[Intel-gfx] [PATCH v3 3/4] PCI/VGA: only deal with VGA class devices

[Sui Jingfeng]

Hi,

On 2023/6/9 03:12, Bjorn Helgaas wrote:
> Start with verb and capitalize to match ("Deal only with ...")
>
> On Thu, Jun 08, 2023 at 07:43:21PM +0800, Sui Jingfeng wrote:
>> From: Sui Jingfeng <suijingfeng at loongson.cn>
>>
>> vgaarb only deal with the VGA devcie(pdev->class == 0x0300), so replace the
>> pci_get_subsys() function with pci_get_class(). Filter the non pci display
>> device out. There no need to process the non display PCI device.
> s/non pci/non-PCI/
> s/non display/non-display/
>
> This is fine, and I'll merge this, but someday I would like to get rid
> of the bus_register_notifier() and call vga_arbiter_add_pci_device()
> and vga_arbiter_del_pci_device() directly from the PCI core.

Nice idea!


But I'm wondering there are traps in this.

The pci_notifier in vgaarb.c is still need on Mips platform.

Because of loading order problems.

On MIPS system, PCI devices are enumerated by pcibios_init(),

which runs after vga_arb_device_init(). This is true until now.

When vga_arb_device_init() function get called, it will capture nothing.

On that time, the system have no PCI device enumerated.

Because of this, there are various problems in the past.

This is the reason we still need the notifier,

we need a way to capture the PCI display device after vgaarb already 
loaded(finished).


On complex case, there are ASpeed BMC, loongson integrated display 
controller,

and radeon discrete video card co-exist on Loongson 3B4000 server.

I have fixed various bug at our downstream(linux-4.19) environment.


I have fixed a bug on downstream involved with aspeed bmc, by workaround[1].

Chen Huacai grabbing my patch[1] and rewrite it, submit together with 
him-self's.

Its fine,  but those patch still look paleness in the front of Loongson 
integrated display controller.

Loongson integrated display controller don't has a dedicated VRAM bar.

vga_is_firmware_default() will lose its effectiveness then.

This is the reason we reveal our patch(0004 in this series) to face 
upstream.

Its not only for loongson integrated display controller through.

Its not uncommon that ARM servers have A aspeed BMC and discrete amdgpu 
or old radeon card.

Let the device drivers gave us a hint probably can help to resolve 
multi-video card co-exist

problem.

Consider that sometime user want to use integrate gpu, sometime user 
want to use discrete gpu.

Also, during driver developing(or debugging),

driver writer may want override the default boot device.


vgaarb probable shouldn't make the decision

without giving the device driver a chance to override.


Beside,  vga_is_firmware_default() only apply for PCI display device.

On ARM64 system, there are a lot of platform device.

If we move this function back to the device driver,

it probably applicable for a platform display controller + one/two PCI 
display devices case.


We find a method at downstream during we get efifb works LoongArch platform.

We can utilize the screen_info, screen_info say where's the firmware 
framebuffer is located at.

Drivers for specific hardware platform perhaps know more clearly than 
vgaarb.

if it is the default boot device.


[1] 
https://lore.kernel.org/all/20210514080025.1828197-6-chenhuacai@loongson.cn/

> Or if you wanted to do that now, that would be even better :)
>
> Bjorn
>
>> Signed-off-by: Sui Jingfeng <suijingfeng at loongson.cn>
>> ---
>>   drivers/pci/vgaarb.c | 22 ++++++++++++----------
>>   1 file changed, 12 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/pci/vgaarb.c b/drivers/pci/vgaarb.c
>> index 7f0347f4f6fd..b0bf4952a95d 100644
>> --- a/drivers/pci/vgaarb.c
>> +++ b/drivers/pci/vgaarb.c
>> @@ -756,10 +756,6 @@ static bool vga_arbiter_add_pci_device(struct pci_dev *pdev)
>>   	struct pci_dev *bridge;
>>   	u16 cmd;
>>   
>> -	/* Only deal with VGA class devices */
>> -	if ((pdev->class >> 8) != PCI_CLASS_DISPLAY_VGA)
>> -		return false;
>> -
>>   	/* Allocate structure */
>>   	vgadev = kzalloc(sizeof(struct vga_device), GFP_KERNEL);
>>   	if (vgadev == NULL) {
>> @@ -1499,7 +1495,9 @@ static int pci_notify(struct notifier_block *nb, unsigned long action,
>>   	struct pci_dev *pdev = to_pci_dev(dev);
>>   	bool notify = false;
>>   
>> -	vgaarb_dbg(dev, "%s\n", __func__);
>> +	/* Only deal with VGA class devices */
>> +	if ((pdev->class >> 8) != PCI_CLASS_DISPLAY_VGA)
>> +		return 0;
>>   
>>   	/* For now we're only intereted in devices added and removed. I didn't
>>   	 * test this thing here, so someone needs to double check for the
>> @@ -1509,6 +1507,8 @@ static int pci_notify(struct notifier_block *nb, unsigned long action,
>>   	else if (action == BUS_NOTIFY_DEL_DEVICE)
>>   		notify = vga_arbiter_del_pci_device(pdev);
>>   
>> +	vgaarb_dbg(dev, "%s: action = %lu\n", __func__, action);
>> +
>>   	if (notify)
>>   		vga_arbiter_notify_clients();
>>   	return 0;
>> @@ -1533,8 +1533,8 @@ static struct miscdevice vga_arb_device = {
>>   
>>   static int __init vga_arb_device_init(void)
>>   {
>> +	struct pci_dev *pdev = NULL;
>>   	int rc;
>> -	struct pci_dev *pdev;
>>   
>>   	rc = misc_register(&vga_arb_device);
>>   	if (rc < 0)
>> @@ -1545,11 +1545,13 @@ static int __init vga_arb_device_init(void)
>>   	/* We add all PCI devices satisfying VGA class in the arbiter by
>>   	 * default
>>   	 */
>> -	pdev = NULL;
>> -	while ((pdev =
>> -		pci_get_subsys(PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID,
>> -			       PCI_ANY_ID, pdev)) != NULL)
>> +	while (1) {
>> +		pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev);
>> +		if (!pdev)
>> +			break;
>> +
>>   		vga_arbiter_add_pci_device(pdev);
>> +	};
>>   
>>   	pr_info("loaded\n");
>>   	return rc;
>> -- 
>> 2.25.1
>>
-- 
Jingfeng