NFC: NCI: make parent aware in PM terms
Subsystems (2)
PATCH 1/1 NFC: NCI: make parent aware in PM terms
The NCI device is a child of an i2c device. If the i2c layer uses runtime PM the power to the NFC device can be cut whenever the i2c layer is done transmitting data to the NFC device.
Under these conditions NFC can not work, as it needs power to wait for reception of packets.
The necessary extension of runtime PM to the NFC device requires that it be activated as a child of the i2c device. It is not necessary to do any runtime PM operations. This will disable runtime PM for this branch of the tree, but otherwise the NFC device is inoperable.
--- drivers/nfc/nxp-nci/i2c.c | 10 ++++++++++ 1 file changed, 10 insertions(+)
drivers/nfc/nxp-nci/i2c.c +10 -0
@@ -18,6 +18,7 @@#include <linux/interrupt.h> #include <linux/module.h> #include <linux/nfc.h> +#include <linux/pm_runtime.h> #include <linux/gpio/consumer.h> #include <asm/unaligned.h>
@@ -261,6 +262,7 @@ static int nxp_nci_i2c_probe(struct i2c_client *client, {struct device *dev = &client->dev;struct nxp_nci_i2c_phy *phy; +struct nfc_dev *ndev;int r;if (!i2c_check_functionality (client->adapter,I2C_FUNC_I2C )) {
@@ -297,6 +299,11 @@ static int nxp_nci_i2c_probe(struct i2c_client *client,if (r <0 )return r; + ndev = phy->ndev->nfc_dev; +pm_runtime_set_active (&ndev->dev); +pm_runtime_enable (&ndev->dev); +pm_runtime_mark_last_busy (&ndev->dev); + r = request_threaded_irq(client->irq, NULL, nxp_nci_i2c_irq_thread_fn, IRQF_TRIGGER_RISING | IRQF_ONESHOT,
@@ -310,9 +317,12 @@ static int nxp_nci_i2c_probe(struct i2c_client *client,static int nxp_nci_i2c_remove (struct i2c_client *client) {struct nxp_nci_i2c_phy *phy =i2c_get_clientdata (client); +struct nfc_dev *ndev = phy->ndev->nfc_dev;nxp_nci_remove (phy->ndev);free_irq (client->irq, phy); +pm_runtime_disable (&ndev->dev); +pm_runtime_set_suspended (&ndev->dev);return 0 ; }
Signature · 1 line
2.26.2
Quoted 7 lines · Oliver Neukum
The NCI device is a child of an i2c device. If the i2c layer uses runtime PM the power to the NFC device can be cut whenever the i2c layer is done transmitting data to the NFC device.
Under these conditions NFC can not work, as it needs power to wait for reception of packets.
Hi,
Thanks for the patch.
However this
is unparseable.
Please wrap commit
message around 75th character:
https://elixir.bootlin.com/linux/v5.13/source/Documentation/process/submitting-patches.rst#L124
The subject prefix: "nfc: nxp-nci: "
Please also Cc all people and lists. You missed here netdev and linux-kernel, so this cannot be applied.
Quoted 35 lines
The necessary extension of runtime PM to the NFC device requires that it be activated as a child of the i2c device. It is not necessary to do any runtime PM operations. This will disable runtime PM for this branch of the tree, but otherwise the NFC device is inoperable.
Signed-off-by Oliver Neukum <[email protected]>--- drivers/nfc/nxp-nci/i2c.c | 10 ++++++++++ 1 file changed, 10 insertions(+)drivers/nfc/nxp-nci/i2c.c +2
-0@@ -18,6 +18,7 @@#include <linux/interrupt.h> #include <linux/module.h> #include <linux/nfc.h> +#include <linux/pm_runtime.h> #include <linux/gpio/consumer.h> #include <asm/unaligned.h> @@ -261,6 +262,7 @@ static int nxp_nci_i2c_probe(struct i2c_client *client, {struct device *dev = &client->dev;struct nxp_nci_i2c_phy *phy; +struct nfc_dev *ndev;int r;if (!i2c_check_functionality (client->adapter,I2C_FUNC_I2C )) {
drivers/nfc/nxp-nci/i2c.c static int nxp_nci_i2c_probe(struct i2c_client *client, · line 305
Quoted 7 lines
drivers/nfc/nxp-nci/i2c.c +4
-0@@ -297,6 +299,11 @@ static int nxp_nci_i2c_probe(struct i2c_client *client,if (r <0 )return r; + ndev = phy->ndev->nfc_dev; +pm_runtime_set_active (&ndev->dev); +pm_runtime_enable (&ndev->dev); +pm_runtime_mark_last_busy (&ndev->dev);
Setting PM for someone else does not look correct. This breaks encapsulation and separation of these files. The NCI device (nxp_nci_probe() and other parts of core.c) should instead manage it's own runtime PM.
Quoted 8 lines
drivers/nfc/nxp-nci/i2c.c +1
-0+ r = request_threaded_irq(client->irq, NULL, nxp_nci_i2c_irq_thread_fn, IRQF_TRIGGER_RISING | IRQF_ONESHOT,@@ -310,9 +317,12 @@ static int nxp_nci_i2c_probe(struct i2c_client *client,static int nxp_nci_i2c_remove (struct i2c_client *client) {
drivers/nfc/nxp-nci/i2c.c static int nxp_nci_i2c_probe(struct i2c_client *client, · line 328
Quoted 10 lines
drivers/nfc/nxp-nci/i2c.c +3
-0struct nxp_nci_i2c_phy *phy =i2c_get_clientdata (client); +struct nfc_dev *ndev = phy->ndev->nfc_dev;nxp_nci_remove (phy->ndev);free_irq (client->irq, phy); +pm_runtime_disable (&ndev->dev); +pm_runtime_set_suspended (&ndev->dev);return 0 ; }
Best regards,
Krzysztof
Quoted 62 lines · Krzysztof Kozlowski
Quoted 7 lines · Oliver Neukum
The NCI device is a child of an i2c device. If the i2c layer uses runtime PM the power to the NFC device can be cut whenever the i2c layer is done transmitting data to the NFC device.
Under these conditions NFC can not work, as it needs power to wait for reception of packets.Hi,
Thanks for the patch.
However this
is unparseable.
Please wrap commit
message around 75th character:
https://elixir.bootlin.com/linux/v5.13/source/Documentation/process/submitting-patches.rst#L124The subject prefix: "nfc: nxp-nci: "
Please also Cc all people and lists. You missed here netdev and linux-kernel, so this cannot be applied.
Quoted 38 lines
The necessary extension of runtime PM to the NFC device requires that it be activated as a child of the i2c device. It is not necessary to do any runtime PM operations. This will disable runtime PM for this branch of the tree, but otherwise the NFC device is inoperable.
Signed-off-by Oliver Neukum <[email protected]>--- drivers/nfc/nxp-nci/i2c.c | 10 ++++++++++ 1 file changed, 10 insertions(+)drivers/nfc/nxp-nci/i2c.c +2
-0@@ -18,6 +18,7 @@#include <linux/interrupt.h> #include <linux/module.h> #include <linux/nfc.h> +#include <linux/pm_runtime.h> #include <linux/gpio/consumer.h> #include <asm/unaligned.h> @@ -261,6 +262,7 @@ static int nxp_nci_i2c_probe(struct i2c_client *client, {struct device *dev = &client->dev;struct nxp_nci_i2c_phy *phy; +struct nfc_dev *ndev;int r;if (!i2c_check_functionality (client->adapter,I2C_FUNC_I2C )) {@@ -297,6 +299,11 @@ static int nxp_nci_i2c_probe(struct i2c_client *client,if (r <0 )return r;
Quoted 10 lines
drivers/nfc/nxp-nci/i2c.c static int nxp_nci_i2c_probe(struct i2c_client *client, · line 305
Quoted 4 lines
drivers/nfc/nxp-nci/i2c.c +4
-0+ ndev = phy->ndev->nfc_dev; +pm_runtime_set_active (&ndev->dev); +pm_runtime_enable (&ndev->dev); +pm_runtime_mark_last_busy (&ndev->dev);Setting PM for someone else does not look correct. This breaks encapsulation and separation of these files. The NCI device (
nxp_nci_probe()and other parts ofcore.c) should instead manage it's own runtime PM.
Except this, the code is really weird... there is no runtime PM in the drivers but this enables it. For what purpose? It also marks it as last busy but there is no autosuspend. I think I missed entirely the purpose of this code.
Best regards,
Krzysztof
On 19.08.21 16:45, Krzysztof Kozlowski wrote
Hi,
Quoted 10 lines
Hi,
Thanks for the patch.
However this
is unparseable.
Please wrap commit
message around 75th character:
https://elixir.bootlin.com/linux/v5.13/source/Documentation/process/submitting-patches.rst#L124The subject prefix: "nfc: nxp-nci: "
OK.
Quoted 2 lines
Please also Cc all people and lists. You missed here netdev and linux-kernel, so this cannot be applied.
Do you really want LKML on CC for all NFC patches?
Quoted 37 lines
Quoted 37 lines
The necessary extension of runtime PM to the NFC device requires that it be activated as a child of the i2c device. It is not necessary to do any runtime PM operations. This will disable runtime PM for this branch of the tree, but otherwise the NFC device is inoperable.
Signed-off-by Oliver Neukum <[email protected]>--- drivers/nfc/nxp-nci/i2c.c | 10 ++++++++++ 1 file changed, 10 insertions(+)drivers/nfc/nxp-nci/i2c.c +2
-0@@ -18,6 +18,7 @@#include <linux/interrupt.h> #include <linux/module.h> #include <linux/nfc.h> +#include <linux/pm_runtime.h> #include <linux/gpio/consumer.h> #include <asm/unaligned.h> @@ -261,6 +262,7 @@ static int nxp_nci_i2c_probe(struct i2c_client *client, {struct device *dev = &client->dev;struct nxp_nci_i2c_phy *phy; +struct nfc_dev *ndev;int r;if (!i2c_check_functionality (client->adapter,I2C_FUNC_I2C )) {@@ -297,6 +299,11 @@ static int nxp_nci_i2c_probe(struct i2c_client *client,if (r <0 )return r;
Quoted 5 lines
drivers/nfc/nxp-nci/i2c.c static int nxp_nci_i2c_probe(struct i2c_client *client, · line 305
Quoted 4 lines
drivers/nfc/nxp-nci/i2c.c +4
-0+ ndev = phy->ndev->nfc_dev; +pm_runtime_set_active (&ndev->dev); +pm_runtime_enable (&ndev->dev); +pm_runtime_mark_last_busy (&ndev->dev);Setting PM for someone else does not look correct. This breaks
As far as I can tell I am doing this for the device noted in nfc_dev. Could you elaborate?
Quoted 3 lines
encapsulation and separation of these files. The NCI device (
nxp_nci_probe()and other parts ofcore.c) should instead manage it's own runtime PM.
That would be better. The problem I encountered is that i2c and USB have different
needs concerning runtime PM. Any ideas?
Regards
Oliver
Quoted 12 lines · Oliver Neukum
On 19.08.21 16:45, Krzysztof Kozlowski wrote
Hi,Quoted 9 lines
Hi,
Thanks for the patch.
However this
is unparseable.
Please wrap commit
message around 75th character:
https://elixir.bootlin.com/linux/v5.13/source/Documentation/process/submitting-patches.rst#L124
Quoted 5 lines
Quoted 1 line
The subject prefix: "nfc: nxp-nci: "
OK.
Quoted 2 lines
Please also Cc all people and lists. You missed here netdev and linux-kernel, so this cannot be applied.
Do you really want LKML on CC for all NFC patches?
I personally like it a lot because I have filters organized with it. Nowadays no one reads LKML itself (too big volume) so it is purely for archiving on lore.kernel.org for searching and for people's filters.
Therefore unless someone here objects, I would prefer to Cc LKML as well. Anyway, netdev is important as it is tracked by patchwork.
Quoted 37 lines
Quoted 37 lines
Quoted 37 lines
The necessary extension of runtime PM to the NFC device requires that it be activated as a child of the i2c device. It is not necessary to do any runtime PM operations. This will disable runtime PM for this branch of the tree, but otherwise the NFC device is inoperable.
Signed-off-by Oliver Neukum <[email protected]>--- drivers/nfc/nxp-nci/i2c.c | 10 ++++++++++ 1 file changed, 10 insertions(+)drivers/nfc/nxp-nci/i2c.c +2
-0@@ -18,6 +18,7 @@#include <linux/interrupt.h> #include <linux/module.h> #include <linux/nfc.h> +#include <linux/pm_runtime.h> #include <linux/gpio/consumer.h> #include <asm/unaligned.h> @@ -261,6 +262,7 @@ static int nxp_nci_i2c_probe(struct i2c_client *client, {struct device *dev = &client->dev;struct nxp_nci_i2c_phy *phy; +struct nfc_dev *ndev;int r;if (!i2c_check_functionality (client->adapter,I2C_FUNC_I2C )) {@@ -297,6 +299,11 @@ static int nxp_nci_i2c_probe(struct i2c_client *client,if (r <0 )return r;
Quoted 7 lines
Quoted 5 lines
drivers/nfc/nxp-nci/i2c.c static int nxp_nci_i2c_probe(struct i2c_client *client, · line 305
Quoted 4 lines
drivers/nfc/nxp-nci/i2c.c +4
-0+ ndev = phy->ndev->nfc_dev; +pm_runtime_set_active (&ndev->dev); +pm_runtime_enable (&ndev->dev); +pm_runtime_mark_last_busy (&ndev->dev);Setting PM for someone else does not look correct. This breaks
As far as I can tell I am doing this for the device noted in
nfc_dev. Could you elaborate?
I meant that it looks unusual that you don't do it for your own device (client->dev) but for device allocated in different unit. Here, you receive client->dev and mostly you should play only with it.
While I am looking at this more, there is another issue actually - you touch runtime PM of NFC/NCI core device but it's the NFC/NCI core who should handle it's own runtime PM.
Quoted 7 lines
Quoted 3 lines
encapsulation and separation of these files. The NCI device (
nxp_nci_probe()and other parts ofcore.c) should instead manage it's own runtime PM.That would be better. The problem I encountered is that i2c and USB have different
needs concerning runtime PM. Any ideas?
I think we look at different code bases. I failed to find the USB device... It's difficult for me to judge how the final runtime PM should look like because actually you don't do any runtime PM here. You just enable it. Where are the callbacks? Where is suspending and resuming (get/put)?
Best regards,
Krzysztof
Quoted 2 lines · Krzysztof Kozlowski
Quoted 1 line · Oliver Neukum
On 19.08.21 16:45, Krzysztof Kozlowski wrote
Hi,
Quoted 6 lines
Quoted 6 lines
I personally like it a lot because I have filters organized with it. Nowadays no one reads LKML itself (too big volume) so it is purely for archiving on lore.kernel.org for searching and for people's filters.
Therefore unless someone here objects, I would prefer to Cc LKML as well. Anyway, netdev is important as it is tracked by patchwork.
very well, It shall be done.
Quoted 9 lines
Quoted 2 lines
As far as I can tell I am doing this for the device noted in
nfc_dev. Could you elaborate?I meant that it looks unusual that you don't do it for your own device (
client->dev) but for device allocated in different unit. Here, you receiveclient->devand mostly you should play only with it.While I am looking at this more, there is another issue actually - you touch runtime PM of NFC/NCI core device but it's the NFC/NCI core who should handle it's own runtime PM.
Well, I looked into that and that was very difficult. The NFC core has no idea what hardware it is running on. On my test hardware as far as I can tell, the power supply of my NFC hardware is just tied to the power supply of the i2c host controller, which is a PCI device.
It goes into D3cold when the i2c bus is not transferring anything. That makes it impossible to actually communicate over NFC.
However, at least on USB sending a device into USB suspend with remote wakeup should work. So I do not see how to put this into NFC core.
Regards
Oliver
Quoted 30 lines · Oliver Neukum
Quoted 2 lines · Krzysztof Kozlowski
Quoted 1 line · Oliver Neukum
On 19.08.21 16:45, Krzysztof Kozlowski wrote
Hi,
Quoted 6 lines
Quoted 6 lines
I personally like it a lot because I have filters organized with it. Nowadays no one reads LKML itself (too big volume) so it is purely for archiving on lore.kernel.org for searching and for people's filters.
Therefore unless someone here objects, I would prefer to Cc LKML as well. Anyway, netdev is important as it is tracked by patchwork.
very well, It shall be done.
Quoted 9 lines
Quoted 2 lines
As far as I can tell I am doing this for the device noted in
nfc_dev. Could you elaborate?I meant that it looks unusual that you don't do it for your own device (
client->dev) but for device allocated in different unit. Here, you receiveclient->devand mostly you should play only with it.While I am looking at this more, there is another issue actually - you touch runtime PM of NFC/NCI core device but it's the NFC/NCI core who should handle it's own runtime PM.
Well, I looked into that and that was very difficult. The NFC core has no idea what hardware it is running on. On my test hardware as far as I can tell, the power supply of my NFC hardware is just tied to the power supply of the i2c host controller, which is a PCI device.
Quoted 6 lines
It goes into D3cold when the i2c bus is not transferring anything. That makes it impossible to actually communicate over NFC.
However, at least on USB sending a device into USB suspend with remote wakeup should work. So I do not see how to put this into NFC core.
If I understand your case correctly, you have a tree of devices:
PCI I2C bus controller
- NXP NFC I2C device
- NFC NCI device
and the only thing you care is to make PCI I2C bus resumed all the time. The simplest seems to add PM runtime to the NXP NFC I2C device (so i2c.c) with get_noresume+active+enable.
Best regards,
Krzysztof