dm: ensure device names are unique

From: Stephen Warren <[email protected]>

It is possible for HW to contain multiple instances of the same device. In this case, the name passed to device_bind() may not be unique across all devices within its uclass. One example is a system with multiple identical PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Enhance the code to detect this situation, and append a sequence number so the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]>

---
Some possible issues with this patch:

1) Doing this in bind() rather than probe() means dev->seq isn't available, so can't be used to generate the unique name. This process should be done during bind() rather than probe() though, since devices can be seen (e.g. by running "dm tree") before they're probed. Perhaps the uclass_resolve_seq() should be called by bind() not probe().

2) uclass_find_device_by_name() needs to look up the uclass pointer again even though device_bind() already knows it.

3) Iterating over the list to find the count of devices in the uclass is a bit annoying. Should the uclass maintain this count so it doesn't need to be re-calculated each time?

Quoted 29 lines · Stephen Warren

From: Stephen Warren <[email protected]>

It is possible for HW to contain multiple instances of the same device. In
this case, the name passed to device_bind() may not be unique across all devices within its uclass. One example is a system with multiple identical
PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Enhance the code to detect this situation, and append a sequence number so
the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]>

---
Some possible issues with this patch:

1) Doing this in bind() rather than probe() means dev->seq isn't available, so can't be used to generate the unique name. This process should be done during bind() rather than probe() though, since devices can
be seen (e.g. by running "dm tree") before they're probed. Perhaps the uclass_resolve_seq() should be called by bind() not probe().

2) uclass_find_device_by_name() needs to look up the uclass pointer again even though device_bind() already knows it.

Quoted 4 lines

3) Iterating over the list to find the count of devices in the uclass is a
bit annoying. Should the uclass maintain this count so it doesn't need to be re-calculated each time?

4) This causes "ut dm autobind" to fail. I'll investigate that locally, but won't bother reposting until the questions above are considered.

Quoted 32 lines · Stephen Warren

Quoted 30 lines · Stephen Warren

From: Stephen Warren <[email protected]>

It is possible for HW to contain multiple instances of the same device. In
this case, the name passed to device_bind() may not be unique across all devices within its uclass. One example is a system with multiple identical
PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Enhance the code to detect this situation, and append a sequence number so
the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]>

---
Some possible issues with this patch:

1) Doing this in bind() rather than probe() means dev->seq isn't available, so can't be used to generate the unique name. This process should be done during bind() rather than probe() though, since devices can
be seen (e.g. by running "dm tree") before they're probed. Perhaps the uclass_resolve_seq() should be called by bind() not probe().

2) uclass_find_device_by_name() needs to look up the uclass pointer again even though device_bind() already knows it.

Quoted 8 lines

Quoted 4 lines

3) Iterating over the list to find the count of devices in the uclass is a
bit annoying. Should the uclass maintain this count so it doesn't need to be re-calculated each time?

4) This causes "ut dm autobind" to fail. I'll investigate that locally, but won't bother reposting until the questions above are considered.

It looks like that's because I need to add "gd->dm_root && " to the start of the following condition:

Quoted 5 lines

drivers/core/device.c +0 -0

@@ -62,6 +62,23 @@ int device_bind(struct udevice *parent, const struct
 driver *drv,
        dev->driver = drv;
        dev->uclass = uc;
 

Review int device_bind(struct udevice *parent, const struct · line 67

Quoted 2 lines

(quoted diff) +1 -0

+       if (!uclass_find_device_by_name(uc->uc_drv->id, dev->name, &dev2))
 {

... otherwise, the uc_class_get/find() nested inside that end up creating a second root uclass.

I can fold that into v2, pending any other comments.

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot

From: Stephen Warren <[email protected]>

It is possible for HW to contain multiple instances of the same device. In this case, the name passed to device_bind() may not be unique across all devices within its uclass. One example is a system with multiple identical PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Enhance the code to detect this situation, and append a sequence number so the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]>

---
Some possible issues with this patch:

1) Doing this in bind() rather than probe() means dev->seq isn't available, so can't be used to generate the unique name. This process should be done during bind() rather than probe() though, since devices can be seen (e.g. by running "dm tree") before they're probed. Perhaps the uclass_resolve_seq() should be called by bind() not probe().

2) uclass_find_device_by_name() needs to look up the uclass pointer again even though device_bind() already knows it.

3) Iterating over the list to find the count of devices in the uclass is a bit annoying. Should the uclass maintain this count so it doesn't need to be re-calculated each time?

 drivers/core/device-remove.c |  4 ++++
 drivers/core/device.c        | 23 ++++++++++++++++++++++-
 include/dm/device.h          |  3 +++
 3 files changed, 29 insertions(+), 1 deletion(-)

drivers/core/device-remove.c +4 -0

index e1714b2202b6..a2400989024a 100644
@@ -112,6 +112,10 @@ int device_unbind(struct udevice *dev)
 
        devres_release_all(dev);
 
+       if (dev->flags & DM_FLAG_ALLOC_NAME) {
+               free((char *)dev->name);
+               dev->name = NULL;
+       }
        free(dev);
 
        return 0;

drivers/core/device.c +22 -1

index 269087a084cf..fcac0073c642 100644
@@ -30,7 +30,7 @@ int device_bind(struct udevice *parent, const struct driver *drv,
                const char *name, void *platdata, int of_offset,
                struct udevice **devp)
 {
-       struct udevice *dev;
+       struct udevice *dev, *dev2;
        struct uclass *uc;
        int size, ret = 0;
 

@@ -62,6 +62,23 @@ int device_bind(struct udevice *parent, const struct driver *drv,
        dev->driver = drv;
        dev->uclass = uc;
 
+       if (!uclass_find_device_by_name(uc->uc_drv->id, dev->name, &dev2)) {
+               char *unique_name;
+               int count;
+               size = strlen(name) + 4 + 1;
+               unique_name = malloc(size);
+               if (!unique_name) {
+                       ret = -ENOMEM;
+                       goto fail_alloc1;
+               }
+               dev->flags |= DM_FLAG_ALLOC_NAME;
+               dev->name = unique_name;
+               count = 0;
+               list_for_each_entry(dev2, &uc->dev_head, uclass_node)
+                       count++;
+               snprintf(unique_name, size, "%s.%d", name, count);
+       }
+
        dev->seq = -1;
        dev->req_seq = -1;
        if (CONFIG_IS_ENABLED(OF_CONTROL) && CONFIG_IS_ENABLED(DM_SEQ_ALIAS)) {

@@ -186,6 +203,10 @@ fail_alloc2:
                dev->platdata = NULL;
        }
 fail_alloc1:
+       if (dev->flags & DM_FLAG_ALLOC_NAME) {
+               free((char *)dev->name);
+               dev->name = NULL;
+       }
        devres_release_all(dev);
 
        free(dev);

include/dm/device.h +3 -1

index dad7591dfacb..976ba72a550c 100644
@@ -41,6 +41,9 @@ struct driver_info;
 /* Device is bound */
 #define DM_FLAG_BOUND                  (1 << 6)
 
+/* DM is responsible for allocating and freeing device name */
+#define DM_FLAG_ALLOC_NAME             (1 << 7)
+
 /**
  * struct udevice - An instance of a driver
  *
--

2.8.1

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot

Quoted 13 lines · Stephen Warren

From: Stephen Warren <[email protected]>

It is possible for HW to contain multiple instances of the same device. In this case, the name passed to device_bind() may not be unique across all devices within its uclass. One example is a system with multiple identical PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Enhance the code to detect this situation, and append a sequence number so the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]>

Quoted 8 lines

---
Some possible issues with this patch:

1) Doing this in bind() rather than probe() means dev->seq isn't available, so can't be used to generate the unique name. This process should be done during bind() rather than probe() though, since devices can be seen (e.g. by running "dm tree") before they're probed. Perhaps the uclass_resolve_seq() should be called by bind() not probe().

We (Simon and I) had discussion about this when I first added support for eth devices. He convinced me the correct time for seq to be evaluated is at probe time. I can dig through the mail history for reasons if you're interested.

Quoted 2 lines

2) uclass_find_device_by_name() needs to look up the uclass pointer again even though device_bind() already knows it.

Maybe we could add a parameter to provide the pointer and if it's NULL, then it gets looked up. Might be a bit noisy change, though. Is that optimization very valuable?

Quoted 3 lines

3) Iterating over the list to find the count of devices in the uclass is a bit annoying. Should the uclass maintain this count so it doesn't need to be re-calculated each time?

These lists aren't long, right? It seems like the optimization to store the value is only helpful if the lists could be expected to be long, or the size is looked up many times.

Quoted 26 lines · Joe Hershberger

Quoted 22 lines · Stephen Warren

From: Stephen Warren <[email protected]>

It is possible for HW to contain multiple instances of the same device. In this case, the name passed to device_bind() may not be unique across all devices within its uclass. One example is a system with multiple identical PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Enhance the code to detect this situation, and append a sequence number so the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]>

---
Some possible issues with this patch:

1) Doing this in bind() rather than probe() means dev->seq isn't available, so can't be used to generate the unique name. This process should be done during bind() rather than probe() though, since devices can be seen (e.g. by running "dm tree") before they're probed. Perhaps the uclass_resolve_seq() should be called by bind() not probe().

Quoted 4 lines

We (Simon and I) had discussion about this when I first added support for eth devices. He convinced me the correct time for seq to be evaluated is at probe time. I can dig through the mail history for reasons if you're interested.

That might be interesting. Right now, about the only thing I see moving the evaluation from probe to bind would do is very marginally increase the amount of work done in bind, which potentially might be skipped if a particular Ethernet device was never probed. I would not expect this to make a noticeable difference, especially since IIRC when the network is first used, all the Ethernet devices are probed anyway, so this just moves work around?

Quoted 5 lines

Quoted 3 lines

2) uclass_find_device_by_name() needs to look up the uclass pointer again even though device_bind() already knows it.

Quoted 3 lines

Maybe we could add a parameter to provide the pointer and if it's NULL, then it gets looked up. Might be a bit noisy change, though. Is that optimization very valuable?

I don't imagine there are too many uclasses, so it's probably not going to save too much time. Still, Simon has in the past objected to code that parses DT scanning the DT to find parent offsets, and this feels like exactly the same kind of thing. I'm not personally concerned about it; simply pointing it out in case anyone wanted that to be addressed.

Quoted 8 lines

Quoted 3 lines

3) Iterating over the list to find the count of devices in the uclass is a bit annoying. Should the uclass maintain this count so it doesn't need to be re-calculated each time?

These lists aren't long, right? It seems like the optimization to store the value is only helpful if the lists could be expected to be long, or the size is looked up many times.

Yes, I don't imagine this would be a particular issue in practice.

It is possible for HW to contain multiple instances of the same device. In this case, the name passed to device_bind() may not be unique across all devices within its uclass. One example is a system with multiple identical PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Enhance the code to detect this situation, and append a sequence number so the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]>

---
Some possible issues with this patch:

1) Doing this in bind() rather than probe() means dev->seq isn't available, so can't be used to generate the unique name. This process should be done during bind() rather than probe() though, since devices can be seen (e.g. by running "dm tree") before they're probed. Perhaps the uclass_resolve_seq() should be called by bind() not probe().

2) uclass_find_device_by_name() needs to look up the uclass pointer again even though device_bind() already knows it.

3) Iterating over the list to find the count of devices in the uclass is a bit annoying. Should the uclass maintain this count so it doesn't need to be re-calculated each time?
---
drivers/core/device-remove.c | 4 ++++
drivers/core/device.c | 23 ++++++++++++++++++++++- include/dm/device.h | 3 +++

3 files changed, 29 insertions(+), 1 deletion(-)

Hi Stephen,

Quoted 7 lines · Stephen Warren

It is possible for HW to contain multiple instances of the same device. In this case, the name passed to device_bind() may not be unique across all devices within its uclass. One example is a system with multiple identical PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Quoted 4 lines

Enhance the code to detect this situation, and append a sequence number so the device name to ensure uniqueness.

Signed-off-by Stephen Warren <swarren at nvidia.com <[email protected]>>

I would rather that the caller handles this. But failing this perhaps a new function that does it? Is this for the Ethernet use case?

Quoted 14 lines · Simon Glass

Hi Stephen,

Quoted 1 line · Stephen Warren

It is possible for HW to contain multiple instances of the same device.

In

Quoted 1 line

this case, the name passed to device_bind() may not be unique across

all

Quoted 1 line

devices within its uclass. One example is a system with multiple

identical

Quoted 4 lines

PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Quoted 8 lines

Quoted 1 line

Enhance the code to detect this situation, and append a sequence

number so

Quoted 3 lines

the device name to ensure uniqueness.

Signed-off-by Stephen Warren <swarren at nvidia.com <[email protected]>>

I would rather that the caller handles this. But failing this perhaps a new function that does it? Is this for the Ethernet use case?

Wouldn't all callers of this function simply call the new function? I'm not aware of any case where the code to avoid duplicate names would not be desired.

I hit this for the Ethernet case, but I believe it applies to any type of device at all; see another possible trigger case in the commit description.

Hi Stephen,

Quoted 29 lines · Stephen Warren

Quoted 14 lines · Simon Glass

Hi Stephen,

Quoted 1 line · Stephen Warren

It is possible for HW to contain multiple instances of the same device.

In

Quoted 1 line

this case, the name passed to device_bind() may not be unique across

all

Quoted 1 line

devices within its uclass. One example is a system with multiple

identical

Quoted 4 lines

PCI Ethernet devices. Another might be a system with multiple identical I2C GPIO expanders, each connected to a separate I2C bus, yet using the same I2C address on that bus and hence having the same DT node name.

Quoted 8 lines

Quoted 1 line

Enhance the code to detect this situation, and append a sequence

number so

Quoted 3 lines

the device name to ensure uniqueness.

Signed-off-by Stephen Warren <swarren at nvidia.com <[email protected]>>

I would rather that the caller handles this. But failing this perhaps a new function that does it? Is this for the Ethernet use case?

Wouldn't all callers of this function simply call the new function? I'm not aware of any case where the code to avoid duplicate names would not be desired.

Quoted 2 lines

I hit this for the Ethernet case, but I believe it applies to any type of device at all; see another possible trigger case in the commit description.

This does not happen with devices from the device tree. It only happens with auto-probed devices. Your I2C GPIO example is odd but I'd rather solve that by using the device tree node name.

Quoted 41 lines · Simon Glass

Hi Stephen,

Quoted 37 lines · Stephen Warren

Quoted 19 lines · Simon Glass

Hi Stephen,

Quoted 1 line · Stephen Warren

It is possible for HW to contain multiple instances of the same

device.
In

Quoted 1 line

this case, the name passed to device_bind() may not be unique across

all

Quoted 1 line

devices within its uclass. One example is a system with multiple

identical

Quoted 1 line

PCI Ethernet devices. Another might be a system with multiple

identical

Quoted 1 line

I2C GPIO expanders, each connected to a separate I2C bus, yet using

the

Quoted 3 lines

same I2C address on that bus and hence having the same DT node name.

Enhance the code to detect this situation, and append a sequence

Quoted 8 lines

number so

Quoted 3 lines

the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]

<[email protected]>>

I would rather that the caller handles this. But failing this perhaps a new function that does it? Is this for the Ethernet use case?

Wouldn't all callers of this function simply call the new function? I'm not
aware of any case where the code to avoid duplicate names would not be desired.

Quoted 8 lines

Quoted 3 lines

I hit this for the Ethernet case, but I believe it applies to any type of device at all; see another possible trigger case in the commit description.

This does not happen with devices from the device tree. It only happens with auto-probed devices. Your I2C GPIO example is odd but I'd rather solve that by using the device tree node name.

DT itself imposes no such rule; node names must be unique only within their parent node but there's no restriction on identical node names appearing in different parts of the tree.

If U-Boot imposes that rule on DT, then there's no way in general that we can guarantee U-Boot will be able to use standard DTs (i.e. identical to those used by Linux or any other OS) for any platform; it'd be another change someone would need to make to transform a DT to be "U-Boot compatible", which rather reduces a potential benefit of DT for U-Boot; being able to just drop a DT in and have it work.

It would be possible for U-Boot to decouple its internal device name from the DT node name. In which case, your statement would work. However, I don't think that's the case at the moment, and in fact it's effectively what this patch is doing, although admittedly there are other ways of doing this.

Hi Stephen,

Quoted 56 lines · Stephen Warren

Quoted 41 lines · Simon Glass

Hi Stephen,

Quoted 37 lines · Stephen Warren

Quoted 19 lines · Simon Glass

Hi Stephen,

Quoted 1 line · Stephen Warren

It is possible for HW to contain multiple instances of the same

device.
In

Quoted 1 line

this case, the name passed to device_bind() may not be unique across

all

Quoted 1 line

devices within its uclass. One example is a system with multiple

identical

Quoted 1 line

PCI Ethernet devices. Another might be a system with multiple

identical

Quoted 1 line

I2C GPIO expanders, each connected to a separate I2C bus, yet using

the

Quoted 3 lines

same I2C address on that bus and hence having the same DT node name.

Enhance the code to detect this situation, and append a sequence

Quoted 8 lines

number so

Quoted 3 lines

the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]

<[email protected]>>

I would rather that the caller handles this. But failing this perhaps a new function that does it? Is this for the Ethernet use case?

Wouldn't all callers of this function simply call the new function? I'm not
aware of any case where the code to avoid duplicate names would not be desired.

Quoted 8 lines

Quoted 3 lines

I hit this for the Ethernet case, but I believe it applies to any type of device at all; see another possible trigger case in the commit description.

This does not happen with devices from the device tree. It only happens with auto-probed devices. Your I2C GPIO example is odd but I'd rather solve that by using the device tree node name.

DT itself imposes no such rule; node names must be unique only within their parent node but there's no restriction on identical node names appearing in different parts of the tree.

Quoted 6 lines

If U-Boot imposes that rule on DT, then there's no way in general that we can guarantee U-Boot will be able to use standard DTs (i.e. identical to those used by Linux or any other OS) for any platform; it'd be another change someone would need to make to transform a DT to be "U-Boot compatible", which rather reduces a potential benefit of DT for U-Boot; being able to just drop a DT in and have it work.

U-Boot does not impose a rule. If you want duplicate device names you can have them. I think it is bad practice though.

Quoted 5 lines

It would be possible for U-Boot to decouple its internal device name from the DT node name. In which case, your statement would work. However, I don't think that's the case at the moment, and in fact it's effectively what this patch is doing, although admittedly there are other ways of doing this.

Anyway I believe my point stands. Whereas users can edit the device tree and avoid conflicts they cannot do this with auto-probed devices. So for that case we should have a way of allocating a name before calling device_bind().

Quoted 94 lines · Simon Glass

Hi Stephen,

Quoted 70 lines · Stephen Warren

Quoted 56 lines · Simon Glass

Hi Stephen,

Quoted 44 lines · Stephen Warren

Quoted 24 lines · Simon Glass

Hi Stephen,

Quoted 1 line · Stephen Warren

It is possible for HW to contain multiple instances of the same

device.
In

Quoted 1 line

this case, the name passed to device_bind() may not be unique

across
all

Quoted 1 line

devices within its uclass. One example is a system with multiple

identical

Quoted 1 line

PCI Ethernet devices. Another might be a system with multiple

identical

Quoted 1 line

I2C GPIO expanders, each connected to a separate I2C bus, yet

using
the

Quoted 1 line

same I2C address on that bus and hence having the same DT node

name.

Quoted 2 lines

Enhance the code to detect this situation, and append a sequence

number so

Quoted 8 lines

Quoted 3 lines

the device name to ensure uniqueness.

Signed-off-by Stephen Warren <[email protected]

<[email protected]>>

I would rather that the caller handles this. But failing this perhaps a
new function that does it? Is this for the Ethernet use case?

Wouldn't all callers of this function simply call the new function? I'm not
aware of any case where the code to avoid duplicate names would not be desired.

Quoted 4 lines

I hit this for the Ethernet case, but I believe it applies to any type of
device at all; see another possible trigger case in the commit description.

Quoted 3 lines

This does not happen with devices from the device tree. It only happens with auto-probed devices. Your I2C GPIO example is odd but I'd rather solve that by using the device tree node name.

DT itself imposes no such rule; node names must be unique only within their
parent node but there's no restriction on identical node names appearing in
different parts of the tree.

Quoted 6 lines

If U-Boot imposes that rule on DT, then there's no way in general that we can guarantee U-Boot will be able to use standard DTs (i.e. identical to those used by Linux or any other OS) for any platform; it'd be another change someone would need to make to transform a DT to be "U-Boot compatible", which rather reduces a potential benefit of DT for U-Boot; being able to just drop a DT in and have it work.

U-Boot does not impose a rule. If you want duplicate device names you can have them. I think it is bad practice though.

Quoted 7 lines

It would be possible for U-Boot to decouple its internal device name from the DT node name. In which case, your statement would work. However, I don't
think that's the case at the moment, and in fact it's effectively what this
patch is doing, although admittedly there are other ways of doing this.

Quoted 4 lines

Anyway I believe my point stands. Whereas users can edit the device tree and avoid conflicts they cannot do this with auto-probed devices. So for that case we should have a way of allocating a name before calling device_bind().

OK. I'm just going to solve this by unplugging the second Ethernet card, or holding this patch locally for the case when I need to test the PCIe port.