All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH v2] hwmon: (pmbus) Clear only CML invalid command fault during probe
@ 2026-11-19  1:32 Arthur Korn
  2026-11-19  2:08 ` sashiko-bot
  0 siblings, 1 reply; 3+ messages in thread
From: Arthur Korn @ 2026-11-19  1:32 UTC (permalink / raw)
  To: Guenter Roeck; +Cc: linux-hwmon, linux-kernel, Arthur Korn

During driver initialization and register capability probing, pmbus_core
and the generic pmbus driver send CLEAR_FAULTS (0x03) to clear any
communication or command faults triggered while probing unsupported
registers. However, CLEAR_FAULTS unconditionally clears all latched
fault status bits across the device, including hardware power fault
registers such as STATUS_WORD, STATUS_VOUT, STATUS_IOUT, and
STATUS_MFR_SPECIFIC. As a result, a BMC reboot or driver re-probe erases
pre-existing voltage regulator fault telemetry before monitoring or
diagnostic tools can inspect it.

Simply omitting fault clearing during probe is not viable because
probing unsupported commands (such as PMBUS_PAGE on single-page PMBus
devices) asserts PB_CML_FAULT_INVALID_COMMAND in STATUS_CML. Leaving
that bit latched causes pmbus_check_status_cml() to fail subsequent
capability checks (including PMBUS_VOUT_MODE and sensor discovery) with
-EIO.

Starting with PMBus Revision 1.2, individual bits in STATUS_CML can be
cleared via Write-1-to-Clear (W1C), a mechanism pmbus_core already uses
in pmbus_get_boolean(). Introduce pmbus_clear_cml_fault() and export
pmbus_clear_cml_faults() in the "PMBUS" module namespace to write
PB_CML_FAULT_INVALID_COMMAND to PMBUS_STATUS_CML during device
initialization, register probing, and pmbus_identify() on PMBus 1.2+
devices while preserving all other latched fault bits. Fall back to
pmbus_clear_fault_page() on PMBus < 1.2 devices or if writing
PMBUS_STATUS_CML fails.

Tested on BMC hardware across 130 hwmon devices (105 PMBus devices
covering tda38740/xdpe1a2g5b, isl68137/raa228234, adm1266,
ltc4286/ltc4287, and generic pmbus regulators including TI
tps546e25/tps546b25 and factory write-protected converters), verifying
that all devices probe cleanly with no -EIO regressions. Also verified
via unbind, fault injection, and re-bind across both driver-specific and
generic pmbus_identify() paths that PB_CML_FAULT_INVALID_COMMAND in
PMBUS_STATUS_CML is cleared on probe while other STATUS_CML and
STATUS_WORD fault bits remain preserved, and verified clean fallback
when writing PMBUS_STATUS_CML is NACKed.

Signed-off-by: Arthur Korn <akorn@google.com>
---
Changes in v2:
- Fall back to pmbus_clear_fault_page() if writing PMBUS_STATUS_CML
  fails (e.g. NACK on devices that do not support W1C on STATUS_CML).
- Export pmbus_clear_cml_faults() in the "PMBUS" namespace and use it in
  pmbus_identify() (pmbus.c) so generic pmbus devices also preserve
  non-CML faults during probe.
- Re-tested on hardware covering generic pmbus_identify() fault
  preservation and the STATUS_CML write-failure fallback path.

 drivers/hwmon/pmbus/pmbus.c      |  2 +-
 drivers/hwmon/pmbus/pmbus.h      |  1 +
 drivers/hwmon/pmbus/pmbus_core.c | 30 +++++++++++++++++++++++++-----
 3 files changed, 27 insertions(+), 6 deletions(-)

diff --git a/drivers/hwmon/pmbus/pmbus.c b/drivers/hwmon/pmbus/pmbus.c
index f8189c11..f239dafb 100644
--- a/drivers/hwmon/pmbus/pmbus.c
+++ b/drivers/hwmon/pmbus/pmbus.c
@@ -113,7 +113,7 @@ static int pmbus_identify(struct i2c_client *client,
 			info->pages = 1;
 		}
 
-		pmbus_clear_faults(client);
+		pmbus_clear_cml_faults(client);
 	}
 
 	if (pmbus_check_byte_register(client, 0, PMBUS_VOUT_MODE)) {
diff --git a/drivers/hwmon/pmbus/pmbus.h b/drivers/hwmon/pmbus/pmbus.h
index 5525c048..8170098f 100644
--- a/drivers/hwmon/pmbus/pmbus.h
+++ b/drivers/hwmon/pmbus/pmbus.h
@@ -581,6 +581,7 @@ int pmbus_write_byte_data(struct i2c_client *client, int page, u8 reg,
 int pmbus_update_byte_data(struct i2c_client *client, int page, u8 reg,
 			   u8 mask, u8 value);
 void pmbus_clear_faults(struct i2c_client *client);
+void pmbus_clear_cml_faults(struct i2c_client *client);
 void pmbus_check_and_notify_faults(struct i2c_client *client);
 bool pmbus_check_byte_register(struct i2c_client *client, int page, int reg);
 bool pmbus_check_word_register(struct i2c_client *client, int page, int reg);
diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c
index b3aa7f1a..a5255ffb 100644
--- a/drivers/hwmon/pmbus/pmbus_core.c
+++ b/drivers/hwmon/pmbus/pmbus_core.c
@@ -717,6 +717,26 @@ void pmbus_clear_faults(struct i2c_client *client)
 }
 EXPORT_SYMBOL_NS_GPL(pmbus_clear_faults, "PMBUS");
 
+static void pmbus_clear_cml_fault(struct i2c_client *client, int page)
+{
+	struct pmbus_data *data = i2c_get_clientdata(client);
+
+	if (data->revision < PMBUS_REV_12 ||
+	    _pmbus_write_byte_data(client, page, PMBUS_STATUS_CML,
+				   PB_CML_FAULT_INVALID_COMMAND) < 0)
+		pmbus_clear_fault_page(client, page);
+}
+
+void pmbus_clear_cml_faults(struct i2c_client *client)
+{
+	struct pmbus_data *data = i2c_get_clientdata(client);
+	int i;
+
+	for (i = 0; i < data->info->pages; i++)
+		pmbus_clear_cml_fault(client, i);
+}
+EXPORT_SYMBOL_NS_GPL(pmbus_clear_cml_faults, "PMBUS");
+
 static int pmbus_check_status_cml(struct i2c_client *client)
 {
 	struct pmbus_data *data = i2c_get_clientdata(client);
@@ -745,7 +765,7 @@ static bool pmbus_check_register(struct i2c_client *client,
 	if (rv < 0 && (data->flags & PMBUS_READ_STATUS_AFTER_FAILED_CHECK))
 		data->read_status(client, -1);
 	if (reg < PMBUS_VIRT_BASE)
-		pmbus_clear_fault_page(client, -1);
+		pmbus_clear_cml_fault(client, -1);
 	return rv >= 0;
 }
 
@@ -762,7 +782,7 @@ static bool pmbus_check_status_register(struct i2c_client *client, int page)
 			status = -EIO;
 	}
 
-	pmbus_clear_fault_page(client, -1);
+	pmbus_clear_cml_fault(client, -1);
 	return status >= 0;
 }
 
@@ -790,7 +810,7 @@ static bool __maybe_unused pmbus_check_block_register(struct i2c_client *client,
 		rv = pmbus_check_status_cml(client);
 	if (rv < 0 && (data->flags & PMBUS_READ_STATUS_AFTER_FAILED_CHECK))
 		data->read_status(client, -1);
-	pmbus_clear_fault_page(client, -1);
+	pmbus_clear_cml_fault(client, -1);
 	return rv >= 0;
 }
 
@@ -3053,9 +3073,9 @@ static int pmbus_init_common(struct i2c_client *client, struct pmbus_data *data,
 	}
 
 	if (data->info->pages)
-		pmbus_clear_faults(client);
+		pmbus_clear_cml_faults(client);
 	else
-		pmbus_clear_fault_page(client, -1);
+		pmbus_clear_cml_fault(client, -1);
 
 	if (info->identify) {
 		ret = (*info->identify)(client, info);
-- 
2.56.1.703.gf848ed807c-goog


^ permalink raw reply related	[flat|nested] 3+ messages in thread

* Re: [PATCH v2] hwmon: (pmbus) Clear only CML invalid command fault during probe
  2026-11-19  1:32 [PATCH v2] hwmon: (pmbus) Clear only CML invalid command fault during probe Arthur Korn
@ 2026-11-19  2:08 ` sashiko-bot
  2026-11-19  2:29   ` Guenter Roeck
  0 siblings, 1 reply; 3+ messages in thread
From: sashiko-bot @ 2026-11-19  2:08 UTC (permalink / raw)
  To: Arthur Korn; +Cc: linux-hwmon

Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [Medium] hwmon: (pmbus) pmbus_write_smbalert_mask() clears all faults
  on probe
- [Medium] hwmon: (pmbus) pmbus_clear_cml_fault() leaves INVALID_DATA
  fault latched
- [Medium] hwmon: (pmbus) pmbus_clear_cml_fault() bypasses driver
  write_byte() hooks
--

commit 94d47b10140ba57df41cbee449647587f625d5f3
Author: Arthur Korn <akorn@google.com>

hwmon: (pmbus) Clear only CML invalid command fault during probe

This commit introduces pmbus_clear_cml_fault() and pmbus_clear_cml_faults()
to clear only PB_CML_FAULT_INVALID_COMMAND in PMBUS_STATUS_CML during probe
on PMBus 1.2+ devices instead of sending PMBUS_CLEAR_FAULTS.

> diff --git a/drivers/hwmon/pmbus/pmbus.c b/drivers/hwmon/pmbus/pmbus.c
> index 28889fb743b3e..e2e49fdd54841 100644
> --- a/drivers/hwmon/pmbus/pmbus.c
> +++ b/drivers/hwmon/pmbus/pmbus.c
> @@ -113,7 +113,7 @@ static int pmbus_identify(struct i2c_client *client,
>  			info->pages = 1;
>  		}
>  
> -		pmbus_clear_faults(client);
> +		pmbus_clear_cml_faults(client);
>  	}
>  
>  	if (pmbus_check_byte_register(client, 0, PMBUS_VOUT_MODE)) {

[Severity: Medium]
Does pmbus_clear_cml_faults() leave PB_CML_FAULT_INVALID_DATA latched
after page auto-detection?

When PMBUS_PAGE is supported, the loop right above probes page numbers
with pmbus_set_page() until writing an out-of-range page index to the
valid PMBUS_PAGE command fails, which latches PB_CML_FAULT_INVALID_DATA
in PMBUS_STATUS_CML.  Because pmbus_clear_cml_fault() only writes
PB_CML_FAULT_INVALID_COMMAND, PB_CML_FAULT_INVALID_DATA and PB_STATUS_CML
remain latched after probe.

> diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c
> index 82dac2c6c579f..74fc8d7481689 100644
> --- a/drivers/hwmon/pmbus/pmbus_core.c
> +++ b/drivers/hwmon/pmbus/pmbus_core.c
> @@ -713,6 +713,26 @@ void pmbus_clear_faults(struct i2c_client *client)
>  }
>  EXPORT_SYMBOL_NS_GPL(pmbus_clear_faults, "PMBUS");
>  
> +static void pmbus_clear_cml_fault(struct i2c_client *client, int page)
> +{
> +	struct pmbus_data *data = i2c_get_clientdata(client);
> +
> +	if (data->revision < PMBUS_REV_12 ||
> +	    _pmbus_write_byte_data(client, page, PMBUS_STATUS_CML,
> +				   PB_CML_FAULT_INVALID_COMMAND) < 0)
> +		pmbus_clear_fault_page(client, page);
> +}

[Severity: Medium]
Does calling _pmbus_write_byte_data() here bypass driver info->write_byte
callbacks that intercept or block fault clearing?

Previously, probe-time fault clearing always went through
pmbus_clear_fault_page() -> _pmbus_write_byte(client, page,
PMBUS_CLEAR_FAULTS).  Several drivers hook write_byte without
implementing write_byte_data: for example, max16601_write_byte() routes
page 2 (PMBUS_PAGE_VIRTUAL) PMBUS_CLEAR_FAULTS to a separate I2C client
(data->vsa), so _pmbus_write_byte_data() writes PMBUS_STATUS_CML to
client (page 0) instead and leaves CML faults on data->vsa uncleared.
Similarly, delta-ahe50dc-fan and max31785 block fault-clearing writes in
write_byte(), and mpq8646 blocks unsupported PMBUS_STATUS_CML reads in
read_byte_data(), none of which implement write_byte_data().

> +
> +void pmbus_clear_cml_faults(struct i2c_client *client)
> +{
> +	struct pmbus_data *data = i2c_get_clientdata(client);
> +	int i;
> +
> +	for (i = 0; i < data->info->pages; i++)
> +		pmbus_clear_cml_fault(client, i);
> +}
> +EXPORT_SYMBOL_NS_GPL(pmbus_clear_cml_faults, "PMBUS");
> +
>  static int pmbus_check_status_cml(struct i2c_client *client)
>  {
>  	struct pmbus_data *data = i2c_get_clientdata(client);

[ ... ]

> @@ -2978,9 +2998,9 @@ static int pmbus_init_common(struct i2c_client *client, struct pmbus_data *data,
>  	}
>  
>  	if (data->info->pages)
> -		pmbus_clear_faults(client);
> +		pmbus_clear_cml_faults(client);
>  	else
> -		pmbus_clear_fault_page(client, -1);
> +		pmbus_clear_cml_fault(client, -1);

[Severity: Medium]
Should pmbus_write_smbalert_mask() also use pmbus_clear_cml_fault()
instead of pmbus_clear_fault_page()?

When client->irq is set, pmbus_do_probe() -> pmbus_irq_setup() calls
pmbus_write_smbalert_mask() across every page:

drivers/hwmon/pmbus/pmbus_core.c:pmbus_write_smbalert_mask() {
    ...
	ret = _pmbus_write_word_data(client, page, PMBUS_SMBALERT_MASK,
				     reg | (val << 8));
	/*
	 * Clear fault systematically in case writing PMBUS_SMBALERT_MASK
	 * is not supported by the chip.
	 */
	pmbus_clear_fault_page(client, page);

	return ret;
}

On any IRQ-enabled PMBus device, this still sends PMBUS_CLEAR_FAULTS on
all pages during probe and erases the latched fault registers that this
patch aims to preserve.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261119013208.4089480-1-akorn@google.com?part=1

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH v2] hwmon: (pmbus) Clear only CML invalid command fault during probe
  2026-11-19  2:08 ` sashiko-bot
@ 2026-11-19  2:29   ` Guenter Roeck
  0 siblings, 0 replies; 3+ messages in thread
From: Guenter Roeck @ 2026-11-19  2:29 UTC (permalink / raw)
  To: sashiko-reviews, Arthur Korn; +Cc: linux-hwmon

Hi Arthur,

On 11/18/26 18:08, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 3 potential
> issues to consider:
> - [Medium] hwmon: (pmbus) pmbus_write_smbalert_mask() clears all faults
>    on probe
> - [Medium] hwmon: (pmbus) pmbus_clear_cml_fault() leaves INVALID_DATA
>    fault latched
> - [Medium] hwmon: (pmbus) pmbus_clear_cml_fault() bypasses driver
>    write_byte() hooks
> --
> 

Looks like Sashiko has a point, including the write_byte hook problem.
I think Google uses MAX16601, and that is a PMBus 1.2 chip, so it may
be affected.

On a side note, it should be possible to run Sashiko locally on your
system, or I can run it before you submit the next version if you
have trouble setting it up. This might avoid a couple of revisions
if/when Sashiko finds new problems.

Thanks
Guenter

> commit 94d47b10140ba57df41cbee449647587f625d5f3
> Author: Arthur Korn <akorn@google.com>
> 
> hwmon: (pmbus) Clear only CML invalid command fault during probe
> 
> This commit introduces pmbus_clear_cml_fault() and pmbus_clear_cml_faults()
> to clear only PB_CML_FAULT_INVALID_COMMAND in PMBUS_STATUS_CML during probe
> on PMBus 1.2+ devices instead of sending PMBUS_CLEAR_FAULTS.
> 
>> diff --git a/drivers/hwmon/pmbus/pmbus.c b/drivers/hwmon/pmbus/pmbus.c
>> index 28889fb743b3e..e2e49fdd54841 100644
>> --- a/drivers/hwmon/pmbus/pmbus.c
>> +++ b/drivers/hwmon/pmbus/pmbus.c
>> @@ -113,7 +113,7 @@ static int pmbus_identify(struct i2c_client *client,
>>   			info->pages = 1;
>>   		}
>>   
>> -		pmbus_clear_faults(client);
>> +		pmbus_clear_cml_faults(client);
>>   	}
>>   
>>   	if (pmbus_check_byte_register(client, 0, PMBUS_VOUT_MODE)) {
> 
> [Severity: Medium]
> Does pmbus_clear_cml_faults() leave PB_CML_FAULT_INVALID_DATA latched
> after page auto-detection?
> 
> When PMBUS_PAGE is supported, the loop right above probes page numbers
> with pmbus_set_page() until writing an out-of-range page index to the
> valid PMBUS_PAGE command fails, which latches PB_CML_FAULT_INVALID_DATA
> in PMBUS_STATUS_CML.  Because pmbus_clear_cml_fault() only writes
> PB_CML_FAULT_INVALID_COMMAND, PB_CML_FAULT_INVALID_DATA and PB_STATUS_CML
> remain latched after probe.
> 
>> diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c
>> index 82dac2c6c579f..74fc8d7481689 100644
>> --- a/drivers/hwmon/pmbus/pmbus_core.c
>> +++ b/drivers/hwmon/pmbus/pmbus_core.c
>> @@ -713,6 +713,26 @@ void pmbus_clear_faults(struct i2c_client *client)
>>   }
>>   EXPORT_SYMBOL_NS_GPL(pmbus_clear_faults, "PMBUS");
>>   
>> +static void pmbus_clear_cml_fault(struct i2c_client *client, int page)
>> +{
>> +	struct pmbus_data *data = i2c_get_clientdata(client);
>> +
>> +	if (data->revision < PMBUS_REV_12 ||
>> +	    _pmbus_write_byte_data(client, page, PMBUS_STATUS_CML,
>> +				   PB_CML_FAULT_INVALID_COMMAND) < 0)
>> +		pmbus_clear_fault_page(client, page);
>> +}
> 
> [Severity: Medium]
> Does calling _pmbus_write_byte_data() here bypass driver info->write_byte
> callbacks that intercept or block fault clearing?
> 
> Previously, probe-time fault clearing always went through
> pmbus_clear_fault_page() -> _pmbus_write_byte(client, page,
> PMBUS_CLEAR_FAULTS).  Several drivers hook write_byte without
> implementing write_byte_data: for example, max16601_write_byte() routes
> page 2 (PMBUS_PAGE_VIRTUAL) PMBUS_CLEAR_FAULTS to a separate I2C client
> (data->vsa), so _pmbus_write_byte_data() writes PMBUS_STATUS_CML to
> client (page 0) instead and leaves CML faults on data->vsa uncleared.
> Similarly, delta-ahe50dc-fan and max31785 block fault-clearing writes in
> write_byte(), and mpq8646 blocks unsupported PMBUS_STATUS_CML reads in
> read_byte_data(), none of which implement write_byte_data().
> 
>> +
>> +void pmbus_clear_cml_faults(struct i2c_client *client)
>> +{
>> +	struct pmbus_data *data = i2c_get_clientdata(client);
>> +	int i;
>> +
>> +	for (i = 0; i < data->info->pages; i++)
>> +		pmbus_clear_cml_fault(client, i);
>> +}
>> +EXPORT_SYMBOL_NS_GPL(pmbus_clear_cml_faults, "PMBUS");
>> +
>>   static int pmbus_check_status_cml(struct i2c_client *client)
>>   {
>>   	struct pmbus_data *data = i2c_get_clientdata(client);
> 
> [ ... ]
> 
>> @@ -2978,9 +2998,9 @@ static int pmbus_init_common(struct i2c_client *client, struct pmbus_data *data,
>>   	}
>>   
>>   	if (data->info->pages)
>> -		pmbus_clear_faults(client);
>> +		pmbus_clear_cml_faults(client);
>>   	else
>> -		pmbus_clear_fault_page(client, -1);
>> +		pmbus_clear_cml_fault(client, -1);
> 
> [Severity: Medium]
> Should pmbus_write_smbalert_mask() also use pmbus_clear_cml_fault()
> instead of pmbus_clear_fault_page()?
> 
> When client->irq is set, pmbus_do_probe() -> pmbus_irq_setup() calls
> pmbus_write_smbalert_mask() across every page:
> 
> drivers/hwmon/pmbus/pmbus_core.c:pmbus_write_smbalert_mask() {
>      ...
> 	ret = _pmbus_write_word_data(client, page, PMBUS_SMBALERT_MASK,
> 				     reg | (val << 8));
> 	/*
> 	 * Clear fault systematically in case writing PMBUS_SMBALERT_MASK
> 	 * is not supported by the chip.
> 	 */
> 	pmbus_clear_fault_page(client, page);
> 
> 	return ret;
> }
> 
> On any IRQ-enabled PMBus device, this still sends PMBUS_CLEAR_FAULTS on
> all pages during probe and erases the latched fault registers that this
> patch aims to preserve.
> 


^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2026-11-19  2:29 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-19  1:32 [PATCH v2] hwmon: (pmbus) Clear only CML invalid command fault during probe Arthur Korn
2026-11-19  2:08 ` sashiko-bot
2026-11-19  2:29   ` Guenter Roeck

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.