[RFC PATCH v3 1/6] dpll: Add DPLL framework base functions
Vadim Fedorenko
vfedorenko at novek.ru
Mon Oct 10 12:54:26 PDT 2022
On 10.10.2022 10:18, Jiri Pirko wrote:
> Mon, Oct 10, 2022 at 03:17:59AM CEST, vfedorenko at novek.ru wrote:
>> From: Vadim Fedorenko <vadfed at fb.com>
>>
>> DPLL framework is used to represent and configure DPLL devices
>> in systems. Each device that has DPLL and can configure sources
>> and outputs can use this framework.
>>
>> Signed-off-by: Vadim Fedorenko <vadfed at fb.com>
>> Co-developed-by: Jakub Kicinski <kuba at kernel.org>
>> Co-developed-by: Arkadiusz Kubalewski <arkadiusz.kubalewski at intel.com>
>> ---
>> MAINTAINERS | 8 +
>> drivers/Kconfig | 2 +
>> drivers/Makefile | 1 +
>> drivers/dpll/Kconfig | 7 +
>> drivers/dpll/Makefile | 7 +
>> drivers/dpll/dpll_core.c | 177 +++++++++++++++
>> drivers/dpll/dpll_core.h | 41 ++++
>> drivers/dpll/dpll_netlink.c | 417 ++++++++++++++++++++++++++++++++++++
>> drivers/dpll/dpll_netlink.h | 7 +
>> include/linux/dpll.h | 29 +++
>> include/uapi/linux/dpll.h | 101 +++++++++
>> 11 files changed, 797 insertions(+)
>> create mode 100644 drivers/dpll/Kconfig
>> create mode 100644 drivers/dpll/Makefile
>> create mode 100644 drivers/dpll/dpll_core.c
>> create mode 100644 drivers/dpll/dpll_core.h
>> create mode 100644 drivers/dpll/dpll_netlink.c
>> create mode 100644 drivers/dpll/dpll_netlink.h
>> create mode 100644 include/linux/dpll.h
>> create mode 100644 include/uapi/linux/dpll.h
>>
>> diff --git a/MAINTAINERS b/MAINTAINERS
>> index 9ca84cb5ab4a..e2f4fede937f 100644
>> --- a/MAINTAINERS
>> +++ b/MAINTAINERS
>> @@ -6301,6 +6301,14 @@ F: Documentation/networking/device_drivers/ethernet/freescale/dpaa2/switch-drive
>> F: drivers/net/ethernet/freescale/dpaa2/dpaa2-switch*
>> F: drivers/net/ethernet/freescale/dpaa2/dpsw*
>>
>> +DPLL CLOCK SUBSYSTEM
>> +M: Vadim Fedorenko <vadfed at fb.com>
>> +L: netdev at vger.kernel.org
>> +S: Maintained
>> +F: drivers/dpll/*
>> +F: include/net/dpll.h
>> +F: include/uapi/linux/dpll.h
>> +
>> DRBD DRIVER
>> M: Philipp Reisner <philipp.reisner at linbit.com>
>> M: Lars Ellenberg <lars.ellenberg at linbit.com>
>> diff --git a/drivers/Kconfig b/drivers/Kconfig
>> index 19ee995bd0ae..a3e00294a995 100644
>> --- a/drivers/Kconfig
>> +++ b/drivers/Kconfig
>> @@ -239,4 +239,6 @@ source "drivers/peci/Kconfig"
>>
>> source "drivers/hte/Kconfig"
>>
>> +source "drivers/dpll/Kconfig"
>> +
>> endmenu
>> diff --git a/drivers/Makefile b/drivers/Makefile
>> index 057857258bfd..78a68f1621cc 100644
>> --- a/drivers/Makefile
>> +++ b/drivers/Makefile
>> @@ -188,3 +188,4 @@ obj-$(CONFIG_COUNTER) += counter/
>> obj-$(CONFIG_MOST) += most/
>> obj-$(CONFIG_PECI) += peci/
>> obj-$(CONFIG_HTE) += hte/
>> +obj-$(CONFIG_DPLL) += dpll/
>> diff --git a/drivers/dpll/Kconfig b/drivers/dpll/Kconfig
>> new file mode 100644
>> index 000000000000..a4cae73f20d3
>> --- /dev/null
>> +++ b/drivers/dpll/Kconfig
>> @@ -0,0 +1,7 @@
>> +# SPDX-License-Identifier: GPL-2.0-only
>> +#
>> +# Generic DPLL drivers configuration
>> +#
>> +
>> +config DPLL
>> + bool
>> diff --git a/drivers/dpll/Makefile b/drivers/dpll/Makefile
>> new file mode 100644
>> index 000000000000..0748c80097e4
>> --- /dev/null
>> +++ b/drivers/dpll/Makefile
>> @@ -0,0 +1,7 @@
>> +# SPDX-License-Identifier: GPL-2.0
>> +#
>> +# Makefile for DPLL drivers.
>> +#
>> +
>> +obj-$(CONFIG_DPLL) += dpll_sys.o
>> +dpll_sys-y += dpll_core.o dpll_netlink.o
>> diff --git a/drivers/dpll/dpll_core.c b/drivers/dpll/dpll_core.c
>> new file mode 100644
>> index 000000000000..7fdee145e82c
>> --- /dev/null
>> +++ b/drivers/dpll/dpll_core.c
>> @@ -0,0 +1,177 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * dpll_core.c - Generic DPLL Management class support.
>> + *
>> + * Copyright (c) 2021 Meta Platforms, Inc. and affiliates
>> + */
>> +
>> +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
>> +
>> +#include <linux/device.h>
>> +#include <linux/err.h>
>> +#include <linux/slab.h>
>> +#include <linux/string.h>
>> +
>> +#include "dpll_core.h"
>> +
>> +static DEFINE_MUTEX(dpll_device_xa_lock);
>> +static DEFINE_XARRAY_FLAGS(dpll_device_xa, XA_FLAGS_ALLOC);
>> +#define DPLL_REGISTERED XA_MARK_1
>> +
>> +#define ASSERT_DPLL_REGISTERED(d) \
>> + WARN_ON_ONCE(!xa_get_mark(&dpll_device_xa, (d)->id, DPLL_REGISTERED))
>> +#define ASSERT_DPLL_NOT_REGISTERED(d) \
>> + WARN_ON_ONCE(xa_get_mark(&dpll_device_xa, (d)->id, DPLL_REGISTERED))
>> +
>> +
>> +int for_each_dpll_device(int id, int (*cb)(struct dpll_device *, void *),
>> + void *data)
>
> Probably better to avoid this "cb" indirection here.
> Perhaps in a similar way devlinks_xa_for_each_registered_get() is done?
>
Thanks for pointing to an example, I will consider avoid indirection.
checkpatch.pl is also complaining about this line, but the same way was used in
some other places that's why I chosed to use it.
>
>> +{
>> + struct dpll_device *dpll;
>> + unsigned long index;
>> + int ret = 0;
>> +
>> + mutex_lock(&dpll_device_xa_lock);
>> + xa_for_each_start(&dpll_device_xa, index, dpll, id) {
>> + if (!xa_get_mark(&dpll_device_xa, index, DPLL_REGISTERED))
>> + continue;
>> + ret = cb(dpll, data);
>> + if (ret)
>> + break;
>> + }
>> + mutex_unlock(&dpll_device_xa_lock);
>> +
>> + return ret;
>> +}
>> +
>> +struct dpll_device *dpll_device_get_by_id(int id)
>> +{
>> + struct dpll_device *dpll = NULL;
>> +
>> + if (xa_get_mark(&dpll_device_xa, id, DPLL_REGISTERED))
>> + dpll = xa_load(&dpll_device_xa, id);
>> + return dpll;
>> +}
>> +
>> +struct dpll_device *dpll_device_get_by_name(const char *name)
>> +{
>> + struct dpll_device *dpll, *ret = NULL;
>> + unsigned long index;
>> +
>> + mutex_lock(&dpll_device_xa_lock);
>> + xa_for_each_marked(&dpll_device_xa, index, dpll, DPLL_REGISTERED) {
>> + if (!strcmp(dev_name(&dpll->dev), name)) {
>> + ret = dpll;
>> + break;
>> + }
>> + }
>> + mutex_unlock(&dpll_device_xa_lock);
>> +
>> + return ret;
>> +}
>> +
>> +void *dpll_priv(struct dpll_device *dpll)
>> +{
>> + return dpll->priv;
>> +}
>> +EXPORT_SYMBOL_GPL(dpll_priv);
>> +
>> +static void dpll_device_release(struct device *dev)
>> +{
>> + struct dpll_device *dpll;
>> +
>> + dpll = to_dpll_device(dev);
>> +
>> + dpll_device_unregister(dpll);
>> + dpll_device_free(dpll);
>> +}
>> +
>> +static struct class dpll_class = {
>> + .name = "dpll",
>> + .dev_release = dpll_device_release,
>> +};
>> +
>> +struct dpll_device *dpll_device_alloc(struct dpll_device_ops *ops, const char *name,
>> + int sources_count, int outputs_count, void *priv)
>> +{
>> + struct dpll_device *dpll;
>> + int ret;
>> +
>> + dpll = kzalloc(sizeof(*dpll), GFP_KERNEL);
>> + if (!dpll)
>> + return ERR_PTR(-ENOMEM);
>> +
>> + mutex_init(&dpll->lock);
>> + dpll->ops = ops;
>> + dpll->dev.class = &dpll_class;
>> + dpll->sources_count = sources_count;
>> + dpll->outputs_count = outputs_count;
>> +
>> + mutex_lock(&dpll_device_xa_lock);
>> + ret = xa_alloc(&dpll_device_xa, &dpll->id, dpll, xa_limit_16b, GFP_KERNEL);
>> + if (ret)
>> + goto error;
>> + dev_set_name(&dpll->dev, "%s%d", name ? name : "dpll", dpll->id);
>> + mutex_unlock(&dpll_device_xa_lock);
>> + dpll->priv = priv;
>> +
>> + return dpll;
>> +
>> +error:
>> + mutex_unlock(&dpll_device_xa_lock);
>> + kfree(dpll);
>> + return ERR_PTR(ret);
>> +}
>> +EXPORT_SYMBOL_GPL(dpll_device_alloc);
>> +
>> +void dpll_device_free(struct dpll_device *dpll)
>> +{
>> + if (!dpll)
>> + return;
>> +
>> + mutex_destroy(&dpll->lock);
>> + kfree(dpll);
>> +}
>> +EXPORT_SYMBOL_GPL(dpll_device_free);
>> +
>> +void dpll_device_register(struct dpll_device *dpll)
>> +{
>> + ASSERT_DPLL_NOT_REGISTERED(dpll);
>> +
>> + mutex_lock(&dpll_device_xa_lock);
>> + xa_set_mark(&dpll_device_xa, dpll->id, DPLL_REGISTERED);
>> + mutex_unlock(&dpll_device_xa_lock);
>> +}
>> +EXPORT_SYMBOL_GPL(dpll_device_register);
>> +
>> +void dpll_device_unregister(struct dpll_device *dpll)
>> +{
>> + ASSERT_DPLL_REGISTERED(dpll);
>> +
>> + mutex_lock(&dpll_device_xa_lock);
>> + xa_erase(&dpll_device_xa, dpll->id);
>> + mutex_unlock(&dpll_device_xa_lock);
>> +}
>> +EXPORT_SYMBOL_GPL(dpll_device_unregister);
>> +
>> +static int __init dpll_init(void)
>> +{
>> + int ret;
>> +
>> + ret = dpll_netlink_init();
>> + if (ret)
>> + goto error;
>> +
>> + ret = class_register(&dpll_class);
>> + if (ret)
>> + goto unregister_netlink;
>> +
>> + return 0;
>> +
>> +unregister_netlink:
>> + dpll_netlink_finish();
>> +error:
>> + mutex_destroy(&dpll_device_xa_lock);
>> + return ret;
>> +}
>> +subsys_initcall(dpll_init);
>> diff --git a/drivers/dpll/dpll_core.h b/drivers/dpll/dpll_core.h
>> new file mode 100644
>> index 000000000000..4b6fc9eb228f
>> --- /dev/null
>> +++ b/drivers/dpll/dpll_core.h
>> @@ -0,0 +1,41 @@
>> +/* SPDX-License-Identifier: GPL-2.0 */
>> +/*
>> + * Copyright (c) 2021 Meta Platforms, Inc. and affiliates
>> + */
>> +
>> +#ifndef __DPLL_CORE_H__
>> +#define __DPLL_CORE_H__
>> +
>> +#include <linux/dpll.h>
>> +
>> +#include "dpll_netlink.h"
>> +
>> +/**
>> + * struct dpll_device - structure for a DPLL device
>> + * @id: unique id number for each edvice
>> + * @dev: &struct device for this dpll device
>> + * @sources_count: amount of input sources this dpll_device supports
>> + * @outputs_count: amount of outputs this dpll_device supports
>> + * @ops: operations this &dpll_device supports
>> + * @lock: mutex to serialize operations
>> + * @priv: pointer to private information of owner
>> + */
>> +struct dpll_device {
>> + int id;
>> + struct device dev;
>> + int sources_count;
>> + int outputs_count;
>> + struct dpll_device_ops *ops;
>> + struct mutex lock;
>> + void *priv;
>> +};
>> +
>> +#define to_dpll_device(_dev) \
>> + container_of(_dev, struct dpll_device, dev)
>> +
>> +int for_each_dpll_device(int id, int (*cb)(struct dpll_device *, void *),
>> + void *data);
>> +struct dpll_device *dpll_device_get_by_id(int id);
>> +struct dpll_device *dpll_device_get_by_name(const char *name);
>> +void dpll_device_unregister(struct dpll_device *dpll);
>> +#endif
>> diff --git a/drivers/dpll/dpll_netlink.c b/drivers/dpll/dpll_netlink.c
>> new file mode 100644
>> index 000000000000..31966e0eec3a
>> --- /dev/null
>> +++ b/drivers/dpll/dpll_netlink.c
>> @@ -0,0 +1,417 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * Generic netlink for DPLL management framework
>> + *
>> + * Copyright (c) 2021 Meta Platforms, Inc. and affiliates
>> + *
>> + */
>> +#include <linux/module.h>
>> +#include <linux/kernel.h>
>> +#include <net/genetlink.h>
>> +#include "dpll_core.h"
>> +
>> +#include <uapi/linux/dpll.h>
>> +
>> +static const struct genl_multicast_group dpll_genl_mcgrps[] = {
>> + { .name = DPLL_CONFIG_DEVICE_GROUP_NAME, },
>> + { .name = DPLL_CONFIG_SOURCE_GROUP_NAME, },
>> + { .name = DPLL_CONFIG_OUTPUT_GROUP_NAME, },
>> + { .name = DPLL_MONITOR_GROUP_NAME, },
>> +};
>> +
>> +static const struct nla_policy dpll_genl_get_policy[] = {
>> + [DPLLA_DEVICE_ID] = { .type = NLA_U32 },
>> + [DPLLA_DEVICE_NAME] = { .type = NLA_STRING,
>> + .len = DPLL_NAME_LENGTH },
>> + [DPLLA_FLAGS] = { .type = NLA_U32 },
>> +};
>> +
>> +static const struct nla_policy dpll_genl_set_source_policy[] = {
>> + [DPLLA_DEVICE_ID] = { .type = NLA_U32 },
>> + [DPLLA_SOURCE_ID] = { .type = NLA_U32 },
>> + [DPLLA_SOURCE_TYPE] = { .type = NLA_U32 },
>> +};
>> +
>> +static const struct nla_policy dpll_genl_set_output_policy[] = {
>> + [DPLLA_DEVICE_ID] = { .type = NLA_U32 },
>> + [DPLLA_OUTPUT_ID] = { .type = NLA_U32 },
>> + [DPLLA_OUTPUT_TYPE] = { .type = NLA_U32 },
>> +};
>> +
>> +struct param {
>
> Namespace prefix please .
>
Yep, will add.
>
>> + struct netlink_callback *cb;
>> + struct dpll_device *dpll;
>> + struct sk_buff *msg;
>
>
>
>> + int dpll_id;
>> + int dpll_source_id;
>> + int dpll_source_type;
>> + int dpll_output_id;
>> + int dpll_output_type;
>
> I don't see where you use these 5.
>
Yep, the implementation was improved, it looks like it's artifacts of the
previous version.
>
>> +};
>> +
>> +struct dpll_dump_ctx {
>> + struct dpll_device *dev;
>> + int flags;
>> + int pos_idx;
>
>
>
>> + int pos_src_idx;
>> + int pos_out_idx;
>
> You don't use these 2.
>
The same again, artifacts.
>
>> +};
>> +
>> +typedef int (*cb_t)(struct param *);
>> +
>> +static struct genl_family dpll_gnl_family;
>> +
>> +static struct dpll_dump_ctx *dpll_dump_context(struct netlink_callback *cb)
>> +{
>> + return (struct dpll_dump_ctx *)cb->ctx;
>> +}
>> +
>> +static int __dpll_cmd_device_dump_one(struct dpll_device *dpll,
>> + struct sk_buff *msg)
>> +{
>> + if (nla_put_u32(msg, DPLLA_DEVICE_ID, dpll->id))
>> + return -EMSGSIZE;
>> +
>> + if (nla_put_string(msg, DPLLA_DEVICE_NAME, dev_name(&dpll->dev)))
>> + return -EMSGSIZE;
>> +
>> + return 0;
>> +}
>> +
>> +static int __dpll_cmd_dump_sources(struct dpll_device *dpll,
>> + struct sk_buff *msg)
>> +{
>> + struct nlattr *src_attr;
>> + int i, ret = 0, type;
>> +
>> + for (i = 0; i < dpll->sources_count; i++) {
>> + src_attr = nla_nest_start(msg, DPLLA_SOURCE);
>> + if (!src_attr) {
>> + ret = -EMSGSIZE;
>> + break;
>> + }
>> + type = dpll->ops->get_source_type(dpll, i);
>> + if (nla_put_u32(msg, DPLLA_SOURCE_ID, i) ||
>> + nla_put_u32(msg, DPLLA_SOURCE_TYPE, type)) {
>> + nla_nest_cancel(msg, src_attr);
>> + ret = -EMSGSIZE;
>> + break;
>> + }
>> + if (dpll->ops->get_source_supported) {
>> + for (type = 0; type <= DPLL_TYPE_MAX; type++) {
>> + ret = dpll->ops->get_source_supported(dpll, i, type);
>
> Okay, this looks weird to me. This implicates that it is possible to
> have for example:
> source index 0 of type 10
> source index 0 of type 11
> Both possible.
>
> However, from how I understand this, each source if of certain fixed type.
> Either it is:
> SyncE port
> 1pps external input (SMA)
> 10MHZ external input (SMA)
> internal oscilator (free-running)
> GNSS (GPS)
>
> So for example:
> index 0, type: 1pps external input (SMA)
> index 1, type: 10MHZ external input (
> index 2, type: SyncE port, netdev ifindex: 20
> index 3, type: SyncE port, netdev ifindex: 30
>
> So 4 "source" objects, each of different type.
> In this case I can imagine that the netlink API might look something
> like:
> -> DPLL_CMD_SOURCE_GET - dump
> ATTR_DEVICE_ID X
>
> <- DPLL_CMD_SOURCE_GET
>
> ATTR_DEVICE_ID X
> ATTR_SOURCE_INDEX 0
> ATTR_SOURCE_TYPE EXT_1PPS
>
> ATTR_DEVICE_ID X
> ATTR_SOURCE_INDEX 1
> ATTR_SOURCE_TYPE EXT_10MHZ
>
> ATTR_DEVICE_ID X
> ATTR_SOURCE_INDEX 2
> ATTR_SOURCE_TYPE SYNCE_ETH_PORT
> ATTR_SOURCE_NETDEV_IFINDEX 20
>
> ATTR_DEVICE_ID X
> ATTR_SOURCE_INDEX 3
> ATTR_SOURCE_TYPE SYNCE_ETH_PORT
> ATTR_SOURCE_NETDEV_IFINDEX 30
>
> You see kernel would dump 4 source objects.
>
I see your point. We do have hardware which allows changing type of SMA
connector, and even the direction, each SMA could be used as input/source or
output of different signals. But there are limitation, like not all SMAs can
produce IRIG-B signal or only some of them can be used to get GNSS 1PPS. The
interface was created to cover such case. I believe we have to improve it to
cover SyncE configuration better, but I personally don't have SyncE hardware
ready to test and that's why I have to rely on suggestions from yours or
Arkadiusz's experience. From what I can see now there is need for special
attribute to link source to net device, and I'm happy to add it. In case of
fixed configuration of sources, the device should provide only one type as
supported and that's it.
>
>
>> + if (ret && nla_put_u32(msg, DPLLA_SOURCE_SUPPORTED, type)) {
>> + ret = -EMSGSIZE;
>> + break;
>> + }
>> + }
>> + ret = 0;
>> + }
>> + nla_nest_end(msg, src_attr);
>> + }
>> +
>> + return ret;
>> +}
>> +
>> +static int __dpll_cmd_dump_outputs(struct dpll_device *dpll,
>> + struct sk_buff *msg)
>> +{
>> + struct nlattr *out_attr;
>> + int i, ret = 0, type;
>> +
>> + for (i = 0; i < dpll->outputs_count; i++) {
>> + out_attr = nla_nest_start(msg, DPLLA_OUTPUT);
>> + if (!out_attr) {
>> + ret = -EMSGSIZE;
>> + break;
>> + }
>> + type = dpll->ops->get_output_type(dpll, i);
>> + if (nla_put_u32(msg, DPLLA_OUTPUT_ID, i) ||
>> + nla_put_u32(msg, DPLLA_OUTPUT_TYPE, type)) {
>> + nla_nest_cancel(msg, out_attr);
>> + ret = -EMSGSIZE;
>> + break;
>> + }
>> + if (dpll->ops->get_output_supported) {
>> + for (type = 0; type <= DPLL_TYPE_MAX; type++) {
>> + ret = dpll->ops->get_output_supported(dpll, i, type);
>> + if (ret && nla_put_u32(msg, DPLLA_OUTPUT_SUPPORTED, type)) {
>
> This I believe is similar to sources, see my comment above.
>
As I said, we have to cover the case when SMAs are flexible to configure.
> I believe we should have separate commands to GET and SET outputs and
> sources. That would make the object separation clear and will also help
> event model. See below I suggestion how output netlink API may look
> like (comment in header file near enum dpll_genl_cmd definition).
>
>
>> + ret = -EMSGSIZE;
>> + break;
>> + }
>> + }
>> + ret = 0;
>> + }
>> + nla_nest_end(msg, out_attr);
>> + }
>> +
>> + return ret;
>> +}
>> +
>> +static int __dpll_cmd_dump_status(struct dpll_device *dpll,
>> + struct sk_buff *msg)
>> +{
>> + int ret;
>> +
>> + if (dpll->ops->get_status) {
>> + ret = dpll->ops->get_status(dpll);
>> + if (nla_put_u32(msg, DPLLA_STATUS, ret))
>> + return -EMSGSIZE;
>> + }
>> +
>> + if (dpll->ops->get_temp) {
>> + ret = dpll->ops->get_temp(dpll);
>> + if (nla_put_u32(msg, DPLLA_TEMP, ret))
>> + return -EMSGSIZE;
>> + }
>> +
>> + if (dpll->ops->get_lock_status) {
>> + ret = dpll->ops->get_lock_status(dpll);
>> + if (nla_put_u32(msg, DPLLA_LOCK_STATUS, ret))
>> + return -EMSGSIZE;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int
>> +dpll_device_dump_one(struct dpll_device *dpll, struct sk_buff *msg,
>> + u32 portid, u32 seq, int flags)
>> +{
>> + struct nlattr *hdr;
>> + int ret;
>> +
>> + hdr = genlmsg_put(msg, portid, seq, &dpll_gnl_family, 0,
>> + DPLL_CMD_DEVICE_GET);
>> + if (!hdr)
>> + return -EMSGSIZE;
>> +
>> + mutex_lock(&dpll->lock);
>> + ret = __dpll_cmd_device_dump_one(dpll, msg);
>> + if (ret)
>> + goto out_unlock;
>> +
>> + if (flags & DPLL_FLAG_SOURCES && dpll->ops->get_source_type) {
>> + ret = __dpll_cmd_dump_sources(dpll, msg);
>> + if (ret)
>> + goto out_unlock;
>> + }
>> +
>> + if (flags & DPLL_FLAG_OUTPUTS && dpll->ops->get_output_type) {
>> + ret = __dpll_cmd_dump_outputs(dpll, msg);
>> + if (ret)
>> + goto out_unlock;
>> + }
>> +
>> + if (flags & DPLL_FLAG_STATUS) {
>> + ret = __dpll_cmd_dump_status(dpll, msg);
>> + if (ret)
>> + goto out_unlock;
>> + }
>> + mutex_unlock(&dpll->lock);
>> + genlmsg_end(msg, hdr);
>> +
>> + return 0;
>> +
>> +out_unlock:
>> + mutex_unlock(&dpll->lock);
>> + genlmsg_cancel(msg, hdr);
>> +
>> + return ret;
>> +}
>> +
>> +static int dpll_genl_cmd_set_source(struct sk_buff *skb, struct genl_info *info)
>> +{
>> + struct dpll_device *dpll = info->user_ptr[0];
>> + struct nlattr **attrs = info->attrs;
>> + int ret = 0, src_id, type;
>
> Enums.
>
Sure.
>> +
>> + if (!attrs[DPLLA_SOURCE_ID] ||
>> + !attrs[DPLLA_SOURCE_TYPE])
>> + return -EINVAL;
>> +
>> + if (!dpll->ops->set_source_type)
>> + return -EOPNOTSUPP;
>> +
>> + src_id = nla_get_u32(attrs[DPLLA_SOURCE_ID]);
>> + type = nla_get_u32(attrs[DPLLA_SOURCE_TYPE]);
>
>
> This looks odd to me. The user should just pass the index of source to
> select. Type should be static, and non-changeable.
>
>
>> +
>> + mutex_lock(&dpll->lock);
>> + ret = dpll->ops->set_source_type(dpll, src_id, type);
>> + mutex_unlock(&dpll->lock);
>> +
>> + return ret;
>> +}
>> +
>> +static int dpll_genl_cmd_set_output(struct sk_buff *skb, struct genl_info *info)
>> +{
>> + struct dpll_device *dpll = info->user_ptr[0];
>> + struct nlattr **attrs = info->attrs;
>> + int ret = 0, out_id, type;
>
> Enums.
>
>
>> +
>> + if (!attrs[DPLLA_OUTPUT_ID] ||
>> + !attrs[DPLLA_OUTPUT_TYPE])
>> + return -EINVAL;
>> +
>> + if (!dpll->ops->set_output_type)
>> + return -EOPNOTSUPP;
>> +
>> + out_id = nla_get_u32(attrs[DPLLA_OUTPUT_ID]);
>> + type = nla_get_u32(attrs[DPLLA_OUTPUT_TYPE]);
>
> Same here, passing type here looks wrong.
>
>
>
>> +
>> + mutex_lock(&dpll->lock);
>> + ret = dpll->ops->set_output_type(dpll, out_id, type);
>> + mutex_unlock(&dpll->lock);
>> +
>> + return ret;
>> +}
>> +
>> +static int dpll_device_loop_cb(struct dpll_device *dpll, void *data)
>> +{
>> + struct dpll_dump_ctx *ctx;
>> + struct param *p = (struct param *)data;
>> +
>> + ctx = dpll_dump_context(p->cb);
>> +
>> + ctx->pos_idx = dpll->id;
>> +
>> + return dpll_device_dump_one(dpll, p->msg, 0, 0, ctx->flags);
>> +}
>> +
>> +static int
>> +dpll_cmd_device_dump(struct sk_buff *skb, struct netlink_callback *cb)
>> +{
>> + struct dpll_dump_ctx *ctx = dpll_dump_context(cb);
>> + struct param p = { .cb = cb, .msg = skb };
>> +
>> + return for_each_dpll_device(ctx->pos_idx, dpll_device_loop_cb, &p);
>> +}
>> +
>> +static int
>> +dpll_genl_cmd_device_get_id(struct sk_buff *skb, struct genl_info *info)
>
> Just "get", no "id" here.
>
Got it.
>
>> +{
>> + struct dpll_device *dpll = info->user_ptr[0];
>> + struct nlattr **attrs = info->attrs;
>> + struct sk_buff *msg;
>> + int flags = 0;
>> + int ret;
>> +
>> + if (attrs[DPLLA_FLAGS])
>> + flags = nla_get_u32(attrs[DPLLA_FLAGS]);
>> +
>> + msg = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
>> + if (!msg)
>> + return -ENOMEM;
>> +
>> + ret = dpll_device_dump_one(dpll, msg, info->snd_portid, info->snd_seq,
>> + flags);
>> + if (ret)
>> + goto out_free_msg;
>> +
>> + return genlmsg_reply(msg, info);
>> +
>> +out_free_msg:
>> + nlmsg_free(msg);
>> + return ret;
>> +
>> +}
>> +
>> +static int dpll_genl_cmd_start(struct netlink_callback *cb)
>> +{
>> + const struct genl_dumpit_info *info = genl_dumpit_info(cb);
>> + struct dpll_dump_ctx *ctx = dpll_dump_context(cb);
>> +
>> + ctx->dev = NULL;
>> + if (info->attrs[DPLLA_FLAGS])
>> + ctx->flags = nla_get_u32(info->attrs[DPLLA_FLAGS]);
>> + else
>> + ctx->flags = 0;
>> + ctx->pos_idx = 0;
>> + ctx->pos_src_idx = 0;
>> + ctx->pos_out_idx = 0;
>> + return 0;
>> +}
>> +
>> +static int dpll_pre_doit(const struct genl_ops *ops, struct sk_buff *skb,
>> + struct genl_info *info)
>> +{
>> + struct dpll_device *dpll_id = NULL, *dpll_name = NULL;
>> +
>> + if (!info->attrs[DPLLA_DEVICE_ID] &&
>> + !info->attrs[DPLLA_DEVICE_NAME])
>> + return -EINVAL;
>> +
>> + if (info->attrs[DPLLA_DEVICE_ID]) {
>> + u32 id = nla_get_u32(info->attrs[DPLLA_DEVICE_ID]);
>> +
>> + dpll_id = dpll_device_get_by_id(id);
>> + if (!dpll_id)
>> + return -ENODEV;
>> + info->user_ptr[0] = dpll_id;
>
> struct dpll_device *dpll should be stored here.
>
>
>> + }
>> + if (info->attrs[DPLLA_DEVICE_NAME]) {
>
> You define new API, have one clear handle for devices. Either name or
> ID. Having both is messy.
>
That was added after the discussion with Jakub and Arkadiusz where we agreed
that the device could be referenced either by index or by name. The example is
that userspace app can easily find specific DPLL device if it knows the name
provided by a driver of that specific device. Without searching through sysfs to
find index value. Later commands could be executed using index once it's known
through CMD_GET_DEVICE/ATTR_DEVICE_NAME.
>
>> + const char *name = nla_data(info->attrs[DPLLA_DEVICE_NAME]);
>> +
>> + dpll_name = dpll_device_get_by_name(name);
>> + if (!dpll_name)
>> + return -ENODEV;
>> +
>> + if (dpll_id && dpll_name != dpll_id)
>> + return -EINVAL;
>> + info->user_ptr[0] = dpll_name;
>
> struct dpll_device *dpll should be stored here.
>
Didn't get you, where should we store it?
dpll_name and dpll_id are of type struct dpll_device*,
and they are compared to avoid situation when both index and name are provided,
but refer to different devices.
>
>
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static const struct genl_ops dpll_genl_ops[] = {
>> + {
>> + .cmd = DPLL_CMD_DEVICE_GET,
>> + .flags = GENL_UNS_ADMIN_PERM,
>> + .start = dpll_genl_cmd_start,
>> + .dumpit = dpll_cmd_device_dump,
>> + .doit = dpll_genl_cmd_device_get_id,
>> + .policy = dpll_genl_get_policy,
>> + .maxattr = ARRAY_SIZE(dpll_genl_get_policy) - 1,
>> + },
>> + {
>> + .cmd = DPLL_CMD_SET_SOURCE_TYPE,
>> + .flags = GENL_UNS_ADMIN_PERM,
>> + .doit = dpll_genl_cmd_set_source,
>> + .policy = dpll_genl_set_source_policy,
>> + .maxattr = ARRAY_SIZE(dpll_genl_set_source_policy) - 1,
>> + },
>> + {
>> + .cmd = DPLL_CMD_SET_OUTPUT_TYPE,
>> + .flags = GENL_UNS_ADMIN_PERM,
>> + .doit = dpll_genl_cmd_set_output,
>> + .policy = dpll_genl_set_output_policy,
>> + .maxattr = ARRAY_SIZE(dpll_genl_set_output_policy) - 1,
>> + },
>> +};
>> +
>> +static struct genl_family dpll_gnl_family __ro_after_init = {
>> + .hdrsize = 0,
>> + .name = DPLL_FAMILY_NAME,
>> + .version = DPLL_VERSION,
>> + .ops = dpll_genl_ops,
>> + .n_ops = ARRAY_SIZE(dpll_genl_ops),
>> + .mcgrps = dpll_genl_mcgrps,
>> + .n_mcgrps = ARRAY_SIZE(dpll_genl_mcgrps),
>> + .pre_doit = dpll_pre_doit,
>
> Have .parallel_ops = true,
> You have dpll->lock, you don't need genl, to protect you.
>
Yep, sure, thanks!
>
>> +};
>> +
>> +int __init dpll_netlink_init(void)
>> +{
>> + return genl_register_family(&dpll_gnl_family);
>> +}
>> +
>> +void dpll_netlink_finish(void)
>> +{
>> + genl_unregister_family(&dpll_gnl_family);
>> +}
>> +
>> +void __exit dpll_netlink_fini(void)
>> +{
>> + dpll_netlink_finish();
>> +}
>> diff --git a/drivers/dpll/dpll_netlink.h b/drivers/dpll/dpll_netlink.h
>> new file mode 100644
>> index 000000000000..e2d100f59dd6
>> --- /dev/null
>> +++ b/drivers/dpll/dpll_netlink.h
>> @@ -0,0 +1,7 @@
>> +/* SPDX-License-Identifier: GPL-2.0 */
>> +/*
>> + * Copyright (c) 2021 Meta Platforms, Inc. and affiliates
>> + */
>> +
>> +int __init dpll_netlink_init(void);
>> +void dpll_netlink_finish(void);
>> diff --git a/include/linux/dpll.h b/include/linux/dpll.h
>> new file mode 100644
>> index 000000000000..9d49b19d03d9
>> --- /dev/null
>> +++ b/include/linux/dpll.h
>> @@ -0,0 +1,29 @@
>> +/* SPDX-License-Identifier: GPL-2.0 */
>> +/*
>> + * Copyright (c) 2021 Meta Platforms, Inc. and affiliates
>> + */
>> +
>> +#ifndef __DPLL_H__
>> +#define __DPLL_H__
>> +
>> +struct dpll_device;
>> +
>> +struct dpll_device_ops {
>> + int (*get_status)(struct dpll_device *dpll);
>> + int (*get_temp)(struct dpll_device *dpll);
>> + int (*get_lock_status)(struct dpll_device *dpll);
>> + int (*get_source_type)(struct dpll_device *dpll, int id);
>> + int (*get_source_supported)(struct dpll_device *dpll, int id, int type);
>> + int (*get_output_type)(struct dpll_device *dpll, int id);
>> + int (*get_output_supported)(struct dpll_device *dpll, int id, int type);
>> + int (*set_source_type)(struct dpll_device *dpll, int id, int val);
>> + int (*set_output_type)(struct dpll_device *dpll, int id, int val);
>
> All int should be enums, when they are really enums. Makes things much
> nicer and easier to see what's what.
>
Yep, will update it.
>
>
>> +};
>> +
>> +struct dpll_device *dpll_device_alloc(struct dpll_device_ops *ops, const char *name,
>> + int sources_count, int outputs_count, void *priv);
>> +void dpll_device_register(struct dpll_device *dpll);
>> +void dpll_device_unregister(struct dpll_device *dpll);
>> +void dpll_device_free(struct dpll_device *dpll);
>> +void *dpll_priv(struct dpll_device *dpll);
>> +#endif
>> diff --git a/include/uapi/linux/dpll.h b/include/uapi/linux/dpll.h
>> new file mode 100644
>> index 000000000000..fcbea5a5e4d6
>> --- /dev/null
>> +++ b/include/uapi/linux/dpll.h
>> @@ -0,0 +1,101 @@
>> +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
>> +#ifndef _UAPI_LINUX_DPLL_H
>> +#define _UAPI_LINUX_DPLL_H
>> +
>> +#define DPLL_NAME_LENGTH 20
>> +
>> +/* Adding event notification support elements */
>> +#define DPLL_FAMILY_NAME "dpll"
>> +#define DPLL_VERSION 0x01
>> +#define DPLL_CONFIG_DEVICE_GROUP_NAME "config"
>> +#define DPLL_CONFIG_SOURCE_GROUP_NAME "source"
>> +#define DPLL_CONFIG_OUTPUT_GROUP_NAME "output"
>> +#define DPLL_MONITOR_GROUP_NAME "monitor"
>
> What is exactly the reason for multiple multicast groups? Why you don't
> use one?
>
Yes, I agree, there is no need to use multiple groups and I will remove them in
the next version.
>
>> +
>> +#define DPLL_FLAG_SOURCES 1
>> +#define DPLL_FLAG_OUTPUTS 2
>> +#define DPLL_FLAG_STATUS 4
>
> I think it is more common to use either 0x prefix or (1<<X) expression
> But I don't think these flags are needed at all, if you have per-object
> messages.
>
>
>> +
>> +/* Attributes of dpll_genl_family */
>> +enum dpll_genl_attr {
>
> I don't see need for "genl" here.
> Also, it is common to have consistency betwee enum name and members name.
> For example:
>
> enum dpll_attr {
> DPLL_ATTR_UNSPEC,
> DPLL_ATTR_DEVICE_ID,
>
> ...
> }
> This applies to all enums in this file.
>
Got it, naming is always hard task. Will update all of them in the next version.
>
>> + DPLLA_UNSPEC,
>> + DPLLA_DEVICE_ID,
>> + DPLLA_DEVICE_NAME,
>> + DPLLA_SOURCE,
>> + DPLLA_SOURCE_ID,
>
> "ID" sounds a bit odd. I think "index" would be more suitable.
>
I just wanted to have shorter names to better fit 80 columns per line restrictions.
>
>> + DPLLA_SOURCE_TYPE,
>> + DPLLA_SOURCE_SUPPORTED,
>> + DPLLA_OUTPUT,
>> + DPLLA_OUTPUT_ID,
>> + DPLLA_OUTPUT_TYPE,
>> + DPLLA_OUTPUT_SUPPORTED,
>> + DPLLA_STATUS,
>> + DPLLA_TEMP,
>> + DPLLA_LOCK_STATUS,
>> + DPLLA_FLAGS,
>> +
>> + __DPLLA_MAX,
>> +};
>> +#define DPLLA_MAX (__DPLLA_MAX - 1)
>> +
>> +/* DPLL status provides information of device status */
>> +enum dpll_genl_status {
>> + DPLL_STATUS_NONE,
>> + DPLL_STATUS_CALIBRATING,
>> + DPLL_STATUS_LOCKED,
>> +
>> + __DPLL_STATUS_MAX,
>> +};
>> +#define DPLL_STATUS_MAX (__DPLL_STATUS_MAX - 1)
>> +
>> +/* DPLL signal types used as source or as output */
>> +enum dpll_genl_signal_type {
>> + DPLL_TYPE_EXT_1PPS,
>> + DPLL_TYPE_EXT_10MHZ,
>> + DPLL_TYPE_SYNCE_ETH_PORT,
>> + DPLL_TYPE_INT_OSCILLATOR,
>> + DPLL_TYPE_GNSS,
>> +
>> + __DPLL_TYPE_MAX,
>> +};
>> +#define DPLL_TYPE_MAX (__DPLL_TYPE_MAX - 1)
>> +
>> +/* DPLL lock status provides information of source used to lock the device */
>> +enum dpll_genl_lock_status {
>> + DPLL_LOCK_STATUS_UNLOCKED,
>> + DPLL_LOCK_STATUS_EXT_1PPS,
>> + DPLL_LOCK_STATUS_EXT_10MHZ,
>> + DPLL_LOCK_STATUS_SYNCE,
>> + DPLL_LOCK_STATUS_INT_OSCILLATOR,
>> + DPLL_LOCK_STATUS_GNSS,
>
> I find it a bit odd and redundant to have lock status here as a separate
> enum. You have a souce selected (either autoselected or manualy
> selected). Then the status is either:
> "UNLOCKED"
> "LOCKED"
> "HOLDOVER"
>
> Or something similar. The point is, don't have the "source type" as a
> part of lock status.
>
Yes, it's a very good idea, I was thinking about the same, but it didn't end up
in the code.
>
>> +
>> + __DPLL_LOCK_STATUS_MAX,
>> +};
>> +#define DPLL_LOCK_STATUS_MAX (__DPLL_LOCK_STATUS_MAX - 1)
>> +
>> +/* Events of dpll_genl_family */
>> +enum dpll_genl_event {
>> + DPLL_EVENT_UNSPEC,
>> + DPLL_EVENT_DEVICE_CREATE, /* DPLL device creation */
>> + DPLL_EVENT_DEVICE_DELETE, /* DPLL device deletion */
>> + DPLL_EVENT_STATUS_LOCKED, /* DPLL device locked to source */
>> + DPLL_EVENT_STATUS_UNLOCKED, /* DPLL device freerun */
>
> Again, redundant I belive. There should be one event,
> inside the message there should be and ATTR of the lock state.
>
> Also, I believe there are 2 options:
> 1) follow the existing netlink models and have:
> DPLL_EVENT_DEVICE_NEW
> - sent for new device
> - sent for change with device
> DPLL_EVENT_DEVICE_DEL
> - sent for removed device
> 2)
> DPLL_EVENT_DEVICE_NEW
> - sent for new device
> DPLL_EVENT_DEVICE_DEL
> - sent for removed device
> DPLL_EVENT_DEVICE_CHANGE
> - sent for change with device
>
> Bot options work fine I belive. The point is, you don't want to have
> "cmd" per one attr change. Changed the device, attrs are passed in one
> message.
>
I will try to change events this way, thanks.
>
>
>> + DPLL_EVENT_SOURCE_CHANGE, /* DPLL device source changed */
>
> Similar here, source of device changed, should be just one attr in
> device message, see above.
>
The WIP right now is going the way to have separate objects representing pins,
which could be used as source/output or even mux device. And the whole part
about sources and outputs is going to change. But I'll keep in mind that
suggestion while implementing this new way.
>
>> + DPLL_EVENT_OUTPUT_CHANGE, /* DPLL device output changed */
>
>
>> +
>> + __DPLL_EVENT_MAX,
>> +};
>> +#define DPLL_EVENT_MAX (__DPLL_EVENT_MAX - 1)
>> +
>> +/* Commands supported by the dpll_genl_family */
>> +enum dpll_genl_cmd {
>> + DPLL_CMD_UNSPEC,
>> + DPLL_CMD_DEVICE_GET, /* List of DPLL devices id */
>> + DPLL_CMD_SET_SOURCE_TYPE, /* Set the DPLL device source type */
>
> This is confusing, you select "source", not "type".
As I said, we can have different types for one source.
>
> Please be consistent in naming:
> DPLL_CMD_*_GET/SET
Sure
>
> Also, to be consistend with other netlink interfaces, we don't need
> cmd per action, rather there should be OBJ_GET (can dump) and OBJ_SET
> commands, like this:
>
> DPLL_CMD_DEVICE_GET (can dump all present devices)
> ATTR_SOURCE_SELECT_MODE (current one)
> ATTR_SOURCE_INDEX (currect one)
> DPLL_CMD_DEVICE_SET
> ATTR_SOURCE_INDEX (to set)
> DPLL_CMD_DEVICE_SET
> ATTR_SOURCE_SELECT_MODE (to set)
>
I believe we have to provide all the possible (or suitable) attributes in SET
command too, to fully configure the source by one command only, right?
And I also think we have to provide special attribute to show which source is
actually used to sync to, like ATTR_CONNECTED.
>
>> + DPLL_CMD_SET_OUTPUT_TYPE, /* Set the DPLL device output type */
>
> Similar to what I suggested for "source", we should
> enable to select the OUTPUT, the type should be static. Also instead,
> I belive we should have a list of outputs and basically just allow
> enable/disable individual outputs:
> DPLL_CMD_OUTPUT_GET (can dump the list of available outputs)
> ATTR_ENABLED (or "CONNECTED" or whatever) of type bool
> DPLL_CMD_OUTPUT_SET
> ATTR_ENABLED (or "CONNECTED" or whatever) of type bool
>
> This is suitable for SyncE for example, when you have multiple netdev ports
> that are connected as "outputs", you can enable exactly those you you want.
> Example:
> # To list the available outputs:
> -> DPLL_CMD_OUTPUT_GET - dump
> ATTR_DEVICE_ID X
>
> <- DPLL_CMD_OUTPUT_GET
> ATTR_DEVICE_ID X
> ATTR_OUTPUT_INDEX 0
> ATTR_OUTPUT_TYPE SYNCE_ETH_PORT
> ATTR_OUTPUT_NETDEV_IFINDEX 20
> ATTR_OUTPUT_ENABLED 0
>
> ATTR_DEVICE_ID X
> ATTR_OUTPUT_INDEX 1
> ATTR_OUTPUT_TYPE SYNCE_ETH_PORT
> ATTR_OUTPUT_NETDEV_IFINDEX 30
> ATTR_OUTPUT_ENABLED 0
>
> # Now enable output with index 0
> -> DPLL_CMD_OUTPUT_SET
> ATTR_DEVICE_ID X
> ATTR_OUTPUT_INDEX 0
> ATTR_OUTPUT_ENABLED 1
>
Well, in case when we have flexible outputs, we can provide this information as
special type, like DPLL_TYPE_DISABLED. Having other types configured on the
output will assume that the OUTPUT is enabled.
>
>> +
>> + __DPLL_CMD_MAX,
>> +};
>> +#define DPLL_CMD_MAX (__DPLL_CMD_MAX - 1)
>> +
>> +#endif /* _UAPI_LINUX_DPLL_H */
>> --
>> 2.27.0
>>
Thanks for actionable feedback, I hope to prepare new version soon with all the
comments and suggestions addressed!
More information about the linux-arm-kernel
mailing list