Re: [PATCH] hwmon: Add driver for fsp-3y PSUs and PDUs

From: Guenter Roeck
Date: Mon Mar 29 2021 - 13:48:23 EST


On 3/29/21 7:38 AM, Václav Kubernát wrote:
> After some testing, I have found out there is a timing issue with this
> device. After setting page, the device doesn't immediately react and
> gives values from the previous page for some time. This is why there
> needs to be a delay between pmbus_set_page and the actual read.
>
> Also, a lot of the standard commands don't work with the devices, so
> they are filtered out in the custom read function.
>

This is not an appropriate patch description. Describe the driver here,
not the workarounds / quirks. The reason for the delay should be a
comment in the patch, not in the patch description.

Also, "don't work" is inappropriate (and, again, does not belong into
the patch description). It is perfectly appropriate for the core
to try those commands to see if they are supported. The only reason
to mask them out would be that the device reacts badly to seeing
them. If that is the case, "don't work" should be replaced with
a more detailed comment in the code. Describe what happens, and why
the commands needs to be caught.


What might be useful is a note indicating if you have a manual for
those power supplies available, or if the driver is based on reverse
engineering.

> Signed-off-by: Václav Kubernát <kubernat@xxxxxxxxx>
> ---
> drivers/hwmon/pmbus/Kconfig | 9 ++
> drivers/hwmon/pmbus/Makefile | 1 +
> drivers/hwmon/pmbus/fsp-3y.c | 164 +++++++++++++++++++++++++++++++++++

Documentation/hwmon/fsp-3y.rst is missing.

> 3 files changed, 174 insertions(+)
> create mode 100644 drivers/hwmon/pmbus/fsp-3y.c
>
> diff --git a/drivers/hwmon/pmbus/Kconfig b/drivers/hwmon/pmbus/Kconfig
> index 03606d4298a4..66d1655b6750 100644
> --- a/drivers/hwmon/pmbus/Kconfig
> +++ b/drivers/hwmon/pmbus/Kconfig
> @@ -56,6 +56,15 @@ config SENSORS_BEL_PFE
> This driver can also be built as a module. If so, the module will
> be called bel-pfe.
>
> +config SENSORS_FSP_3Y
> + tristate "FSP/3Y-Power power supplies"
> + help
> + If you say yes here you get hardware monitoring support for
> + FSP/3Y-Power hot-swap power supplies.
> +

This should list the supported models - if not here, then at least in the
(missing) documentation.

> + This driver can also be built as a module. If so, the module will
> + be called fsp-3y.
> +
> config SENSORS_IBM_CFFPS
> tristate "IBM Common Form Factor Power Supply"
> depends on LEDS_CLASS
> diff --git a/drivers/hwmon/pmbus/Makefile b/drivers/hwmon/pmbus/Makefile
> index 6a4ba0fdc1db..bfe218ad898f 100644
> --- a/drivers/hwmon/pmbus/Makefile
> +++ b/drivers/hwmon/pmbus/Makefile
> @@ -8,6 +8,7 @@ obj-$(CONFIG_SENSORS_PMBUS) += pmbus.o
> obj-$(CONFIG_SENSORS_ADM1266) += adm1266.o
> obj-$(CONFIG_SENSORS_ADM1275) += adm1275.o
> obj-$(CONFIG_SENSORS_BEL_PFE) += bel-pfe.o
> +obj-$(CONFIG_SENSORS_FSP_3Y) += fsp-3y.o
> obj-$(CONFIG_SENSORS_IBM_CFFPS) += ibm-cffps.o
> obj-$(CONFIG_SENSORS_INSPUR_IPSPS) += inspur-ipsps.o
> obj-$(CONFIG_SENSORS_IR35221) += ir35221.o
> diff --git a/drivers/hwmon/pmbus/fsp-3y.c b/drivers/hwmon/pmbus/fsp-3y.c
> new file mode 100644
> index 000000000000..7f3c3de3a1e6
> --- /dev/null
> +++ b/drivers/hwmon/pmbus/fsp-3y.c
> @@ -0,0 +1,164 @@
> +// SPDX-License-Identifier: GPL-2.0-or-later
> +/*
> + * Hardware monitoring driver for FSP 3Y-Power PSUs
> + *
> + * Copyright (c) 2021 Václav Kubernát, CESNET
> + */
> +
> +#include <linux/kernel.h>
> +#include <linux/module.h>
> +#include <linux/delay.h>
> +#include <linux/i2c.h>

Alphabetic include file order, please.

> +#include "pmbus.h"
> +
> +#define YM2151_PAGE_12V 0x00
> +#define YM2151_PAGE_5V 0x20
> +#define YH5151E_PAGE_12V 0x00
> +#define YH5151E_PAGE_5V 0x10
> +#define YH5151E_PAGE_3V3 0x11
> +
> +enum chips {
> + ym2151e,
> + yh5151e
> +};
> +
> +static int set_page(struct i2c_client *client, int page)
> +{
> + int rv;
> +
> + rv = i2c_smbus_read_byte_data(client, PMBUS_PAGE);
> +
Please no empty line here.

You might want to consider caching the current page to avoid having to read it
for each access, similar to the code implemented in the pmbus core.

> + if (rv < 0)
> + return rv;
> +
> + if (rv != page) {
> + rv = pmbus_set_page(client, page, 0xff);
> + if (rv < 0)
> + return rv;
> +
> + msleep(20);

Please use usleep_range(), and make sure that this huge delay is actually needed.

> + }
> +
> + return 0;
> +}
> +
> +static int fsp3y_read_byte_data(struct i2c_client *client, int page, int reg)
> +{
> + int rv;
> +
> + rv = set_page(client, page);
> + if (rv < 0)
> + return rv;
> +
> + return i2c_smbus_read_byte_data(client, reg);
> +}
> +
> +static int fsp3y_read_word_data(struct i2c_client *client, int page, int phase, int reg)
> +{
> + int rv;
> +
> + if (reg >= PMBUS_VIRT_BASE)
> + return -ENXIO;
> +
> + switch (reg) {
> + case PMBUS_OT_WARN_LIMIT:
> + case PMBUS_OT_FAULT_LIMIT:
> + case PMBUS_UT_WARN_LIMIT:
> + case PMBUS_UT_FAULT_LIMIT:
> + case PMBUS_VIN_UV_WARN_LIMIT:
> + case PMBUS_VIN_UV_FAULT_LIMIT:
> + case PMBUS_VIN_OV_FAULT_LIMIT:
> + case PMBUS_VIN_OV_WARN_LIMIT:
> + case PMBUS_IOUT_OC_WARN_LIMIT:
> + case PMBUS_IOUT_UC_FAULT_LIMIT:
> + case PMBUS_IOUT_OC_FAULT_LIMIT:
> + case PMBUS_IIN_OC_WARN_LIMIT:
> + case PMBUS_IIN_OC_FAULT_LIMIT:
> + case PMBUS_VOUT_UV_WARN_LIMIT:
> + case PMBUS_VOUT_UV_FAULT_LIMIT:
> + case PMBUS_VOUT_OV_WARN_LIMIT:
> + case PMBUS_VOUT_OV_FAULT_LIMIT:
> + case PMBUS_MFR_VIN_MIN:
> + case PMBUS_MFR_VIN_MAX:
> + case PMBUS_MFR_IIN_MAX:
> + case PMBUS_MFR_VOUT_MIN:
> + case PMBUS_MFR_VOUT_MAX:
> + case PMBUS_MFR_IOUT_MAX:
> + case PMBUS_MFR_PIN_MAX:
> + case PMBUS_POUT_MAX:
> + case PMBUS_POUT_OP_WARN_LIMIT:
> + case PMBUS_POUT_OP_FAULT_LIMIT:
> + case PMBUS_MFR_MAX_TEMP_1:
> + case PMBUS_MFR_MAX_TEMP_2:
> + case PMBUS_MFR_MAX_TEMP_3:
> + case PMBUS_MFR_POUT_MAX:
> + return -ENXIO;
> + }

If that many commands indeed cause trouble (ie cause the device
to get into a bad state), it might be better to list the _supported_
commands instead of the unsupported ones. There is no guarantee
that the core won't start to send other commands to the device
in the future.

The underlying question is if those commands are indeed not supported,
or if they report values in an unexpected format (ie not linear11).
The data format that is auto-selected below (because it is not specified)
is "linear". Is this what the device actually uses ? If not, just disabling
reading the limits without explanation what exactly "does not work" is
inappropriate.

> +
> + rv = set_page(client, page);
> + if (rv < 0)
> + return rv;
> +
> + return i2c_smbus_read_word_data(client, reg);
> +}
> +
> +struct pmbus_driver_info fsp3y_info[] = {
> + [ym2151e] = {
> + .pages = 0x21,
> + .func[YM2151_PAGE_12V] =
> + PMBUS_HAVE_VOUT | PMBUS_HAVE_IOUT |
> + PMBUS_HAVE_PIN | PMBUS_HAVE_POUT |
> + PMBUS_HAVE_TEMP | PMBUS_HAVE_TEMP2 |
> + PMBUS_HAVE_VIN | PMBUS_HAVE_IIN |
> + PMBUS_HAVE_FAN12,
> + .func[YM2151_PAGE_5V] =
> + PMBUS_HAVE_VOUT | PMBUS_HAVE_IOUT,
> + PMBUS_HAVE_IIN,

It doesn't really make sense to claim support for 0x21 = 33
pages, especially since the pmbus core (and the pmbus standard)
only supports 32 pages. Since page handling is all local anyway,
I would suggest to claim two pages and map the logical page
to the physical page in the set_page command.

How does this work (compile) anyway ? .func[] array size
is 32, meaning .func[0x20] goes beyond the end of the array.
The compiler should complain about that.

Wait, how does this even instantiate ? The PMBus core
should reject a page count larger than 32, and pmbus_do_probe()
should return -ENODEV. How did you test this code ?

> + .read_word_data = fsp3y_read_word_data,
> + .read_byte_data = fsp3y_read_byte_data,
> + },
> + [yh5151e] = {
> + .pages = 0x12,

Same as above.

> + .func[YH5151E_PAGE_12V] =
> + PMBUS_HAVE_VOUT | PMBUS_HAVE_IOUT |
> + PMBUS_HAVE_POUT |
> + PMBUS_HAVE_TEMP | PMBUS_HAVE_TEMP2 | PMBUS_HAVE_TEMP3,
> + .func[YH5151E_PAGE_5V] =
> + PMBUS_HAVE_VOUT | PMBUS_HAVE_IOUT |
> + PMBUS_HAVE_POUT,
> + .func[YH5151E_PAGE_3V3] =
> + PMBUS_HAVE_VOUT | PMBUS_HAVE_IOUT |
> + PMBUS_HAVE_POUT,
> + .read_word_data = fsp3y_read_word_data,
> + .read_byte_data = fsp3y_read_byte_data,
> + }
> +};
> +
> +static int fsp3y_probe(struct i2c_client *client,
> + const struct i2c_device_id *id)
> +{

This vendor sells dozens of different power supplies. Apparently
they do not have compatible PMBus attributes (or at least the pages
are not compatible to each other). Given that, I think there should
be a model validation here.

This is even more important since an earlier discussion suggests that
at least some of the 3Y power supplies use LINEAR11 instead of LINEAR16
for output voltages (eg YH5301-1EAR, FSP550-50ERS). We need to ensure
that affected power supplies are not enabled with this driver, and that
the enabled power supplies have been tested and are not only confirmed
to work and report correct data.

> + return pmbus_do_probe(client, &fsp3y_info[id->driver_data]);
> +}
> +
> +static const struct i2c_device_id pmbus_id[] = {
> + {"fsp3y_ym2151e", ym2151e},
> + {"fsp3y_yh5151e", yh5151e},> + {}
> +};
> +
> +MODULE_DEVICE_TABLE(i2c, pmbus_id);
> +
> +/* This is the driver that will be inserted */
> +static struct i2c_driver fsp3y_driver = {
> + .driver = {
> + .name = "fsp3y",
> + },
> + .probe = fsp3y_probe,

Please use the .probe_new callback.

> + .id_table = pmbus_id
> +};
> +
> +module_i2c_driver(fsp3y_driver);
> +
> +MODULE_AUTHOR("Václav Kubernát");
> +MODULE_DESCRIPTION("PMBus driver for FSP/3Y-Power power supplies");
> +MODULE_LICENSE("GPL");
>