From: Guenter Roeck <linux@roeck-us.net>
To: Eddie James <eajames@linux.vnet.ibm.com>, linux-kernel@vger.kernel.org
Cc: linux-hwmon@vger.kernel.org, devicetree@vger.kernel.org,
jdelvare@suse.com, mark.rutland@arm.com, robh+dt@kernel.org,
gregkh@linuxfoundation.org, cbostic@linux.vnet.ibm.com,
jk@ozlabs.org, joel@jms.id.au, andrew@aj.id.au,
"Edward A. James" <eajames@us.ibm.com>
Subject: Re: [PATCH 2/7] drivers/hwmon/occ: Add command transport method for P8 and P9
Date: Sat, 24 Jun 2017 07:49:40 -0700 [thread overview]
Message-ID: <660f0d44-907d-72b7-440c-b2bbec0f880d@roeck-us.net> (raw)
In-Reply-To: <1498171716-26620-3-git-send-email-eajames@linux.vnet.ibm.com>
On 06/22/2017 03:48 PM, Eddie James wrote:
> From: "Edward A. James" <eajames@us.ibm.com>
>
> For the P8 OCC, add the procedure to send a command to the OCC over I2C
> bus. This involves writing the OCC command registers with serial
> communication operations (SCOMs) interpreted by the I2C slave. For the
> P9 OCC, add a procedure to use the OCC in-kernel API to send a command
> to the OCC through the SBE engine.
>
> Signed-off-by: Edward A. James <eajames@us.ibm.com>
> ---
> drivers/hwmon/occ/common.h | 13 ++++
> drivers/hwmon/occ/p8_i2c.c | 166 ++++++++++++++++++++++++++++++++++++++++++++-
> drivers/hwmon/occ/p9_sbe.c | 66 +++++++++++++++++-
> 3 files changed, 243 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/hwmon/occ/common.h b/drivers/hwmon/occ/common.h
> index dca642a..0c3f26f 100644
> --- a/drivers/hwmon/occ/common.h
> +++ b/drivers/hwmon/occ/common.h
> @@ -15,6 +15,19 @@
>
> #define OCC_RESP_DATA_BYTES 4089
>
> +#define OCC_TIMEOUT_MS 5000
Five seconds ? Isn't that a bit excessive ?
> +#define OCC_CMD_IN_PRG_MS 100
> +
> +/* OCC return codes */
> +#define RESP_RETURN_CMD_IN_PRG 0xFF
> +#define RESP_RETURN_SUCCESS 0
> +#define RESP_RETURN_CMD_INVAL 0x11
> +#define RESP_RETURN_CMD_LEN 0x12
> +#define RESP_RETURN_DATA_INVAL 0x13
> +#define RESP_RETURN_CHKSUM 0x14
> +#define RESP_RETURN_OCC_ERR 0x15
> +#define RESP_RETURN_STATE 0x16
> +
Why are those return codes defined here ? They must be set somewhere
outside this driver, and I would expect them to be defined there.
Also see below - the missing context makes it all but impossible
th review this patch series.
> /* Same response format for all OCC versions.
> * Allocate the largest possible response.
> */
> diff --git a/drivers/hwmon/occ/p8_i2c.c b/drivers/hwmon/occ/p8_i2c.c
> index 5075146..d6d70ce 100644
> --- a/drivers/hwmon/occ/p8_i2c.c
> +++ b/drivers/hwmon/occ/p8_i2c.c
> @@ -7,6 +7,7 @@
> * (at your option) any later version.
> */
>
> +#include <asm/unaligned.h>
Alphabetic order, though it is ok to list linux.. includes first followed by
asm... includes followed by local includes.
> #include "common.h"
> #include <linux/i2c.h>
> #include <linux/init.h>
> @@ -19,9 +20,172 @@ struct p8_i2c_occ {
>
> #define to_p8_i2c_occ(x) container_of((x), struct p8_i2c_occ, occ)
>
> +static int p8_i2c_occ_getscom(struct i2c_client *client, u32 address, u8 *data)
> +{
> + ssize_t rc;
> + __be64 buf_be;
_be is redundant. Yes, you have buf as well, but that is really not needed.
> + u64 buf;
> + struct i2c_msg msgs[2];
> +
> + /* p8 i2c slave requires shift */
> + address <<= 1;
> +
> + msgs[0].addr = client->addr;
> + msgs[0].flags = client->flags & I2C_M_TEN;
> + msgs[0].len = sizeof(u32);
> + msgs[0].buf = (char *)&address;
No endianness concerns here ?
> +
> +
checkpatch --strict, please
> + msgs[1].addr = client->addr;
> + msgs[1].flags = (client->flags & I2C_M_TEN) | I2C_M_RD;
> + msgs[1].len = sizeof(u64);
> + msgs[1].buf = (char *)&buf_be;
> +
> + rc = i2c_transfer(client->adapter, msgs, 2);
> + if (rc < 0)
> + return rc;
> +
> + buf = be64_to_cpu(buf_be);
> + memcpy(data, &buf, sizeof(u64));
*(u64 *)data = be64_to_cpu(buf_be);
> +
> + return 0;
> +}
> +
> +static int p8_i2c_occ_putscom(struct i2c_client *client, u32 address, u8 *data)
> +{
> + u32 buf[3];
> + ssize_t rc;
> +
> + /* p8 i2c slave requires shift */
> + address <<= 1;
> +
> + buf[0] = address;
> + memcpy(&buf[1], &data[4], sizeof(u32));
> + memcpy(&buf[2], data, sizeof(u32));
No endianness concerns ? Presumably not, but that might warrant a comment
above, with the function declaration.
> +
> + rc = i2c_master_send(client, (const char *)buf, sizeof(buf));
> + if (rc < 0)
> + return rc;
> + else if (rc != sizeof(buf))
> + return -EIO;
> +
> + return 0;
> +}
> +
> +static int p8_i2c_occ_putscom_u32(struct i2c_client *client, u32 address,
> + u32 data0, u32 data1)
> +{
> + u8 buf[8];
> +
> + memcpy(buf, &data0, 4);
> + memcpy(buf + 4, &data1, 4);
> +
> + return p8_i2c_occ_putscom(client, address, buf);
> +}
> +
> +static int p8_i2c_occ_putscom_be(struct i2c_client *client, u32 address,
> + u8 *data)
> +{
> + __be32 data0, data1;
> +
> + memcpy(&data0, data, 4);
> + memcpy(&data1, data + 4, 4);
> +
> + return p8_i2c_occ_putscom_u32(client, address, be32_to_cpu(data0),
> + be32_to_cpu(data1));
> +}
> +
> static int p8_i2c_occ_send_cmd(struct occ *occ, u8 *cmd)
> {
> - return -EOPNOTSUPP;
> + int i, rc;
> + unsigned long start;
> + u16 data_length;
> + struct p8_i2c_occ *p8_i2c_occ = to_p8_i2c_occ(occ);
> + struct i2c_client *client = p8_i2c_occ->client;
> + struct occ_response *resp = &occ->resp;
> +
> + start = jiffies;
> +
> + /* set sram address for command */
> + rc = p8_i2c_occ_putscom_u32(client, 0x6B070, 0xFFFF6000, 0);
Please declare those magics as defines to give readers an idea
what this is about.
> + if (rc)
> + goto err;
> +
> + /* write command (must already be BE), i2c expects cpu-endian */
> + rc = p8_i2c_occ_putscom_be(client, 0x6B075, cmd);
> + if (rc)
> + goto err;
> +
> + /* trigger OCC attention */
> + rc = p8_i2c_occ_putscom_u32(client, 0x6B035, 0x20010000, 0);
> + if (rc)
> + goto err;
> +
> +retry:
> + /* set sram address for response */
> + rc = p8_i2c_occ_putscom_u32(client, 0x6B070, 0xFFFF7000, 0);
> + if (rc)
> + goto err;
> +
> + /* read response */
> + rc = p8_i2c_occ_getscom(client, 0x6B075, (u8 *)resp);
> + if (rc)
> + goto err;
> +
> + /* check the OCC response */
> + switch (resp->return_status) {
> + case RESP_RETURN_CMD_IN_PRG:
> + if (time_after(jiffies,
> + start + msecs_to_jiffies(OCC_TIMEOUT_MS)))
> + rc = -EALREADY;
> + else {
> + set_current_state(TASK_INTERRUPTIBLE);
> + schedule_timeout(msecs_to_jiffies(OCC_CMD_IN_PRG_MS));
> +
> + goto retry;
Please refactor as loop.
> + }
> + break;
> + case RESP_RETURN_SUCCESS:
> + rc = 0;
> + break;
> + case RESP_RETURN_CMD_INVAL:
> + case RESP_RETURN_CMD_LEN:
> + case RESP_RETURN_DATA_INVAL:
> + case RESP_RETURN_CHKSUM:
> + rc = -EINVAL;
> + break;
> + case RESP_RETURN_OCC_ERR:
> + rc = -EREMOTE;
"Object is remote"
> + break;
> + default:
> + rc = -EFAULT;
"Bad address" ?
> + }
> +
> + if (rc < 0) {
> + dev_warn(&client->dev, "occ bad response: %d\n",
> + resp->return_status);
> + return rc;
> + }
> +
> + data_length = get_unaligned_be16(&resp->data_length_be);
> + if (data_length > OCC_RESP_DATA_BYTES) {
> + dev_warn(&client->dev, "occ bad data length: %d\n",
> + data_length);
> + return -EDOM;
"Math argument out of domain" ?
Your error codes seem to be kind of random.
> + }
> +
> + /* read remaining response */
> + for (i = 8; i < data_length + 7; i += 8) {
> + rc = p8_i2c_occ_getscom(client, 0x6B075, ((u8 *)resp) + i);
> + if (rc)
> + goto err;
> + }
> +
> + return data_length + 7;
> +
> +err:
> + dev_err(&client->dev, "i2c scom op failed rc: %d\n", rc);
> + return rc;
> }
>
> static int p8_i2c_occ_probe(struct i2c_client *client,
> diff --git a/drivers/hwmon/occ/p9_sbe.c b/drivers/hwmon/occ/p9_sbe.c
> index 0cef428..981c53f 100644
> --- a/drivers/hwmon/occ/p9_sbe.c
> +++ b/drivers/hwmon/occ/p9_sbe.c
> @@ -22,7 +22,71 @@ struct p9_sbe_occ {
>
> static int p9_sbe_occ_send_cmd(struct occ *occ, u8 *cmd)
> {
> - return -EOPNOTSUPP;
> + int rc;
> + unsigned long start;
> + struct occ_client *client;
> + struct occ_response *resp = &occ->resp;
> + struct p9_sbe_occ *p9_sbe_occ = to_p9_sbe_occ(occ);
> +
> + start = jiffies;
> +
> +retry:
> + client = occ_drv_open(p9_sbe_occ->sbe, 0);
Where does this come from ? Ah, I see, there is an "include <linux/occ.h>",
but no mention where it comes from.
I'll stop reviewing the series here. I can not review it without complete context.
> + if (!client)
> + return -ENODEV;
> +
> + /* skip first byte (sequence number), OCC driver handles it */
> + rc = occ_drv_write(client, (const char *)&cmd[1], 7);
> + if (rc < 0)
> + goto err;
> +
> + rc = occ_drv_read(client, (char *)resp, sizeof(*resp));
> + if (rc < 0)
> + goto err;
> +
> + occ_drv_release(client);
Might as well cann release() first to save the double call below.
> +
> + /* check the OCC response */
> + switch (resp->return_status) {
> + case RESP_RETURN_CMD_IN_PRG:
> + if (time_after(jiffies,
> + start + msecs_to_jiffies(OCC_TIMEOUT_MS)))
> + rc = -EALREADY;
This is another odd return value. Normally one would expect something like
-ETIMEDOUT.
> + else {
> + set_current_state(TASK_INTERRUPTIBLE);
> + schedule_timeout(msecs_to_jiffies(OCC_CMD_IN_PRG_MS));
> +
Does the called code generate an interrupt if the command is complete ?
Makes me wonder why the write/read sequence isn't just synchronous.
Again, missing context.
> + goto retry;
Please refactor as loop.
> + }
> + break;
> + case RESP_RETURN_SUCCESS:
> + rc = 0;
> + break;
> + case RESP_RETURN_CMD_INVAL:
> + case RESP_RETURN_CMD_LEN:
> + case RESP_RETURN_DATA_INVAL:
> + case RESP_RETURN_CHKSUM:
> + rc = -EINVAL;
> + break;
> + case RESP_RETURN_OCC_ERR:
> + rc = -EREMOTE;
> + break;
> + default:
> + rc = -EFAULT;
> + }
> +
> + if (rc < 0) {
> + dev_warn(occ->bus_dev, "occ bad response: %d\n",
> + resp->return_status);
> + return rc;
> + }
> +
> + return 0;
> +
> +err:
> + occ_drv_release(client);
> + dev_err(occ->bus_dev, "occ bus op failed rc: %d\n", rc);
Is all this noise really needed ? I can understand it for debugging,
but for released code ?
> + return rc;
> }
>
> static int p9_sbe_occ_probe(struct platform_device *pdev)
>
next prev parent reply other threads:[~2017-06-24 14:49 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-06-22 22:48 [PATCH 0/7] drivers/hwmon: Add On-Chip Controller (OCC) hwmon driver Eddie James
2017-06-22 22:48 ` [PATCH 1/7] " Eddie James
2017-06-24 14:15 ` Guenter Roeck
2017-06-26 15:06 ` Eddie James
2017-06-26 16:38 ` Guenter Roeck
2017-06-26 19:01 ` Rob Herring
2017-06-22 22:48 ` [PATCH 2/7] drivers/hwmon/occ: Add command transport method for P8 and P9 Eddie James
2017-06-24 14:49 ` Guenter Roeck [this message]
2017-06-26 15:30 ` Eddie James
2017-06-26 16:55 ` Guenter Roeck
2017-06-22 22:48 ` [PATCH 3/7] drivers/hwmon/occ: Parse OCC poll response Eddie James
2017-06-22 22:48 ` [PATCH 4/7] drivers/hwmon/occ: Add sensor types and versions Eddie James
2017-06-22 22:48 ` [PATCH 5/7] drivers/hwmon/occ: Add sensor attributes and register hwmon device Eddie James
2017-06-22 22:48 ` [PATCH 6/7] drivers/hwmon/occ: Add non-hwmon attributes Eddie James
2017-06-22 22:48 ` [PATCH 7/7] drivers/hwmon/occ: Add error handling Eddie James
2017-06-23 4:52 ` [PATCH 0/7] drivers/hwmon: Add On-Chip Controller (OCC) hwmon driver Guenter Roeck
2017-06-23 13:38 ` Eddie James
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=660f0d44-907d-72b7-440c-b2bbec0f880d@roeck-us.net \
--to=linux@roeck-us.net \
--cc=andrew@aj.id.au \
--cc=cbostic@linux.vnet.ibm.com \
--cc=devicetree@vger.kernel.org \
--cc=eajames@linux.vnet.ibm.com \
--cc=eajames@us.ibm.com \
--cc=gregkh@linuxfoundation.org \
--cc=jdelvare@suse.com \
--cc=jk@ozlabs.org \
--cc=joel@jms.id.au \
--cc=linux-hwmon@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=mark.rutland@arm.com \
--cc=robh+dt@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).