On Mon, 22 Jun 2020 at 04:25, Yoshihiro Shimoda
[off-list ref] wrote:
quoted hunk
If pm_suspend_via_firmware() returns true, the system will be able
to cut both vcc and vccq in the suspend. So, call
mmc_poweroff_nofity() if pm_suspend_via_firmware() returns true.
Note that we should not update the MMC_CAP2_FULL_PWR_CYCLE caps
because the mmc_select_voltage() checks the caps when attaches
a mmc/sd.
Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
---
drivers/mmc/core/mmc.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
Sorry, but this doesn't work.
Even if PSCI is a generic FW interface, it doesn't mean that all PSCI
implementations will cut the vcc and vccq for the MMC card at system
suspend.
Instead, you need to decide this based on some specific DT property.
Perhaps in conjunction with using pm_suspend_via_firmware().
Hi Ulf,
On Wed, Jun 24, 2020 at 12:06 PM Ulf Hansson [off-list ref] wrote:
On Mon, 22 Jun 2020 at 04:25, Yoshihiro Shimoda
[off-list ref] wrote:
quoted
If pm_suspend_via_firmware() returns true, the system will be able
to cut both vcc and vccq in the suspend. So, call
mmc_poweroff_nofity() if pm_suspend_via_firmware() returns true.
Note that we should not update the MMC_CAP2_FULL_PWR_CYCLE caps
because the mmc_select_voltage() checks the caps when attaches
a mmc/sd.
Sorry, but this doesn't work.
Even if PSCI is a generic FW interface, it doesn't mean that all PSCI
implementations will cut the vcc and vccq for the MMC card at system
suspend.
Indeed, there's nothing guaranteed here. Nor documented how it should
behave. Basically the firmware is free to power off the SoC. Or not do that.
"If firmware is involved, all odds are off".
Instead, you need to decide this based on some specific DT property.
Perhaps in conjunction with using pm_suspend_via_firmware().
Last time I was involved in a discussion about this, the PSCI people
didn't want to add any properties describing particular PSCI behavior...
"If firmware is involved, all odds are off".
So the only safe thing to do is to expect the worst, and prepare for it...
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
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
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Geert Uytterhoeven, Sent: Wednesday, June 24, 2020 8:13 PM
Hi Ulf,
On Wed, Jun 24, 2020 at 12:06 PM Ulf Hansson [off-list ref] wrote:
quoted
On Mon, 22 Jun 2020 at 04:25, Yoshihiro Shimoda
[off-list ref] wrote:
quoted
If pm_suspend_via_firmware() returns true, the system will be able
to cut both vcc and vccq in the suspend. So, call
mmc_poweroff_nofity() if pm_suspend_via_firmware() returns true.
Note that we should not update the MMC_CAP2_FULL_PWR_CYCLE caps
because the mmc_select_voltage() checks the caps when attaches
a mmc/sd.
Sorry, but this doesn't work.
Even if PSCI is a generic FW interface, it doesn't mean that all PSCI
implementations will cut the vcc and vccq for the MMC card at system
suspend.
Indeed, there's nothing guaranteed here. Nor documented how it should
behave. Basically the firmware is free to power off the SoC. Or not do that.
"If firmware is involved, all odds are off".
I thought we could be guaranteed. But, I understood we could not be guaranteed...
quoted
Instead, you need to decide this based on some specific DT property.
Perhaps in conjunction with using pm_suspend_via_firmware().
Last time I was involved in a discussion about this, the PSCI people
didn't want to add any properties describing particular PSCI behavior...
"If firmware is involved, all odds are off".
So the only safe thing to do is to expect the worst, and prepare for it...
A headache point is an eMMC device consumes much power if that the system
doesn't cut the vcc and vccq and doesn’t enter the sleep mode.
In other words, in power consumption point of view, this patch will
cause a regression in such a case...
By the way, about adding specific DT property, the regulator can have
regulator-off-in-suspend property in regulator-state-mem subnode.
For now, we doesn't seem to get the property from a regulator consumer though.
So, I'll try to add an API of regulator for it.
Best regards,
Yoshihiro Shimoda
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
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
Hi Shimoda-san,
CC broonie
On Thu, Jun 25, 2020 at 8:31 AM Yoshihiro Shimoda
[off-list ref] wrote:
quoted
From: Geert Uytterhoeven, Sent: Wednesday, June 24, 2020 8:13 PM
On Wed, Jun 24, 2020 at 12:06 PM Ulf Hansson [off-list ref] wrote:
quoted
On Mon, 22 Jun 2020 at 04:25, Yoshihiro Shimoda
[off-list ref] wrote:
quoted
If pm_suspend_via_firmware() returns true, the system will be able
to cut both vcc and vccq in the suspend. So, call
mmc_poweroff_nofity() if pm_suspend_via_firmware() returns true.
Note that we should not update the MMC_CAP2_FULL_PWR_CYCLE caps
because the mmc_select_voltage() checks the caps when attaches
a mmc/sd.
Sorry, but this doesn't work.
Even if PSCI is a generic FW interface, it doesn't mean that all PSCI
implementations will cut the vcc and vccq for the MMC card at system
suspend.
Indeed, there's nothing guaranteed here. Nor documented how it should
behave. Basically the firmware is free to power off the SoC. Or not do that.
"If firmware is involved, all odds are off".
I thought we could be guaranteed. But, I understood we could not be guaranteed...
quoted
quoted
Instead, you need to decide this based on some specific DT property.
Perhaps in conjunction with using pm_suspend_via_firmware().
Last time I was involved in a discussion about this, the PSCI people
didn't want to add any properties describing particular PSCI behavior...
"If firmware is involved, all odds are off".
So the only safe thing to do is to expect the worst, and prepare for it...
A headache point is an eMMC device consumes much power if that the system
doesn't cut the vcc and vccq and doesn’t enter the sleep mode.
In other words, in power consumption point of view, this patch will
cause a regression in such a case...
Indeed.
By the way, about adding specific DT property, the regulator can have
regulator-off-in-suspend property in regulator-state-mem subnode.
For now, we doesn't seem to get the property from a regulator consumer though.
So, I'll try to add an API of regulator for it.
Oh right, the eMMC is described in DT as being connected to two
regulators.
Note that the semantics of regulator-off-in-suspend are that the
regulator should be disabled (by the regulator core) during suspend, not
that the regulator is disabled during suspend by a third party.
No idea if that will work with a fixed-regulator without GPIO control,
but of course you can try.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
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
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
+ Rob
On Thu, 25 Jun 2020 at 11:26, Geert Uytterhoeven [off-list ref] wrote:
Hi Shimoda-san,
CC broonie
On Thu, Jun 25, 2020 at 8:31 AM Yoshihiro Shimoda
[off-list ref] wrote:
quoted
quoted
From: Geert Uytterhoeven, Sent: Wednesday, June 24, 2020 8:13 PM
On Wed, Jun 24, 2020 at 12:06 PM Ulf Hansson [off-list ref] wrote:
quoted
On Mon, 22 Jun 2020 at 04:25, Yoshihiro Shimoda
[off-list ref] wrote:
quoted
If pm_suspend_via_firmware() returns true, the system will be able
to cut both vcc and vccq in the suspend. So, call
mmc_poweroff_nofity() if pm_suspend_via_firmware() returns true.
Note that we should not update the MMC_CAP2_FULL_PWR_CYCLE caps
because the mmc_select_voltage() checks the caps when attaches
a mmc/sd.
Sorry, but this doesn't work.
Even if PSCI is a generic FW interface, it doesn't mean that all PSCI
implementations will cut the vcc and vccq for the MMC card at system
suspend.
Indeed, there's nothing guaranteed here. Nor documented how it should
behave. Basically the firmware is free to power off the SoC. Or not do that.
"If firmware is involved, all odds are off".
I thought we could be guaranteed. But, I understood we could not be guaranteed...
quoted
quoted
Instead, you need to decide this based on some specific DT property.
Perhaps in conjunction with using pm_suspend_via_firmware().
Last time I was involved in a discussion about this, the PSCI people
didn't want to add any properties describing particular PSCI behavior...
"If firmware is involved, all odds are off".
So the only safe thing to do is to expect the worst, and prepare for it...
A headache point is an eMMC device consumes much power if that the system
doesn't cut the vcc and vccq and doesn’t enter the sleep mode.
In other words, in power consumption point of view, this patch will
cause a regression in such a case...
Indeed.
quoted
By the way, about adding specific DT property, the regulator can have
regulator-off-in-suspend property in regulator-state-mem subnode.
For now, we doesn't seem to get the property from a regulator consumer though.
So, I'll try to add an API of regulator for it.
Oh right, the eMMC is described in DT as being connected to two
regulators.
Note that the semantics of regulator-off-in-suspend are that the
regulator should be disabled (by the regulator core) during suspend, not
that the regulator is disabled during suspend by a third party.
No idea if that will work with a fixed-regulator without GPIO control,
but of course you can try.
For mmc we already have a DT property, "full-pwr-cycle". Which tells
whether the host is able to completely power-cycle the card (some
host's manage power internally).
However, maybe the proper thing here would be to add a property of
regulator instead. If that doesn't make sense, I am also open to add a
new MMC property, "full-pwr-cycle-in-suspend".
Kind regards
Uffe
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Ulf Hansson, Sent: Friday, June 26, 2020 8:03 PM
+ Rob
On Thu, 25 Jun 2020 at 11:26, Geert Uytterhoeven [off-list ref] wrote:
quoted
Hi Shimoda-san,
CC broonie
On Thu, Jun 25, 2020 at 8:31 AM Yoshihiro Shimoda
[off-list ref] wrote:
quoted
quoted
From: Geert Uytterhoeven, Sent: Wednesday, June 24, 2020 8:13 PM
On Wed, Jun 24, 2020 at 12:06 PM Ulf Hansson [off-list ref] wrote:
quoted
On Mon, 22 Jun 2020 at 04:25, Yoshihiro Shimoda
[off-list ref] wrote:
quoted
If pm_suspend_via_firmware() returns true, the system will be able
to cut both vcc and vccq in the suspend. So, call
mmc_poweroff_nofity() if pm_suspend_via_firmware() returns true.
Note that we should not update the MMC_CAP2_FULL_PWR_CYCLE caps
because the mmc_select_voltage() checks the caps when attaches
a mmc/sd.
Sorry, but this doesn't work.
Even if PSCI is a generic FW interface, it doesn't mean that all PSCI
implementations will cut the vcc and vccq for the MMC card at system
suspend.
Indeed, there's nothing guaranteed here. Nor documented how it should
behave. Basically the firmware is free to power off the SoC. Or not do that.
"If firmware is involved, all odds are off".
I thought we could be guaranteed. But, I understood we could not be guaranteed...
quoted
quoted
Instead, you need to decide this based on some specific DT property.
Perhaps in conjunction with using pm_suspend_via_firmware().
Last time I was involved in a discussion about this, the PSCI people
didn't want to add any properties describing particular PSCI behavior...
"If firmware is involved, all odds are off".
So the only safe thing to do is to expect the worst, and prepare for it...
A headache point is an eMMC device consumes much power if that the system
doesn't cut the vcc and vccq and doesn’t enter the sleep mode.
In other words, in power consumption point of view, this patch will
cause a regression in such a case...
Indeed.
quoted
By the way, about adding specific DT property, the regulator can have
regulator-off-in-suspend property in regulator-state-mem subnode.
For now, we doesn't seem to get the property from a regulator consumer though.
So, I'll try to add an API of regulator for it.
Oh right, the eMMC is described in DT as being connected to two
regulators.
Note that the semantics of regulator-off-in-suspend are that the
regulator should be disabled (by the regulator core) during suspend, not
that the regulator is disabled during suspend by a third party.
No idea if that will work with a fixed-regulator without GPIO control,
but of course you can try.
For mmc we already have a DT property, "full-pwr-cycle". Which tells
whether the host is able to completely power-cycle the card (some
host's manage power internally).
However, maybe the proper thing here would be to add a property of
regulator instead.
My v4 patch was using a regulator property. But since I could not get
approval, we could not use this way, I think.
If that doesn't make sense, I am also open to add a
new MMC property, "full-pwr-cycle-in-suspend".
On Thu, 2 Jul 2020 at 14:37, Yoshihiro Shimoda
[off-list ref] wrote:
Hi Ulf,
quoted
From: Ulf Hansson, Sent: Friday, June 26, 2020 8:03 PM
+ Rob
On Thu, 25 Jun 2020 at 11:26, Geert Uytterhoeven [off-list ref] wrote:
quoted
Hi Shimoda-san,
CC broonie
On Thu, Jun 25, 2020 at 8:31 AM Yoshihiro Shimoda
[off-list ref] wrote:
quoted
quoted
From: Geert Uytterhoeven, Sent: Wednesday, June 24, 2020 8:13 PM
On Wed, Jun 24, 2020 at 12:06 PM Ulf Hansson [off-list ref] wrote:
quoted
On Mon, 22 Jun 2020 at 04:25, Yoshihiro Shimoda
[off-list ref] wrote:
quoted
If pm_suspend_via_firmware() returns true, the system will be able
to cut both vcc and vccq in the suspend. So, call
mmc_poweroff_nofity() if pm_suspend_via_firmware() returns true.
Note that we should not update the MMC_CAP2_FULL_PWR_CYCLE caps
because the mmc_select_voltage() checks the caps when attaches
a mmc/sd.
Sorry, but this doesn't work.
Even if PSCI is a generic FW interface, it doesn't mean that all PSCI
implementations will cut the vcc and vccq for the MMC card at system
suspend.
Indeed, there's nothing guaranteed here. Nor documented how it should
behave. Basically the firmware is free to power off the SoC. Or not do that.
"If firmware is involved, all odds are off".
I thought we could be guaranteed. But, I understood we could not be guaranteed...
quoted
quoted
Instead, you need to decide this based on some specific DT property.
Perhaps in conjunction with using pm_suspend_via_firmware().
Last time I was involved in a discussion about this, the PSCI people
didn't want to add any properties describing particular PSCI behavior...
"If firmware is involved, all odds are off".
So the only safe thing to do is to expect the worst, and prepare for it...
A headache point is an eMMC device consumes much power if that the system
doesn't cut the vcc and vccq and doesn’t enter the sleep mode.
In other words, in power consumption point of view, this patch will
cause a regression in such a case...
Indeed.
quoted
By the way, about adding specific DT property, the regulator can have
regulator-off-in-suspend property in regulator-state-mem subnode.
For now, we doesn't seem to get the property from a regulator consumer though.
So, I'll try to add an API of regulator for it.
Oh right, the eMMC is described in DT as being connected to two
regulators.
Note that the semantics of regulator-off-in-suspend are that the
regulator should be disabled (by the regulator core) during suspend, not
that the regulator is disabled during suspend by a third party.
No idea if that will work with a fixed-regulator without GPIO control,
but of course you can try.
For mmc we already have a DT property, "full-pwr-cycle". Which tells
whether the host is able to completely power-cycle the card (some
host's manage power internally).
However, maybe the proper thing here would be to add a property of
regulator instead.
My v4 patch was using a regulator property. But since I could not get
approval, we could not use this way, I think.
quoted
If that doesn't make sense, I am also open to add a
new MMC property, "full-pwr-cycle-in-suspend".
I thought this way was better. However, I'm wondering if adding such a new MMC
property to issue Power Off Notification in mmc_suspend() is really better
than the current implementation. This is because we don't have any completely
solutions now:
- Depend on board configuration (The board doesn't have "full-pwr-cycle").
- Depend on firmware on board [2]. So, even if adding a new MMC property,
it cannot sync the firmware condition. In fact, this is possible to
cause regression in power consumption point of view [3].
This is a generic problem with FWs.
I guess one could try to update the DTB using DT overlays, in case
some FW version changes.
- My environment (PSCI which is one of firmware) doesn't support
any ability to sync between the firmware and OS for now [4].
Yes, I see.
It seems like Geert has tried different approaches to convince PSCI
folkz, to find a solution for this, but it seems like none have make
it yet.
But, what do you think?
From the mmc DT property point of view, I am fine adding a
"full-pwr-cycle-in-suspend" - and then also update the mmc core's
behavior to use the eMMC power-off-notify command at system suspend.
However, I don't have a strong opinion, as the current solution with
the eMMC sleep command also seems to work.
My main concern with the current approach though, is not about wasting
energy, but rather that we are not doing a graceful shutdown of the
eMMC device. Instead we just cut VCCQ while the eMMC is "sleeping",
which in worst case could lead to internal data corruptions, but also
increased re-initialization time at system suspend.
From: Ulf Hansson, Sent: Monday, July 6, 2020 10:11 PM
On Thu, 2 Jul 2020 at 14:37, Yoshihiro Shimoda
[off-list ref] wrote:
quoted
Hi Ulf,
quoted
From: Ulf Hansson, Sent: Friday, June 26, 2020 8:03 PM
+ Rob
On Thu, 25 Jun 2020 at 11:26, Geert Uytterhoeven [off-list ref] wrote:
quoted
Hi Shimoda-san,
CC broonie
On Thu, Jun 25, 2020 at 8:31 AM Yoshihiro Shimoda
[off-list ref] wrote:
quoted
quoted
From: Geert Uytterhoeven, Sent: Wednesday, June 24, 2020 8:13 PM
On Wed, Jun 24, 2020 at 12:06 PM Ulf Hansson [off-list ref] wrote:
quoted
On Mon, 22 Jun 2020 at 04:25, Yoshihiro Shimoda
[off-list ref] wrote:
quoted
If pm_suspend_via_firmware() returns true, the system will be able
to cut both vcc and vccq in the suspend. So, call
mmc_poweroff_nofity() if pm_suspend_via_firmware() returns true.
Note that we should not update the MMC_CAP2_FULL_PWR_CYCLE caps
because the mmc_select_voltage() checks the caps when attaches
a mmc/sd.
Sorry, but this doesn't work.
Even if PSCI is a generic FW interface, it doesn't mean that all PSCI
implementations will cut the vcc and vccq for the MMC card at system
suspend.
Indeed, there's nothing guaranteed here. Nor documented how it should
behave. Basically the firmware is free to power off the SoC. Or not do that.
"If firmware is involved, all odds are off".
I thought we could be guaranteed. But, I understood we could not be guaranteed...
quoted
quoted
Instead, you need to decide this based on some specific DT property.
Perhaps in conjunction with using pm_suspend_via_firmware().
Last time I was involved in a discussion about this, the PSCI people
didn't want to add any properties describing particular PSCI behavior...
"If firmware is involved, all odds are off".
So the only safe thing to do is to expect the worst, and prepare for it...
A headache point is an eMMC device consumes much power if that the system
doesn't cut the vcc and vccq and doesn’t enter the sleep mode.
In other words, in power consumption point of view, this patch will
cause a regression in such a case...
Indeed.
quoted
By the way, about adding specific DT property, the regulator can have
regulator-off-in-suspend property in regulator-state-mem subnode.
For now, we doesn't seem to get the property from a regulator consumer though.
So, I'll try to add an API of regulator for it.
Oh right, the eMMC is described in DT as being connected to two
regulators.
Note that the semantics of regulator-off-in-suspend are that the
regulator should be disabled (by the regulator core) during suspend, not
that the regulator is disabled during suspend by a third party.
No idea if that will work with a fixed-regulator without GPIO control,
but of course you can try.
As other email thread of this trial as v4, I could not get approval [1].
[1]
For mmc we already have a DT property, "full-pwr-cycle". Which tells
whether the host is able to completely power-cycle the card (some
host's manage power internally).
However, maybe the proper thing here would be to add a property of
regulator instead.
My v4 patch was using a regulator property. But since I could not get
approval, we could not use this way, I think.
quoted
If that doesn't make sense, I am also open to add a
new MMC property, "full-pwr-cycle-in-suspend".
I thought this way was better. However, I'm wondering if adding such a new MMC
property to issue Power Off Notification in mmc_suspend() is really better
than the current implementation. This is because we don't have any completely
solutions now:
- Depend on board configuration (The board doesn't have "full-pwr-cycle").
- Depend on firmware on board [2]. So, even if adding a new MMC property,
it cannot sync the firmware condition. In fact, this is possible to
cause regression in power consumption point of view [3].
This is a generic problem with FWs.
I guess one could try to update the DTB using DT overlays, in case
some FW version changes.
It's interesting.
quoted
- My environment (PSCI which is one of firmware) doesn't support
any ability to sync between the firmware and OS for now [4].
Yes, I see.
It seems like Geert has tried different approaches to convince PSCI
folkz, to find a solution for this, but it seems like none have make
it yet.
quoted
But, what do you think?
From the mmc DT property point of view, I am fine adding a
"full-pwr-cycle-in-suspend" - and then also update the mmc core's
behavior to use the eMMC power-off-notify command at system suspend.
However, I don't have a strong opinion, as the current solution with
the eMMC sleep command also seems to work.
My main concern with the current approach though, is not about wasting
energy, but rather that we are not doing a graceful shutdown of the
eMMC device. Instead we just cut VCCQ while the eMMC is "sleeping",
which in worst case could lead to internal data corruptions, but also
increased re-initialization time at system suspend.
Thank you very much for your comments! I understood your main concern.
So, I'm thinking adding a new MMC property "full-pwr-cycle-in-suspend"
is useful now. After that, user could choose a graceful shutdown instead
of "sleeping" to avoid internal data corruptions in worst case at system
suspend. I'll submit such a patch.
Best regards,
Yoshihiro Shimoda