Re: [PATCH 2/6] PCI/VGA: Deal with PCI VGA compatible devices only

[Sui Jingfeng <sui.jingfeng@linux.dev>]

[-- Attachment #1: Type: text/plain, Size: 5568 bytes --]

Hi,

On 2023/7/17 17:51, suijingfeng wrote:
>
> Hi,
>
>
> On 2023/7/11 21:43, Sui Jingfeng wrote:
>> From: Sui Jingfeng<suijingfeng@loongson.cn>
>>
>> Currently, vgaarb only cares about PCI VGA-compatible class devices.
>>
>> While vga_arbiter_del_pci_device() gets called unbalanced when some PCI
>> device is about to be removed. This happens even during the boot process.
>>
>> Another reason is that the vga_arb_device_init() function is not efficient.
>> Since we only care about VGA-compatible devices (pdev->class == 0x030000),
>> We could filter the unqualified devices out in the vga_arb_device_init()
>> function. While the current implementation is to search all PCI devices
>> in a system, this is not necessary.
>>
>> Reviewed-by: Mario Limonciello<mario.limonciello@amd.com>
>> Signed-off-by: Sui Jingfeng<suijingfeng@loongson.cn>
>> ---
>>   drivers/pci/vgaarb.c | 25 +++++++++++++------------
>>   1 file changed, 13 insertions(+), 12 deletions(-)
>>
>> diff --git a/drivers/pci/vgaarb.c b/drivers/pci/vgaarb.c
>> index c1bc6c983932..021116ed61cb 100644
>> --- a/drivers/pci/vgaarb.c
>> +++ b/drivers/pci/vgaarb.c
>> @@ -754,10 +754,6 @@ 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) {
>> @@ -1502,6 +1498,10 @@ static int pci_notify(struct notifier_block *nb, unsigned long action,
>>   
>>   	vgaarb_dbg(dev, "%s\n", __func__);
>>   
>> +	/* Deal with VGA compatible devices only */
>> +	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
>>   	 * cases of hotplugable vga cards. */
>> @@ -1534,8 +1534,8 @@ 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)
>> @@ -1543,13 +1543,14 @@ static int __init vga_arb_device_init(void)
>>   
>>   	bus_register_notifier(&pci_bus_type, &pci_notifier);
>>   
>> -	/* 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)
>> -		vga_arbiter_add_pci_device(pdev);
>> +	/*
>> +	 * We add all PCI VGA compatible devices in the arbiter by default
>> +	 */
>> +	do {
>> +		pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev);
>> +		if (pdev)
>> +			vga_arbiter_add_pci_device(pdev);
>> +	} while (pdev);
> I suddenly remember one more thing that I forget to explain.
>
> In a previous thread[1] of previous version of this series,
>
> Maciej seems argue that PCI_CLASS_NOT_DEFINED_VGA should be handled also.
>
> But, even I try to handled PCI_CLASS_NOT_DEFINED_VGA at here,
>
> this card still will not be annotate as boot_vga,
>
> because the pci_dev_attrs_are_visible() function only consider VGA compatible devices
>
> which (pdev->class >> 8 == PCI_CLASS_DISPLAY_VGA) is true. See [2].
>
>
> Intel integrated GPU is more intelligent,
>
> which could change their class of the PCI(e) device to 0x038000 when
>
> multiple GPU co-exist. Even though the GPU can display. This is configurable by
>
> the firmware, but this is trying to escape the arbitration by changing is PCI id.

"by changing is PCI id" -> "by changing its PCI(e) class number".

For example, the intel GPU will change their PCI class number from 0x030000 to 0x038000,

if a user choose the dGPU as primary by setting their UEFI firmware from the BIOS.


But other GPUs may not support to change their PCI class number.

>
> So, PCI devices belong to the PCI_CLASS_DISPLAY_OTHER, PCI_CLASS_DISPLAY_3D and PCI_CLASS_DISPLAY_XGA
>
> can not participate in the arbitration. They are all will be get filtered.

I means that PCI devices with their PCI class number equal  to

PCI_CLASS_DISPLAY_OTHER, PCI_CLASS_DISPLAY_3D and PCI_CLASS_DISPLAY_XGA

will get ignored by vgaarb currently.

Even we throw other devices(DISPLAY_OTHER, DISPLAY_3D, DISPLAY_XGA) into 
the arbitrator,

We still not make a meaningful progress, I need also to change the code 
in link [2]

to make the boot_vga visible to userspace. Because X server is the end 
consumer.

This already form an ABI.  So change the code here alone is not enough 
to expand vgaarb.

>
> So, this is a limitation of the vgaarb itself.
>
> While I could help to improve it in the future, what do you think?
> Is my express clear?
Am I express my thoughts clearly ?
>
> [1] 
> https://lkml.kernel.org/nouveau/alpine.DEB.2.21.2306190339590.14084@angie.orcam.me.uk/#t
>
> [2] 
> https://elixir.bootlin.com/linux/latest/source/drivers/pci/pci-sysfs.c#L1544
>
Alex acclaims that "we won't need vgaarb if the platform has no VGA devices", see [3].

while this is not 100% correct, because X server is the primary consumer of the boot_vga flag,

the convention usage of the userspace and the kernel space is already established.

So without we can craft something new easily without significant change.



[3]https://lkml.kernel.org/nouveau/CADnq5_PFoM2O8mCd6+VFfu9Nc-Hg_HTnwEMxrq0FGRpva1kKiA@mail.gmail.com/


>>   
>>   	pr_info("loaded\n");
>>   	return rc;

[-- Attachment #2: Type: text/html, Size: 7870 bytes --]