Date:   Mon, 27 Apr 2020 17:06:03 +0200
From:   Hans de Goede <hdegoede@...hat.com>
To:     Andy Shevchenko <andy.shevchenko@...il.com>,
        Wolfram Sang <wsa@...-dreams.de>
Cc:     "Rafael J . Wysocki" <rjw@...ysocki.net>,
        Len Brown <lenb@...nel.org>,
        Darren Hart <dvhart@...radead.org>,
        Andy Shevchenko <andy@...radead.org>,
        ACPI Devel Maling List <linux-acpi@...r.kernel.org>,
        Platform Driver <platform-driver-x86@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        linux-i2c@...r.kernel.org
Subject: Re: [PATCH 1/2] platform/x86: i2c-multi-instantiate: Add flag for
 passing fwnode

Hi,

On 4/27/20 3:18 PM, Andy Shevchenko wrote:
> On Mon, Apr 27, 2020 at 3:51 PM Hans de Goede <hdegoede@...hat.com> wrote:
>>
>> Hi,
>>
>> On 4/26/20 7:59 PM, Andy Shevchenko wrote:
>>> On Sun, Apr 26, 2020 at 1:47 PM Hans de Goede <hdegoede@...hat.com> wrote:
>>>>
>>>> In some cases the driver for the i2c_client-s which i2c-multi-instantiate
>>>> instantiates may need access some fields / methods from to the ACPI fwnode
>>>> for which i2c_clients are being instantiated.
>>>>
>>>> An example of this are CPLM3218 ACPI device-s. These contain CPM0 and
>>>> CPM1 packages with various information (e.g. register init values) which
>>>> the driver needs.
>>>>
>>>> Passing the fwnode through the i2c_board_info struct also gives the
>>>> i2c-core access to it, and if we do not pass an IRQ then the i2c-core
>>>> will use the fwnode to get an IRQ, see i2c_acpi_get_irq().
>>>
>>> I'm wondering, can we rather do it in the same way like we do for
>>> GPIO/APIC case here.
>>> Introduce IRQ_RESOURCE_SHARED (or so) and
>>>
>>> case _SHARED:
>>>    irq = i2c_acpi_get_irq();
>>> ...
>>>
>>> ?
>>
>> I think you are miss-understanding the problem. The problem is not that
>> we want to share the IRQ, the problem is that we want to pass the single
>> IRQ in the resources to only 1 of the instantiated I2C-clients. But if we
>> do not pass an IRQ (we leave it at 0) and we do pass the fwnode then
>> i2c-core-base.c will see that there is an ACPI-node attached to the
>> device and will call i2c_acpi_get_irq().
> 
> Do we know ahead which device should take IRQ resource and which should not?
> Can we use current _NONE flag for them?

The problem is not internal to i2c-multi-instantiate.c, the problem
(once we pass a fwnode) is the API between i2c-multi-instantiate.c and
the i2c-core. For the IRQ_RESOURCE_NONE case i2c-multi-instantiate.c
sets board_info.irq to 0, which is the correct way to specify that
we do not have an IRQ, but if don't pass an IRQ then the i2c-core
will try to find one itself.  And once we pass the fwnode, then
the "try to find one itself" code will call i2c_acpi_get_irq()
and find the same IRQ for clients we instantiate, leading to
the earlier mentioned IRQ conflict.

<adding Wolfram + i2c lists to the Cc>

We could set board_info.irq to -ENOENT to indicate that there should
not be an irq. But that will get passed to various i2c-drivers, many of
which check for an irq like this:

	if (client->irq) {
		...
	}

We can avoid this, without needing to change all the drivers
by making the i2c-core check for board_info.irq < 0 to skip its
own "try to find IRQ" code and then set client->irq to 0 after
that check, rather then setting it to board_info.irq = -ENOENT.

If we do that then we can unconditionally pass the fwnode in
the i2c-multi-instantiate code.

Regards,

Hans





>> So the solution is definitely not calling i2c_acpi_get_irq() inside
>> i2c-multi-instantiate.c we want to avoid the i2c_acpi_get_irq(),
>> leaving the other 2 clients for the BSG1160 device without an IRQ
>> and thus avoiding the IRQ mismatch (it is a mismatch because the
>> drivers do not set the shared flag; and that is ok, we do not want
>> to share the IRQ, it is just for the accelerometer AFAIK).
> 
>>>> This is a problem when there is only an IRQ for 1 of the clients described
>>>> in the ACPI device we are instantiating clients for. If we unconditionally
>>>> pass the fwnode, then i2c_acpi_get_irq() will assign the same IRQ to all
>>>> clients instantiated, leading to kernel-oopses like this (BSG1160 device):
>>>>
>>>> [   27.340557] genirq: Flags mismatch irq 76. 00002001 (bmc150_magn_event) vs. 00000001 (bmc150_accel_event)
>>>> [   27.340567] Call Trace:
>>>> ...
>>>>
>>>> So we cannot simply always pass the fwnode. This commit adds a PASS_FWNODE
>>>> flag, which can be used to pass the fwnode in cases where we do not have
>>>> the IRQ problem and the driver for the instantiated client(s) needs access
>>>> to the fwnode.
>>>>
>>>> Signed-off-by: Hans de Goede <hdegoede@...hat.com>
>>>> ---
>>>>    drivers/platform/x86/i2c-multi-instantiate.c | 6 ++++++
>>>>    1 file changed, 6 insertions(+)
>>>>
>>>> diff --git a/drivers/platform/x86/i2c-multi-instantiate.c b/drivers/platform/x86/i2c-multi-instantiate.c
>>>> index 6acc8457866e..dcafb1a29d17 100644
>>>> --- a/drivers/platform/x86/i2c-multi-instantiate.c
>>>> +++ b/drivers/platform/x86/i2c-multi-instantiate.c
>>>> @@ -20,6 +20,8 @@
>>>>    #define IRQ_RESOURCE_GPIO      1
>>>>    #define IRQ_RESOURCE_APIC      2
>>>>
>>>> +#define PASS_FWNODE            BIT(2)
>>>> +
>>>>    struct i2c_inst_data {
>>>>           const char *type;
>>>>           unsigned int flags;
>>>> @@ -93,6 +95,10 @@ static int i2c_multi_inst_probe(struct platform_device *pdev)
>>>>                   snprintf(name, sizeof(name), "%s-%s.%d", dev_name(dev),
>>>>                            inst_data[i].type, i);
>>>>                   board_info.dev_name = name;
>>>> +
>>>> +               if (inst_data[i].flags & PASS_FWNODE)
>>>> +                       board_info.fwnode = dev->fwnode;
>>>> +
>>>>                   switch (inst_data[i].flags & IRQ_RESOURCE_TYPE) {
>>>>                   case IRQ_RESOURCE_GPIO:
>>>>                           ret = acpi_dev_gpio_irq_get(adev, inst_data[i].irq_idx);
>>>> --
>>>> 2.26.0
>>>>
>>>
>>>
>>
> 
> 

Powered by blists - more mailing lists