Message ID | 20200426104713.216896-2-hdegoede@redhat.com (mailing list archive) |
---|---|
State | Changes Requested, archived |
Headers | show |
Series | ACPI / scan: Create platform device for CPLM3218 ACPI nodes | expand |
On Sun, Apr 26, 2020 at 1:47 PM Hans de Goede <hdegoede@redhat.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(); ... ? > > 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@redhat.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 >
Hi, On 4/26/20 7:59 PM, Andy Shevchenko wrote: > On Sun, Apr 26, 2020 at 1:47 PM Hans de Goede <hdegoede@redhat.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(). 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). Regards, Hans > >> >> 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@redhat.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 >> > >
On Mon, Apr 27, 2020 at 3:51 PM Hans de Goede <hdegoede@redhat.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@redhat.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? > 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@redhat.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 > >> > > > > >
Hi, On 4/27/20 3:18 PM, Andy Shevchenko wrote: > On Mon, Apr 27, 2020 at 3:51 PM Hans de Goede <hdegoede@redhat.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@redhat.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@redhat.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 >>>> >>> >>> >> > >
On Mon, Apr 27, 2020 at 6:06 PM Hans de Goede <hdegoede@redhat.com> wrote: > On 4/27/20 3:18 PM, Andy Shevchenko wrote: > > On Mon, Apr 27, 2020 at 3:51 PM Hans de Goede <hdegoede@redhat.com> wrote: > >> On 4/26/20 7:59 PM, Andy Shevchenko wrote: > >>> On Sun, Apr 26, 2020 at 1:47 PM Hans de Goede <hdegoede@redhat.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. I'm missing something here. Why we need to pass an fwnode in the first place? Seems you would like to access to methods from the driver. But if you simple enumerate the driver in ACPI multi-instantiate won't be needed. As far as I understand, the actual driver consumes *both* I²C resources. It's not a multi-instantiate in this case.
Hi, On 4/27/20 7:33 PM, Andy Shevchenko wrote: > On Mon, Apr 27, 2020 at 6:06 PM Hans de Goede <hdegoede@redhat.com> wrote: >> On 4/27/20 3:18 PM, Andy Shevchenko wrote: >>> On Mon, Apr 27, 2020 at 3:51 PM Hans de Goede <hdegoede@redhat.com> wrote: >>>> On 4/26/20 7:59 PM, Andy Shevchenko wrote: >>>>> On Sun, Apr 26, 2020 at 1:47 PM Hans de Goede <hdegoede@redhat.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. > > I'm missing something here. Why we need to pass an fwnode in the first place? > Seems you would like to access to methods from the driver. Right, the cm32181 code needs access to the CPM0 and CPM1 ACPI objects, which requires access to the fwnode. > But if you simple enumerate the driver in ACPI multi-instantiate won't > be needed. > > As far as I understand, the actual driver consumes *both* I²C > resources. It's not a multi-instantiate in this case. On systems where there are 2 resources, the driver only attaches to the second resouce. It does detect when it gets called for the first resource (it detects the ARA address) and then returns -ENODEV. Another approach might be for the driver to call i2c_acpi_new_device itself when it detects the ARA address, but that is quite ugly, then we get: -ACPI subsys instantiates i2c-client -cm32181_probe -cm32181_probe instantiates i2c-client for second resource -cm32181 probe (for second resource) -cm32181 probe returns 0 -cm32181 probe returns -ENODEV So the end result is the same (2 clients instantiated, one bound to the cm32181 driver). But the nested probe calls to me look quite ugly and since this solution actually still does multi-instantiation using i2c-multi-inst seems like the more clean solution to me. Note that we need to likely solve the fwnode passing problem sooner or later anyways. One of these days a driver for an i2c-client instantiated by the i2c-multi-inst code is going to need access to some methods or objects from the ACPI device. Since you do not like the PASS_FWNODE flag, one solution would be this change: diff --git a/drivers/i2c/i2c-core-base.c b/drivers/i2c/i2c-core-base.c index a66912782064..365864e8bfd5 100644 --- a/drivers/i2c/i2c-core-base.c +++ b/drivers/i2c/i2c-core-base.c @@ -341,12 +341,12 @@ static int i2c_device_probe(struct device *dev) if (irq == -EPROBE_DEFER) return irq; - if (irq < 0) - irq = 0; - client->irq = irq; } + if (client->irq < 0) + client->irq = 0; + /* * An I2C ID table is not mandatory, if and only if, a suitable OF * or ACPI ID table is supplied for the probing device. This allows us to set board_info.irq to -ENOENT in the i2c-multi-inst code, causing the core to skip trying to get the irq from the fwnode itself, while still making drivers see 0 as irq value (which they expect when there is no irq). With this change i2c-multi-inst can pass the fwnode unconditionally. Regards, Hans
Hi, On 4/27/20 7:55 PM, Hans de Goede wrote: > Hi, > > On 4/27/20 7:33 PM, Andy Shevchenko wrote: >> On Mon, Apr 27, 2020 at 6:06 PM Hans de Goede <hdegoede@redhat.com> wrote: >>> On 4/27/20 3:18 PM, Andy Shevchenko wrote: >>>> On Mon, Apr 27, 2020 at 3:51 PM Hans de Goede <hdegoede@redhat.com> wrote: >>>>> On 4/26/20 7:59 PM, Andy Shevchenko wrote: >>>>>> On Sun, Apr 26, 2020 at 1:47 PM Hans de Goede <hdegoede@redhat.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. >> >> I'm missing something here. Why we need to pass an fwnode in the first place? >> Seems you would like to access to methods from the driver. > > Right, the cm32181 code needs access to the CPM0 and CPM1 ACPI > objects, which requires access to the fwnode. > >> But if you simple enumerate the driver in ACPI multi-instantiate won't >> be needed. > >> As far as I understand, the actual driver consumes *both* I²C >> resources. It's not a multi-instantiate in this case. > > On systems where there are 2 resources, the driver only attaches > to the second resouce. It does detect when it gets called for > the first resource (it detects the ARA address) and then returns > -ENODEV. > > Another approach might be for the driver to call i2c_acpi_new_device > itself when it detects the ARA address, but that is quite ugly, then > we get: > > -ACPI subsys instantiates i2c-client > -cm32181_probe > -cm32181_probe instantiates i2c-client for second resource > -cm32181 probe (for second resource) > -cm32181 probe returns 0 > -cm32181 probe returns -ENODEV > > So the end result is the same (2 clients instantiated, one > bound to the cm32181 driver). But the nested probe calls to me > look quite ugly and since this solution actually still does > multi-instantiation using i2c-multi-inst seems like the more > clean solution to me. Ok so thinking more about this, you are right. The ACPI resources describe a single chip here, so this really should not use the i2c-multi-instantiate code. Self-nack for this series. I still think we may want to eventually pass the fwnode through to the clients instantiated from the i2c-multi-instantiate code, so my below proposal still stands: > Note that we need to likely solve the fwnode passing problem > sooner or later anyways. One of these days a driver for an > i2c-client instantiated by the i2c-multi-inst code is going > to need access to some methods or objects from the ACPI > device. > > Since you do not like the PASS_FWNODE flag, one solution > would be this change: > > diff --git a/drivers/i2c/i2c-core-base.c b/drivers/i2c/i2c-core-base.c > index a66912782064..365864e8bfd5 100644 > --- a/drivers/i2c/i2c-core-base.c > +++ b/drivers/i2c/i2c-core-base.c > @@ -341,12 +341,12 @@ static int i2c_device_probe(struct device *dev) > if (irq == -EPROBE_DEFER) > return irq; > > - if (irq < 0) > - irq = 0; > - > client->irq = irq; > } > > + if (client->irq < 0) > + client->irq = 0; > + > /* > * An I2C ID table is not mandatory, if and only if, a suitable OF > * or ACPI ID table is supplied for the probing device. > > > This allows us to set board_info.irq to -ENOENT in the i2c-multi-inst > code, causing the core to skip trying to get the irq from the fwnode > itself, while still making drivers see 0 as irq value (which they > expect when there is no irq). > > With this change i2c-multi-inst can pass the fwnode unconditionally. Regards, Hans
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);
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(). 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@redhat.com> --- drivers/platform/x86/i2c-multi-instantiate.c | 6 ++++++ 1 file changed, 6 insertions(+)