Skip to contentSkip to conversation

net: Select PTP_1588_CLOCK in PTP-specific drivers

PATCH In review Clay McClure 4 participants 6 messages Apr 28, 2020 – May 1, 2020

Subsystems (6)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net: Select PTP_1588_CLOCK in PTP-specific drivers

Commit d1cbfd771ce8 ("ptp_clock: Allow for it to be optional") changed all PTP-capable Ethernet drivers from select PTP_1588_CLOCK to imply PTP_1588_CLOCK, "in order to break the hard dependency between the PTP clock subsystem and ethernet drivers capable of being clock providers." As a result it is possible to build PTP-capable Ethernet drivers without the PTP subsystem by deselecting PTP_1588_CLOCK. Drivers are required to handle the missing dependency gracefully.

Some PTP-capable Ethernet drivers (e.g., TI_CPSW) factor their PTP code out into separate drivers (e.g., TI_CPTS_MOD). The above commit also changed these PTP-specific drivers to imply PTP_1588_CLOCK, making it possible to build them without the PTP subsystem. But as Grygorii Strashko noted in [1]:

Quoted 3 lines · Grygorii Strashko

Another question is that CPTS completely nonfunctional in this case and it was never expected that somebody will even try to use/run such configuration (except for random build purposes).

In my view, enabling a PTP-specific driver without the PTP subsystem is a configuration error made possible by the above commit. Kconfig should not allow users to create a configuration with missing dependencies that results in "completely nonfunctional" drivers.

I audited all network drivers that call ptp_clock_register() and found six that look like PTP-specific drivers that are likely nonfunctional without PTP_1588_CLOCK:

    NET_DSA_MV88E6XXX_PTP
    NET_DSA_SJA1105_PTP
    MACB_USE_HWSTAMP
    CAVIUM_PTP
    TI_CPTS_MOD
    PTP_1588_CLOCK_IXP46X

Note how they all reference PTP or timestamping in their name; this is a clue that they depend on PTP_1588_CLOCK.

Change these drivers back [2] to select PTP_1588_CLOCK. Note that this requires also selecting POSIX_TIMERS, a transitive dependency of PTP_1588_CLOCK.

[1]: https://lkml.org/lkml/2020/4/22/1056

[2]: NET_DSA_SJA1105_PTP had never declared any type of dependency on PTP_1588_CLOCK (imply or otherwise); adding it here seems appropriate.

Cc Arnd Bergmann <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Nicolas Pitre <[email protected]>
Cc Grygorii Strashko <[email protected]>
Fixes d1cbfd771ce8 ("ptp_clock: Allow for it to be optional")
---
 drivers/net/dsa/mv88e6xxx/Kconfig    | 3 ++-
 drivers/net/dsa/sja1105/Kconfig      | 2 ++
 drivers/net/ethernet/cadence/Kconfig | 3 ++-
 drivers/net/ethernet/cavium/Kconfig  | 3 ++-
 drivers/net/ethernet/ti/Kconfig      | 2 +-
 drivers/net/ethernet/xscale/Kconfig  | 3 ++-
 6 files changed, 11 insertions(+), 5 deletions(-)
drivers/net/dsa/mv88e6xxx/Kconfig +2 -1
index 6435020d690d..6e8250599eb1 100644
@@ -25,7 +25,8 @@ config NET_DSA_MV88E6XXX_PTP
 	default n
 	depends on NET_DSA_MV88E6XXX_GLOBAL2
 	imply NETWORK_PHY_TIMESTAMPING
-	imply PTP_1588_CLOCK
+	select PTP_1588_CLOCK
+	select POSIX_TIMERS
 	help
 	  Say Y to enable PTP hardware timestamping on Marvell 88E6xxx switch
 	  chips that support it.
drivers/net/dsa/sja1105/Kconfig +2 -0
index 0fe1ae173aa1..84349b6c8c44 100644
@@ -20,6 +20,8 @@ tristate "NXP SJA1105 Ethernet switch family support"
 config NET_DSA_SJA1105_PTP
 	bool "Support for the PTP clock on the NXP SJA1105 Ethernet switch"
 	depends on NET_DSA_SJA1105
+	select PTP_1588_CLOCK
+	select POSIX_TIMERS
 	help
 	  This enables support for timestamping and PTP clock manipulations in
 	  the SJA1105 DSA driver.
drivers/net/ethernet/cadence/Kconfig +2 -1
index 53b50c24d9c9..bc792c334903 100644
@@ -36,7 +36,8 @@ config MACB_USE_HWSTAMP
 	bool "Use IEEE 1588 hwstamp"
 	depends on MACB
 	default y
-	imply PTP_1588_CLOCK
+	select PTP_1588_CLOCK
+	select POSIX_TIMERS
 	---help---
 	  Enable IEEE 1588 Precision Time Protocol (PTP) support for MACB.
 
drivers/net/ethernet/cavium/Kconfig +2 -1
index 6a700d34019e..bae58c488792 100644
@@ -54,7 +54,8 @@ config	THUNDER_NIC_RGX
 config CAVIUM_PTP
 	tristate "Cavium PTP coprocessor as PTP clock"
 	depends on 64BIT && PCI
-	imply PTP_1588_CLOCK
+	select PTP_1588_CLOCK
+	select POSIX_TIMERS
 	---help---
 	  This driver adds support for the Precision Time Protocol Clocks and
 	  Timestamping coprocessor (PTP) found on Cavium processors.
drivers/net/ethernet/ti/Kconfig +1 -1
index 89cec778cf2d..1177953790c5 100644
@@ -92,7 +92,7 @@ config TI_CPTS_MOD
 	depends on TI_CPTS
 	default y if TI_CPSW=y || TI_KEYSTONE_NETCP=y || TI_CPSW_SWITCHDEV=y
 	select NET_PTP_CLASSIFY
-	imply PTP_1588_CLOCK
+	select PTP_1588_CLOCK
 	default m
 
 config TI_K3_AM65_CPSW_NUSS
drivers/net/ethernet/xscale/Kconfig +2 -1
index 98aa7b8ddb06..82d4ec3c6398 100644
@@ -30,7 +30,8 @@ config IXP4XX_ETH
 config PTP_1588_CLOCK_IXP46X
 	tristate "Intel IXP46x as PTP clock"
 	depends on IXP4XX_ETH
-	depends on PTP_1588_CLOCK
+	select PTP_1588_CLOCK
+	select POSIX_TIMERS
 	default y
 	help
 	  This driver adds support for using the IXP46X as a PTP
Signature · 1 line
2.20.1
#1 lore raw

Hi Clay,

Thanks for your patch!

Quoted 36 lines · Clay McClure

Commit d1cbfd771ce8 ("ptp_clock: Allow for it to be optional") changed all PTP-capable Ethernet drivers from select PTP_1588_CLOCK to imply PTP_1588_CLOCK, "in order to break the hard dependency between the PTP clock subsystem and ethernet drivers capable of being clock providers." As a result it is possible to build PTP-capable Ethernet drivers without the PTP subsystem by deselecting PTP_1588_CLOCK. Drivers are required to handle the missing dependency gracefully.

Some PTP-capable Ethernet drivers (e.g., TI_CPSW) factor their PTP code out into separate drivers (e.g., TI_CPTS_MOD). The above commit also changed these PTP-specific drivers to imply PTP_1588_CLOCK, making it possible to build them without the PTP subsystem. But as Grygorii Strashko noted in [1]:

Quoted 3 lines · Grygorii Strashko

Another question is that CPTS completely nonfunctional in this case and it was never expected that somebody will even try to use/run such configuration (except for random build purposes).

In my view, enabling a PTP-specific driver without the PTP subsystem is a configuration error made possible by the above commit. Kconfig should not allow users to create a configuration with missing dependencies that results in "completely nonfunctional" drivers.

I audited all network drivers that call ptp_clock_register() and found six that look like PTP-specific drivers that are likely nonfunctional without PTP_1588_CLOCK:

    NET_DSA_MV88E6XXX_PTP
    NET_DSA_SJA1105_PTP
    MACB_USE_HWSTAMP
    CAVIUM_PTP
    TI_CPTS_MOD
    PTP_1588_CLOCK_IXP46X
Quoted 6 lines

Note how they all reference PTP or timestamping in their name; this is a clue that they depend on PTP_1588_CLOCK.

Change these drivers back [2] to select PTP_1588_CLOCK. Note that this requires also selecting POSIX_TIMERS, a transitive dependency of PTP_1588_CLOCK.

If these drivers have a hard dependency on PTP_1588_CLOCK, IMHO they should depend on PTP_1588_CLOCK, not select PTP_1588_CLOCK.

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
#2 lore raw
Quoted 40 lines · Geert Uytterhoeven
Quoted 39 lines · Clay McClure

Commit d1cbfd771ce8 ("ptp_clock: Allow for it to be optional") changed all PTP-capable Ethernet drivers from select PTP_1588_CLOCK to imply PTP_1588_CLOCK, "in order to break the hard dependency between the PTP clock subsystem and ethernet drivers capable of being clock providers." As a result it is possible to build PTP-capable Ethernet drivers without the PTP subsystem by deselecting PTP_1588_CLOCK. Drivers are required to handle the missing dependency gracefully.

Some PTP-capable Ethernet drivers (e.g., TI_CPSW) factor their PTP code out into separate drivers (e.g., TI_CPTS_MOD). The above commit also changed these PTP-specific drivers to imply PTP_1588_CLOCK, making it possible to build them without the PTP subsystem. But as Grygorii Strashko noted in [1]:

Quoted 3 lines · Grygorii Strashko

Another question is that CPTS completely nonfunctional in this case and it was never expected that somebody will even try to use/run such configuration (except for random build purposes).

In my view, enabling a PTP-specific driver without the PTP subsystem is a configuration error made possible by the above commit. Kconfig should not allow users to create a configuration with missing dependencies that results in "completely nonfunctional" drivers.

I audited all network drivers that call ptp_clock_register() and found six that look like PTP-specific drivers that are likely nonfunctional without PTP_1588_CLOCK:

    NET_DSA_MV88E6XXX_PTP
    NET_DSA_SJA1105_PTP
    MACB_USE_HWSTAMP
    CAVIUM_PTP
    TI_CPTS_MOD
    PTP_1588_CLOCK_IXP46X

Note how they all reference PTP or timestamping in their name; this is a clue that they depend on PTP_1588_CLOCK.

Quoted 6 lines
Quoted 3 lines

Change these drivers back [2] to select PTP_1588_CLOCK. Note that this requires also selecting POSIX_TIMERS, a transitive dependency of PTP_1588_CLOCK.

If these drivers have a hard dependency on PTP_1588_CLOCK, IMHO they should depend on PTP_1588_CLOCK, not select PTP_1588_CLOCK.

Agreed. Note that for drivers that only optionally use the PTP_1588_CLOCK support, we probably want 'depends on PTP_1588_CLOCK || !PTP_1588_CLOCK' (or the syntax replacing it eventually), to avoid the case where a built-in driver fails to use a modular ptp implementation.

Arnd

#3 lore raw
Quoted 11 lines · Arnd Bergmann
Quoted 7 lines · Geert Uytterhoeven
Quoted 3 lines · Clay McClure

Change these drivers back [2] to select PTP_1588_CLOCK. Note that this requires also selecting POSIX_TIMERS, a transitive dependency of PTP_1588_CLOCK.

If these drivers have a hard dependency on PTP_1588_CLOCK, IMHO they should depend on PTP_1588_CLOCK, not select PTP_1588_CLOCK.

Agreed.

Thanks for reviewing the patch. I'll post v2 using depends on shortly.

Quoted 4 lines

Note that for drivers that only optionally use the PTP_1588_CLOCK support, we probably want 'depends on PTP_1588_CLOCK || !PTP_1588_CLOCK' (or the syntax replacing it eventually), to avoid the case where a built-in driver fails to use a modular ptp implementation.

I see some drivers are starting to do just that, e.g.:

commit 96c34151d157 ("net/mlx5: Kconfig: convert imply usage to weak dependency")

I can post a patch this weekend converting the rest of the drivers.

Signature · 1 line
Clay
#4 lore raw

PATCH v2 1/1 net: Make PTP-specific drivers depend on PTP_1588_CLOCK

Commit d1cbfd771ce8 ("ptp_clock: Allow for it to be optional") changed all PTP-capable Ethernet drivers from select PTP_1588_CLOCK to imply PTP_1588_CLOCK, "in order to break the hard dependency between the PTP clock subsystem and ethernet drivers capable of being clock providers." As a result it is possible to build PTP-capable Ethernet drivers without the PTP subsystem by deselecting PTP_1588_CLOCK. Drivers are required to handle the missing dependency gracefully.

Some PTP-capable Ethernet drivers (e.g., TI_CPSW) factor their PTP code out into separate drivers (e.g., TI_CPTS_MOD). The above commit also changed these PTP-specific drivers to imply PTP_1588_CLOCK, making it possible to build them without the PTP subsystem. But as Grygorii Strashko noted in [1]:

Quoted 3 lines · Grygorii Strashko

Another question is that CPTS completely nonfunctional in this case and it was never expected that somebody will even try to use/run such configuration (except for random build purposes).

In my view, enabling a PTP-specific driver without the PTP subsystem is a configuration error made possible by the above commit. Kconfig should not allow users to create a configuration with missing dependencies that results in "completely nonfunctional" drivers.

I audited all network drivers that call ptp_clock_register() but merely imply PTP_1588_CLOCK and found five PTP-specific drivers that are likely nonfunctional without PTP_1588_CLOCK:

    NET_DSA_MV88E6XXX_PTP
    NET_DSA_SJA1105_PTP
    MACB_USE_HWSTAMP
    CAVIUM_PTP
    TI_CPTS_MOD

Note how these symbols all reference PTP or timestamping in their name; this is a clue that they depend on PTP_1588_CLOCK.

Change them from imply PTP_1588_CLOCK [2] to depends on PTP_1588_CLOCK. I'm not using select PTP_1588_CLOCK here because PTP_1588_CLOCK has its own dependencies, which select would not transitively apply.

Additionally, remove the select NET_PTP_CLASSIFY from CPTS_TI_MOD; PTP_1588_CLOCK already selects that.

[1]: https://lore.kernel.org/lkml/[email protected]/

[2]: NET_DSA_SJA1105_PTP had never declared any type of dependency on PTP_1588_CLOCK (imply or otherwise); adding a depends on PTP_1588_CLOCK here seems appropriate.

Cc Arnd Bergmann <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Nicolas Pitre <[email protected]>
Cc Grygorii Strashko <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Fixes d1cbfd771ce8 ("ptp_clock: Allow for it to be optional")
---
 drivers/net/dsa/mv88e6xxx/Kconfig    | 2 +-
 drivers/net/dsa/sja1105/Kconfig      | 1 +
 drivers/net/ethernet/cadence/Kconfig | 2 +-
 drivers/net/ethernet/cavium/Kconfig  | 2 +-
 drivers/net/ethernet/ti/Kconfig      | 3 +--
 5 files changed, 5 insertions(+), 5 deletions(-)
drivers/net/dsa/mv88e6xxx/Kconfig +1 -1
index 6435020d690d..51185e4d7d15 100644
@@ -24,8 +24,8 @@ config NET_DSA_MV88E6XXX_PTP
 	bool "PTP support for Marvell 88E6xxx"
 	default n
 	depends on NET_DSA_MV88E6XXX_GLOBAL2
+	depends on PTP_1588_CLOCK
 	imply NETWORK_PHY_TIMESTAMPING
-	imply PTP_1588_CLOCK
 	help
 	  Say Y to enable PTP hardware timestamping on Marvell 88E6xxx switch
 	  chips that support it.
drivers/net/dsa/sja1105/Kconfig +1 -0
index 0fe1ae173aa1..68c3086af9af 100644
@@ -20,6 +20,7 @@ tristate "NXP SJA1105 Ethernet switch family support"
 config NET_DSA_SJA1105_PTP
 	bool "Support for the PTP clock on the NXP SJA1105 Ethernet switch"
 	depends on NET_DSA_SJA1105
+	depends on PTP_1588_CLOCK
 	help
 	  This enables support for timestamping and PTP clock manipulations in
 	  the SJA1105 DSA driver.
drivers/net/ethernet/cadence/Kconfig +1 -1
index 53b50c24d9c9..2c4c12b03502 100644
@@ -35,8 +35,8 @@ config MACB
 config MACB_USE_HWSTAMP
 	bool "Use IEEE 1588 hwstamp"
 	depends on MACB
+	depends on PTP_1588_CLOCK
 	default y
-	imply PTP_1588_CLOCK
 	---help---
 	  Enable IEEE 1588 Precision Time Protocol (PTP) support for MACB.
 
drivers/net/ethernet/cavium/Kconfig +1 -1
index 6a700d34019e..4520e7ee00fe 100644
@@ -54,7 +54,7 @@ config	THUNDER_NIC_RGX
 config CAVIUM_PTP
 	tristate "Cavium PTP coprocessor as PTP clock"
 	depends on 64BIT && PCI
-	imply PTP_1588_CLOCK
+	depends on PTP_1588_CLOCK
 	---help---
 	  This driver adds support for the Precision Time Protocol Clocks and
 	  Timestamping coprocessor (PTP) found on Cavium processors.
drivers/net/ethernet/ti/Kconfig +1 -2
index 89cec778cf2d..8e348780efb6 100644
@@ -90,9 +90,8 @@ config TI_CPTS
 config TI_CPTS_MOD
 	tristate
 	depends on TI_CPTS
+	depends on PTP_1588_CLOCK
 	default y if TI_CPSW=y || TI_KEYSTONE_NETCP=y || TI_CPSW_SWITCHDEV=y
-	select NET_PTP_CLASSIFY
-	imply PTP_1588_CLOCK
 	default m
 
 config TI_K3_AM65_CPSW_NUSS
Signature · 1 line
2.20.1
maintainer Applied
#5 lore raw

Re: [PATCH v2] net: Make PTP-specific drivers depend on PTP_1588_CLOCK

From: Clay McClure <[email protected]>
Date: Wed, 29 Apr 2020 00:59:00 -0700

Quoted 7 lines

Commit d1cbfd771ce8 ("ptp_clock: Allow for it to be optional") changed all PTP-capable Ethernet drivers from select PTP_1588_CLOCK to imply PTP_1588_CLOCK, "in order to break the hard dependency between the PTP clock subsystem and ethernet drivers capable of being clock providers." As a result it is possible to build PTP-capable Ethernet drivers without the PTP subsystem by deselecting PTP_1588_CLOCK. Drivers are required to handle the missing dependency gracefully.

...

Quoted 9 lines

Note how these symbols all reference PTP or timestamping in their name; this is a clue that they depend on PTP_1588_CLOCK.

Change them from imply PTP_1588_CLOCK [2] to depends on PTP_1588_CLOCK. I'm not using select PTP_1588_CLOCK here because PTP_1588_CLOCK has its own dependencies, which select would not transitively apply.

Additionally, remove the select NET_PTP_CLASSIFY from CPTS_TI_MOD; PTP_1588_CLOCK already selects that.

...

Applied, thanks.