Re: [PATCH] acpi: device: Fix check for sequence number
Hi Wolfgang,
On Thu, Sep 10, 2020 at 3:01 PM Wolfgang Wallner
wrote:
>
> Hi Simon,
>
> -"Simon Glass" schrieb: -----
> > Betreff: Re: [PATCH] acpi: device: Fix check for sequence number
> >
> > Hi Wolfgang,
> >
> > On Thu, 13 Aug 2020 at 01:23, Wolfgang Wallner
> > wrote:
> > >
> > > Hi Simon,
> > >
> > > -"Simon Glass" schrieb: -
> > > > Betreff: Re: [PATCH] acpi: device: Fix check for sequence number
> > > >
> > > > Hi Wolfgang,
> > > >
> > > > On Thu, 30 Jul 2020 at 06:47, Wolfgang Wallner
> > > > wrote:
> > > > >
> > > > > Currently the function acpi_check_seq() checks whether dev->req_seq is
> > > > > unequal to "-1", but it should actually check dev->seq. Change it to
> > > > > check dev->seq.
> > > > >
> > > > > For req_seq the value "-1" would be a valid (meaning 'any'), while for
> > > > > seq the value "-1" means 'none' and is not valid.
> > > > >
> > > > > Quoting the description of udevice in device.h:
> > > > > * @req_seq: Requested sequence number for this device (-1 = any)
> > > > > * @seq: Allocated sequence number for this device (-1 = none).
> > > > > * This is set up when the device is probed and will be unique
> > > > > * within the device's uclass.
> > > > >
> > > > > Signed-off-by: Wolfgang Wallner
> > > > >
> > > > > Fixes: commit fefac0b0643b ("dm: acpi: Enhance acpi_get_name()")
> > > > >
> > > > > ---
> > > > >
> > > > > lib/acpi/acpi_device.c | 4 ++--
> > > > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > > > >
> > > >
> > > > What problem are you seeing without this patch?
> > >
> > > I see the following warning: "Device 'serial@18,2' has no seq".
> > >
> > > In my case req_seq for the UART is -1 ("any"), while seq is 0.
> > > Why would we check for req_seq and not for seq in this function?
> > >
> > > > At present the ACPI device may not always be probed, and probing is
> > > > when the sequence number is currently set up.
> > >
> > > In my case the UART is already probed before the ACPI tables are
> > > generated.
> >
> > I would expect req_seq to be set to the UART number, i.e. the value of
> > the alias (uart0, uart1) that points to the node.
> >
> > I wonder why that doesn't work in your case?
>
> I did not have an alias for my serial. I have added one and now it works as
> expected.
>
> I misunderstood how that code is expected to work. Thanks for the explanation,
> now it makes sense.
>
> This also means that my patch is wrong and should be dropped.
> @Bin: please drop this patch.
Thanks. Have updated the status of this patch in patchwork.
>
> > Are you sure that all UARTs are probed before ACPI tables are created?
> > Normally U-Boot would only probe the one being used for the console.
> >
> > >
> > > >
> > > > I have been thinking about dropping req_seq and doing everything when
> > > > the device is bound, but haven't dug into it in detail yet.
>
Regards,
Bin
Re: [PATCH] acpi: device: Fix check for sequence number
Hi Simon,
-"Simon Glass" schrieb: -
> Betreff: Re: [PATCH] acpi: device: Fix check for sequence number
>
> Hi Wolfgang,
>
> On Thu, 13 Aug 2020 at 01:23, Wolfgang Wallner
> wrote:
> >
> > Hi Simon,
> >
> > -----"Simon Glass" schrieb: -
> > > Betreff: Re: [PATCH] acpi: device: Fix check for sequence number
> > >
> > > Hi Wolfgang,
> > >
> > > On Thu, 30 Jul 2020 at 06:47, Wolfgang Wallner
> > > wrote:
> > > >
> > > > Currently the function acpi_check_seq() checks whether dev->req_seq is
> > > > unequal to "-1", but it should actually check dev->seq. Change it to
> > > > check dev->seq.
> > > >
> > > > For req_seq the value "-1" would be a valid (meaning 'any'), while for
> > > > seq the value "-1" means 'none' and is not valid.
> > > >
> > > > Quoting the description of udevice in device.h:
> > > > * @req_seq: Requested sequence number for this device (-1 = any)
> > > > * @seq: Allocated sequence number for this device (-1 = none).
> > > > * This is set up when the device is probed and will be unique
> > > > * within the device's uclass.
> > > >
> > > > Signed-off-by: Wolfgang Wallner
> > > >
> > > > Fixes: commit fefac0b0643b ("dm: acpi: Enhance acpi_get_name()")
> > > >
> > > > ---
> > > >
> > > > lib/acpi/acpi_device.c | 4 ++--
> > > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > > >
> > >
> > > What problem are you seeing without this patch?
> >
> > I see the following warning: "Device 'serial@18,2' has no seq".
> >
> > In my case req_seq for the UART is -1 ("any"), while seq is 0.
> > Why would we check for req_seq and not for seq in this function?
> >
> > > At present the ACPI device may not always be probed, and probing is
> > > when the sequence number is currently set up.
> >
> > In my case the UART is already probed before the ACPI tables are generated.
>
> I would expect req_seq to be set to the UART number, i.e. the value of
> the alias (uart0, uart1) that points to the node.
>
> I wonder why that doesn't work in your case?
I did not have an alias for my serial. I have added one and now it works as
expected.
I misunderstood how that code is expected to work. Thanks for the explanation,
now it makes sense.
This also means that my patch is wrong and should be dropped.
@Bin: please drop this patch.
> Are you sure that all UARTs are probed before ACPI tables are created?
> Normally U-Boot would only probe the one being used for the console.
>
> >
> > >
> > > I have been thinking about dropping req_seq and doing everything when
> > > the device is bound, but haven't dug into it in detail yet.
regards, Wolfgang
Re: [PATCH] acpi: device: Fix check for sequence number
Hi Wolfgang,
On Thu, 13 Aug 2020 at 01:23, Wolfgang Wallner
wrote:
>
> Hi Simon,
>
> -"Simon Glass" schrieb: -----
> > Betreff: Re: [PATCH] acpi: device: Fix check for sequence number
> >
> > Hi Wolfgang,
> >
> > On Thu, 30 Jul 2020 at 06:47, Wolfgang Wallner
> > wrote:
> > >
> > > Currently the function acpi_check_seq() checks whether dev->req_seq is
> > > unequal to "-1", but it should actually check dev->seq. Change it to
> > > check dev->seq.
> > >
> > > For req_seq the value "-1" would be a valid (meaning 'any'), while for
> > > seq the value "-1" means 'none' and is not valid.
> > >
> > > Quoting the description of udevice in device.h:
> > > * @req_seq: Requested sequence number for this device (-1 = any)
> > > * @seq: Allocated sequence number for this device (-1 = none).
> > > * This is set up when the device is probed and will be unique
> > > * within the device's uclass.
> > >
> > > Signed-off-by: Wolfgang Wallner
> > >
> > > Fixes: commit fefac0b0643b ("dm: acpi: Enhance acpi_get_name()")
> > >
> > > ---
> > >
> > > lib/acpi/acpi_device.c | 4 ++--
> > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> >
> > What problem are you seeing without this patch?
>
> I see the following warning: "Device 'serial@18,2' has no seq".
>
> In my case req_seq for the UART is -1 ("any"), while seq is 0.
> Why would we check for req_seq and not for seq in this function?
>
> > At present the ACPI device may not always be probed, and probing is
> > when the sequence number is currently set up.
>
> In my case the UART is already probed before the ACPI tables are generated.
I would expect req_seq to be set to the UART number, i.e. the value of
the alias (uart0, uart1) that points to the node.
I wonder why that doesn't work in your case?
Are you sure that all UARTs are probed before ACPI tables are created?
Normally U-Boot would only probe the one being used for the console.
>
> >
> > I have been thinking about dropping req_seq and doing everything when
> > the device is bound, but haven't dug into it in detail yet.
>
> regards, Wolfgang
Regards,
Simon
Re: [PATCH] acpi: device: Fix check for sequence number
Hi Simon,
-"Simon Glass" schrieb: -
> Betreff: Re: [PATCH] acpi: device: Fix check for sequence number
>
> Hi Wolfgang,
>
> On Thu, 30 Jul 2020 at 06:47, Wolfgang Wallner
> wrote:
> >
> > Currently the function acpi_check_seq() checks whether dev->req_seq is
> > unequal to "-1", but it should actually check dev->seq. Change it to
> > check dev->seq.
> >
> > For req_seq the value "-1" would be a valid (meaning 'any'), while for
> > seq the value "-1" means 'none' and is not valid.
> >
> > Quoting the description of udevice in device.h:
> > * @req_seq: Requested sequence number for this device (-1 = any)
> > * @seq: Allocated sequence number for this device (-1 = none).
> > * This is set up when the device is probed and will be unique
> > * within the device's uclass.
> >
> > Signed-off-by: Wolfgang Wallner
> >
> > Fixes: commit fefac0b0643b ("dm: acpi: Enhance acpi_get_name()")
> >
> > ---
> >
> > lib/acpi/acpi_device.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
>
> What problem are you seeing without this patch?
I see the following warning: "Device 'serial@18,2' has no seq".
In my case req_seq for the UART is -1 ("any"), while seq is 0.
Why would we check for req_seq and not for seq in this function?
> At present the ACPI device may not always be probed, and probing is
> when the sequence number is currently set up.
In my case the UART is already probed before the ACPI tables are generated.
>
> I have been thinking about dropping req_seq and doing everything when
> the device is bound, but haven't dug into it in detail yet.
regards, Wolfgang
Re: [PATCH] acpi: device: Fix check for sequence number
Hi Wolfgang,
On Thu, 30 Jul 2020 at 06:47, Wolfgang Wallner
wrote:
>
> Currently the function acpi_check_seq() checks whether dev->req_seq is
> unequal to "-1", but it should actually check dev->seq. Change it to
> check dev->seq.
>
> For req_seq the value "-1" would be a valid (meaning 'any'), while for
> seq the value "-1" means 'none' and is not valid.
>
> Quoting the description of udevice in device.h:
> * @req_seq: Requested sequence number for this device (-1 = any)
> * @seq: Allocated sequence number for this device (-1 = none).
> * This is set up when the device is probed and will be unique
> * within the device's uclass.
>
> Signed-off-by: Wolfgang Wallner
>
> Fixes: commit fefac0b0643b ("dm: acpi: Enhance acpi_get_name()")
>
> ---
>
> lib/acpi/acpi_device.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
What problem are you seeing without this patch?
At present the ACPI device may not always be probed, and probing is
when the sequence number is currently set up.
I have been thinking about dropping req_seq and doing everything when
the device is bound, but haven't dug into it in detail yet.
Regards,
SImon
