| Message ID | 1593516846-28189-1-git-send-email-vasundhara-v.volam@broadcom.com |
|---|---|
| State | New |
| Headers | show |
| Series | [RFC,v2,net-next] devlink: Add reset subcommand. | expand |
On 7/1/2020 2:47 AM, Jiri Pirko wrote: > Wed, Jul 01, 2020 at 11:25:50AM CEST, vasundhara-v.volam@broadcom.com wrote: >> On Wed, Jul 1, 2020 at 11:21 AM Jiri Pirko <jiri@resnulli.us> wrote: >>> >>> Tue, Jun 30, 2020 at 05:15:18PM CEST, vasundhara-v.volam@broadcom.com wrote: >>>> On Tue, Jun 30, 2020 at 6:23 PM Jiri Pirko <jiri@resnulli.us> wrote: >>>>> >>>>> Tue, Jun 30, 2020 at 01:34:06PM CEST, vasundhara-v.volam@broadcom.com wrote: >>>>>> Advanced NICs support live reset of some of the hardware >>>>>> components, that resets the device immediately with all the >>>>>> host drivers loaded. >>>>>> >>>>>> Add devlink reset subcommand to support live and deferred modes >>>>>> of reset. It allows to reset the hardware components of the >>>>>> entire device and supports the following fields: >>>>>> >>>>>> component: >>>>>> ---------- >>>>>> 1. MGMT : Management processor. >>>>>> 2. DMA : DMA engine. >>>>>> 3. RAM : RAM shared between multiple components. >>>>>> 4. AP : Application processor. >>>>>> 5. ROCE : RoCE management processor. >>>>>> 6. All : All possible components. >>>>>> >>>>>> Drivers are allowed to reset only a subset of requested components. >>>>> >>>>> I don't understand why would user ever want to do this. He does not care >>>>> about some magic hw entities. He just expects the hw to work. I don't >>>>> undestand the purpose of exposing something like this. Could you please >>>>> explain in details? Thanks! >>>>> >>>> If a user requests multiple components and if the driver is only able >>>> to honor a subset, the driver will return the components unset which >>>> it is able to reset. For example, if a user requests MGMT, RAM and >>>> ROCE components to be reset and driver resets only MGMT and ROCE. >>>> Driver will unset only MGMT and ROCE bits and notifies the user that >>>> RAM is not reset. >>>> >>>> This will be useful for drivers to reset only a subset of components >>>> requested instead of returning error or silently doing only a subset >>>> of components. >>>> >>>> Also, this will be helpful as user will not know the components >>>> supported by different vendors. >>> >>> Your reply does not seem to be related to my question :/ >> I thought that you were referring to: "Drivers are allowed to reset >> only a subset of requested components." >> >> or were you referring to components? If yes, the user can select the >> components that he wants to go for reset. This will be useful in the >> case where, if the user flashed only a certain component and he wants >> to reset that particular component. For example, in the case of SOC >> there are 2 components: MGMT and AP. If a user flashes only >> application processor, he can choose to reset only application >> processor. > > We already have notion of "a component" in "devlink dev flash". I think > that the reset component name should be in-sync with the flash. > Right. We should re-use the component names from devlink dev info here (just as we do in devlink dev flash). > Thinking about it a bit more, we can extend the flash command by "reset" > attribute that would indicate use wants to do flash&reset right away. > If we add this to reload I'm not sure it's necessary. The devlink application could be configured to request a reload after the update completes. > Also, thinking how this all aligns with "devlink dev reload" which we > currently have. The purpose of it is to re-instantiate driver instances, > but in case of mlxsw it means friggering FW reset as well. > > Moshe (cced) is now working on "devlink dev reload" extension that would > allow user to ask for a certain level of reload: driver instances only, > fw reset too, live fw patching, etc. Have patches for this been posted at all? > > Not sure how this overlaps with your intentions. I think it would be > great to see Moshe's RFC here as well so we can aligh the efforts. > Yes.
On Fri, 10 Jul 2020 11:16:51 -0700 Jacob Keller wrote: > > We already have notion of "a component" in "devlink dev flash". I think > > that the reset component name should be in-sync with the flash. > > Right. We should re-use the component names from devlink dev info here > (just as we do in devlink dev flash). Let's remember that the SW did not eat the entire world just yet ;) There are still HW components which don't have FW (and therefore FW component to flash) that folks may want to reset.. that's what the ethtool reset API was cut to.
On 7/10/2020 1:39 PM, Jakub Kicinski wrote: > On Fri, 10 Jul 2020 11:16:51 -0700 Jacob Keller wrote: >>> We already have notion of "a component" in "devlink dev flash". I think >>> that the reset component name should be in-sync with the flash. >> >> Right. We should re-use the component names from devlink dev info here >> (just as we do in devlink dev flash). > > Let's remember that the SW did not eat the entire world just yet ;) > > There are still HW components which don't have FW (and therefore FW > component to flash) that folks may want to reset.. that's what the > ethtool reset API was cut to. > True enough. I was thinking only in terms of resetting to load new firmware, which is a different but related problem. Thanks, Jake
On Wed, Jul 1, 2020 at 3:17 PM Jiri Pirko <jiri@resnulli.us> wrote: > > Wed, Jul 01, 2020 at 11:25:50AM CEST, vasundhara-v.volam@broadcom.com wrote: > >On Wed, Jul 1, 2020 at 11:21 AM Jiri Pirko <jiri@resnulli.us> wrote: > >> > >> Tue, Jun 30, 2020 at 05:15:18PM CEST, vasundhara-v.volam@broadcom.com wrote: > >> >On Tue, Jun 30, 2020 at 6:23 PM Jiri Pirko <jiri@resnulli.us> wrote: > >> >> > >> >> Tue, Jun 30, 2020 at 01:34:06PM CEST, vasundhara-v.volam@broadcom.com wrote: > >> >> >Advanced NICs support live reset of some of the hardware > >> >> >components, that resets the device immediately with all the > >> >> >host drivers loaded. > >> >> > > >> >> >Add devlink reset subcommand to support live and deferred modes > >> >> >of reset. It allows to reset the hardware components of the > >> >> >entire device and supports the following fields: > >> >> > > >> >> >component: > >> >> >---------- > >> >> >1. MGMT : Management processor. > >> >> >2. DMA : DMA engine. > >> >> >3. RAM : RAM shared between multiple components. > >> >> >4. AP : Application processor. > >> >> >5. ROCE : RoCE management processor. > >> >> >6. All : All possible components. > >> >> > > >> >> >Drivers are allowed to reset only a subset of requested components. > >> >> > >> >> I don't understand why would user ever want to do this. He does not care > >> >> about some magic hw entities. He just expects the hw to work. I don't > >> >> undestand the purpose of exposing something like this. Could you please > >> >> explain in details? Thanks! > >> >> > >> >If a user requests multiple components and if the driver is only able > >> >to honor a subset, the driver will return the components unset which > >> >it is able to reset. For example, if a user requests MGMT, RAM and > >> >ROCE components to be reset and driver resets only MGMT and ROCE. > >> >Driver will unset only MGMT and ROCE bits and notifies the user that > >> >RAM is not reset. > >> > > >> >This will be useful for drivers to reset only a subset of components > >> >requested instead of returning error or silently doing only a subset > >> >of components. > >> > > >> >Also, this will be helpful as user will not know the components > >> >supported by different vendors. > >> > >> Your reply does not seem to be related to my question :/ > >I thought that you were referring to: "Drivers are allowed to reset > >only a subset of requested components." > > > >or were you referring to components? If yes, the user can select the > >components that he wants to go for reset. This will be useful in the > >case where, if the user flashed only a certain component and he wants > >to reset that particular component. For example, in the case of SOC > >there are 2 components: MGMT and AP. If a user flashes only > >application processor, he can choose to reset only application > >processor. > > We already have notion of "a component" in "devlink dev flash". I think > that the reset component name should be in-sync with the flash. > > Thinking about it a bit more, we can extend the flash command by "reset" > attribute that would indicate use wants to do flash&reset right away. > > Also, thinking how this all aligns with "devlink dev reload" which we > currently have. The purpose of it is to re-instantiate driver instances, > but in case of mlxsw it means friggering FW reset as well. > > Moshe (cced) is now working on "devlink dev reload" extension that would > allow user to ask for a certain level of reload: driver instances only, > fw reset too, live fw patching, etc. > > Not sure how this overlaps with your intentions. I think it would be > great to see Moshe's RFC here as well so we can aligh the efforts. Are the patches posted yet? > > > > > >> > >> > >> > > >> >Thanks, > >> >Vasundhara > >> > > >> >> > >> >> > > >> >> >width: > >> >> >------ > >> >> >1. single - single host. > >> >> >2. multi - Multi host. > >> >> > > >> >> >mode: > >> >> >----- > >> >> >1. deferred - Reset will happen after unloading all the host drivers > >> >> > on the device. This is be default reset type, if user > >> >> > does not specify the type. > >> >> >2. live - Reset will happen immediately with all host drivers loaded > >> >> > in real time. If the live reset is not supported, driver > >> >> > will return the error. > >> >> > > >> >> >This patch is a proposal in continuation to discussion to the > >> >> >following thread: > >> >> > > >> >> >"[PATCH v3 net-next 0/6] bnxt_en: Add 'enable_live_dev_reset' and 'allow_live_dev_reset' generic devlink params." > >> >> > > >> >> >and here is the URL to the patch series: > >> >> > > >> >> >https://patchwork.ozlabs.org/project/netdev/list/?series=180426&state=* > >> >> > > >> >> >If the proposal looks good, I will re-send the whole patchset > >> >> >including devlink changes and driver usage. > >> >> > > >> >> >Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com> > >> >> >Reviewed-by: Michael Chan <michael.chan@broadcom.com> > >> >> >--- > >> >> >v2: > >> >> >- Switch RAM and AP component definitions. > >> >> >- Remove IRQ, FILTER, OFFLOAD, MAC, PHY components as they are port > >> >> >specific components. > >> >> >- Rename function to host in width parameter. > >> >> >--- > >> >> > Documentation/networking/devlink/devlink-reset.rst | 50 +++++++++++++ > >> >> > include/net/devlink.h | 2 + > >> >> > include/uapi/linux/devlink.h | 46 ++++++++++++ > >> >> > net/core/devlink.c | 85 ++++++++++++++++++++++ > >> >> > 4 files changed, 183 insertions(+) > >> >> > create mode 100644 Documentation/networking/devlink/devlink-reset.rst > >> >> > > >> >> >diff --git a/Documentation/networking/devlink/devlink-reset.rst b/Documentation/networking/devlink/devlink-reset.rst > >> >> >new file mode 100644 > >> >> >index 0000000..652800d > >> >> >--- /dev/null > >> >> >+++ b/Documentation/networking/devlink/devlink-reset.rst > >> >> >@@ -0,0 +1,50 @@ > >> >> >+.. SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) > >> >> >+ > >> >> >+.. _devlink_reset: > >> >> >+ > >> >> >+============= > >> >> >+Devlink reset > >> >> >+============= > >> >> >+ > >> >> >+The ``devlink-reset`` API allows reset the hardware components of the device. After the reset, > >> >> >+device loads the pending updated firmware image. > >> >> >+Example use:: > >> >> >+ > >> >> >+ $ devlink dev reset pci/0000:05:00.0 components COMPONENTS > >> >> >+ > >> >> >+Note that user can mention multiple components. > >> >> >+ > >> >> >+================ > >> >> >+Reset components > >> >> >+================ > >> >> >+ > >> >> >+List of available components:: > >> >> >+ > >> >> >+``DEVLINK_RESET_COMP_MGMT`` - Management processor. > >> >> >+``DEVLINK_RESET_COMP_DMA`` - DMA engine. > >> >> >+``DEVLINK_RESET_COMP_RAM`` - RAM shared between multiple components. > >> >> >+``DEVLINK_RESET_COMP_AP`` - Application processor. > >> >> >+``DEVLINK_RESET_COMP_ROCE`` - RoCE management processor. > >> >> >+``DEVLINK_RESET_COMP_ALL`` - All components. > >> >> >+ > >> >> >+=========== > >> >> >+Reset width > >> >> >+=========== > >> >> >+ > >> >> >+List of available widths:: > >> >> >+ > >> >> >+``DEVLINK_RESET_WIDTH_SINGLE`` - Device is used by single dedicated host. > >> >> >+``DEVLINK_RESET_WIDTH_MULTI`` - Device is shared across multiple hosts. > >> >> >+ > >> >> >+Note that if user specifies DEVLINK_RESET_WIDTH_SINGLE in a multi-host environment, driver returns > >> >> >+error if it does not support resetting a single host. > >> >> >+ > >> >> >+=========== > >> >> >+Reset modes > >> >> >+=========== > >> >> >+ > >> >> >+List of available reset modes:: > >> >> >+ > >> >> >+``DEVLINK_RESET_MODE_DEFERRED`` - Reset happens after all host drivers are unloaded on the device. > >> >> >+``DEVLINK_RESET_MODE_LIVE`` - Reset happens immediately, with all loaded host drivers in real > >> >> >+ time. > >> >> >diff --git a/include/net/devlink.h b/include/net/devlink.h > >> >> >index 428f55f..a71c8f5 100644 > >> >> >--- a/include/net/devlink.h > >> >> >+++ b/include/net/devlink.h > >> >> >@@ -1129,6 +1129,8 @@ struct devlink_ops { > >> >> > int (*port_function_hw_addr_set)(struct devlink *devlink, struct devlink_port *port, > >> >> > const u8 *hw_addr, int hw_addr_len, > >> >> > struct netlink_ext_ack *extack); > >> >> >+ int (*reset)(struct devlink *devlink, u32 *components, u8 width, u8 mode, > >> >> >+ struct netlink_ext_ack *extack); > >> >> > }; > >> >> > > >> >> > static inline void *devlink_priv(struct devlink *devlink) > >> >> >diff --git a/include/uapi/linux/devlink.h b/include/uapi/linux/devlink.h > >> >> >index 87c83a8..6f32c00 100644 > >> >> >--- a/include/uapi/linux/devlink.h > >> >> >+++ b/include/uapi/linux/devlink.h > >> >> >@@ -122,6 +122,9 @@ enum devlink_command { > >> >> > DEVLINK_CMD_TRAP_POLICER_NEW, > >> >> > DEVLINK_CMD_TRAP_POLICER_DEL, > >> >> > > >> >> >+ DEVLINK_CMD_RESET, > >> >> >+ DEVLINK_CMD_RESET_STATUS, /* notification only */ > >> >> >+ > >> >> > /* add new commands above here */ > >> >> > __DEVLINK_CMD_MAX, > >> >> > DEVLINK_CMD_MAX = __DEVLINK_CMD_MAX - 1 > >> >> >@@ -265,6 +268,44 @@ enum devlink_trap_type { > >> >> > DEVLINK_TRAP_TYPE_CONTROL, > >> >> > }; > >> >> > > >> >> >+/** > >> >> >+ * enum devlink_reset_component - Reset components. > >> >> >+ * @DEVLINK_RESET_COMP_MGMT: Management processor. > >> >> >+ * @DEVLINK_RESET_COMP_DMA: DMA engine. > >> >> >+ * @DEVLINK_RESET_COMP_RAM: RAM shared between multiple components. > >> >> >+ * @DEVLINK_RESET_COMP_AP: Application processor. > >> >> >+ * @DEVLINK_RESET_COMP_ROCE: RoCE management processor. > >> >> >+ * @DEVLINK_RESET_COMP_ALL: All components. > >> >> >+ */ > >> >> >+enum devlink_reset_component { > >> >> >+ DEVLINK_RESET_COMP_MGMT = (1 << 0), > >> >> >+ DEVLINK_RESET_COMP_DMA = (1 << 1), > >> >> >+ DEVLINK_RESET_COMP_RAM = (1 << 2), > >> >> >+ DEVLINK_RESET_COMP_AP = (1 << 3), > >> >> >+ DEVLINK_RESET_COMP_ROCE = (1 << 4), > >> >> >+ DEVLINK_RESET_COMP_ALL = 0xffffffff, > >> >> >+}; > >> >> >+ > >> >> >+/** > >> >> >+ * enum devlink_reset_width - Number of hosts effected by reset. > >> >> >+ * @DEVLINK_RESET_WIDTH_SINGLE: Device is used by single dedicated host. > >> >> >+ * @DEVLINK_RESET_WIDTH_MULTI: Device is shared across multiple hosts. > >> >> >+ */ > >> >> >+enum devlink_reset_width { > >> >> >+ DEVLINK_RESET_WIDTH_SINGLE = 0, > >> >> >+ DEVLINK_RESET_WIDTH_MULTI = 1, > >> >> >+}; > >> >> >+ > >> >> >+/** > >> >> >+ * enum devlink_reset_mode - Modes of reset. > >> >> >+ * @DEVLINK_RESET_MODE_DEFERRED: Reset will happen after host drivers are unloaded. > >> >> >+ * @DEVLINK_RESET_MODE_LIVE: All host drivers also will be reset without reloading manually. > >> >> >+ */ > >> >> >+enum devlink_reset_mode { > >> >> >+ DEVLINK_RESET_MODE_DEFERRED = 0, > >> >> >+ DEVLINK_RESET_MODE_LIVE = 1, > >> >> >+}; > >> >> >+ > >> >> > enum { > >> >> > /* Trap can report input port as metadata */ > >> >> > DEVLINK_ATTR_TRAP_METADATA_TYPE_IN_PORT, > >> >> >@@ -455,6 +496,11 @@ enum devlink_attr { > >> >> > > >> >> > DEVLINK_ATTR_INFO_BOARD_SERIAL_NUMBER, /* string */ > >> >> > > >> >> >+ DEVLINK_ATTR_RESET_COMPONENTS, /* u32 */ > >> >> >+ DEVLINK_ATTR_RESET_WIDTH, /* u8 */ > >> >> >+ DEVLINK_ATTR_RESET_MODE, /* u8 */ > >> >> >+ DEVLINK_ATTR_RESET_STATUS_MSG, /* string */ > >> >> >+ > >> >> > /* add new attributes above here, update the policy in devlink.c */ > >> >> > > >> >> > __DEVLINK_ATTR_MAX, > >> >> >diff --git a/net/core/devlink.c b/net/core/devlink.c > >> >> >index 6ae3680..c0eebc5 100644 > >> >> >--- a/net/core/devlink.c > >> >> >+++ b/net/core/devlink.c > >> >> >@@ -6797,6 +6797,82 @@ static int devlink_nl_cmd_trap_policer_set_doit(struct sk_buff *skb, > >> >> > return devlink_trap_policer_set(devlink, policer_item, info); > >> >> > } > >> >> > > >> >> >+static int devlink_nl_reset_fill(struct sk_buff *msg, struct devlink *devlink, > >> >> >+ const char *status_msg, u32 components) > >> >> >+{ > >> >> >+ void *hdr; > >> >> >+ > >> >> >+ hdr = genlmsg_put(msg, 0, 0, &devlink_nl_family, 0, DEVLINK_CMD_RESET_STATUS); > >> >> >+ if (!hdr) > >> >> >+ return -EMSGSIZE; > >> >> >+ > >> >> >+ if (devlink_nl_put_handle(msg, devlink)) > >> >> >+ goto nla_put_failure; > >> >> >+ > >> >> >+ if (status_msg && nla_put_string(msg, DEVLINK_ATTR_RESET_STATUS_MSG, status_msg)) > >> >> >+ goto nla_put_failure; > >> >> >+ > >> >> >+ if (nla_put_u32(msg, DEVLINK_ATTR_RESET_COMPONENTS, components)) > >> >> >+ goto nla_put_failure; > >> >> >+ > >> >> >+ genlmsg_end(msg, hdr); > >> >> >+ return 0; > >> >> >+ > >> >> >+nla_put_failure: > >> >> >+ genlmsg_cancel(msg, hdr); > >> >> >+ return -EMSGSIZE; > >> >> >+} > >> >> >+ > >> >> >+static void __devlink_reset_notify(struct devlink *devlink, const char *status_msg, u32 components) > >> >> >+{ > >> >> >+ struct sk_buff *msg; > >> >> >+ int err; > >> >> >+ > >> >> >+ msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); > >> >> >+ if (!msg) > >> >> >+ return; > >> >> >+ > >> >> >+ err = devlink_nl_reset_fill(msg, devlink, status_msg, components); > >> >> >+ if (err) > >> >> >+ goto out; > >> >> >+ > >> >> >+ genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink), msg, 0, > >> >> >+ DEVLINK_MCGRP_CONFIG, GFP_KERNEL); > >> >> >+ return; > >> >> >+ > >> >> >+out: > >> >> >+ nlmsg_free(msg); > >> >> >+} > >> >> >+ > >> >> >+static int devlink_nl_cmd_reset(struct sk_buff *skb, struct genl_info *info) > >> >> >+{ > >> >> >+ struct devlink *devlink = info->user_ptr[0]; > >> >> >+ u32 components, req_comps; > >> >> >+ struct nlattr *nla_type; > >> >> >+ u8 width, mode; > >> >> >+ int err; > >> >> >+ > >> >> >+ if (!devlink->ops->reset) > >> >> >+ return -EOPNOTSUPP; > >> >> >+ > >> >> >+ if (!info->attrs[DEVLINK_ATTR_RESET_COMPONENTS]) > >> >> >+ return -EINVAL; > >> >> >+ components = nla_get_u32(info->attrs[DEVLINK_ATTR_RESET_COMPONENTS]); > >> >> >+ > >> >> >+ nla_type = info->attrs[DEVLINK_ATTR_RESET_WIDTH]; > >> >> >+ width = nla_type ? nla_get_u8(nla_type) : DEVLINK_RESET_WIDTH_SINGLE; > >> >> >+ > >> >> >+ nla_type = info->attrs[DEVLINK_ATTR_RESET_MODE]; > >> >> >+ mode = nla_type ? nla_get_u8(nla_type) : DEVLINK_RESET_MODE_DEFERRED; > >> >> >+ > >> >> >+ req_comps = components; > >> >> >+ __devlink_reset_notify(devlink, "Reset request", components); > >> >> >+ err = devlink->ops->reset(devlink, &components, width, mode, info->extack); > >> >> >+ __devlink_reset_notify(devlink, "Components reset", req_comps & ~components); > >> >> >+ > >> >> >+ return err; > >> >> >+} > >> >> >+ > >> >> > static const struct nla_policy devlink_nl_policy[DEVLINK_ATTR_MAX + 1] = { > >> >> > [DEVLINK_ATTR_UNSPEC] = { .strict_start_type = > >> >> > DEVLINK_ATTR_TRAP_POLICER_ID }, > >> >> >@@ -6842,6 +6918,9 @@ static int devlink_nl_cmd_trap_policer_set_doit(struct sk_buff *skb, > >> >> > [DEVLINK_ATTR_TRAP_POLICER_RATE] = { .type = NLA_U64 }, > >> >> > [DEVLINK_ATTR_TRAP_POLICER_BURST] = { .type = NLA_U64 }, > >> >> > [DEVLINK_ATTR_PORT_FUNCTION] = { .type = NLA_NESTED }, > >> >> >+ [DEVLINK_ATTR_RESET_COMPONENTS] = { .type = NLA_U32 }, > >> >> >+ [DEVLINK_ATTR_RESET_WIDTH] = { .type = NLA_U8 }, > >> >> >+ [DEVLINK_ATTR_RESET_MODE] = { .type = NLA_U8 }, > >> >> > }; > >> >> > > >> >> > static const struct genl_ops devlink_nl_ops[] = { > >> >> >@@ -7190,6 +7269,12 @@ static int devlink_nl_cmd_trap_policer_set_doit(struct sk_buff *skb, > >> >> > .flags = GENL_ADMIN_PERM, > >> >> > .internal_flags = DEVLINK_NL_FLAG_NEED_DEVLINK, > >> >> > }, > >> >> >+ { > >> >> >+ .cmd = DEVLINK_CMD_RESET, > >> >> >+ .doit = devlink_nl_cmd_reset, > >> >> >+ .flags = GENL_ADMIN_PERM, > >> >> >+ .internal_flags = DEVLINK_NL_FLAG_NEED_DEVLINK, > >> >> >+ }, > >> >> > }; > >> >> > > >> >> > static struct genl_family devlink_nl_family __ro_after_init = { > >> >> >-- > >> >> >1.8.3.1 > >> >> >
Tue, Jul 21, 2020 at 11:51:21AM CEST, vasundhara-v.volam@broadcom.com wrote: >On Wed, Jul 1, 2020 at 3:17 PM Jiri Pirko <jiri@resnulli.us> wrote: >> >> Wed, Jul 01, 2020 at 11:25:50AM CEST, vasundhara-v.volam@broadcom.com wrote: >> >On Wed, Jul 1, 2020 at 11:21 AM Jiri Pirko <jiri@resnulli.us> wrote: >> >> >> >> Tue, Jun 30, 2020 at 05:15:18PM CEST, vasundhara-v.volam@broadcom.com wrote: >> >> >On Tue, Jun 30, 2020 at 6:23 PM Jiri Pirko <jiri@resnulli.us> wrote: >> >> >> >> >> >> Tue, Jun 30, 2020 at 01:34:06PM CEST, vasundhara-v.volam@broadcom.com wrote: >> >> >> >Advanced NICs support live reset of some of the hardware >> >> >> >components, that resets the device immediately with all the >> >> >> >host drivers loaded. >> >> >> > >> >> >> >Add devlink reset subcommand to support live and deferred modes >> >> >> >of reset. It allows to reset the hardware components of the >> >> >> >entire device and supports the following fields: >> >> >> > >> >> >> >component: >> >> >> >---------- >> >> >> >1. MGMT : Management processor. >> >> >> >2. DMA : DMA engine. >> >> >> >3. RAM : RAM shared between multiple components. >> >> >> >4. AP : Application processor. >> >> >> >5. ROCE : RoCE management processor. >> >> >> >6. All : All possible components. >> >> >> > >> >> >> >Drivers are allowed to reset only a subset of requested components. >> >> >> >> >> >> I don't understand why would user ever want to do this. He does not care >> >> >> about some magic hw entities. He just expects the hw to work. I don't >> >> >> undestand the purpose of exposing something like this. Could you please >> >> >> explain in details? Thanks! >> >> >> >> >> >If a user requests multiple components and if the driver is only able >> >> >to honor a subset, the driver will return the components unset which >> >> >it is able to reset. For example, if a user requests MGMT, RAM and >> >> >ROCE components to be reset and driver resets only MGMT and ROCE. >> >> >Driver will unset only MGMT and ROCE bits and notifies the user that >> >> >RAM is not reset. >> >> > >> >> >This will be useful for drivers to reset only a subset of components >> >> >requested instead of returning error or silently doing only a subset >> >> >of components. >> >> > >> >> >Also, this will be helpful as user will not know the components >> >> >supported by different vendors. >> >> >> >> Your reply does not seem to be related to my question :/ >> >I thought that you were referring to: "Drivers are allowed to reset >> >only a subset of requested components." >> > >> >or were you referring to components? If yes, the user can select the >> >components that he wants to go for reset. This will be useful in the >> >case where, if the user flashed only a certain component and he wants >> >to reset that particular component. For example, in the case of SOC >> >there are 2 components: MGMT and AP. If a user flashes only >> >application processor, he can choose to reset only application >> >processor. >> >> We already have notion of "a component" in "devlink dev flash". I think >> that the reset component name should be in-sync with the flash. >> >> Thinking about it a bit more, we can extend the flash command by "reset" >> attribute that would indicate use wants to do flash&reset right away. >> >> Also, thinking how this all aligns with "devlink dev reload" which we >> currently have. The purpose of it is to re-instantiate driver instances, >> but in case of mlxsw it means friggering FW reset as well. >> >> Moshe (cced) is now working on "devlink dev reload" extension that would >> allow user to ask for a certain level of reload: driver instances only, >> fw reset too, live fw patching, etc. >> >> Not sure how this overlaps with your intentions. I think it would be >> great to see Moshe's RFC here as well so we can aligh the efforts. >Are the patches posted yet? I don't think so. Moshe?
On 7/21/2020 3:19 PM, Jiri Pirko wrote: > Tue, Jul 21, 2020 at 11:51:21AM CEST, vasundhara-v.volam@broadcom.com wrote: >> On Wed, Jul 1, 2020 at 3:17 PM Jiri Pirko <jiri@resnulli.us> wrote: >>> Wed, Jul 01, 2020 at 11:25:50AM CEST, vasundhara-v.volam@broadcom.com wrote: >>>> On Wed, Jul 1, 2020 at 11:21 AM Jiri Pirko <jiri@resnulli.us> wrote: >>>>> Tue, Jun 30, 2020 at 05:15:18PM CEST, vasundhara-v.volam@broadcom.com wrote: >>>>>> On Tue, Jun 30, 2020 at 6:23 PM Jiri Pirko <jiri@resnulli.us> wrote: >>>>>>> Tue, Jun 30, 2020 at 01:34:06PM CEST, vasundhara-v.volam@broadcom.com wrote: >>>>>>>> Advanced NICs support live reset of some of the hardware >>>>>>>> components, that resets the device immediately with all the >>>>>>>> host drivers loaded. >>>>>>>> >>>>>>>> Add devlink reset subcommand to support live and deferred modes >>>>>>>> of reset. It allows to reset the hardware components of the >>>>>>>> entire device and supports the following fields: >>>>>>>> >>>>>>>> component: >>>>>>>> ---------- >>>>>>>> 1. MGMT : Management processor. >>>>>>>> 2. DMA : DMA engine. >>>>>>>> 3. RAM : RAM shared between multiple components. >>>>>>>> 4. AP : Application processor. >>>>>>>> 5. ROCE : RoCE management processor. >>>>>>>> 6. All : All possible components. >>>>>>>> >>>>>>>> Drivers are allowed to reset only a subset of requested components. >>>>>>> I don't understand why would user ever want to do this. He does not care >>>>>>> about some magic hw entities. He just expects the hw to work. I don't >>>>>>> undestand the purpose of exposing something like this. Could you please >>>>>>> explain in details? Thanks! >>>>>>> >>>>>> If a user requests multiple components and if the driver is only able >>>>>> to honor a subset, the driver will return the components unset which >>>>>> it is able to reset. For example, if a user requests MGMT, RAM and >>>>>> ROCE components to be reset and driver resets only MGMT and ROCE. >>>>>> Driver will unset only MGMT and ROCE bits and notifies the user that >>>>>> RAM is not reset. >>>>>> >>>>>> This will be useful for drivers to reset only a subset of components >>>>>> requested instead of returning error or silently doing only a subset >>>>>> of components. >>>>>> >>>>>> Also, this will be helpful as user will not know the components >>>>>> supported by different vendors. >>>>> Your reply does not seem to be related to my question :/ >>>> I thought that you were referring to: "Drivers are allowed to reset >>>> only a subset of requested components." >>>> >>>> or were you referring to components? If yes, the user can select the >>>> components that he wants to go for reset. This will be useful in the >>>> case where, if the user flashed only a certain component and he wants >>>> to reset that particular component. For example, in the case of SOC >>>> there are 2 components: MGMT and AP. If a user flashes only >>>> application processor, he can choose to reset only application >>>> processor. >>> We already have notion of "a component" in "devlink dev flash". I think >>> that the reset component name should be in-sync with the flash. >>> >>> Thinking about it a bit more, we can extend the flash command by "reset" >>> attribute that would indicate use wants to do flash&reset right away. >>> >>> Also, thinking how this all aligns with "devlink dev reload" which we >>> currently have. The purpose of it is to re-instantiate driver instances, >>> but in case of mlxsw it means friggering FW reset as well. >>> >>> Moshe (cced) is now working on "devlink dev reload" extension that would >>> allow user to ask for a certain level of reload: driver instances only, >>> fw reset too, live fw patching, etc. >>> >>> Not sure how this overlaps with your intentions. I think it would be >>> great to see Moshe's RFC here as well so we can aligh the efforts. >> Are the patches posted yet? > I don't think so. > > Moshe? Not yet, still in internal review. If won't pass by EOW I will send part of it as RFC.
diff --git a/Documentation/networking/devlink/devlink-reset.rst b/Documentation/networking/devlink/devlink-reset.rst new file mode 100644 index 0000000..652800d --- /dev/null +++ b/Documentation/networking/devlink/devlink-reset.rst @@ -0,0 +1,50 @@ +.. SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) + +.. _devlink_reset: + +============= +Devlink reset +============= + +The ``devlink-reset`` API allows reset the hardware components of the device. After the reset, +device loads the pending updated firmware image. +Example use:: + + $ devlink dev reset pci/0000:05:00.0 components COMPONENTS + +Note that user can mention multiple components. + +================ +Reset components +================ + +List of available components:: + +``DEVLINK_RESET_COMP_MGMT`` - Management processor. +``DEVLINK_RESET_COMP_DMA`` - DMA engine. +``DEVLINK_RESET_COMP_RAM`` - RAM shared between multiple components. +``DEVLINK_RESET_COMP_AP`` - Application processor. +``DEVLINK_RESET_COMP_ROCE`` - RoCE management processor. +``DEVLINK_RESET_COMP_ALL`` - All components. + +=========== +Reset width +=========== + +List of available widths:: + +``DEVLINK_RESET_WIDTH_SINGLE`` - Device is used by single dedicated host. +``DEVLINK_RESET_WIDTH_MULTI`` - Device is shared across multiple hosts. + +Note that if user specifies DEVLINK_RESET_WIDTH_SINGLE in a multi-host environment, driver returns +error if it does not support resetting a single host. + +=========== +Reset modes +=========== + +List of available reset modes:: + +``DEVLINK_RESET_MODE_DEFERRED`` - Reset happens after all host drivers are unloaded on the device. +``DEVLINK_RESET_MODE_LIVE`` - Reset happens immediately, with all loaded host drivers in real + time. diff --git a/include/net/devlink.h b/include/net/devlink.h index 428f55f..a71c8f5 100644 --- a/include/net/devlink.h +++ b/include/net/devlink.h @@ -1129,6 +1129,8 @@ struct devlink_ops { int (*port_function_hw_addr_set)(struct devlink *devlink, struct devlink_port *port, const u8 *hw_addr, int hw_addr_len, struct netlink_ext_ack *extack); + int (*reset)(struct devlink *devlink, u32 *components, u8 width, u8 mode, + struct netlink_ext_ack *extack); }; static inline void *devlink_priv(struct devlink *devlink) diff --git a/include/uapi/linux/devlink.h b/include/uapi/linux/devlink.h index 87c83a8..6f32c00 100644 --- a/include/uapi/linux/devlink.h +++ b/include/uapi/linux/devlink.h @@ -122,6 +122,9 @@ enum devlink_command { DEVLINK_CMD_TRAP_POLICER_NEW, DEVLINK_CMD_TRAP_POLICER_DEL, + DEVLINK_CMD_RESET, + DEVLINK_CMD_RESET_STATUS, /* notification only */ + /* add new commands above here */ __DEVLINK_CMD_MAX, DEVLINK_CMD_MAX = __DEVLINK_CMD_MAX - 1 @@ -265,6 +268,44 @@ enum devlink_trap_type { DEVLINK_TRAP_TYPE_CONTROL, }; +/** + * enum devlink_reset_component - Reset components. + * @DEVLINK_RESET_COMP_MGMT: Management processor. + * @DEVLINK_RESET_COMP_DMA: DMA engine. + * @DEVLINK_RESET_COMP_RAM: RAM shared between multiple components. + * @DEVLINK_RESET_COMP_AP: Application processor. + * @DEVLINK_RESET_COMP_ROCE: RoCE management processor. + * @DEVLINK_RESET_COMP_ALL: All components. + */ +enum devlink_reset_component { + DEVLINK_RESET_COMP_MGMT = (1 << 0), + DEVLINK_RESET_COMP_DMA = (1 << 1), + DEVLINK_RESET_COMP_RAM = (1 << 2), + DEVLINK_RESET_COMP_AP = (1 << 3), + DEVLINK_RESET_COMP_ROCE = (1 << 4), + DEVLINK_RESET_COMP_ALL = 0xffffffff, +}; + +/** + * enum devlink_reset_width - Number of hosts effected by reset. + * @DEVLINK_RESET_WIDTH_SINGLE: Device is used by single dedicated host. + * @DEVLINK_RESET_WIDTH_MULTI: Device is shared across multiple hosts. + */ +enum devlink_reset_width { + DEVLINK_RESET_WIDTH_SINGLE = 0, + DEVLINK_RESET_WIDTH_MULTI = 1, +}; + +/** + * enum devlink_reset_mode - Modes of reset. + * @DEVLINK_RESET_MODE_DEFERRED: Reset will happen after host drivers are unloaded. + * @DEVLINK_RESET_MODE_LIVE: All host drivers also will be reset without reloading manually. + */ +enum devlink_reset_mode { + DEVLINK_RESET_MODE_DEFERRED = 0, + DEVLINK_RESET_MODE_LIVE = 1, +}; + enum { /* Trap can report input port as metadata */ DEVLINK_ATTR_TRAP_METADATA_TYPE_IN_PORT, @@ -455,6 +496,11 @@ enum devlink_attr { DEVLINK_ATTR_INFO_BOARD_SERIAL_NUMBER, /* string */ + DEVLINK_ATTR_RESET_COMPONENTS, /* u32 */ + DEVLINK_ATTR_RESET_WIDTH, /* u8 */ + DEVLINK_ATTR_RESET_MODE, /* u8 */ + DEVLINK_ATTR_RESET_STATUS_MSG, /* string */ + /* add new attributes above here, update the policy in devlink.c */ __DEVLINK_ATTR_MAX, diff --git a/net/core/devlink.c b/net/core/devlink.c index 6ae3680..c0eebc5 100644 --- a/net/core/devlink.c +++ b/net/core/devlink.c @@ -6797,6 +6797,82 @@ static int devlink_nl_cmd_trap_policer_set_doit(struct sk_buff *skb, return devlink_trap_policer_set(devlink, policer_item, info); } +static int devlink_nl_reset_fill(struct sk_buff *msg, struct devlink *devlink, + const char *status_msg, u32 components) +{ + void *hdr; + + hdr = genlmsg_put(msg, 0, 0, &devlink_nl_family, 0, DEVLINK_CMD_RESET_STATUS); + if (!hdr) + return -EMSGSIZE; + + if (devlink_nl_put_handle(msg, devlink)) + goto nla_put_failure; + + if (status_msg && nla_put_string(msg, DEVLINK_ATTR_RESET_STATUS_MSG, status_msg)) + goto nla_put_failure; + + if (nla_put_u32(msg, DEVLINK_ATTR_RESET_COMPONENTS, components)) + goto nla_put_failure; + + genlmsg_end(msg, hdr); + return 0; + +nla_put_failure: + genlmsg_cancel(msg, hdr); + return -EMSGSIZE; +} + +static void __devlink_reset_notify(struct devlink *devlink, const char *status_msg, u32 components) +{ + struct sk_buff *msg; + int err; + + msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); + if (!msg) + return; + + err = devlink_nl_reset_fill(msg, devlink, status_msg, components); + if (err) + goto out; + + genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink), msg, 0, + DEVLINK_MCGRP_CONFIG, GFP_KERNEL); + return; + +out: + nlmsg_free(msg); +} + +static int devlink_nl_cmd_reset(struct sk_buff *skb, struct genl_info *info) +{ + struct devlink *devlink = info->user_ptr[0]; + u32 components, req_comps; + struct nlattr *nla_type; + u8 width, mode; + int err; + + if (!devlink->ops->reset) + return -EOPNOTSUPP; + + if (!info->attrs[DEVLINK_ATTR_RESET_COMPONENTS]) + return -EINVAL; + components = nla_get_u32(info->attrs[DEVLINK_ATTR_RESET_COMPONENTS]); + + nla_type = info->attrs[DEVLINK_ATTR_RESET_WIDTH]; + width = nla_type ? nla_get_u8(nla_type) : DEVLINK_RESET_WIDTH_SINGLE; + + nla_type = info->attrs[DEVLINK_ATTR_RESET_MODE]; + mode = nla_type ? nla_get_u8(nla_type) : DEVLINK_RESET_MODE_DEFERRED; + + req_comps = components; + __devlink_reset_notify(devlink, "Reset request", components); + err = devlink->ops->reset(devlink, &components, width, mode, info->extack); + __devlink_reset_notify(devlink, "Components reset", req_comps & ~components); + + return err; +} + static const struct nla_policy devlink_nl_policy[DEVLINK_ATTR_MAX + 1] = { [DEVLINK_ATTR_UNSPEC] = { .strict_start_type = DEVLINK_ATTR_TRAP_POLICER_ID }, @@ -6842,6 +6918,9 @@ static int devlink_nl_cmd_trap_policer_set_doit(struct sk_buff *skb, [DEVLINK_ATTR_TRAP_POLICER_RATE] = { .type = NLA_U64 }, [DEVLINK_ATTR_TRAP_POLICER_BURST] = { .type = NLA_U64 }, [DEVLINK_ATTR_PORT_FUNCTION] = { .type = NLA_NESTED }, + [DEVLINK_ATTR_RESET_COMPONENTS] = { .type = NLA_U32 }, + [DEVLINK_ATTR_RESET_WIDTH] = { .type = NLA_U8 }, + [DEVLINK_ATTR_RESET_MODE] = { .type = NLA_U8 }, }; static const struct genl_ops devlink_nl_ops[] = { @@ -7190,6 +7269,12 @@ static int devlink_nl_cmd_trap_policer_set_doit(struct sk_buff *skb, .flags = GENL_ADMIN_PERM, .internal_flags = DEVLINK_NL_FLAG_NEED_DEVLINK, }, + { + .cmd = DEVLINK_CMD_RESET, + .doit = devlink_nl_cmd_reset, + .flags = GENL_ADMIN_PERM, + .internal_flags = DEVLINK_NL_FLAG_NEED_DEVLINK, + }, }; static struct genl_family devlink_nl_family __ro_after_init = {
Advanced NICs support live reset of some of the hardware components, that resets the device immediately with all the host drivers loaded. Add devlink reset subcommand to support live and deferred modes of reset. It allows to reset the hardware components of the entire device and supports the following fields: component: ---------- 1. MGMT : Management processor. 2. DMA : DMA engine. 3. RAM : RAM shared between multiple components. 4. AP : Application processor. 5. ROCE : RoCE management processor. 6. All : All possible components. Drivers are allowed to reset only a subset of requested components. width: ------ 1. single - single host. 2. multi - Multi host. mode: ----- 1. deferred - Reset will happen after unloading all the host drivers on the device. This is be default reset type, if user does not specify the type. 2. live - Reset will happen immediately with all host drivers loaded in real time. If the live reset is not supported, driver will return the error. This patch is a proposal in continuation to discussion to the following thread: "[PATCH v3 net-next 0/6] bnxt_en: Add 'enable_live_dev_reset' and 'allow_live_dev_reset' generic devlink params." and here is the URL to the patch series: https://patchwork.ozlabs.org/project/netdev/list/?series=180426&state=* If the proposal looks good, I will re-send the whole patchset including devlink changes and driver usage. Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com> Reviewed-by: Michael Chan <michael.chan@broadcom.com> --- v2: - Switch RAM and AP component definitions. - Remove IRQ, FILTER, OFFLOAD, MAC, PHY components as they are port specific components. - Rename function to host in width parameter. --- Documentation/networking/devlink/devlink-reset.rst | 50 +++++++++++++ include/net/devlink.h | 2 + include/uapi/linux/devlink.h | 46 ++++++++++++ net/core/devlink.c | 85 ++++++++++++++++++++++ 4 files changed, 183 insertions(+) create mode 100644 Documentation/networking/devlink/devlink-reset.rst