diff mbox series

[V4] PCI: loongson: Skip scanning unavailable child devices

Message ID	20221108064240.8030-1-liupeibao@loongson.cn
State	New
Headers	Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20; From: Liu Peibao <liupeibao@loongson.cn> To: Bjorn Helgaas <bhelgaas@google.com>, Rob Herring <robh+dt@kernel.org>, Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>, Lorenzo Pieralisi <lpieralisi@kernel.org>, =?utf-8?q?Krzysztof_Wilczy=C5=84?= =?utf-8?q?ski?= <kw@linux.com>, Jiaxun Yang <jiaxun.yang@flygoat.com>, Christophe JAILLET <christophe.jaillet@wanadoo.fr> Cc: Huacai Chen <chenhuacai@loongson.cn>, Jianmin Lv <lvjianmin@loongson.cn>, Yinbo Zhu <zhuyinbo@loongson.cn>, Liu Peibao <liupeibao@loongson.cn>, linux-pci@vger.kernel.org, linux-kernel@vger.kernel.org Subject: [PATCH V4] PCI: loongson: Skip scanning unavailable child devices Date: Tue, 8 Nov 2022 14:42:40 +0800 Message-Id: <20221108064240.8030-1-liupeibao@loongson.cn> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Precedence: bulk
Series	[V4] PCI: loongson: Skip scanning unavailable child devices \| [V4] PCI: loongson: Skip scanning unavailable child devices

Commit Message

Liu Peibao Nov. 8, 2022, 6:42 a.m. UTC

  The PCI Controller of 2k1000 could not mask devices by setting vender ID or
device ID in configuration space header as invalid values. When there are
pins shareable between the platform device and PCI device, if the platform
device is preferred, we should not scan this PCI device. In the above
scene, add `status = "disabled"` property in DT node of this PCI device.

Signed-off-by: Liu Peibao <liupeibao@loongson.cn>
---
V3 -> V4: 1. get rid of the masklist and search the status property
	  directly.
          2. check the status property only when accessing the vendor ID.
V2 -> V3: 1. use list_for_each_entry() for more clearly.
          2. fix wrong use of sizeof().
V1 -> V2: use existing property "status" instead of adding new property.

 drivers/pci/controller/pci-loongson.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

Comments

Bjorn Helgaas Nov. 10, 2022, 9:07 p.m. UTC | #1

On Tue, Nov 08, 2022 at 02:42:40PM +0800, Liu Peibao wrote:
> The PCI Controller of 2k1000 could not mask devices by setting vender ID or
> device ID in configuration space header as invalid values. When there are
> pins shareable between the platform device and PCI device, if the platform
> device is preferred, we should not scan this PCI device. In the above
> scene, add `status = "disabled"` property in DT node of this PCI device.
> 
> Signed-off-by: Liu Peibao <liupeibao@loongson.cn>
> ---
> V3 -> V4: 1. get rid of the masklist and search the status property
> 	  directly.
>           2. check the status property only when accessing the vendor ID.
> V2 -> V3: 1. use list_for_each_entry() for more clearly.
>           2. fix wrong use of sizeof().
> V1 -> V2: use existing property "status" instead of adding new property.
> 
>  drivers/pci/controller/pci-loongson.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/drivers/pci/controller/pci-loongson.c b/drivers/pci/controller/pci-loongson.c
> index 05c50408f13b..efca0b3b5a29 100644
> --- a/drivers/pci/controller/pci-loongson.c
> +++ b/drivers/pci/controller/pci-loongson.c
> @@ -194,6 +194,17 @@ static void __iomem *pci_loongson_map_bus(struct pci_bus *bus,
>  			return NULL;
>  	}
>  
> +#ifdef CONFIG_OF
> +	/* Don't access disabled devices. */
> +	if (pci_is_root_bus(bus) && where == PCI_VENDOR_ID) {
> +		struct device_node *dn;
> +
> +		dn = of_pci_find_child_device(bus->dev.of_node, devfn);
> +		if (dn && !of_device_is_available(dn))
> +			return NULL;
> +	}
> +#endif

Looks nice and simple, thanks for trying this out.

>  	/* CFG0 can only access standard space */
>  	if (where < PCI_CFG_SPACE_SIZE && priv->cfg0_base)
>  		return cfg0_map(priv, bus, devfn, where);
> -- 
> 2.20.1
>

Jiaxun Yang Nov. 10, 2022, 11 p.m. UTC | #2

在2022年11月10日十一月 下午9:07，Bjorn Helgaas写道：
> On Tue, Nov 08, 2022 at 02:42:40PM +0800, Liu Peibao wrote:
>> The PCI Controller of 2k1000 could not mask devices by setting vender ID or
>> device ID in configuration space header as invalid values. When there are
>> pins shareable between the platform device and PCI device, if the platform
>> device is preferred, we should not scan this PCI device. In the above
>> scene, add `status = "disabled"` property in DT node of this PCI device.
>> 
>> Signed-off-by: Liu Peibao <liupeibao@loongson.cn>
>> ---
>> V3 -> V4: 1. get rid of the masklist and search the status property
>> 	  directly.
>>           2. check the status property only when accessing the vendor ID.
>> V2 -> V3: 1. use list_for_each_entry() for more clearly.
>>           2. fix wrong use of sizeof().
>> V1 -> V2: use existing property "status" instead of adding new property.
>> 
>>  drivers/pci/controller/pci-loongson.c | 11 +++++++++++
>>  1 file changed, 11 insertions(+)
>> 
>> diff --git a/drivers/pci/controller/pci-loongson.c b/drivers/pci/controller/pci-loongson.c
>> index 05c50408f13b..efca0b3b5a29 100644
>> --- a/drivers/pci/controller/pci-loongson.c
>> +++ b/drivers/pci/controller/pci-loongson.c
>> @@ -194,6 +194,17 @@ static void __iomem *pci_loongson_map_bus(struct pci_bus *bus,
>>  			return NULL;
>>  	}
>>  
>> +#ifdef CONFIG_OF
>> +	/* Don't access disabled devices. */
>> +	if (pci_is_root_bus(bus) && where == PCI_VENDOR_ID) {
>> +		struct device_node *dn;
>> +
>> +		dn = of_pci_find_child_device(bus->dev.of_node, devfn);
>> +		if (dn && !of_device_is_available(dn))
>> +			return NULL;
>> +	}
>> +#endif
>
> Looks nice and simple, thanks for trying this out.

Should we make this into common PCI code?
I guess Loongson won’t be the last platform having such problem.


>
>>  	/* CFG0 can only access standard space */
>>  	if (where < PCI_CFG_SPACE_SIZE && priv->cfg0_base)
>>  		return cfg0_map(priv, bus, devfn, where);
>> -- 
>> 2.20.1
>>

Bjorn Helgaas Nov. 10, 2022, 11:13 p.m. UTC | #3

On Thu, Nov 10, 2022 at 11:00:45PM +0000, Jiaxun Yang wrote:
> 在2022年11月10日十一月 下午9:07，Bjorn Helgaas写道：
> > On Tue, Nov 08, 2022 at 02:42:40PM +0800, Liu Peibao wrote:
> >> The PCI Controller of 2k1000 could not mask devices by setting vender ID or
> >> device ID in configuration space header as invalid values. When there are
> >> pins shareable between the platform device and PCI device, if the platform
> >> device is preferred, we should not scan this PCI device. In the above
> >> scene, add `status = "disabled"` property in DT node of this PCI device.
> >> 
> >> Signed-off-by: Liu Peibao <liupeibao@loongson.cn>
> >> ---
> >> V3 -> V4: 1. get rid of the masklist and search the status property
> >> 	  directly.
> >>           2. check the status property only when accessing the vendor ID.
> >> V2 -> V3: 1. use list_for_each_entry() for more clearly.
> >>           2. fix wrong use of sizeof().
> >> V1 -> V2: use existing property "status" instead of adding new property.
> >> 
> >>  drivers/pci/controller/pci-loongson.c | 11 +++++++++++
> >>  1 file changed, 11 insertions(+)
> >> 
> >> diff --git a/drivers/pci/controller/pci-loongson.c b/drivers/pci/controller/pci-loongson.c
> >> index 05c50408f13b..efca0b3b5a29 100644
> >> --- a/drivers/pci/controller/pci-loongson.c
> >> +++ b/drivers/pci/controller/pci-loongson.c
> >> @@ -194,6 +194,17 @@ static void __iomem *pci_loongson_map_bus(struct pci_bus *bus,
> >>  			return NULL;
> >>  	}
> >>  
> >> +#ifdef CONFIG_OF
> >> +	/* Don't access disabled devices. */
> >> +	if (pci_is_root_bus(bus) && where == PCI_VENDOR_ID) {
> >> +		struct device_node *dn;
> >> +
> >> +		dn = of_pci_find_child_device(bus->dev.of_node, devfn);
> >> +		if (dn && !of_device_is_available(dn))
> >> +			return NULL;
> >> +	}
> >> +#endif
> >
> > Looks nice and simple, thanks for trying this out.
> 
> Should we make this into common PCI code?
> I guess Loongson won’t be the last platform having such problem.

I think we should wait until somebody else has this problem.

It's not a completely trivial situation because if the device uses PCI
memory or I/O space, we have to worry about how that space is handled.
Does the BIOS assign that space?  Is it included in the host bridge
_CRS or "ranges" properties?  If the device is below any PCI bridges
(I don't think that's the case in your situation), how does the space
it requires get routed through the bridges?

It would be nice to say something in this commit log about whether
these are issues on your platform.

> >>  	/* CFG0 can only access standard space */
> >>  	if (where < PCI_CFG_SPACE_SIZE && priv->cfg0_base)
> >>  		return cfg0_map(priv, bus, devfn, where);
> >> -- 
> >> 2.20.1
> >>
> 
> -- 
> - Jiaxun

Jiaxun Yang Nov. 11, 2022, 4:06 p.m. UTC | #4

在 2022/11/10 23:13, Bjorn Helgaas 写道:
> On Thu, Nov 10, 2022 at 11:00:45PM +0000, Jiaxun Yang wrote:
>> 在2022年11月10日十一月 下午9:07，Bjorn Helgaas写道：
>>> On Tue, Nov 08, 2022 at 02:42:40PM +0800, Liu Peibao wrote:
>>>> The PCI Controller of 2k1000 could not mask devices by setting vender ID or
>>>> device ID in configuration space header as invalid values. When there are
>>>> pins shareable between the platform device and PCI device, if the platform
>>>> device is preferred, we should not scan this PCI device. In the above
>>>> scene, add `status = "disabled"` property in DT node of this PCI device.
>>>>
>>>> Signed-off-by: Liu Peibao <liupeibao@loongson.cn>
>>>> ---
>>>> V3 -> V4: 1. get rid of the masklist and search the status property
>>>> 	  directly.
>>>>            2. check the status property only when accessing the vendor ID.
>>>> V2 -> V3: 1. use list_for_each_entry() for more clearly.
>>>>            2. fix wrong use of sizeof().
>>>> V1 -> V2: use existing property "status" instead of adding new property.
>>>>
>>>>   drivers/pci/controller/pci-loongson.c | 11 +++++++++++
>>>>   1 file changed, 11 insertions(+)
>>>>
>>>> diff --git a/drivers/pci/controller/pci-loongson.c b/drivers/pci/controller/pci-loongson.c
>>>> index 05c50408f13b..efca0b3b5a29 100644
>>>> --- a/drivers/pci/controller/pci-loongson.c
>>>> +++ b/drivers/pci/controller/pci-loongson.c
>>>> @@ -194,6 +194,17 @@ static void __iomem *pci_loongson_map_bus(struct pci_bus *bus,
>>>>   			return NULL;
>>>>   	}
>>>>   
>>>> +#ifdef CONFIG_OF
>>>> +	/* Don't access disabled devices. */
>>>> +	if (pci_is_root_bus(bus) && where == PCI_VENDOR_ID) {
>>>> +		struct device_node *dn;
>>>> +
>>>> +		dn = of_pci_find_child_device(bus->dev.of_node, devfn);
>>>> +		if (dn && !of_device_is_available(dn))
>>>> +			return NULL;
>>>> +	}
>>>> +#endif
>>> Looks nice and simple, thanks for trying this out.
>> Should we make this into common PCI code?
>> I guess Loongson won’t be the last platform having such problem.
> I think we should wait until somebody else has this problem.
>
> It's not a completely trivial situation because if the device uses PCI
> memory or I/O space, we have to worry about how that space is handled.
> Does the BIOS assign that space?  Is it included in the host bridge
> _CRS or "ranges" properties?  If the device is below any PCI bridges
> (I don't think that's the case in your situation), how does the space
> it requires get routed through the bridges?

I believe in this case the address is assigned by BIOS and they are out 
of ranges
properties of host bridge. Those are all on chip devices so there won't 
be any
bridges.

@Peibao, can you please confirm this? I was never able to boot mainline 
kernel
on my LS2K board.

Thanks.
- Jiaxun


>
> It would be nice to say something in this commit log about whether
> these are issues on your platform.
>
>>>>   	/* CFG0 can only access standard space */
>>>>   	if (where < PCI_CFG_SPACE_SIZE && priv->cfg0_base)
>>>>   		return cfg0_map(priv, bus, devfn, where);
>>>> -- 
>>>> 2.20.1
>>>>
>> -- 
>> - Jiaxun

Liu Peibao Nov. 14, 2022, 4:03 a.m. UTC | #5

On 11/12/22 12:06 AM, Jiaxun Yang wrote:
> 
> 
> 在 2022/11/10 23:13, Bjorn Helgaas 写道:
>> On Thu, Nov 10, 2022 at 11:00:45PM +0000, Jiaxun Yang wrote:
>>> 在2022年11月10日十一月 下午9:07，Bjorn Helgaas写道：
>>>> On Tue, Nov 08, 2022 at 02:42:40PM +0800, Liu Peibao wrote:
>>>>> The PCI Controller of 2k1000 could not mask devices by setting vender ID or
>>>>> device ID in configuration space header as invalid values. When there are
>>>>> pins shareable between the platform device and PCI device, if the platform
>>>>> device is preferred, we should not scan this PCI device. In the above
>>>>> scene, add `status = "disabled"` property in DT node of this PCI device.
>>>>>
>>>>> Signed-off-by: Liu Peibao <liupeibao@loongson.cn>
>>>>> ---
>>>>> V3 -> V4: 1. get rid of the masklist and search the status property
>>>>>       directly.
>>>>>            2. check the status property only when accessing the vendor ID.
>>>>> V2 -> V3: 1. use list_for_each_entry() for more clearly.
>>>>>            2. fix wrong use of sizeof().
>>>>> V1 -> V2: use existing property "status" instead of adding new property.
>>>>>
>>>>>   drivers/pci/controller/pci-loongson.c | 11 +++++++++++
>>>>>   1 file changed, 11 insertions(+)
>>>>>
>>>>> diff --git a/drivers/pci/controller/pci-loongson.c b/drivers/pci/controller/pci-loongson.c
>>>>> index 05c50408f13b..efca0b3b5a29 100644
>>>>> --- a/drivers/pci/controller/pci-loongson.c
>>>>> +++ b/drivers/pci/controller/pci-loongson.c
>>>>> @@ -194,6 +194,17 @@ static void __iomem *pci_loongson_map_bus(struct pci_bus *bus,
>>>>>               return NULL;
>>>>>       }
>>>>>   +#ifdef CONFIG_OF
>>>>> +    /* Don't access disabled devices. */
>>>>> +    if (pci_is_root_bus(bus) && where == PCI_VENDOR_ID) {
>>>>> +        struct device_node *dn;
>>>>> +
>>>>> +        dn = of_pci_find_child_device(bus->dev.of_node, devfn);
>>>>> +        if (dn && !of_device_is_available(dn))
>>>>> +            return NULL;
>>>>> +    }
>>>>> +#endif
>>>> Looks nice and simple, thanks for trying this out.
>>> Should we make this into common PCI code?
>>> I guess Loongson won’t be the last platform having such problem.
>> I think we should wait until somebody else has this problem.
>>
>> It's not a completely trivial situation because if the device uses PCI
>> memory or I/O space, we have to worry about how that space is handled.
>> Does the BIOS assign that space?  Is it included in the host bridge
>> _CRS or "ranges" properties?  If the device is below any PCI bridges
>> (I don't think that's the case in your situation), how does the space
>> it requires get routed through the bridges?
> 
> I believe in this case the address is assigned by BIOS and they are out of ranges
> properties of host bridge. Those are all on chip devices so there won't be any
> bridges.
> 
> @Peibao, can you please confirm this? I was never able to boot mainline kernel
> on my LS2K board.
> 
> Thanks.
> - Jiaxun
> 

@Jiaxun,

Yes, the same as you said totally.

Did you notice the following patch? The liointc of LS2K can't work after
commit b2c4c3969fd7 ("irqchip/loongson-liointc: irqchip add 2.0 version"),
which may cause the booting failure. 
https://lore.kernel.org/all/20221104110712.23300-1-liupeibao@loongson.cn/

In fact, I am developing this on the LoongArch compatible board LS2K1000LA.
I boot the mainline kernel basing my FDT supporting patch set for LoongArch
and the BIOS following current LoongArch booting protocol :).

BR,
Peibao

> 
>>
>> It would be nice to say something in this commit log about whether
>> these are issues on your platform.
>>

@Bjorn,

Thanks!

I will update commit log in the next patch to clear the issue on our platform,
which is absolutely what Jiaxun has described above.

BR,
Peibao

diff mbox series

Patch

diff --git a/drivers/pci/controller/pci-loongson.c b/drivers/pci/controller/pci-loongson.c
index 05c50408f13b..efca0b3b5a29 100644
--- a/drivers/pci/controller/pci-loongson.c
+++ b/drivers/pci/controller/pci-loongson.c
@@ -194,6 +194,17 @@  static void __iomem *pci_loongson_map_bus(struct pci_bus *bus,
 			return NULL;
 	}
 
+#ifdef CONFIG_OF
+	/* Don't access disabled devices. */
+	if (pci_is_root_bus(bus) && where == PCI_VENDOR_ID) {
+		struct device_node *dn;
+
+		dn = of_pci_find_child_device(bus->dev.of_node, devfn);
+		if (dn && !of_device_is_available(dn))
+			return NULL;
+	}
+#endif
+
 	/* CFG0 can only access standard space */
 	if (where < PCI_CFG_SPACE_SIZE && priv->cfg0_base)
 		return cfg0_map(priv, bus, devfn, where);