| Message ID | 20211214121638.138784-4-philippe.schenker@toradex.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | Add Possiblity to Reset PHY After Power-up | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Clearly marked for net-next |
| netdev/fixes_present | success | Fixes tag not required for -next series |
| netdev/subject_prefix | success | Link |
| netdev/cover_letter | success | Series has a cover letter |
| netdev/patch_count | success | Link |
| netdev/header_inline | success | No static functions without inline keyword in header files |
| netdev/build_32bit | success | Errors and warnings before: 0 this patch: 0 |
| netdev/cc_maintainers | success | CCed 4 of 4 maintainers |
| netdev/build_clang | success | Errors and warnings before: 0 this patch: 0 |
| netdev/module_param | success | Was 0 now: 0 |
| netdev/verify_signedoff | success | Signed-off-by tag matches author and committer |
| netdev/verify_fixes | success | No Fixes tag |
| netdev/build_allmodconfig_warn | success | Errors and warnings before: 0 this patch: 0 |
| netdev/checkpatch | success | total: 0 errors, 0 warnings, 0 checks, 7 lines checked |
| netdev/kdoc | success | Errors and warnings before: 0 this patch: 0 |
| netdev/source_inline | success | Was 0 now: 0 |
On Tue, Dec 14, 2021 at 01:16:38PM +0100, Philippe Schenker wrote: > Reset the eth PHY after resume in case the power was switched off > during suspend, this is required by some PHYs if the reset signal > is controlled by software. > > Signed-off-by: Philippe Schenker <philippe.schenker@toradex.com> > > --- > > drivers/net/ethernet/freescale/fec_main.c | 1 + Hi Philippe What i don't particularly like about this is that the MAC driver is doing it. Meaning if this PHY is used with any other MAC, the same code needs adding there. Is there a way we can put this into phylib? Maybe as part of phy_init_hw()? Humm, actually, thinking aloud: int phy_init_hw(struct phy_device *phydev) { int ret = 0; /* Deassert the reset signal */ phy_device_reset(phydev, 0); So maybe in the phy driver, add a suspend handler, which asserts the reset. This call here will take it out of reset, so applying the reset you need? Andrew
On Tue, Dec 14, 2021 at 07:54:54PM +0100, Andrew Lunn wrote: > On Tue, Dec 14, 2021 at 01:16:38PM +0100, Philippe Schenker wrote: > > Reset the eth PHY after resume in case the power was switched off > > during suspend, this is required by some PHYs if the reset signal > > is controlled by software. > > > > Signed-off-by: Philippe Schenker <philippe.schenker@toradex.com> > > > > --- > > > > drivers/net/ethernet/freescale/fec_main.c | 1 + > > Hi Philippe > > What i don't particularly like about this is that the MAC driver is > doing it. Meaning if this PHY is used with any other MAC, the same > code needs adding there. > > Is there a way we can put this into phylib? Maybe as part of > phy_init_hw()? Humm, actually, thinking aloud: > > int phy_init_hw(struct phy_device *phydev) > { > int ret = 0; > > /* Deassert the reset signal */ > phy_device_reset(phydev, 0); > > So maybe in the phy driver, add a suspend handler, which asserts the > reset. This call here will take it out of reset, so applying the reset > you need? It seems to be a combination issue - it's the fact that the power is turned off and the fact that the reset needs to be applied. If other PHYs such as AR8035 are subjected to this, they appear to have a requirement that reset is asserted when power is applied and kept asserted until the clock has stabilised and a certain time has elapsed. As I've already highlighted, we do not want to be asserting the reset signal in phy_init_hw() - doing so would mean that any PHY with a GPIO reset gets reset whenever the PHY is connected to the MAC - which can be whenever the interface is brought up. That will introduce a multi- second delay to bringing up the network.
Hello Andrew, On Tue, Dec 14, 2021 at 07:54:54PM +0100, Andrew Lunn wrote: > What i don't particularly like about this is that the MAC driver is > doing it. Meaning if this PHY is used with any other MAC, the same > code needs adding there. This is exactly the same case as phy_reset_after_clk_enable() [1][2], to me it does not look that bad. > So maybe in the phy driver, add a suspend handler, which asserts the > reset. This call here will take it out of reset, so applying the reset > you need? Asserting the reset in the phylib in suspend path is a bad idea, in the general case in which the PHY is powered in suspend the power-consumption is likely to be higher if the device is in reset compared to software power-down using the BMCR register (at least for the PHY datasheet I checked). What we could do is to call phy_device_reset in the fec driver suspend path when we know we are going to disable the regulator, I do not like it, but it would solve the issue. --- a/drivers/net/ethernet/freescale/fec_main.c +++ b/drivers/net/ethernet/freescale/fec_main.c @@ -4064,7 +4064,11 @@ static int __maybe_unused fec_suspend(struct device *dev) rtnl_unlock(); if (fep->reg_phy && !(fep->wol_flag & FEC_WOL_FLAG_ENABLE)) + { regulator_disable(fep->reg_phy); + phy_device_reset(ndev->phydev, 1); + } + /* SOC supply clock to phy, when clock is disabled, phy link down * SOC control phy regulator, when regulator is disabled, phy link down Francesco [1] https://lore.kernel.org/netdev/20171211121700.10200-1-dev@g0hl1n.net/ [2] 1b0a83ac04e3 ("net: fec: add phy_reset_after_clk_enable() support")
On Tue, Dec 14, 2021 at 11:35:48PM +0100, Francesco Dolcini wrote: > Hello Andrew, > > On Tue, Dec 14, 2021 at 07:54:54PM +0100, Andrew Lunn wrote: > > What i don't particularly like about this is that the MAC driver is > > doing it. Meaning if this PHY is used with any other MAC, the same > > code needs adding there. > This is exactly the same case as phy_reset_after_clk_enable() [1][2], to > me it does not look that bad. > > > So maybe in the phy driver, add a suspend handler, which asserts the > > reset. This call here will take it out of reset, so applying the reset > > you need? > Asserting the reset in the phylib in suspend path is a bad idea, in the > general case in which the PHY is powered in suspend the > power-consumption is likely to be higher if the device is in reset > compared to software power-down using the BMCR register (at least for > the PHY datasheet I checked). Maybe i don't understand your hardware. You have a regulator providing power of the PHY. You have a reset, i guess a GPIO, connected to the reset pin of the PHY. What you could do is: PHY driver suspend handler does a phy_device_reset(ndev->phydev, 1) to put the PHY into reset. MAC driver disables the regulator. Power consumption should now be 0, since it does not have any power. On resume, the MAC enables the regulator. At this point, the PHY gets power, but is still held in reset. It is now consuming power, but not doing anything. The MAC calls phy_hw_init(), which calls phy_device_reset(ndev->phydev, 0), taking the PHY out of reset. Hopefully, this release from reset is enough to make the PHY work. Doing it like this also addresses Russell point. phy_hw_init() is not putting the device into reset, it is only taking it out of reset, if it happens to be already in reset. So we are not slowing down link up for everybody. Andrew
Hi Francesco, > -----Original Message----- > From: Francesco Dolcini <francesco.dolcini@toradex.com> > Sent: 2021年12月15日 6:36 > To: Andrew Lunn <andrew@lunn.ch> > Cc: Philippe Schenker <philippe.schenker@toradex.com>; > netdev@vger.kernel.org; Joakim Zhang <qiangqing.zhang@nxp.com>; David > S . Miller <davem@davemloft.net>; Russell King <linux@armlinux.org.uk>; > Heiner Kallweit <hkallweit1@gmail.com>; Francesco Dolcini > <francesco.dolcini@toradex.com>; Jakub Kicinski <kuba@kernel.org>; Fabio > Estevam <festevam@gmail.com>; Fugang Duan <fugang.duan@nxp.com>; > linux-kernel@vger.kernel.org > Subject: Re: [PATCH net-next 3/3] net: fec: reset phy on resume after > power-up > > Hello Andrew, > > On Tue, Dec 14, 2021 at 07:54:54PM +0100, Andrew Lunn wrote: > > What i don't particularly like about this is that the MAC driver is > > doing it. Meaning if this PHY is used with any other MAC, the same > > code needs adding there. > This is exactly the same case as phy_reset_after_clk_enable() [1][2], to me it > does not look that bad. > > > So maybe in the phy driver, add a suspend handler, which asserts the > > reset. This call here will take it out of reset, so applying the reset > > you need? > Asserting the reset in the phylib in suspend path is a bad idea, in the general > case in which the PHY is powered in suspend the power-consumption is likely > to be higher if the device is in reset compared to software power-down using > the BMCR register (at least for the PHY datasheet I checked). > > What we could do is to call phy_device_reset in the fec driver suspend path > when we know we are going to disable the regulator, I do not like it, but it > would solve the issue. > > --- a/drivers/net/ethernet/freescale/fec_main.c > +++ b/drivers/net/ethernet/freescale/fec_main.c > @@ -4064,7 +4064,11 @@ static int __maybe_unused fec_suspend(struct > device *dev) > rtnl_unlock(); > > if (fep->reg_phy && !(fep->wol_flag & FEC_WOL_FLAG_ENABLE)) > + { > regulator_disable(fep->reg_phy); > + phy_device_reset(ndev->phydev, 1); > + } > + > > /* SOC supply clock to phy, when clock is disabled, phy link down > * SOC control phy regulator, when regulator is disabled, phy link > down As I mentioned before, both mac and phylib have not taken PHY reset into consideration during system suspend/resume scenario. As Andrew suggested, you could move this into phy driver suspend function, this is a corner case. One point I don't understand, why do you reject to assert reset signal during system suspended? Best Regards, Joakim Zhang > Francesco > > [1] > https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Flore.k > ernel.org%2Fnetdev%2F20171211121700.10200-1-dev%40g0hl1n.net%2F&a > mp;data=04%7C01%7Cqiangqing.zhang%40nxp.com%7Cf7140fe971544fe8d2 > 2608d9bf521517%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C6377 > 51181527979233%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiL > CJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=itV > m0jroQ0MzDG5Ipqs3OY0F5SY%2FkbdFRWauNKq2XiQ%3D&reserved=0 > [2] 1b0a83ac04e3 ("net: fec: add phy_reset_after_clk_enable() support")
On Wed, Dec 15, 2021 at 10:36:52AM +0100, Andrew Lunn wrote: > On Tue, Dec 14, 2021 at 11:35:48PM +0100, Francesco Dolcini wrote: > > Hello Andrew, > > > > On Tue, Dec 14, 2021 at 07:54:54PM +0100, Andrew Lunn wrote: > > > What i don't particularly like about this is that the MAC driver is > > > doing it. Meaning if this PHY is used with any other MAC, the same > > > code needs adding there. > > This is exactly the same case as phy_reset_after_clk_enable() [1][2], to > > me it does not look that bad. > > > > > So maybe in the phy driver, add a suspend handler, which asserts the > > > reset. This call here will take it out of reset, so applying the reset > > > you need? > > Asserting the reset in the phylib in suspend path is a bad idea, in the > > general case in which the PHY is powered in suspend the > > power-consumption is likely to be higher if the device is in reset > > compared to software power-down using the BMCR register (at least for > > the PHY datasheet I checked). > > Maybe i don't understand your hardware. > > You have a regulator providing power of the PHY. > > You have a reset, i guess a GPIO, connected to the reset pin of the > PHY. > > What you could do is: > > PHY driver suspend handler does a phy_device_reset(ndev->phydev, 1) > to put the PHY into reset. > > MAC driver disables the regulator. > > Power consumption should now be 0, since it does not have any power. > > On resume, the MAC enables the regulator. At this point, the PHY gets > power, but is still held in reset. It is now consuming power, but not > doing anything. The MAC calls phy_hw_init(), which calls > phy_device_reset(ndev->phydev, 0), taking the PHY out of reset. > > Hopefully, this release from reset is enough to make the PHY work. > > Doing it like this also addresses Russell point. phy_hw_init() is not > putting the device into reset, it is only taking it out of reset, if > it happens to be already in reset. So we are not slowing down link up > for everybody. Here's another question which no one seems to have considered. If the PHY power source can be controlled, why doesn't the firmware describe the power supply for the PHY, and why doesn't the PHY driver control the PHY power source? Why is that in the SoC network driver?
On Wed, Dec 15, 2021 at 12:01:39PM +0100, Francesco Dolcini wrote: > Any agreement on how to move forward? ... > 3. move regulator to phy/micrel.c and assert reset in the phy driver resume > callback whoops, s/resume/suspend/
> This is all correct and will solve the issue, however ... > > The problem I see is that nor the phylib nor the PHY driver is aware > that the PHY was powered down, if we unconditionally assert the reset in > the suspend callback in the PHY driver/lib this will affect in a bad > case the most common use case in which we keep the PHY powered in > suspend. We know if the PHY should be left up because of WoL. So that is not an issue. We can also put the PHY into lower power mode, before making the call to put the PHY into reset. If the reset is not implemented, the PHY stays in low power mode. If it is implemented, it is both in lower power mode and held in reset. And if the regulator is provided, the power will go off. > The reason is that the power consumption in reset is higher in reset > compared to the normal PHY software power down. Does the datasheet have numbers for in lower power mode and held in reset? We only have an issue if held in reset when in low power mode consumes more power than simply in low power mode. Andrew
On Wed, Dec 15, 2021 at 08:34:00PM +0100, Andrew Lunn wrote: > > The reason is that the power consumption in reset is higher in reset > > compared to the normal PHY software power down. > > Does the datasheet have numbers for in lower power mode and held in > reset? We only have an issue if held in reset when in low power mode > consumes more power than simply in low power mode. The numbers for KSZ9131, Table 6-3: Power Consumption from datasheet [0] 61.2mW in reset, 24.4mW in software power down (3.3VDD) 40.9mW in reset, 12.5mW in software power down (2.5VDD) Francesco [0] https://ww1.microchip.com/downloads/en/DeviceDoc/00002841B.pdf
> -----Original Message----- > From: Francesco Dolcini <francesco.dolcini@toradex.com> > Sent: 2021年12月15日 19:02 > To: Andrew Lunn <andrew@lunn.ch>; Russell King (Oracle) > <linux@armlinux.org.uk>; Joakim Zhang <qiangqing.zhang@nxp.com> > Cc: Russell King (Oracle) <linux@armlinux.org.uk>; Francesco Dolcini > <francesco.dolcini@toradex.com>; Philippe Schenker > <philippe.schenker@toradex.com>; netdev@vger.kernel.org; Joakim Zhang > <qiangqing.zhang@nxp.com>; David S . Miller <davem@davemloft.net>; > Heiner Kallweit <hkallweit1@gmail.com>; Jakub Kicinski <kuba@kernel.org>; > Fabio Estevam <festevam@gmail.com>; linux-kernel@vger.kernel.org > Subject: Re: [PATCH net-next 3/3] net: fec: reset phy on resume after > power-up > [...] > On Wed, Dec 15, 2021 at 10:25:14AM +0000, Joakim Zhang wrote: > > As I mentioned before, both mac and phylib have not taken PHY reset > > into consideration during system suspend/resume scenario. As Andrew > > suggested, you could move this into phy driver suspend function, this > > is a corner case. One point I don't understand, why do you reject to > > assert reset signal during system suspended? > See my answer to Andrew above, in short asserting the reset without > disabling the regulator will create a regression on the power consumption. As I can see, when system suspended, PHY is totally powered down, since you disable the regulator. At this situation, if you assert reset signal, you mean it will increase the power consumption? PHY is totally powered down, why assert reset signal still affect PHY? Best Regards, Joakim Zhang
On Thu, Dec 16, 2021 at 04:52:39AM +0000, Joakim Zhang wrote: > As I can see, when system suspended, PHY is totally powered down, > since you disable the regulator. At this situation, if you > assert reset signal, you mean it will increase the power > consumption? PHY is totally powered down, why assert reset > signal still affect PHY? In general there are *other* use cases in which the PHY is powered in suspend. We should not create a regression there. Francesco
On Thu, Dec 16, 2021 at 08:52:16AM +0100, Francesco Dolcini wrote: > On Thu, Dec 16, 2021 at 04:52:39AM +0000, Joakim Zhang wrote: > > As I can see, when system suspended, PHY is totally powered down, > > since you disable the regulator. At this situation, if you > > assert reset signal, you mean it will increase the power > > consumption? PHY is totally powered down, why assert reset > > signal still affect PHY? > In general there are *other* use cases in which the PHY is powered in > suspend. We should not create a regression there. Yes, this is the sticking point. We can do what you want, but potentially, the change affects others. I think you need to move the regulator into phylib, so the PHY driver can do the right thing. It is really the only entity which knows what is the correct thing to do. Andrew
On Thu, Dec 16, 2021 at 11:24:24AM +0100, Andrew Lunn wrote: > I think you need to move the regulator into phylib, so the PHY driver > can do the right thing. It is really the only entity which knows what > is the correct thing to do. Do you believe that the right place is the phylib and not the phy driver? Is this generic enough? Francesco
On Thu, Dec 16, 2021 at 12:24:33PM +0100, Francesco Dolcini wrote: > On Thu, Dec 16, 2021 at 11:24:24AM +0100, Andrew Lunn wrote: > > I think you need to move the regulator into phylib, so the PHY driver > > can do the right thing. It is really the only entity which knows what > > is the correct thing to do. > Do you believe that the right place is the phylib and not the phy driver? > Is this generic enough? It is split. phylib can do the lookup in DT, get the regulator and provide a helper to enable/disable it. So very similar to the reset. The phy driver would then use the helpers. It probably needs to look into the phydev structure to see what is actually available, is there a reset, a regulator etc, and then decide what is best to do given the available resources. Andrew
On Thu, Dec 16, 2021 at 12:28:19PM +0100, Andrew Lunn wrote: > On Thu, Dec 16, 2021 at 12:24:33PM +0100, Francesco Dolcini wrote: > > On Thu, Dec 16, 2021 at 11:24:24AM +0100, Andrew Lunn wrote: > > > I think you need to move the regulator into phylib, so the PHY driver > > > can do the right thing. It is really the only entity which knows what > > > is the correct thing to do. > > > Do you believe that the right place is the phylib and not the phy driver? > > Is this generic enough? > > It is split. phylib can do the lookup in DT, get the regulator and > provide a helper to enable/disable it. So very similar to the reset. Sounds good. Can we safely assume that we do have at most one regulator for the phy? Francesco
> Can we safely assume that we do have at most one regulator for the phy?
Seems like a reasonable assumption.
Andrew
diff --git a/drivers/net/ethernet/freescale/fec_main.c b/drivers/net/ethernet/freescale/fec_main.c index 1b1f7f2a6130..c29eddbb0155 100644 --- a/drivers/net/ethernet/freescale/fec_main.c +++ b/drivers/net/ethernet/freescale/fec_main.c @@ -4086,6 +4086,7 @@ static int __maybe_unused fec_resume(struct device *dev) ret = regulator_enable(fep->reg_phy); if (ret) return ret; + phy_reset_after_power_on(ndev->phydev); } rtnl_lock();
Reset the eth PHY after resume in case the power was switched off during suspend, this is required by some PHYs if the reset signal is controlled by software. Signed-off-by: Philippe Schenker <philippe.schenker@toradex.com> --- drivers/net/ethernet/freescale/fec_main.c | 1 + 1 file changed, 1 insertion(+)