Re: [PATCH v9 5/6] fpga: dfl: add support for N3000 Nios private feature

From: Xu Yilun
Date: Mon Oct 12 2020 - 02:57:06 EST

Next message: Xing Zhengjun: "Re: [LKP] [mm] 5ef64cc898: vm-scalability.throughput -20.6% regression"
Previous message: Ira Weiny: "Re: [PATCH RFC PKS/PMEM 22/58] fs/f2fs: Utilize new kmap_thread()"
In reply to: Tom Rix: "Re: [PATCH v9 5/6] fpga: dfl: add support for N3000 Nios private feature"
Next in thread: Tom Rix: "Re: [PATCH v9 5/6] fpga: dfl: add support for N3000 Nios private feature"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Sat, Oct 10, 2020 at 02:21:45PM -0700, Tom Rix wrote:
>
> On 10/10/20 12:09 AM, Xu Yilun wrote:
> > This patch adds support for the Nios handshake private feature on Intel
> > PAC (Programmable Acceleration Card) N3000.
> >
> > The Nios is the embedded processor on the FPGA card. This private feature
> > provides a handshake interface to FPGA Nios firmware, which receives
> > retimer configuration command from host and executes via an internal SPI
> > master (spi-altera). When Nios finishes the configuration, host takes over
> > the ownership of the SPI master to control an Intel MAX10 BMC (Board
> > Management Controller) Chip on the SPI bus.
> >
> > For Nios firmware handshake part, this driver requests the retimer
> > configuration for Nios firmware on probe, and adds some sysfs nodes for
> > user to query the onboard retimer's working mode and Nios firmware
> > version.
> >
> > For SPI part, this driver adds a spi-altera platform device as well as
> > the MAX10 BMC spi slave info. A spi-altera driver will be matched to
> > handle the following SPI work.
> >
> > [mdf@xxxxxxxxxx: Fixed up MAINTAINERS file to include added ABI doc]
> > Signed-off-by: Xu Yilun <yilun.xu@xxxxxxxxx>
> > Signed-off-by: Wu Hao <hao.wu@xxxxxxxxx>
> > Signed-off-by: Matthew Gerlach <matthew.gerlach@xxxxxxxxxxxxxxx>
> > Signed-off-by: Russ Weight <russell.h.weight@xxxxxxxxx>
> > ---
> > v3: add the doc for this driver
> > minor fixes for comments from Tom
> > v4: move the err log in regmap implementation, and delete
> > n3000_nios_writel/readl(), they have nothing to wrapper now.
> > some minor fixes and comments improvement.
> > v5: fix the output of fec_mode sysfs inf to "no" on 10G configuration,
> > cause no FEC mode could be configured for 10G.
> > rename the dfl_n3000_nios_* to n3000_nios_*
> > improves comments.
> > v6: fix the output of fec_mode sysfs inf to "not supported" if in 10G,
> > or the firmware version major < 3.
> > minor fixes and improves comments.
> > v7: improves comments.
> > v8: add sysfs interfaces for retimer mode, also doc update.
> > delete duplicated sysfs interfaces description in doc.
> > minor fixes.
> > v9: delete the retimer FEC mode configuration via module_parameter.
> > update the kernel version of the sysfs interfaces in Doc.
> > merge the patch "Make m10_n3000_info static" (https://lore.kernel.org/linux-fpga/52d8411e-13d8-1e91-756d-131802f5f445@xxxxxxxxxx/T/#t)
> > remove the tags of Maintainers, this implementation is changed.
> > ---
> > .../ABI/testing/sysfs-bus-dfl-devices-n3000-nios | 47 ++
> > MAINTAINERS | 2 +-
> > drivers/fpga/Kconfig | 11 +
> > drivers/fpga/Makefile | 2 +
> > drivers/fpga/dfl-n3000-nios.c | 566 +++++++++++++++++++++
> > 5 files changed, 627 insertions(+), 1 deletion(-)
> > create mode 100644 Documentation/ABI/testing/sysfs-bus-dfl-devices-n3000-nios
> > create mode 100644 drivers/fpga/dfl-n3000-nios.c
> >
> > diff --git a/Documentation/ABI/testing/sysfs-bus-dfl-devices-n3000-nios b/Documentation/ABI/testing/sysfs-bus-dfl-devices-n3000-nios
> > new file mode 100644
> > index 0000000..f2f6e5e
> > --- /dev/null
> > +++ b/Documentation/ABI/testing/sysfs-bus-dfl-devices-n3000-nios
> > @@ -0,0 +1,47 @@
> > +What: /sys/bus/dfl/devices/dfl_dev.X/fec_mode
> > +Date: Oct 2020
> > +KernelVersion: 5.11
> > +Contact: Xu Yilun <yilun.xu@xxxxxxxxx>
> > +Description: Read-only. It returns the FEC mode of the 25G links of the
> Could drop the 'It' and start with 'Returns' here and similar

Yes, I can change it.

> > + ethernet retimers configured by Nios firmware. "rs" for Reed
> > + Solomon FEC, "kr" for Fire Code FEC, "no" for NO FEC.
> > + "not supported" if the FEC mode setting is not supported, this
> setting ? the description says 'Read-only'

The setting is done by Nios firmware, host cannot change it at runtime
so it is 'Read-only'. It is previously described in this section. Is it
good?

> > + happens when the Nios firmware version major < 3, or no link is
> > + configured to 25G.
> > + Format: string
> > +
> > +What: /sys/bus/dfl/devices/dfl_dev.X/retimer_A_mode
> > +Date: Oct 2020
> > +KernelVersion: 5.11
> > +Contact: Xu Yilun <yilun.xu@xxxxxxxxx>
> > +Description: Read-only. It returns the enumeration value of the working mode
> > + of the retimer A configured by the Nios firmware. The value is
> > + read out from shared registers filled by the Nios firmware. Now
> > + the values could be:
> > +
> > + - "0": Reset
> > + - "1": 4x10G
> > + - "2": 4x25G
> > + - "3": 2x25G
> > + - "4": 2x25G+2x10G
> > + - "5": 1x25G
> > +
> > + If the Nios firmware is updated in future to support more
> > + retimer modes, more enumeration value is expected.
> > + Format: 0x%x
> > +
> > +What: /sys/bus/dfl/devices/dfl_dev.X/retimer_B_mode
> > +Date: Oct 2020
> > +KernelVersion: 5.11
> > +Contact: Xu Yilun <yilun.xu@xxxxxxxxx>
> > +Description: Read-only. It returns the enumeration value of the working mode
> > + of the retimer B configured by the Nios firmware. The value
> > + format is the same as retimer_A_mode.
> > +
> > +What: /sys/bus/dfl/devices/dfl_dev.X/nios_fw_version
> > +Date: Oct 2020
> > +KernelVersion: 5.11
> > +Contact: Xu Yilun <yilun.xu@xxxxxxxxx>
> > +Description: Read-only. It returns the version of the Nios firmware in FPGA.
> Can drop 'in FPGA'.

The PAC N3000 card has 2 Nios firmwares. The one for this driver is
embedded in A10 FPGA. Another one is embedded in MAX10 BMC chip.

So to be clear, I think we may keep the 'in FPGA'.

> > + Its format is "major.minor.patch".
> > + Format: %x.%x.%x
> > diff --git a/MAINTAINERS b/MAINTAINERS
> > index 9d46097..3a81cac 100644
> > --- a/MAINTAINERS
> > +++ b/MAINTAINERS
> > @@ -6881,7 +6881,7 @@ M: Wu Hao <hao.wu@xxxxxxxxx>
> > R: Tom Rix <trix@xxxxxxxxxx>
> > L: linux-fpga@xxxxxxxxxxxxxxx
> > S: Maintained
> > -F: Documentation/ABI/testing/sysfs-bus-dfl
> > +F: Documentation/ABI/testing/sysfs-bus-dfl*
> > F: Documentation/fpga/dfl.rst
> > F: drivers/fpga/dfl*
> > F: include/linux/fpga/dfl.h
> > diff --git a/drivers/fpga/Kconfig b/drivers/fpga/Kconfig
> > index 7cd5a29..38c7130 100644
> > --- a/drivers/fpga/Kconfig
> > +++ b/drivers/fpga/Kconfig
> > @@ -191,6 +191,17 @@ config FPGA_DFL_AFU
> > to the FPGA infrastructure via a Port. There may be more than one
> > Port/AFU per DFL based FPGA device.
> >
> > +config FPGA_DFL_NIOS_INTEL_PAC_N3000
> > + tristate "FPGA DFL NIOS Driver for Intel PAC N3000"
> > + depends on FPGA_DFL
> Would 'depends on FPGA_DFL_PCI' be better ?

The relationship between the 2 config is, the n3000 nios private feature
now only appears on PAC N3000 which is a PCIe card.

But the code has no dependency to FPGA_DFL_PCI. Is it necessary we add the
'depends on'?

> > + select REGMAP
> > + help
> Last 3 lines have whitespace issues, need to convert spaces to tabs

Thanks for catching this.

> > + This is the driver for the N3000 Nios private feature on Intel
> > + PAC (Programmable Acceleration Card) N3000. It communicates
> > + with the embedded Nios processor to configure the retimers on
> > + the card. It also instantiates the SPI master (spi-altera) for
> > + the card's BMC (Board Management Controller) control.
>
> can drop 'control'
>
> If you read out 'card's board mangement controller control' it sounds awkward.

It's true. I'll drop it.

>
> > +
> > config FPGA_DFL_PCI
> > tristate "FPGA DFL PCIe Device Driver"
> > depends on PCI && FPGA_DFL
> > diff --git a/drivers/fpga/Makefile b/drivers/fpga/Makefile
> > index d8e21df..18dc9885 100644
> > --- a/drivers/fpga/Makefile
> > +++ b/drivers/fpga/Makefile
> > @@ -44,5 +44,7 @@ dfl-fme-objs += dfl-fme-perf.o
> > dfl-afu-objs := dfl-afu-main.o dfl-afu-region.o dfl-afu-dma-region.o
> > dfl-afu-objs += dfl-afu-error.o
> >
> > +obj-$(CONFIG_FPGA_DFL_NIOS_INTEL_PAC_N3000) += dfl-n3000-nios.o
> > +
> > # Drivers for FPGAs which implement DFL
> > obj-$(CONFIG_FPGA_DFL_PCI) += dfl-pci.o
> > diff --git a/drivers/fpga/dfl-n3000-nios.c b/drivers/fpga/dfl-n3000-nios.c
> > new file mode 100644
> > index 0000000..e7c3862
> > --- /dev/null
> > +++ b/drivers/fpga/dfl-n3000-nios.c
> > @@ -0,0 +1,566 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * DFL device driver for Nios private feature on Intel PAC (Programmable
> > + * Acceleration Card) N3000
> > + *
> > + * Copyright (C) 2019-2020 Intel Corporation, Inc.
> > + *
> > + * Authors:
> > + * Wu Hao <hao.wu@xxxxxxxxx>
> > + * Xu Yilun <yilun.xu@xxxxxxxxx>
> > + */
> > +#include <linux/bitfield.h>
> > +#include <linux/errno.h>
> > +#include <linux/fpga/dfl.h>
> > +#include <linux/io.h>
> > +#include <linux/io-64-nonatomic-lo-hi.h>
> > +#include <linux/kernel.h>
> > +#include <linux/module.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/regmap.h>
> > +#include <linux/stddef.h>
> > +#include <linux/spi/altera.h>
> > +#include <linux/spi/spi.h>
> > +#include <linux/types.h>
> > +
> > +/* N3000 Nios private feature registers */
> > +#define NIOS_SPI_PARAM 0x8
>
> The #defines should have a consistent prefix that is unlikely to conflict.
>
> Maybe N3000_NIOS_

Yes, I'll try to refactor the macro names.
The register names in spec are NIOS_SPI_xx, how about we compress the prefixes,
'N3000_NS' & 'n3000_ns' for 'N3000 NIOS SPI'

>
> > +#define PARAM_SHIFT_MODE_MSK BIT_ULL(1)
> > +#define PARAM_SHIFT_MODE_MSB 0
> > +#define PARAM_SHIFT_MODE_LSB 1
> > +#define PARAM_DATA_WIDTH GENMASK_ULL(7, 2)
> > +#define PARAM_NUM_CS GENMASK_ULL(13, 8)
> > +#define PARAM_CLK_POL BIT_ULL(14)
> > +#define PARAM_CLK_PHASE BIT_ULL(15)
> > +#define PARAM_PERIPHERAL_ID GENMASK_ULL(47, 32)
> > +
> > +#define NIOS_SPI_CTRL 0x10
> > +#define CTRL_WR_DATA GENMASK_ULL(31, 0)
> > +#define CTRL_ADDR GENMASK_ULL(44, 32)
> > +#define CTRL_CMD_MSK GENMASK_ULL(63, 62)
> > +#define CTRL_CMD_NOP 0
> > +#define CTRL_CMD_RD 1
> > +#define CTRL_CMD_WR 2
> > +
> > +#define NIOS_SPI_STAT 0x18
> > +#define STAT_RD_DATA GENMASK_ULL(31, 0)
> > +#define STAT_RW_VAL BIT_ULL(32)
> > +
> > +/* Nios handshake registers, indirect access */
> > +#define NIOS_INIT 0x1000
> > +#define NIOS_INIT_DONE BIT(0)
> > +#define NIOS_INIT_START BIT(1)
> > +/* Mode for retimer A, link 0, the same below */
> > +#define REQ_FEC_MODE_A0_MSK GENMASK(9, 8)
> > +#define REQ_FEC_MODE_A1_MSK GENMASK(11, 10)
> > +#define REQ_FEC_MODE_A2_MSK GENMASK(13, 12)
> > +#define REQ_FEC_MODE_A3_MSK GENMASK(15, 14)
> > +#define REQ_FEC_MODE_B0_MSK GENMASK(17, 16)
> > +#define REQ_FEC_MODE_B1_MSK GENMASK(19, 18)
> > +#define REQ_FEC_MODE_B2_MSK GENMASK(21, 20)
> > +#define REQ_FEC_MODE_B3_MSK GENMASK(23, 22)
> > +#define REQ_FEC_MODE_NO 0x0
> > +#define REQ_FEC_MODE_KR 0x1
> > +#define REQ_FEC_MODE_RS 0x2
> > +
> > +#define NIOS_FW_VERSION 0x1004
> > +#define NIOS_FW_VERSION_PATCH GENMASK(23, 20)
> > +#define NIOS_FW_VERSION_MINOR GENMASK(27, 24)
> > +#define NIOS_FW_VERSION_MAJOR GENMASK(31, 28)
> > +
> > +/* The retimers we use on Intel PAC N3000 is Parkvale, abbreviated to PKVL */
> Parkvale is codename, why not just RETIMER_ ?

The register fields names in Hardware spec is PKVL_, so I mean to keep
align with the spec.

> > +#define PKVL_A_MODE_STS 0x1020
> > +#define PKVL_B_MODE_STS 0x1024
> > +#define PKVL_MODE_STS_GROUP_MSK GENMASK(15, 8)
> > +#define PKVL_MODE_STS_GROUP_OK 0x0
> > +#define PKVL_MODE_STS_ID_MSK GENMASK(7, 0)
> > +/* When GROUP MASK field == GROUP_OK */
> > +#define PKVL_MODE_ID_RESET 0x0
> > +#define PKVL_MODE_ID_4X10G 0x1
> > +#define PKVL_MODE_ID_4X25G 0x2
> > +#define PKVL_MODE_ID_2X25G 0x3
> > +#define PKVL_MODE_ID_2X25G_2X10G 0x4
> > +#define PKVL_MODE_ID_1X25G 0x5
> > +
> > +#define NS_REGBUS_WAIT_TIMEOUT 10000 /* loop count */
>
> NS ? Expand.

I could explain the NS - NIOS SPI in comments, and is N3000_NS_REGBUS_RETRY_COUNT good?

>
> 'WAIT_TIMEOUT' should be changed to 'RETRY_COUNT' to reflect its use

It makes sense.

>
> > +#define NIOS_INIT_TIMEOUT 10000000 /* usec */
> > +#define NIOS_INIT_TIME_INTV 100000 /* usec */
> > +
> > +struct n3000_nios {
> > + void __iomem *base;
> > + struct regmap *regmap;
> > + struct device *dev;
> > + struct platform_device *altera_spi;
> > +};
> > +
> > +static ssize_t nios_fw_version_show(struct device *dev,
> > + struct device_attribute *attr, char *buf)
> > +{
> > + struct n3000_nios *ns = dev_get_drvdata(dev);
> > + unsigned int val;
> > + int ret;
> > +
> > + ret = regmap_read(ns->regmap, NIOS_FW_VERSION, &val);
> > + if (ret)
> > + return ret;
> > +
> > + return sprintf(buf, "%x.%x.%x\n",
> > + (u8)FIELD_GET(NIOS_FW_VERSION_MAJOR, val),
> > + (u8)FIELD_GET(NIOS_FW_VERSION_MINOR, val),
> > + (u8)FIELD_GET(NIOS_FW_VERSION_PATCH, val));
> > +}
> > +static DEVICE_ATTR_RO(nios_fw_version);
> > +
> > +#define IS_MODE_STATUS_OK(mode_stat) \
> > + (FIELD_GET(PKVL_MODE_STS_GROUP_MSK, (mode_stat)) == \
> > + PKVL_MODE_STS_GROUP_OK)
> > +
> > +#define IS_RETIMER_FEC_SUPPORTED(retimer_mode) \
> > + ((retimer_mode) != PKVL_MODE_ID_RESET && \
> > + (retimer_mode) != PKVL_MODE_ID_4X10G)
> > +
> > +static int get_retimer_mode(struct n3000_nios *ns, unsigned int mode_stat_reg,
> > + unsigned int *retimer_mode)
> > +{
> > + unsigned int val;
> > + int ret;
> > +
> > + ret = regmap_read(ns->regmap, mode_stat_reg, &val);
> > + if (ret)
> > + return ret;
> > +
> > + if (!IS_MODE_STATUS_OK(val))
> > + return -EFAULT;
> > +
> > + *retimer_mode = FIELD_GET(PKVL_MODE_STS_ID_MSK, val);
> > +
> > + return 0;
> > +}
> > +
> > +static ssize_t retimer_A_mode_show(struct device *dev,
> > + struct device_attribute *attr, char *buf)
> > +{
> > + struct n3000_nios *ns = dev_get_drvdata(dev);
> > + unsigned int mode;
> > + int ret;
> > +
> > + ret = get_retimer_mode(ns, PKVL_A_MODE_STS, &mode);
> > + if (ret)
> > + return ret;
> > +
> > + return sprintf(buf, "0x%x\n", mode);
> > +}
> > +static DEVICE_ATTR_RO(retimer_A_mode);
> > +
> > +static ssize_t retimer_B_mode_show(struct device *dev,
> > + struct device_attribute *attr, char *buf)
> > +{
> > + struct n3000_nios *ns = dev_get_drvdata(dev);
> > + unsigned int mode;
> > + int ret;
> > +
> > + ret = get_retimer_mode(ns, PKVL_B_MODE_STS, &mode);
> > + if (ret)
> > + return ret;
> > +
> > + return sprintf(buf, "0x%x\n", mode);
> > +}
> > +static DEVICE_ATTR_RO(retimer_B_mode);
> > +
> > +static ssize_t fec_mode_show(struct device *dev,
> > + struct device_attribute *attr, char *buf)
> > +{
> > + unsigned int val, retimer_a_mode, retimer_b_mode, fec_mode;
> > + struct n3000_nios *ns = dev_get_drvdata(dev);
> > + int ret;
> > +
> > + /* FEC mode setting is not supported in early FW versions */
> > + ret = regmap_read(ns->regmap, NIOS_FW_VERSION, &val);
> > + if (ret)
> > + return ret;
> > +
> > + if (FIELD_GET(NIOS_FW_VERSION_MAJOR, val) < 3)
> > + return sprintf(buf, "not supported\n");
> > +
> > + /* If no 25G links, FEC mode setting is not supported either */
> > + ret = get_retimer_mode(ns, PKVL_A_MODE_STS, &retimer_a_mode);
> > + if (ret)
> > + return ret;
> > +
> > + ret = get_retimer_mode(ns, PKVL_B_MODE_STS, &retimer_b_mode);
> > + if (ret)
> > + return ret;
> > +
> > + if (!IS_RETIMER_FEC_SUPPORTED(retimer_a_mode) &&
> > + !IS_RETIMER_FEC_SUPPORTED(retimer_b_mode))
> > + return sprintf(buf, "not supported\n");
> > +
> > + /* get the valid FEC mode for 25G links */
> > + ret = regmap_read(ns->regmap, NIOS_INIT, &val);
> > + if (ret)
> > + return ret;
> > +
> > + /*
> > + * FEC mode should always be the same for all links, as we set them
> > + * in this way.
> > + */
> > + fec_mode = FIELD_GET(REQ_FEC_MODE_A0_MSK, val);
> > + if (fec_mode != FIELD_GET(REQ_FEC_MODE_A1_MSK, val) ||
> > + fec_mode != FIELD_GET(REQ_FEC_MODE_A2_MSK, val) ||
> > + fec_mode != FIELD_GET(REQ_FEC_MODE_A3_MSK, val) ||
> > + fec_mode != FIELD_GET(REQ_FEC_MODE_B0_MSK, val) ||
> > + fec_mode != FIELD_GET(REQ_FEC_MODE_B1_MSK, val) ||
> > + fec_mode != FIELD_GET(REQ_FEC_MODE_B2_MSK, val) ||
> > + fec_mode != FIELD_GET(REQ_FEC_MODE_B3_MSK, val))
> > + return -EFAULT;
>
> a precomputed table setting the fields would reduce the this to a lookup and 1 check.
>
> similar table could be used below for the setting.

Yes, I can do it.

>
> > +
> > + switch (fec_mode) {
> > + case REQ_FEC_MODE_NO:
> > + return sprintf(buf, "no\n");
> > + case REQ_FEC_MODE_KR:
> > + return sprintf(buf, "kr\n");
> > + case REQ_FEC_MODE_RS:
> > + return sprintf(buf, "rs\n");
> > + }
> > +
> > + return -EFAULT;
> > +}
> > +static DEVICE_ATTR_RO(fec_mode);
> > +
> > +static struct attribute *n3000_nios_attrs[] = {
> > + &dev_attr_nios_fw_version.attr,
> > + &dev_attr_retimer_A_mode.attr,
> > + &dev_attr_retimer_B_mode.attr,
> > + &dev_attr_fec_mode.attr,
> > + NULL,
> > +};
> > +ATTRIBUTE_GROUPS(n3000_nios);
> > +
> > +static bool init_error_detected(struct n3000_nios *ns)
> > +{
> > + unsigned int val;
> > +
> > + if (regmap_read(ns->regmap, PKVL_A_MODE_STS, &val))
> > + return true;
> > +
> > + if (!IS_MODE_STATUS_OK(val))
> > + return true;
> > +
> > + if (regmap_read(ns->regmap, PKVL_B_MODE_STS, &val))
> > + return true;
> > +
> > + if (!IS_MODE_STATUS_OK(val))
> > + return true;
> > +
> > + return false;
> > +}
> > +
> > +static void dump_error_stat(struct n3000_nios *ns)
> > +{
> > + unsigned int val;
> > +
> > + if (regmap_read(ns->regmap, PKVL_A_MODE_STS, &val))
> > + return;
> > +
> > + dev_err(ns->dev, "PKVL_A_MODE_STS 0x%x\n", val);
> > +
> > + if (regmap_read(ns->regmap, PKVL_B_MODE_STS, &val))
> > + return;
> > +
> > + dev_err(ns->dev, "PKVL_B_MODE_STS 0x%x\n", val);
> > +}
> > +
> > +static int n3000_nios_init_done_check(struct n3000_nios *ns)
> > +{
> > + struct device *dev = ns->dev;
> > + unsigned int val, mode;
> > + int ret;
> > +
> > + /*
> > + * this SPI is shared by Nios core inside FPGA, Nios will use this SPI
> 'by the Nios core inside the FPGA ... '

Yes.

> > + * master to do some one time initialization after power up, and then
> > + * release the control to OS. driver needs to poll on INIT_DONE to
> 'The driver..'

Yes.

> > + * see when driver could take the control.
> > + *
> > + * Please note that after Nios firmware version 3.0.0, INIT_START is
> > + * introduced, so driver needs to trigger START firstly and then check
> > + * INIT_DONE.
> > + */
> > +
> > + ret = regmap_read(ns->regmap, NIOS_FW_VERSION, &val);
> > + if (ret)
> > + return ret;
> > +
> > + /*
> > + * If Nios version register is totally uninitialized(== 0x0), then the
> > + * Nios firmware is missing. So host could take control of SPI master
> > + * safely, but initialization work for Nios is not done. To restore the
> > + * card, we need to reprogram a new Nios firmware via the BMC chip on
> > + * SPI bus. So the driver doesn't error out, it continues to create the
> > + * spi controller device and spi_board_info for BMC.
> > + */
> > + if (val == 0) {
> > + dev_err(dev, "Nios version reg = 0x%x, skip INIT_DONE check, but the retimer may be uninitialized\n",
> > + val);
> > + return 0;
> > + }
> > +
> > + if (FIELD_GET(NIOS_FW_VERSION_MAJOR, val) >= 3) {
> > + /* read NIOS_INIT to check if retimer initialization is done */
> > + ret = regmap_read(ns->regmap, NIOS_INIT, &val);
> > + if (ret)
> > + return ret;
> > +
> > + /* check if retimers are initialized already */
> > + if (val & NIOS_INIT_DONE || val & NIOS_INIT_START)
> val & (NIOS_INIT_DONE | NIOS_INIT_START)

Yes.

> > + goto nios_init_done;
> > +
> > + /* configure FEC mode per module param */
> > + val = NIOS_INIT_START;
> > +
> > + /*
> > + * When the retimer is to be set to 10G mode, there is no FEC
> > + * mode setting, so the REQ_FEC_MODE field will be ignored by
> > + * Nios firmware in this case. But we should still fill the FEC
> > + * mode field cause host could not get the retimer working mode
> > + * until the Nios init is done.
> > + *
> > + * For now the driver doesn't support the retimer FEC mode
> > + * switching per user's request. It is always set to Reed
> > + * Solomon FEC.
> > + */
> > + mode = REQ_FEC_MODE_RS;
> > +
> > + /* set the same FEC mode for all links */
> > + val |= FIELD_PREP(REQ_FEC_MODE_A0_MSK, mode) |
> > + FIELD_PREP(REQ_FEC_MODE_A1_MSK, mode) |
> > + FIELD_PREP(REQ_FEC_MODE_A2_MSK, mode) |
> > + FIELD_PREP(REQ_FEC_MODE_A3_MSK, mode) |
> > + FIELD_PREP(REQ_FEC_MODE_B0_MSK, mode) |
> > + FIELD_PREP(REQ_FEC_MODE_B1_MSK, mode) |
> > + FIELD_PREP(REQ_FEC_MODE_B2_MSK, mode) |
> > + FIELD_PREP(REQ_FEC_MODE_B3_MSK, mode);

I'll also optimize the code here.

> > +
> > + ret = regmap_write(ns->regmap, NIOS_INIT, val);
> > + if (ret)
> > + return ret;
> > + }
> > +
> > +nios_init_done:
> > + /* polls on NIOS_INIT_DONE */
> > + ret = regmap_read_poll_timeout(ns->regmap, NIOS_INIT, val,
> > + val & NIOS_INIT_DONE,
> > + NIOS_INIT_TIME_INTV,
> > + NIOS_INIT_TIMEOUT);
> > + if (ret) {
> > + dev_err(dev, "NIOS_INIT_DONE %s\n",
> > + (ret == -ETIMEDOUT) ? "timed out" : "check error");
> > + goto dump_sts;
> > + }
> > +
> > + /*
> > + * After INIT_DONE is detected, it still needs to check if any error
> > + * detected.
> > + */
> > + if (init_error_detected(ns)) {
> > + /*
> > + * If the retimer configuration is failed, the Nios firmware
> > + * will still release the spi controller for host to
> > + * communicate with the BMC. It makes possible for people to
> > + * reprogram a new Nios firmware and restore the card. So the
> > + * driver doesn't error out, it continues to create the spi
> > + * controller device and spi_board_info for BMC.
> > + */
> > + dev_err(dev, "NIOS_INIT_DONE OK, but err found during init\n");
> > + goto dump_sts;
> > + }
> > + return 0;
> > +
> > +dump_sts:
> > + dump_error_stat(ns);
> Could init_error_detected and dump_error_stat be combined ?

Yes, I could try to refactor it.

> > +
> > + return ret;
> > +}
> > +
> > +static struct spi_board_info m10_n3000_info = {
> > + .modalias = "m10-n3000",
> > + .max_speed_hz = 12500000,
> > + .bus_num = 0,
> > + .chip_select = 0,
> > +};
> > +
> > +static int create_altera_spi_controller(struct n3000_nios *ns)
> > +{
> > + struct altera_spi_platform_data pdata = { 0 };
> > + struct platform_device_info pdevinfo = { 0 };
> > + void __iomem *base = ns->base;
> > + u64 v;
> > +
> > + v = readq(base + NIOS_SPI_PARAM);
> > +
> > + pdata.mode_bits = SPI_CS_HIGH;
> > + if (FIELD_GET(PARAM_CLK_POL, v))
> > + pdata.mode_bits |= SPI_CPOL;
> > + if (FIELD_GET(PARAM_CLK_PHASE, v))
> > + pdata.mode_bits |= SPI_CPHA;
> > +
> > + pdata.num_chipselect = FIELD_GET(PARAM_NUM_CS, v);
> > + pdata.bits_per_word_mask =
> > + SPI_BPW_RANGE_MASK(1, FIELD_GET(PARAM_DATA_WIDTH, v));
> > +
> > + pdata.num_devices = 1;
> > + pdata.devices = &m10_n3000_info;
> > +
> > + dev_dbg(ns->dev, "%s cs %u bpm 0x%x mode 0x%x\n", __func__,
> > + pdata.num_chipselect, pdata.bits_per_word_mask,
> > + pdata.mode_bits);
> > +
> > + pdevinfo.name = "subdev_spi_altera";
> > + pdevinfo.id = PLATFORM_DEVID_AUTO;
> > + pdevinfo.parent = ns->dev;
> > + pdevinfo.data = &pdata;
> > + pdevinfo.size_data = sizeof(pdata);
> > +
> > + ns->altera_spi = platform_device_register_full(&pdevinfo);
> > + return PTR_ERR_OR_ZERO(ns->altera_spi);
> > +}
> > +
> > +static void destroy_altera_spi_controller(struct n3000_nios *ns)
> > +{
> > + platform_device_unregister(ns->altera_spi);
> > +}
> > +
> > +/* ns is the abbreviation of nios_spi */
> > +static int ns_poll_stat_timeout(void __iomem *base, u64 *v)
> > +{
> > + int loops = NS_REGBUS_WAIT_TIMEOUT;
> > +
> > + /*
> > + * We don't use the time based timeout here for performance.
> > + *
> > + * The regbus read/write is on the critical path of Intel PAC N3000
> > + * image programing. The time based timeout checking will add too much
> > + * overhead on it. Usually the state changes in 1 or 2 loops on the
> > + * test server, and we set 10000 times loop here for safety.
> > + */
> > + do {
> > + *v = readq(base + NIOS_SPI_STAT);
> > + if (*v & STAT_RW_VAL)
> > + break;
> > + cpu_relax();
> > + } while (--loops);
> This becomes an infinite loop if loops <= 0

The loops count is always initialized at the start of the function.
Do you mean we should lower the impact when the developer inputs the
wrong loop count?

Is it better we change to:

unsigned int loops = NS_REGBUS_WAIT_TIMEOUT;

while (loops--) {

}

> > +
> > + return loops ? 0 : -ETIMEDOUT;
> > +}
> > +
> > +static int ns_reg_write(void *context, unsigned int reg, unsigned int val)
> > +{
> > + struct n3000_nios *ns = context;
> > + u64 v = 0;
> > + int ret;
> > +
> > + v |= FIELD_PREP(CTRL_CMD_MSK, CTRL_CMD_WR);
> > + v |= FIELD_PREP(CTRL_ADDR, reg);
> > + v |= FIELD_PREP(CTRL_WR_DATA, val);
> > + writeq(v, ns->base + NIOS_SPI_CTRL);
> > +
> > + ret = ns_poll_stat_timeout(ns->base, &v);
> > + if (ret)
> > + dev_err(ns->dev, "fail to write reg 0x%x val 0x%x: %d\n",
> > + reg, val, ret);
> > +
> > + return ret;
> > +}
> > +
> > +static int ns_reg_read(void *context, unsigned int reg, unsigned int *val)
> > +{
> > + struct n3000_nios *ns = context;
> > + u64 v = 0;
> > + int ret;
> > +
> > + v |= FIELD_PREP(CTRL_CMD_MSK, CTRL_CMD_RD);
>
> v = , and save the intialiation
>
> or v = ( ... | ... )

Yes.

>
> > + v |= FIELD_PREP(CTRL_ADDR, reg);
> > + writeq(v, ns->base + NIOS_SPI_CTRL);
> > +
> > + ret = ns_poll_stat_timeout(ns->base, &v);
> > + if (ret)
> > + dev_err(ns->dev, "fail to read reg 0x%x: %d\n", reg, ret);
> > + else
> > + *val = FIELD_GET(STAT_RD_DATA, v);
> > +
> > + return ret;
> > +}
> > +
> > +static const struct regmap_config ns_regbus_cfg = {
> > + .reg_bits = 32,
> > + .reg_stride = 4,
> > + .val_bits = 32,
> > + .fast_io = true,
> > +
> > + .reg_write = ns_reg_write,
> > + .reg_read = ns_reg_read,
> > +};
> > +
> > +static int n3000_nios_probe(struct dfl_device *ddev)
> > +{
> > + struct device *dev = &ddev->dev;
> > + struct n3000_nios *ns;
> > + int ret;
> > +
> > + ns = devm_kzalloc(dev, sizeof(*ns), GFP_KERNEL);
> > + if (!ns)
> > + return -ENOMEM;
> > +
> > + dev_set_drvdata(&ddev->dev, ns);
> > +
> > + ns->dev = dev;
> > +
> > + ns->base = devm_ioremap_resource(&ddev->dev, &ddev->mmio_res);
> > + if (IS_ERR(ns->base))
> > + return PTR_ERR(ns->base);
> > +
> > + ns->regmap = devm_regmap_init(dev, NULL, ns, &ns_regbus_cfg);
> > + if (IS_ERR(ns->regmap))
> > + return PTR_ERR(ns->regmap);
> > +
> > + ret = n3000_nios_init_done_check(ns);
> > + if (ret)
> > + return ret;
> > +
> > + ret = create_altera_spi_controller(ns);
> > + if (ret)
> > + dev_err(dev, "altera spi controller create failed: %d\n", ret);
> > +
> > + return ret;
> > +}
> > +
> > +static void n3000_nios_remove(struct dfl_device *ddev)
> > +{
> > + struct n3000_nios *ns = dev_get_drvdata(&ddev->dev);
> > +
> > + destroy_altera_spi_controller(ns);
> > +}
> > +
> > +#define FME_FEATURE_ID_N3000_NIOS 0xd
> > +
> > +static const struct dfl_device_id n3000_nios_ids[] = {
> > + { FME_ID, FME_FEATURE_ID_N3000_NIOS },
> > + { }
> > +};
> > +
> > +static struct dfl_driver n3000_nios_driver = {
> > + .drv = {
> > + .name = "n3000-nios",
>
> "dfl-n3000-nios" ?

Yes. I think it is better.

Thanks,
Yilun

>
> Tom
>
> > + .dev_groups = n3000_nios_groups,
> > + },
> > + .id_table = n3000_nios_ids,
> > + .probe = n3000_nios_probe,
> > + .remove = n3000_nios_remove,
> > +};
> > +
> > +module_dfl_driver(n3000_nios_driver);
> > +
> > +MODULE_DEVICE_TABLE(dfl, n3000_nios_ids);
> > +MODULE_DESCRIPTION("Driver for Nios private feature on Intel PAC N3000");
> > +MODULE_AUTHOR("Intel Corporation");
> > +MODULE_LICENSE("GPL v2");

Next message: Xing Zhengjun: "Re: [LKP] [mm] 5ef64cc898: vm-scalability.throughput -20.6% regression"
Previous message: Ira Weiny: "Re: [PATCH RFC PKS/PMEM 22/58] fs/f2fs: Utilize new kmap_thread()"
In reply to: Tom Rix: "Re: [PATCH v9 5/6] fpga: dfl: add support for N3000 Nios private feature"
Next in thread: Tom Rix: "Re: [PATCH v9 5/6] fpga: dfl: add support for N3000 Nios private feature"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]