gve: unify driver name usage
Subsystems (3)
PATCH 1/1 gve: unify driver name usage
Current codebase contained the usage of two different names for this driver (i.e., gvnic and gve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name of gve. It's more reasonable to align the name of the driver with the module.
--- drivers/net/ethernet/google/gve/gve.h | 1 + drivers/net/ethernet/google/gve/gve_adminq.c | 10 +++++----- drivers/net/ethernet/google/gve/gve_ethtool.c | 2 +- drivers/net/ethernet/google/gve/gve_main.c | 11 ++++++----- 4 files changed, 13 insertions(+), 11 deletions(-)
drivers/net/ethernet/google/gve/gve.h +1 -0
@@ -964,5 +964,6 @@ void gve_handle_report_stats(struct gve_priv *priv);/* exported by ethtool.c */ extern const struct ethtool_ops gve_ethtool_ops;/* needed by ethtool */ +extern char gve_driver_name[];extern const char gve_version_str[];#endif /* _GVE_H_ */
drivers/net/ethernet/google/gve/gve_adminq.c +5 -5
@@ -899,7 +899,7 @@ int gve_adminq_verify_driver_compatibility(struct gve_priv *priv,int gve_adminq_report_link_speed (struct gve_priv *priv) { -union gve_adminq_command gvnic_cmd; +union gve_adminq_command gve_cmd;dma_addr_t link_speed_region_bus;__be64 *link_speed_region;int err;
@@ -911,12 +911,12 @@ int gve_adminq_report_link_speed(struct gve_priv *priv)if (!link_speed_region)return -ENOMEM ; -memset (&gvnic_cmd,0 ,sizeof (gvnic_cmd)); - gvnic_cmd.opcode =cpu_to_be32 (GVE_ADMINQ_REPORT_LINK_SPEED ); - gvnic_cmd.report_link_speed.link_speed_address = +memset (&gve_cmd,0 ,sizeof (gve_cmd)); + gve_cmd.opcode =cpu_to_be32 (GVE_ADMINQ_REPORT_LINK_SPEED ); + gve_cmd.report_link_speed.link_speed_address =cpu_to_be64 (link_speed_region_bus); - err =gve_adminq_execute_cmd (priv, &gvnic_cmd); + err =gve_adminq_execute_cmd (priv, &gve_cmd); priv->link_speed =be64_to_cpu (*link_speed_region); dma_free_coherent(&priv->pdev->dev, sizeof(*link_speed_region), link_speed_region,
drivers/net/ethernet/google/gve/gve_ethtool.c +1 -1
@@ -15,7 +15,7 @@ static void gve_get_drvinfo(struct net_device *netdev, {struct gve_priv *priv =netdev_priv (netdev); -strscpy (info->driver,"gve" ,sizeof (info->driver)); +strscpy (info->driver, gve_driver_name,sizeof (info->driver));strscpy (info->version, gve_version_str,sizeof (info->version));strscpy (info->bus_info,pci_name (priv->pdev),sizeof (info->bus_info)); }
drivers/net/ethernet/google/gve/gve_main.c +6 -5
@@ -33,6 +33,7 @@#define MIN_TX_TIMEOUT_GAP (1000 * 10)#define DQO_TX_MAX 0x3FFFF +char gve_driver_name[] ="gve" ;const char gve_version_str[] =GVE_VERSION ;static const char gve_version_prefix[] =GVE_VERSION_PREFIX ;
@@ -2200,7 +2201,7 @@ static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent)if (err)return err; - err =pci_request_regions (pdev,"gvnic-cfg" ); + err =pci_request_regions (pdev, gve_driver_name);if (err) goto abort_with_enabled;
@@ -2393,8 +2394,8 @@ static const struct pci_device_id gve_id_table[] = { { } }; -static struct pci_driver gvnic_driver = { - .name ="gvnic" , +static struct pci_driver gve_driver = { + .name = gve_driver_name, .id_table = gve_id_table, .probe = gve_probe, .remove = gve_remove,
@@ -2405,10 +2406,10 @@ static struct pci_driver gvnic_driver = {#endif }; -module_pci_driver (gvnic_driver); +module_pci_driver (gve_driver);MODULE_DEVICE_TABLE (pci, gve_id_table);MODULE_AUTHOR ("Google, Inc." ); -MODULE_DESCRIPTION ("gVNIC Driver" ); +MODULE_DESCRIPTION ("Google Virtual NIC Driver" );MODULE_LICENSE ("Dual MIT/GPL" );MODULE_VERSION (GVE_VERSION );
Signature · 1 line
2.25.1
Quoted 11 lines · Junfeng Guo
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.Fixes 893ce44df565 ("gve: Add basic driver framework for Compute Engine Virtual NIC")Signed-off-by Junfeng Guo <[email protected]>---
The patch makes the driver strings much more consistent. Looks good to me.
Thanks,
Quoted 27 lines
-----Original Message-----
From: Kubiak, Michal <[email protected]>
Sent: Saturday, July 8, 2023 02:27
To: Guo, Junfeng <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Wang, Haiyue <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH net] gve: unify driver name usageQuoted 7 lines · Junfeng Guo
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.Fixes 893ce44df565 ("gve: Add basic driver framework for ComputeEngine Virtual NIC")
Quoted 4 lines
Signed-off-by Junfeng Guo <[email protected]>---
Quoted 5 lines
The patch makes the driver strings much more consistent. Looks good to me.
Thanks,
Reviewed-by Michal Kubiak <[email protected]>
Thanks for the review!
Jakub Kicinski
maintainerQuoted 9 lines · On Fri, 7 Jul 2023 18:37:10 +0800 Junfeng Guo wrote:
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.Fixes 893ce44df565 ("gve: Add basic driver framework for Compute Engine Virtual NIC")Signed-off-by Junfeng Guo <[email protected]>
Google's maintainers definitely need to agree to this, because it's a user visible change. It can very well break someone's scripts.
Quoted 28 lines
drivers/net/ethernet/google/gve/gve.h +1
-0@@ -964,5 +964,6 @@ void gve_handle_report_stats(struct gve_priv *priv);/* exported by ethtool.c */ extern const struct ethtool_ops gve_ethtool_ops;/* needed by ethtool */ +extern char gve_driver_name[];extern const char gve_version_str[];#endif /* _GVE_H_ */ drivers/net/ethernet/google/gve/gve_adminq.c +1
-1@@ -899,7 +899,7 @@ int gve_adminq_verify_driver_compatibility(struct gve_priv *priv,int gve_adminq_report_link_speed (struct gve_priv *priv) { -union gve_adminq_command gvnic_cmd; +union gve_adminq_command gve_cmd;dma_addr_t link_speed_region_bus;__be64 *link_speed_region;int err;@@ -911,12 +911,12 @@ int gve_adminq_report_link_speed(struct gve_priv *priv)if (!link_speed_region)return -ENOMEM ;
drivers/net/ethernet/google/gve/gve_adminq.c int gve_adminq_report_link_speed(struct gve_priv *priv) · line 919
Quoted 10 lines
drivers/net/ethernet/google/gve/gve_adminq.c +4
-4-memset (&gvnic_cmd,0 ,sizeof (gvnic_cmd)); - gvnic_cmd.opcode =cpu_to_be32 (GVE_ADMINQ_REPORT_LINK_SPEED ); - gvnic_cmd.report_link_speed.link_speed_address = +memset (&gve_cmd,0 ,sizeof (gve_cmd)); + gve_cmd.opcode =cpu_to_be32 (GVE_ADMINQ_REPORT_LINK_SPEED ); + gve_cmd.report_link_speed.link_speed_address =cpu_to_be64 (link_speed_region_bus); - err =gve_adminq_execute_cmd (priv, &gvnic_cmd); + err =gve_adminq_execute_cmd (priv, &gve_cmd);
What's the problem with the variable being called gvnic_cmd ? Please limit renames, if you want this to be a fix.
Signature · 1 line
pw-bot: cr
Quoted 20 lines
-----Original Message-----
From: Jakub Kicinski <[email protected]>
Sent: Saturday, July 8, 2023 06:20
To: Guo, Junfeng <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Wang, Haiyue <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH net] gve: unify driver name usageQuoted 6 lines · On Fri, 7 Jul 2023 18:37:10 +0800 Junfeng Guo wrote:
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.
Quoted 7 lines
Quoted 1 line
Fixes 893ce44df565 ("gve: Add basic driver framework for ComputeEngine Virtual NIC")
Quoted 2 lines
Signed-off-by Junfeng Guo <[email protected]>Google's maintainers definitely need to agree to this, because it's a user visible change. It can very well break someone's scripts.
Sure, exactly!
Google's maintainers are also CC'ed to give decisive comments for this.
This patch is to align the inconsistencies for the driver name and the kernel module name, which may also help improve some scripts to avoid using exception mapping with two different names.
Quoted 41 lines
Review
Quoted 2 lines
(quoted diff) +0
-0
b/drivers/net/ethernet/google/gve/gve.hdrivers/net/ethernet/google/gve/gve.h void gve_handle_report_stats(struct gve_priv
Quoted 4 lines
index
98eb78d98e9f..4b425bf71ede100644drivers/net/ethernet/google/gve/gve.h +0
-0@@ -964,5 +964,6 @@ void gve_handle_report_stats(struct gve_priv*priv);
Review
Quoted 7 lines
drivers/net/ethernet/google/gve/gve.h +1
-0/* exported by ethtool.c */ extern const struct ethtool_ops gve_ethtool_ops;/* needed by ethtool */ +extern char gve_driver_name[];extern const char gve_version_str[];#endif /* _GVE_H_ */ (quoted diff) +0
-0
b/drivers/net/ethernet/google/gve/gve_adminq.cdrivers/net/ethernet/google/gve/gve_adminq.c int
Quoted 4 lines
index
252974202a3f..ae8f8c935bbe100644drivers/net/ethernet/google/gve/gve_adminq.c +0
-0@@ -899,7 +899,7 @@ intgve_adminq_verify_driver_compatibility(struct
gve_priv*priv,drivers/net/ethernet/google/gve/gve_adminq.c int gve_adminq_report_link_speed(struct
Quoted 9 lines
drivers/net/ethernet/google/gve/gve_adminq.c +1
-1int gve_adminq_report_link_speed (struct gve_priv *priv) { -union gve_adminq_command gvnic_cmd; +union gve_adminq_command gve_cmd;dma_addr_t link_speed_region_bus;__be64 *link_speed_region;int err;@@ -911,12 +911,12 @@ int gve_adminq_report_link_speed(struct
gve_priv*priv)drivers/net/ethernet/google/gve/gve_adminq.c int gve_adminq_report_link_speed(struct gve_priv *priv)
Quoted 5 lines
drivers/net/ethernet/google/gve/gve_adminq.c +0
-2if (!link_speed_region)return -ENOMEM ; -memset (&gvnic_cmd,0 ,sizeof (gvnic_cmd)); - gvnic_cmd.opcode =
cpu_to_be32(GVE_ADMINQ_REPORT_LINK_SPEED);drivers/net/ethernet/google/gve/gve_adminq.c int gve_adminq_report_link_speed(struct gve_priv *priv)
Quoted 3 lines
drivers/net/ethernet/google/gve/gve_adminq.c +2
-1- gvnic_cmd.report_link_speed.link_speed_address = +memset (&gve_cmd,0 ,sizeof (gve_cmd)); + gve_cmd.opcode =
cpu_to_be32(GVE_ADMINQ_REPORT_LINK_SPEED);
Quoted 8 lines
drivers/net/ethernet/google/gve/gve_adminq.c int gve_adminq_report_link_speed(struct gve_priv *priv) · line 919
Quoted 5 lines
drivers/net/ethernet/google/gve/gve_adminq.c +2
-1+ gve_cmd.report_link_speed.link_speed_address =cpu_to_be64 (link_speed_region_bus); - err =gve_adminq_execute_cmd (priv, &gvnic_cmd); + err =gve_adminq_execute_cmd (priv, &gve_cmd);What's the problem with the variable being called
gvnic_cmd? Please limit renames, if you want this to be a fix.
Thanks for the comments!
Yes, this is not very related to the fixing purpose. Will remove this part in the coming version. Thanks!
Quoted 2 lines
--
pw-bot: cr
PATCH v2 1/1 gve: unify driver name usage
In reply to Junfeng GuoCurrent codebase contained the usage of two different names for this driver (i.e., gvnic and gve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name of gve. It's more reasonable to align the name of the driver with the module.
---
v2:
- Remove un-related variable name chang, just keep it as it is.
--- drivers/net/ethernet/google/gve/gve.h | 1 + drivers/net/ethernet/google/gve/gve_ethtool.c | 2 +- drivers/net/ethernet/google/gve/gve_main.c | 11 ++++++----- 3 files changed, 8 insertions(+), 6 deletions(-)
drivers/net/ethernet/google/gve/gve.h +1 -0
@@ -964,5 +964,6 @@ void gve_handle_report_stats(struct gve_priv *priv);/* exported by ethtool.c */ extern const struct ethtool_ops gve_ethtool_ops;/* needed by ethtool */ +extern char gve_driver_name[];extern const char gve_version_str[];#endif /* _GVE_H_ */
drivers/net/ethernet/google/gve/gve_ethtool.c +1 -1
@@ -15,7 +15,7 @@ static void gve_get_drvinfo(struct net_device *netdev, {struct gve_priv *priv =netdev_priv (netdev); -strscpy (info->driver,"gve" ,sizeof (info->driver)); +strscpy (info->driver, gve_driver_name,sizeof (info->driver));strscpy (info->version, gve_version_str,sizeof (info->version));strscpy (info->bus_info,pci_name (priv->pdev),sizeof (info->bus_info)); }
drivers/net/ethernet/google/gve/gve_main.c +6 -5
@@ -33,6 +33,7 @@#define MIN_TX_TIMEOUT_GAP (1000 * 10)#define DQO_TX_MAX 0x3FFFF +char gve_driver_name[] ="gve" ;const char gve_version_str[] =GVE_VERSION ;static const char gve_version_prefix[] =GVE_VERSION_PREFIX ;
@@ -2200,7 +2201,7 @@ static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent)if (err)return err; - err =pci_request_regions (pdev,"gvnic-cfg" ); + err =pci_request_regions (pdev, gve_driver_name);if (err) goto abort_with_enabled;
@@ -2393,8 +2394,8 @@ static const struct pci_device_id gve_id_table[] = { { } }; -static struct pci_driver gvnic_driver = { - .name ="gvnic" , +static struct pci_driver gve_driver = { + .name = gve_driver_name, .id_table = gve_id_table, .probe = gve_probe, .remove = gve_remove,
@@ -2405,10 +2406,10 @@ static struct pci_driver gvnic_driver = {#endif }; -module_pci_driver (gvnic_driver); +module_pci_driver (gve_driver);MODULE_DEVICE_TABLE (pci, gve_id_table);MODULE_AUTHOR ("Google, Inc." ); -MODULE_DESCRIPTION ("gVNIC Driver" ); +MODULE_DESCRIPTION ("Google Virtual NIC Driver" );MODULE_LICENSE ("Dual MIT/GPL" );MODULE_VERSION (GVE_VERSION );
Signature · 1 line
2.25.1
Quoted 7 lines · Junfeng Guo
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.Fixes 893ce44df565 ("gve: Add basic driver framework for Compute Engine Virtual NIC")
You are posting this for net, and have a fixes tag. Please take a look at:
https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html
Do you think it fulfils those rules?
Andrew
This patch was applied to netdev/net.git (main) by David S. Miller <[email protected]>:
Hello:
This patch was applied to netdev/net.git (main)
by David S. Miller <[email protected]>:
Quoted 11 lines · On Sat, 8 Jul 2023 11:14:51 +0800 you wrote:
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.Fixes 893ce44df565 ("gve: Add basic driver framework for Compute Engine Virtual NIC")Signed-off-by Junfeng Guo <[email protected]>[...]
Here is the summary with links:
- [net,v2] gve: unify driver name usage
https://git.kernel.org/netdev/net/c/9d0aba98316d
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
From: Junfeng Guo <[email protected]>
Date: Fri, 7 Jul 2023 18:37:10 +0800
Quoted 5 lines
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.
[...]
drivers/net/ethernet/google/gve/gve_main.c static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent) · line 2204
Quoted 6 lines
drivers/net/ethernet/google/gve/gve_main.c +1
-1@@ -2200,7 +2201,7 @@ static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent)if (err)return err; - err =pci_request_regions (pdev,"gvnic-cfg" ); + err =pci_request_regions (pdev, gve_driver_name);
I won't repeat others' comments, but will comment on this. Passing just driver name with no unique identifiers makes it very confusing to read /proc/iomem et al.
Imagine you have 2 NICs in your system. Then, in /proc/iomem you will have:
gve 0x00001000-0x00002000
gve 0x00004000-0x00005000
Can you say which region belongs to which NIC? Nope.
If you really want to make this more "user friendly", you should make it possible for users to distinguish different NICs in your system. The easiest way:
err = pci_request_regions(pdev, pci_name(pdev));
But you're not limited to this. Just make it unique.
(as a net-next commit obv)
drivers/net/ethernet/google/gve/gve_main.c static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent) · line 2207
Quoted 3 lines
drivers/net/ethernet/google/gve/gve_main.c +0
-0if (err) goto abort_with_enabled;
[...]
Thanks,
Olek
Quoted 32 lines
-----Original Message-----
From: Lobakin, Aleksander <[email protected]>
Sent: Tuesday, July 11, 2023 21:14
To: Guo, Junfeng <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Wang, Haiyue <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH net] gve: unify driver name usageFrom: Junfeng Guo <[email protected]>
Date: Fri, 7 Jul 2023 18:37:10 +0800Quoted 5 lines
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.[...]
drivers/net/ethernet/google/gve/gve_main.c static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent) · line 2204
Quoted 6 lines
drivers/net/ethernet/google/gve/gve_main.c +1
-1@@ -2200,7 +2201,7 @@ static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent)if (err)return err; - err =pci_request_regions (pdev,"gvnic-cfg" ); + err =pci_request_regions (pdev, gve_driver_name);I won't repeat others' comments, but will comment on this. Passing just driver name with no unique identifiers makes it very confusing to read
/proc/iomemet al.
Imagine you have 2 NICs in your system. Then, in/proc/iomemyou will have:
Quoted 3 lines
gve 0x00001000-0x00002000
gve 0x00004000-0x00005000
Looks like, in real world, it is PCI BAR tree layers, take Intel ice as an example:
err = pcim_iomap_regions(pdev, BIT(ICE_BAR0), dev_driver_string(dev));
3b4000000000-3b7fffffffff : PCI Bus 0000:b7
3b7ffa000000-3b7ffe4fffff : PCI Bus 0000:b8
3b7ffa000000-3b7ffbffffff : 0000:b8:00.0 <--- Different NIC, has different BDF
3b7ffa000000-3b7ffbffffff : ice <--- The region name is driver name.
3b7ffc000000-3b7ffdffffff : 0000:b8:00.0
3b7ffe000000-3b7ffe00ffff : 0000:b8:00.0
3b7ffe010000-3b7ffe40ffff : 0000:b8:00.0google/gve/gve_main.c:2203: err = pci_request_regions(pdev, "gvnic-cfg");hisilicon/hns3/hns3pf/hclge_main.c:11350: ret = pci_request_regions(pdev, HCLGE_DRIVER_NAME);hisilicon/hns3/hns3vf/hclgevf_main.c:2588: ret = pci_request_regions(pdev, HCLGEVF_DRIVER_NAME);huawei/hinic/hinic_main.c:1362: err = pci_request_regions(pdev, HINIC_DRV_NAME);intel/ixgbevf/ixgbevf_main.c:4544: err = pci_request_regions(pdev, ixgbevf_driver_name);intel/ixgbevf/ixgbevf_main.c:4546: dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);intel/e100.c:2865: if ((err = pci_request_regions(pdev, DRV_NAME))) {intel/igbvf/netdev.c:2732: err = pci_request_regions(pdev, igbvf_driver_name);intel/iavf/iavf_main.c:4849: err = pci_request_regions(pdev, iavf_driver_name);intel/iavf/iavf_main.c:4852: "pci_request_regions failed 0x%x\n", err);
jme.c:2939: rc = pci_request_regions(pdev, DRV_NAME);
marvell/octeontx2/nic/otx2_pf.c:2793: err = pci_request_regions(pdev, DRV_NAME); marvell/octeontx2/nic/otx2_vf.c:534: err = pci_request_regions(pdev, DRV_NAME); marvell/octeontx2/af/mcs.c:1516: err = pci_request_regions(pdev, DRV_NAME); marvell/octeontx2/af/rvu.c:3238: err = pci_request_regions(pdev, DRV_NAME); marvell/octeontx2/af/cgx.c:1831: err = pci_request_regions(pdev, DRV_NAME);
Quoted 18 lines
Can you say which region belongs to which NIC? Nope.
If you really want to make this more "user friendly", you should make it possible for users to distinguish different NICs in your system. The easiest way:err = pci_request_regions(pdev, pci_name(pdev));But you're not limited to this. Just make it unique.
(as a net-next commit obv)
drivers/net/ethernet/google/gve/gve_main.c static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent) · line 2207
Quoted 3 lines
drivers/net/ethernet/google/gve/gve_main.c +0
-0if (err) goto abort_with_enabled;[...]
Thanks,
Olek
From: Wang, Haiyue <[email protected]>
Date: Tue, 11 Jul 2023 19:24:36 +0200
Quoted 36 lines
Quoted 32 lines
-----Original Message-----
From: Lobakin, Aleksander <[email protected]>
Sent: Tuesday, July 11, 2023 21:14
To: Guo, Junfeng <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Wang, Haiyue <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH net] gve: unify driver name usageFrom: Junfeng Guo <[email protected]>
Date: Fri, 7 Jul 2023 18:37:10 +0800Quoted 5 lines
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.[...]
drivers/net/ethernet/google/gve/gve_main.c static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent) · line 2204
Quoted 6 lines
drivers/net/ethernet/google/gve/gve_main.c +1
-1@@ -2200,7 +2201,7 @@ static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent)if (err)return err; - err =pci_request_regions (pdev,"gvnic-cfg" ); + err =pci_request_regions (pdev, gve_driver_name);I won't repeat others' comments, but will comment on this. Passing just driver name with no unique identifiers makes it very confusing to read
/proc/iomemet al.
Imagine you have 2 NICs in your system. Then, in/proc/iomemyou will have:Quoted 3 lines
gve 0x00001000-0x00002000
gve 0x00004000-0x00005000
Quoted 8 lines
Looks like, in real world, it is PCI BAR tree layers, take Intel ice as an example:
err =
pcim_iomap_regions(pdev, BIT(ICE_BAR0), dev_driver_string(dev));3b4000000000-3b7fffffffff : PCI Bus 0000:b7
3b7ffa000000-3b7ffe4fffff : PCI Bus 0000:b8
3b7ffa000000-3b7ffbffffff : 0000:b8:00.0 <--- Different NIC, has different BDF
3b7ffa000000-3b7ffbffffff : ice <--- The region name is driver name.
I didn't say Intel drivers do that better ¯\_(ツ)_/¯
Why rely on that the kernel or something else will beautify the output for you or that the user won't do grep <drvname> /proc/iomem or something else? Or do that just because "look, it's done the same way in other drivers", which were taken into the tree years ago and sometimes with no detailed review?
There are efforts[0] time to time[1] to convert precisely what you are doing into what I'm asking for. Do they exist by mistake or...?
(the second link also shows that even pci_name() is not enough when you
map several BARs of the same device, but that's not the case this time)
Quoted 15 lines
3b7ffc000000-3b7ffdffffff : 0000:b8:00.0 3b7ffe000000-3b7ffe00ffff : 0000:b8:00.0 3b7ffe010000-3b7ffe40ffff : 0000:b8:00.0
google/gve/gve_main.c:2203: err =pci_request_regions(pdev, "gvnic-cfg");hisilicon/hns3/hns3pf/hclge_main.c:11350: ret =pci_request_regions(pdev, HCLGE_DRIVER_NAME);hisilicon/hns3/hns3vf/hclgevf_main.c:2588: ret =pci_request_regions(pdev, HCLGEVF_DRIVER_NAME);huawei/hinic/hinic_main.c:1362: err =pci_request_regions(pdev, HINIC_DRV_NAME);intel/ixgbevf/ixgbevf_main.c:4544: err =pci_request_regions(pdev, ixgbevf_driver_name);intel/ixgbevf/ixgbevf_main.c:4546:dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);intel/e100.c:2865: if ((err =pci_request_regions(pdev, DRV_NAME))) {intel/igbvf/netdev.c:2732: err =pci_request_regions(pdev, igbvf_driver_name);intel/iavf/iavf_main.c:4849: err =pci_request_regions(pdev, iavf_driver_name);intel/iavf/iavf_main.c:4852: "pci_request_regionsfailed 0x%x\n", err);
jme.c:2939: rc =pci_request_regions(pdev, DRV_NAME);
Quoted 25 lines
marvell/octeontx2/nic/otx2_pf.c:2793: err = pci_request_regions(pdev, DRV_NAME); marvell/octeontx2/nic/otx2_vf.c:534: err = pci_request_regions(pdev, DRV_NAME); marvell/octeontx2/af/mcs.c:1516: err = pci_request_regions(pdev, DRV_NAME); marvell/octeontx2/af/rvu.c:3238: err = pci_request_regions(pdev, DRV_NAME); marvell/octeontx2/af/cgx.c:1831: err = pci_request_regions(pdev, DRV_NAME);Quoted 18 lines
Can you say which region belongs to which NIC? Nope.
If you really want to make this more "user friendly", you should make it possible for users to distinguish different NICs in your system. The easiest way:err = pci_request_regions(pdev, pci_name(pdev));But you're not limited to this. Just make it unique.
(as a net-next commit obv)
drivers/net/ethernet/google/gve/gve_main.c static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent) · line 2207
Quoted 3 lines
drivers/net/ethernet/google/gve/gve_main.c +0
-0if (err) goto abort_with_enabled;[...]
Thanks,
Olek
[0]
https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=0af6e21eed2778e68139941389460e2a00d6ef8e
[1]
https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=35bd8c07db2ce8fd2834ef866240613a4ef982e7
Thanks,
Olek
Quoted 75 lines
-----Original Message-----
From: Lobakin, Aleksander <[email protected]>
Sent: Thursday, July 13, 2023 00:55
To: Wang, Haiyue <[email protected]>
Cc: Guo, Junfeng <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH net] gve: unify driver name usageFrom: Wang, Haiyue <[email protected]>
Date: Tue, 11 Jul 2023 19:24:36 +0200Quoted 36 lines
Quoted 32 lines
-----Original Message-----
From: Lobakin, Aleksander <[email protected]>
Sent: Tuesday, July 11, 2023 21:14
To: Guo, Junfeng <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Wang, Haiyue <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH net] gve: unify driver name usageFrom: Junfeng Guo <[email protected]>
Date: Fri, 7 Jul 2023 18:37:10 +0800Quoted 5 lines
Current codebase contained the usage of two different names for this driver (i.e.,
gvnicandgve), which is quite unfriendly for users to use, especially when trying to bind or unbind the driver manually. The corresponding kernel module is registered with the name ofgve. It's more reasonable to align the name of the driver with the module.[...]
drivers/net/ethernet/google/gve/gve_main.c static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent) · line 2204
Quoted 6 lines
drivers/net/ethernet/google/gve/gve_main.c +1
-1@@ -2200,7 +2201,7 @@ static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent)if (err)return err; - err =pci_request_regions (pdev,"gvnic-cfg" ); + err =pci_request_regions (pdev, gve_driver_name);I won't repeat others' comments, but will comment on this. Passing just driver name with no unique identifiers makes it very confusing to read
/proc/iomemet al.
Imagine you have 2 NICs in your system. Then, in/proc/iomemyou will have:Quoted 3 lines
gve 0x00001000-0x00002000
gve 0x00004000-0x00005000Quoted 8 lines
Looks like, in real world, it is PCI BAR tree layers, take Intel ice as an example:
err =
pcim_iomap_regions(pdev, BIT(ICE_BAR0), dev_driver_string(dev));3b4000000000-3b7fffffffff : PCI Bus 0000:b7
3b7ffa000000-3b7ffe4fffff : PCI Bus 0000:b8
3b7ffa000000-3b7ffbffffff : 0000:b8:00.0 <--- Different NIC, has different BDF
3b7ffa000000-3b7ffbffffff : ice <--- The region name is driver name.I didn't say Intel drivers do that better ¯\_(ツ)_/¯
Why rely on that the kernel or something else will beautify the output for you or that the user won't do
grep <drvname> /proc/iomemor something else? Or do that just because "look, it's done the same way in other drivers", which were taken into the tree years ago and sometimes with no detailed review?
There are efforts[0] time to time[1] to convert precisely what you are doing into what I'm asking for. Do they exist by mistake or...?(the second link also shows that even
pci_name()is not enough when you
map several BARs of the same device, but that's not the case this time)Quoted 3 lines
Quoted 3 lines
Thanks,
Olek
Quoted 6 lines
[0]
https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-
next.git/commit/?id=0af6e21eed2778e68139941389460e2a00d6ef8e
[1]
https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-
next.git/commit/?id=35bd8c07db2ce8fd2834ef866240613a4ef982e7
I see, yes, making sense to beautify the output, and is nice for 'grep ..' to debug quickly, thanks for sharing more experience. :-)
Quoted 2 lines
Thanks,
Olek