Michal Simek michal.simek@xilinx.com wrote:

On 31. 01. 19 11:04, Simon Glass wrote:

...

Hi Michal,

On Fri, 18 Jan 2019 at 08:13, Michal Simek michal.simek@xilinx.com

wrote:

...

...

The same functionality was added to Linux for i2c bus registration

with this

...

...

commit message:

" of: base: add function to get highest id of an alias stem

I2C supports adding adapters using either a dynamic or fixed id. The latter is provided by aliases in the DT case. To prevent id collisions of those two types, install this function which gives us the highest fixed id, so we can then let the dynamically created ones come after this highest number.

Signed-off-by: Wolfram Sang wsa+renesas@sang-engineering.com Acked-by: Rob Herring robh@kernel.org Signed-off-by: Wolfram Sang wsa@the-dreams.de "

Add it also to U-Boot for DM I2C support.

Signed-off-by: Michal Simek michal.simek@xilinx.com

drivers/core/of_access.c | 18 ++++++++++++++++++ include/dm/of_access.h | 9 +++++++++ 2 files changed, 27 insertions(+)

diff --git a/drivers/core/of_access.c b/drivers/core/of_access.c index 14c020a687b7..7c2df2354109 100644 --- a/drivers/core/of_access.c +++ b/drivers/core/of_access.c @@ -812,6 +812,24 @@ int of_alias_get_id(const struct device_node *np,

const char *stem)

...

...

    return id;

}

+int of_alias_get_highest_id(const char *stem) +{

•   struct alias_prop *app;
  

•   int id = -ENODEV;
  

•   mutex_lock(&of_mutex);
  

•   list_for_each_entry(app, &aliases_lookup, link) {
  

•           if (strcmp(app->stem, stem) != 0)
  

•                   continue;
  

•           if (app->id > id)
  

•                   id = app->id;
  

•   }
  

•   mutex_unlock(&of_mutex);
  

•   return id;
  

+}

struct device_node *of_get_stdout(void) { return of_stdout; diff --git a/include/dm/of_access.h b/include/dm/of_access.h index 5ed1a0cdb427..5cbfd220bfd4 100644 --- a/include/dm/of_access.h +++ b/include/dm/of_access.h @@ -425,6 +425,15 @@ int of_alias_scan(void); int of_alias_get_id(const struct device_node *np, const char *stem);

/**

• • of_alias_get_highest_id - Get highest alias id for the given stem
• • @stem: Alias stem to be examined
• • The function travels the lookup table to get the highest alias id

for the

...

...

• • given alias stem. It returns the alias id if found.

@return

...

• */

+int of_alias_get_highest_id(const char *stem);

+/**

• of_get_stdout() - Get node to use for stdout
• @return node referred to by stdout-path alias, or NULL if none

-- 1.9.1

Can we place have a test that calls this for a few values?

Also see uclass_find_next_free_req_seq() and uclass_resolve_seq() which is what U-Boot does today. It is first in, first served. This is supposed to mean that devices without an alias don't get a req_seq value, meaning that the seq value is set to the next available value.

Let me play with it and see that calling sequence.

...

I suspect it is OK to change the behaviour though. It might affect some boards which rely on not having aliases but still getting bus numbers.

Do you mean with I2C? Or different buses?

I don't think there is any affected board with DM_I2C and OF_CONTROL because unlisted aliases are using req_seq -1 now.

Yes I think that is right, actually.

Regards, SImon

Hi Michal,

On Fri, 18 Jan 2019 at 08:13, Michal Simek michal.simek@xilinx.com wrote:

Find out the highest alias ID used for certain subsystem. This call will be used for alocating IDs for i2c buses which are not described in DT.

Signed-off-by: Michal Simek michal.simek@xilinx.com

include/fdtdec.h | 13 +++++++++++++ lib/fdtdec.c | 33 +++++++++++++++++++++++++++++++++ 2 files changed, 46 insertions(+)

diff --git a/include/fdtdec.h b/include/fdtdec.h index f1bcbf837ffb..c2dd87ede226 100644 --- a/include/fdtdec.h +++ b/include/fdtdec.h @@ -626,6 +626,19 @@ int fdtdec_get_alias_seq(const void *blob, const char *base, int node, int *seqp);

/**

• • Get the highest alias number for susbystem.
• • It parses all aliases and find out highest recorded alias for subsystem.
• • Aliases are of the form <base><num> where <num> is the sequence number.
• • @param blob Device tree blob (if NULL, then error is returned)
• • @param base Base name for alias susbystem (before the number)
• • @return 0 highest alias ID, -1 if not found
• */

+int fdtdec_get_alias_highest_id(const void *blob, const char *base);

+/**

• Get a property from the /chosen node
• @param blob Device tree blob (if NULL, then NULL is returned)

diff --git a/lib/fdtdec.c b/lib/fdtdec.c index 18663ce6bdac..55811975ef54 100644 --- a/lib/fdtdec.c +++ b/lib/fdtdec.c @@ -549,6 +549,39 @@ int fdtdec_get_alias_seq(const void *blob, const char *base, int offset, return -ENOENT; }

+int fdtdec_get_alias_highest_id(const void *blob, const char *base) +{

•   int base_len = strlen(base);
  

•   int prop_offset;
  

•   int aliases;
  

•   int max = -1;
  

•   debug("Looking for highest alias id for '%s'\n", base);
  

•   aliases = fdt_path_offset(blob, "/aliases");
  

•   for (prop_offset = fdt_first_property_offset(blob, aliases);
  

•        prop_offset > 0;
  

•        prop_offset = fdt_next_property_offset(blob, prop_offset)) {
  

•           const char *prop;
  

•           const char *name;
  

•           int len, val;
  

•           prop = fdt_getprop_by_offset(blob, prop_offset, &name, &len);
  

•           debug("   - %s, %s\n", name, prop);
  

•           if (*prop != '/' || prop[len - 1] ||
  

•               strncmp(name, base, base_len))
  

•                   continue;
  

•           val = trailing_strtol(name);
  

•           if (val > max) {
  

•                   debug("Found seq %d\n", val);
  

•                   max = val;
  

•           }
  

•   }
  

•   return max;
  

This looks right to me. Can you please add a test that calls this for a few sandbox aliases?

+}

const char *fdtdec_get_chosen_prop(const void *blob, const char *name) { int chosen_node; -- 1.9.1

Regards, Simon

On Fri, 18 Jan 2019 at 08:13, Michal Simek michal.simek@xilinx.com wrote:

Find out the highest alias ID used for certain subsystem. This call will be used for alocating IDs for i2c buses which are not described in DT.

Signed-off-by: Michal Simek michal.simek@xilinx.com

include/fdtdec.h | 13 +++++++++++++ lib/fdtdec.c | 33 +++++++++++++++++++++++++++++++++ 2 files changed, 46 insertions(+)

diff --git a/include/fdtdec.h b/include/fdtdec.h index f1bcbf837ffb..c2dd87ede226 100644 --- a/include/fdtdec.h +++ b/include/fdtdec.h @@ -626,6 +626,19 @@ int fdtdec_get_alias_seq(const void *blob, const char *base, int node, int *seqp);

/**

• • Get the highest alias number for susbystem.
• • It parses all aliases and find out highest recorded alias for subsystem.
• • Aliases are of the form <base><num> where <num> is the sequence number.
• • @param blob Device tree blob (if NULL, then error is returned)
• • @param base Base name for alias susbystem (before the number)
• • @return 0 highest alias ID, -1 if not found
• */

+int fdtdec_get_alias_highest_id(const void *blob, const char *base);

+/**

• Get a property from the /chosen node
• @param blob Device tree blob (if NULL, then NULL is returned)

diff --git a/lib/fdtdec.c b/lib/fdtdec.c index 18663ce6bdac..55811975ef54 100644 --- a/lib/fdtdec.c +++ b/lib/fdtdec.c @@ -549,6 +549,39 @@ int fdtdec_get_alias_seq(const void *blob, const char *base, int offset, return -ENOENT; }

+int fdtdec_get_alias_highest_id(const void *blob, const char *base) +{

•   int base_len = strlen(base);
  

•   int prop_offset;
  

•   int aliases;
  

•   int max = -1;
  

•   debug("Looking for highest alias id for '%s'\n", base);
  

•   aliases = fdt_path_offset(blob, "/aliases");
  

•   for (prop_offset = fdt_first_property_offset(blob, aliases);
  

•        prop_offset > 0;
  

•        prop_offset = fdt_next_property_offset(blob, prop_offset)) {
  

•           const char *prop;
  

•           const char *name;
  

•           int len, val;
  

•           prop = fdt_getprop_by_offset(blob, prop_offset, &name, &len);
  

•           debug("   - %s, %s\n", name, prop);
  

•           if (*prop != '/' || prop[len - 1] ||
  

•               strncmp(name, base, base_len))
  

•                   continue;
  

•           val = trailing_strtol(name);
  

•           if (val > max) {
  

•                   debug("Found seq %d\n", val);
  

•                   max = val;
  

•           }
  

•   }
  

•   return max;
  

+}

const char *fdtdec_get_chosen_prop(const void *blob, const char *name) { int chosen_node; -- 1.9.1

On 31. 01. 19 11:04, Simon Glass wrote:

Hi Michal,

On Fri, 18 Jan 2019 at 08:13, Michal Simek michal.simek@xilinx.com wrote:

...

There is a need to find out the first free i2c ID which can be used for i2s buses (including i2c buses connected to i2c mux). Do it early in init and share this variable with other i2c classes for uniq bus identification.

Signed-off-by: Michal Simek michal.simek@xilinx.com

drivers/i2c/i2c-uclass.c | 19 +++++++++++++++++++ include/i2c.h | 3 +++ 2 files changed, 22 insertions(+)

diff --git a/drivers/i2c/i2c-uclass.c b/drivers/i2c/i2c-uclass.c index 975318e5f254..aaece115f02f 100644 --- a/drivers/i2c/i2c-uclass.c +++ b/drivers/i2c/i2c-uclass.c @@ -23,6 +23,8 @@ enum { PIN_COUNT, };

+int i2c_highest_id = -1;

/* Useful debugging function */ void i2c_dump_msgs(struct i2c_msg *msg, int nmsgs) { @@ -619,10 +621,27 @@ static int i2c_child_post_bind(struct udevice *dev) #endif }

+int i2c_uclass_init(struct uclass *class) +{ +#ifdef CONFIG_OF_LIVE

•   int ret;
  

•   ret = of_alias_get_highest_id("i2c");
  

•   if (ret >= 0)
  

•           i2c_highest_id = ret;
  

+#else

•   i2c_highest_id = fdtdec_get_alias_highest_id(gd->fdt_blob, "i2c");
  

+#endif

This should call dev_read_alias_highest_id() or similar, which does the logic you have here.

not a problem with this.

...

•   debug("%s: i2c highest_id %d\n", __func__, i2c_highest_id);
  

•   return 0;
  

+}

UCLASS_DRIVER(i2c) = { .id = UCLASS_I2C, .name = "i2c", .flags = DM_UC_FLAG_SEQ_ALIAS,

•   .init           = i2c_uclass_init,
  

#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) .post_bind = dm_scan_fdt_dev, #endif diff --git a/include/i2c.h b/include/i2c.h index ccffc1955275..c69e26deaeed 100644 --- a/include/i2c.h +++ b/include/i2c.h @@ -16,6 +16,9 @@ #ifndef _I2C_H_ #define _I2C_H_

+/* Stores maximum highest i2c alias number */ +extern int i2c_highest_id;

This should be in uclass-private data, not a global.

In this case there are two uclasses. UCLASS_I2C and UCLASS_I2C_MUX which has to share the same numbers. I would expect that private data will be private just for one not for both. Reading

But also I feel that the behaviour needs to be consistent for all U-Boot buses, not just I2C. So can we modify the existing core DM code instead? I think a change to uclass_resolve_seq() might be enough.

I need to check calling sequence to see if this is properly timed or not.

Thanks, Michal

so, 2. úno 2019 7:09 odesílatel Simon Glass sjg@chromium.org napsal:

Hi Michal,

On Thu, 31 Jan 2019 at 04:03, Michal Simek michal.simek@xilinx.com wrote:

...

On 31. 01. 19 11:04, Simon Glass wrote:

...

Hi Michal,

On Fri, 18 Jan 2019 at 08:13, Michal Simek michal.simek@xilinx.com

wrote:

...

...

...

There is a need to find out the first free i2c ID which can be used

for

...

...

...

i2s buses (including i2c buses connected to i2c mux). Do it early in init and share this variable with other i2c classes for uniq bus identification.

Signed-off-by: Michal Simek michal.simek@xilinx.com

drivers/i2c/i2c-uclass.c | 19 +++++++++++++++++++ include/i2c.h | 3 +++ 2 files changed, 22 insertions(+)

diff --git a/drivers/i2c/i2c-uclass.c b/drivers/i2c/i2c-uclass.c index 975318e5f254..aaece115f02f 100644 --- a/drivers/i2c/i2c-uclass.c +++ b/drivers/i2c/i2c-uclass.c @@ -23,6 +23,8 @@ enum { PIN_COUNT, };

+int i2c_highest_id = -1;

/* Useful debugging function */ void i2c_dump_msgs(struct i2c_msg *msg, int nmsgs) { @@ -619,10 +621,27 @@ static int i2c_child_post_bind(struct udevice

*dev)

...

...

...

#endif }

+int i2c_uclass_init(struct uclass *class) +{ +#ifdef CONFIG_OF_LIVE

•   int ret;
  

•   ret = of_alias_get_highest_id("i2c");
  

•   if (ret >= 0)
  

•           i2c_highest_id = ret;
  

+#else

•   i2c_highest_id = fdtdec_get_alias_highest_id(gd->fdt_blob,
  

"i2c");

...

...

...

+#endif

This should call dev_read_alias_highest_id() or similar, which does the logic you have here.

not a problem with this.

...

...

•   debug("%s: i2c highest_id %d\n", __func__, i2c_highest_id);
  

•   return 0;
  

+}

UCLASS_DRIVER(i2c) = { .id = UCLASS_I2C, .name = "i2c", .flags = DM_UC_FLAG_SEQ_ALIAS,

•   .init           = i2c_uclass_init,
  

#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) .post_bind = dm_scan_fdt_dev, #endif diff --git a/include/i2c.h b/include/i2c.h index ccffc1955275..c69e26deaeed 100644 --- a/include/i2c.h +++ b/include/i2c.h @@ -16,6 +16,9 @@ #ifndef _I2C_H_ #define _I2C_H_

+/* Stores maximum highest i2c alias number */ +extern int i2c_highest_id;

This should be in uclass-private data, not a global.

In this case there are two uclasses. UCLASS_I2C and UCLASS_I2C_MUX which has to share the same numbers. I would expect that private data will be private just for one not for both. Reading

Hmm I hadn't realised that. Perhaps you can wrap it in a function which looks up the I2C uclass and the private data. We should try to avoid using BSS so that I2C can be used before relocation, etc.

I was wrong on this one. Také a look at v2.

M

...

...

But also I feel that the behaviour needs to be consistent for all U-Boot buses, not just I2C. So can we modify the existing core DM code instead? I think a change to uclass_resolve_seq() might be enough.

I need to check calling sequence to see if this is properly timed or not.

OK.

Regards, Simon _______________________________________________ U-Boot mailing list U-Boot@lists.denx.de https://lists.denx.de/listinfo/u-boot

Hi Heiko,

On 18. 01. 19 16:13, Michal Simek wrote:

Before this patch is applied all i2c sub-buses are using number -1 and they can't be addresses(switch to). If all busses are listed in DT alias they will get proper numbers and U-Boot can work with them. In Linux buses which are not listed in DT alias get uniq number which starts from the first highest free ID.

This is the behavior on ZynqMP zcu100-revA before this patch is applied.

Bus 0: i2c@ff020000 20: gpio@20, offset len 1, flags 0 21: gpio@21, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus -1: i2c@0 Bus -1: i2c@1 Bus -1: i2c@2 Bus 1: i2c@ff030000 74: i2c-mux@74, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus -1: i2c@0 Bus -1: i2c@1 Bus -1: i2c@2 Bus -1: i2c@3 Bus -1: i2c@4 Bus -1: i2c@0 Bus -1: i2c@1 Bus -1: i2c@2 Bus -1: i2c@3 Bus -1: i2c@4 Bus -1: i2c@5 Bus -1: i2c@6 Bus -1: i2c@7

When the patch is applied also i2c-mux busses are listed properly.

ZynqMP> i2c bus Bus 0: i2c@ff020000 20: gpio@20, offset len 1, flags 0 21: gpio@21, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus 2: i2c@ff020000->i2c-mux@75->i2c@0 Bus 3: i2c@ff020000->i2c-mux@75->i2c@1 Bus 4: i2c@ff020000->i2c-mux@75->i2c@2 Bus 1: i2c@ff030000 (active 1) 74: i2c-mux@74, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus 5: i2c@ff030000->i2c-mux@74->i2c@0 (active 5) 54: generic_54, offset len 1, flags 0 Bus 6: i2c@ff030000->i2c-mux@74->i2c@1 Bus 7: i2c@ff030000->i2c-mux@74->i2c@2 Bus 8: i2c@ff030000->i2c-mux@74->i2c@3 Bus 9: i2c@ff030000->i2c-mux@74->i2c@4 Bus 10: i2c@ff030000->i2c-mux@75->i2c@0 Bus 11: i2c@ff030000->i2c-mux@75->i2c@1 Bus 12: i2c@ff030000->i2c-mux@75->i2c@2 Bus 13: i2c@ff030000->i2c-mux@75->i2c@3 Bus 14: i2c@ff030000->i2c-mux@75->i2c@4 Bus 15: i2c@ff030000->i2c-mux@75->i2c@5 Bus 16: i2c@ff030000->i2c-mux@75->i2c@6 Bus 17: i2c@ff030000->i2c-mux@75->i2c@7

Signed-off-by: Michal Simek michal.simek@xilinx.com

zcu102-revA

Before this patch applied for !DM case with static description ZynqMP> i2c bus Bus 0: zynq_0 Bus 1: zynq_0->PCA9544A@0x75:0 Bus 2: zynq_0->PCA9544A@0x75:1 Bus 3: zynq_0->PCA9544A@0x75:2 Bus 4: zynq_1 Bus 5: zynq_1->PCA9548@0x74:0 Bus 6: zynq_1->PCA9548@0x74:1 Bus 7: zynq_1->PCA9548@0x74:2 Bus 8: zynq_1->PCA9548@0x74:3 Bus 9: zynq_1->PCA9548@0x74:4 Bus 10: zynq_1->PCA9548@0x75:0 Bus 11: zynq_1->PCA9548@0x75:1 Bus 12: zynq_1->PCA9548@0x75:2 Bus 13: zynq_1->PCA9548@0x75:3 Bus 14: zynq_1->PCA9548@0x75:4 Bus 15: zynq_1->PCA9548@0x75:5 Bus 16: zynq_1->PCA9548@0x75:6 Bus 17: zynq_1->PCA9548@0x75:7

When Patch is applied with OF_LIVE - of_alias_get_highest_id() is used ZynqMP> i2c bus Bus 0: i2c@ff020000 75: i2c-mux@75, offset len 1, flags 0 Bus 2: i2c@ff020000->i2c-mux@75->i2c@0 Bus 3: i2c@ff020000->i2c-mux@75->i2c@1 Bus 4: i2c@ff020000->i2c-mux@75->i2c@2 Bus 1: i2c@ff030000 74: i2c-mux@74, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus 5: i2c@ff030000->i2c-mux@74->i2c@0 Bus 6: i2c@ff030000->i2c-mux@74->i2c@1 Bus 7: i2c@ff030000->i2c-mux@74->i2c@2 Bus 8: i2c@ff030000->i2c-mux@74->i2c@3 Bus 9: i2c@ff030000->i2c-mux@74->i2c@4 Bus 10: i2c@ff030000->i2c-mux@75->i2c@0 Bus 11: i2c@ff030000->i2c-mux@75->i2c@1 Bus 12: i2c@ff030000->i2c-mux@75->i2c@2 Bus 13: i2c@ff030000->i2c-mux@75->i2c@3 Bus 14: i2c@ff030000->i2c-mux@75->i2c@4 Bus 15: i2c@ff030000->i2c-mux@75->i2c@5 Bus 16: i2c@ff030000->i2c-mux@75->i2c@6 Bus 17: i2c@ff030000->i2c-mux@75->i2c@7

For !OF_LIVE - hardcoded number is used ZynqMP> i2c bus Bus 0: i2c@ff020000 75: i2c-mux@75, offset len 1, flags 0 Bus 21: i2c@ff020000->i2c-mux@75->i2c@0 Bus 22: i2c@ff020000->i2c-mux@75->i2c@1 Bus 23: i2c@ff020000->i2c-mux@75->i2c@2 Bus 1: i2c@ff030000 74: i2c-mux@74, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus 24: i2c@ff030000->i2c-mux@74->i2c@0 Bus 25: i2c@ff030000->i2c-mux@74->i2c@1 Bus 26: i2c@ff030000->i2c-mux@74->i2c@2 Bus 27: i2c@ff030000->i2c-mux@74->i2c@3 Bus 28: i2c@ff030000->i2c-mux@74->i2c@4 Bus 29: i2c@ff030000->i2c-mux@75->i2c@0 Bus 30: i2c@ff030000->i2c-mux@75->i2c@1 Bus 31: i2c@ff030000->i2c-mux@75->i2c@2 Bus 32: i2c@ff030000->i2c-mux@75->i2c@3 Bus 33: i2c@ff030000->i2c-mux@75->i2c@4 Bus 34: i2c@ff030000->i2c-mux@75->i2c@5 Bus 35: i2c@ff030000->i2c-mux@75->i2c@6 Bus 36: i2c@ff030000->i2c-mux@75->i2c@7

---

drivers/i2c/muxes/i2c-mux-uclass.c | 35 +++++++++++++++++++++++++++++++++-- 1 file changed, 33 insertions(+), 2 deletions(-)

diff --git a/drivers/i2c/muxes/i2c-mux-uclass.c b/drivers/i2c/muxes/i2c-mux-uclass.c index a680ee176253..cb69d053fd59 100644 --- a/drivers/i2c/muxes/i2c-mux-uclass.c +++ b/drivers/i2c/muxes/i2c-mux-uclass.c @@ -9,6 +9,7 @@ #include <errno.h> #include <i2c.h> #include <dm/lists.h> +#include <dm/of_access.h> #include <dm/root.h>

/** @@ -59,13 +60,42 @@ static int i2c_mux_post_bind(struct udevice *mux) dev_for_each_subnode(node, mux) { struct udevice *dev; const char *name;

• const char *arrow = "->";
  

• char *full_name;
  

• int parent_name_len, arrow_len, mux_name_len, name_len;
  
  

  name = ofnode_get_name(node);

• ret = device_bind_driver_to_node(mux, "i2c_mux_bus_drv", name,
  

• 				 node, &dev);
  

• /* Calculate lenghts of strings */
  

• parent_name_len = strlen(mux->parent->name);
  

• arrow_len = strlen(arrow);
  

• mux_name_len = strlen(mux->name);
  

• name_len = strlen(name);
  

• full_name = calloc(1, parent_name_len + arrow_len +
  

• 		   mux_name_len + arrow_len + name_len + 1);
  

• if (!full_name)
  

• 	return -ENOMEM;
  

• /* Compose bus name */
  

• strcat(full_name, mux->parent->name);
  

• strcat(full_name, arrow);
  

• strcat(full_name, mux->name);
  

• strcat(full_name, arrow);
  

• strcat(full_name, name);
  

• ret = device_bind_driver_to_node(mux, "i2c_mux_bus_drv",
  

• 				 full_name, node, &dev);
  

  debug(" - bind ret=%d, %s\n", ret, dev ? dev->name : NULL); if (ret) return ret;

• /* If dt alias is not found start to allocate new IDs */
  

• if (dev->req_seq == -1)
  

• 	dev->req_seq = ++i2c_highest_id;
  

• debug("%s: MUX bus %d: %s\n", __func__, dev->req_seq,
  

•       full_name);
  

  }

  return 0;

@@ -192,6 +222,7 @@ U_BOOT_DRIVER(i2c_mux_bus) = {

UCLASS_DRIVER(i2c_mux) = { .id = UCLASS_I2C_MUX,

• .flags = DM_UC_FLAG_SEQ_ALIAS, .name = "i2c_mux", .post_bind = i2c_mux_post_bind, .post_probe = i2c_mux_post_probe,

Can you please take a look at this series?

Thanks, Michal

On 31. 01. 19 11:04, Simon Glass wrote:

Hi Michal,

On Fri, 18 Jan 2019 at 08:13, Michal Simek michal.simek@xilinx.com wrote:

...

Before this patch is applied all i2c sub-buses are using number -1 and they can't be addresses(switch to). If all busses are listed in DT alias they will get proper numbers and U-Boot can work with them. In Linux buses which are not listed in DT alias get uniq number which starts from the first highest free ID.

This is the behavior on ZynqMP zcu100-revA before this patch is applied.

Bus 0: i2c@ff020000 20: gpio@20, offset len 1, flags 0 21: gpio@21, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus -1: i2c@0 Bus -1: i2c@1 Bus -1: i2c@2 Bus 1: i2c@ff030000 74: i2c-mux@74, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus -1: i2c@0 Bus -1: i2c@1 Bus -1: i2c@2 Bus -1: i2c@3 Bus -1: i2c@4 Bus -1: i2c@0 Bus -1: i2c@1 Bus -1: i2c@2 Bus -1: i2c@3 Bus -1: i2c@4 Bus -1: i2c@5 Bus -1: i2c@6 Bus -1: i2c@7

When the patch is applied also i2c-mux busses are listed properly.

ZynqMP> i2c bus Bus 0: i2c@ff020000 20: gpio@20, offset len 1, flags 0 21: gpio@21, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus 2: i2c@ff020000->i2c-mux@75->i2c@0 Bus 3: i2c@ff020000->i2c-mux@75->i2c@1 Bus 4: i2c@ff020000->i2c-mux@75->i2c@2 Bus 1: i2c@ff030000 (active 1) 74: i2c-mux@74, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus 5: i2c@ff030000->i2c-mux@74->i2c@0 (active 5) 54: generic_54, offset len 1, flags 0 Bus 6: i2c@ff030000->i2c-mux@74->i2c@1 Bus 7: i2c@ff030000->i2c-mux@74->i2c@2 Bus 8: i2c@ff030000->i2c-mux@74->i2c@3 Bus 9: i2c@ff030000->i2c-mux@74->i2c@4 Bus 10: i2c@ff030000->i2c-mux@75->i2c@0 Bus 11: i2c@ff030000->i2c-mux@75->i2c@1 Bus 12: i2c@ff030000->i2c-mux@75->i2c@2 Bus 13: i2c@ff030000->i2c-mux@75->i2c@3 Bus 14: i2c@ff030000->i2c-mux@75->i2c@4 Bus 15: i2c@ff030000->i2c-mux@75->i2c@5 Bus 16: i2c@ff030000->i2c-mux@75->i2c@6 Bus 17: i2c@ff030000->i2c-mux@75->i2c@7

Signed-off-by: Michal Simek michal.simek@xilinx.com

zcu102-revA

Before this patch applied for !DM case with static description ZynqMP> i2c bus Bus 0: zynq_0 Bus 1: zynq_0->PCA9544A@0x75:0 Bus 2: zynq_0->PCA9544A@0x75:1 Bus 3: zynq_0->PCA9544A@0x75:2 Bus 4: zynq_1 Bus 5: zynq_1->PCA9548@0x74:0 Bus 6: zynq_1->PCA9548@0x74:1 Bus 7: zynq_1->PCA9548@0x74:2 Bus 8: zynq_1->PCA9548@0x74:3 Bus 9: zynq_1->PCA9548@0x74:4 Bus 10: zynq_1->PCA9548@0x75:0 Bus 11: zynq_1->PCA9548@0x75:1 Bus 12: zynq_1->PCA9548@0x75:2 Bus 13: zynq_1->PCA9548@0x75:3 Bus 14: zynq_1->PCA9548@0x75:4 Bus 15: zynq_1->PCA9548@0x75:5 Bus 16: zynq_1->PCA9548@0x75:6 Bus 17: zynq_1->PCA9548@0x75:7

When Patch is applied with OF_LIVE - of_alias_get_highest_id() is used ZynqMP> i2c bus Bus 0: i2c@ff020000 75: i2c-mux@75, offset len 1, flags 0 Bus 2: i2c@ff020000->i2c-mux@75->i2c@0 Bus 3: i2c@ff020000->i2c-mux@75->i2c@1 Bus 4: i2c@ff020000->i2c-mux@75->i2c@2 Bus 1: i2c@ff030000 74: i2c-mux@74, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus 5: i2c@ff030000->i2c-mux@74->i2c@0 Bus 6: i2c@ff030000->i2c-mux@74->i2c@1 Bus 7: i2c@ff030000->i2c-mux@74->i2c@2 Bus 8: i2c@ff030000->i2c-mux@74->i2c@3 Bus 9: i2c@ff030000->i2c-mux@74->i2c@4 Bus 10: i2c@ff030000->i2c-mux@75->i2c@0 Bus 11: i2c@ff030000->i2c-mux@75->i2c@1 Bus 12: i2c@ff030000->i2c-mux@75->i2c@2 Bus 13: i2c@ff030000->i2c-mux@75->i2c@3 Bus 14: i2c@ff030000->i2c-mux@75->i2c@4 Bus 15: i2c@ff030000->i2c-mux@75->i2c@5 Bus 16: i2c@ff030000->i2c-mux@75->i2c@6 Bus 17: i2c@ff030000->i2c-mux@75->i2c@7

For !OF_LIVE - hardcoded number is used ZynqMP> i2c bus Bus 0: i2c@ff020000 75: i2c-mux@75, offset len 1, flags 0 Bus 21: i2c@ff020000->i2c-mux@75->i2c@0 Bus 22: i2c@ff020000->i2c-mux@75->i2c@1 Bus 23: i2c@ff020000->i2c-mux@75->i2c@2 Bus 1: i2c@ff030000 74: i2c-mux@74, offset len 1, flags 0 75: i2c-mux@75, offset len 1, flags 0 Bus 24: i2c@ff030000->i2c-mux@74->i2c@0 Bus 25: i2c@ff030000->i2c-mux@74->i2c@1 Bus 26: i2c@ff030000->i2c-mux@74->i2c@2 Bus 27: i2c@ff030000->i2c-mux@74->i2c@3 Bus 28: i2c@ff030000->i2c-mux@74->i2c@4 Bus 29: i2c@ff030000->i2c-mux@75->i2c@0 Bus 30: i2c@ff030000->i2c-mux@75->i2c@1 Bus 31: i2c@ff030000->i2c-mux@75->i2c@2 Bus 32: i2c@ff030000->i2c-mux@75->i2c@3 Bus 33: i2c@ff030000->i2c-mux@75->i2c@4 Bus 34: i2c@ff030000->i2c-mux@75->i2c@5 Bus 35: i2c@ff030000->i2c-mux@75->i2c@6 Bus 36: i2c@ff030000->i2c-mux@75->i2c@7

---

drivers/i2c/muxes/i2c-mux-uclass.c | 35 +++++++++++++++++++++++++++++++++-- 1 file changed, 33 insertions(+), 2 deletions(-)

This is quite a complicated issue...

and it is time to talk about it.

...

diff --git a/drivers/i2c/muxes/i2c-mux-uclass.c b/drivers/i2c/muxes/i2c-mux-uclass.c index a680ee176253..cb69d053fd59 100644 --- a/drivers/i2c/muxes/i2c-mux-uclass.c +++ b/drivers/i2c/muxes/i2c-mux-uclass.c @@ -9,6 +9,7 @@ #include <errno.h> #include <i2c.h> #include <dm/lists.h> +#include <dm/of_access.h> #include <dm/root.h>

/** @@ -59,13 +60,42 @@ static int i2c_mux_post_bind(struct udevice *mux) dev_for_each_subnode(node, mux) { struct udevice *dev; const char *name;

•           const char *arrow = "->";
  

•           char *full_name;
  

•           int parent_name_len, arrow_len, mux_name_len, name_len;
  
            name = ofnode_get_name(node);
  

•           ret = device_bind_driver_to_node(mux, "i2c_mux_bus_drv", name,
  

•                                            node, &dev);
  

•           /* Calculate lenghts of strings */
  

•           parent_name_len = strlen(mux->parent->name);
  

•           arrow_len = strlen(arrow);
  

•           mux_name_len = strlen(mux->name);
  

•           name_len = strlen(name);
  

•           full_name = calloc(1, parent_name_len + arrow_len +
  

•                              mux_name_len + arrow_len + name_len + 1);
  

•           if (!full_name)
  

•                   return -ENOMEM;
  

•           /* Compose bus name */
  

•           strcat(full_name, mux->parent->name);
  

•           strcat(full_name, arrow);
  

•           strcat(full_name, mux->name);
  

•           strcat(full_name, arrow);
  

•           strcat(full_name, name);
  

•           ret = device_bind_driver_to_node(mux, "i2c_mux_bus_drv",
  

•                                            full_name, node, &dev);
  

Why do we use the full name here? We can create this by looking at the parents if needed.

If you look at i2c mux description and you have 3-4 i2c muxes with proper sub bus node name like i2c@0 for all first busses you need to distinguish that. Linux is doing that too.

Because from this - you have no idea which bus is which. Bus -1: i2c@1 Bus -1: i2c@2 Bus -1: i2c@3 Bus -1: i2c@4 Bus -1: i2c@0 Bus -1: i2c@1 Bus -1: i2c@2 Bus -1: i2c@3 Bus -1: i2c@4 Bus -1: i2c@5 Bus -1: i2c@6 Bus -1: i2c@7

That code above is just composing bus name by looking at parent + actual node name. If you see better way how this can be done please let me know.

...

            debug("   - bind ret=%d, %s\n", ret, dev ? dev->name : NULL);
            if (ret)
                    return ret;

•           /* If dt alias is not found start to allocate new IDs */
  

•           if (dev->req_seq == -1)
  

•                   dev->req_seq = ++i2c_highest_id;
  

As in the other patch, if we change this, I think we should do it in DM core, by adjusting uclass_resolve_seq(). Would that work?

I have played with it and uclass_resolve_seq as is now is called when device is probed. But IMHO you need to have bus number before probe to have it listed to know where you should move by i2c dev command.

It means setting up req->seq should be done in bind phase.

Let me send you v2 of this where I fix some stuff based on your comments.

Thanks, Michal