Re: [PATCH] PCI: optimize proc sequential file read

[Guixin Liu <kanie@linux.alibaba.com>]

在 2024/10/22 23:44, Bjorn Helgaas 写道:
> [+cc Greg]
>
> On Fri, Oct 18, 2024 at 01:47:28PM +0800, Guixin Liu wrote:
>> PCI driver will traverse pci device list in pci_seq_start in every
>> sequential file reading, use xarry to store all pci devices to
>> accelerate finding the start.
>>   /* iterator */
>>   static void *pci_seq_start(struct seq_file *m, loff_t *pos)
>>   {
>> -	struct pci_dev *dev = NULL;
>> +	struct pci_dev *dev;
>>   	loff_t n = *pos;
>>   
>> -	for_each_pci_dev(dev) {
>> -		if (!n--)
>> -			break;
>> -	}
> Maybe another approach would be to use pci_get_device() directly
> instead of for_each_pci_dev().
>
> pci_get_device() takes a "from" starting point, so instead of keeping
> track of the index "n", you could keep track of the current pci_dev.

You mean let struct seq_file to keep the pci_dev? Well, we also need keep

the pci_dev which held by seq_file in klist_devices, because finding 
next rely on

it, this make things complicated.

In addition, in pci_seq_start(), the pos may not increased one by one, so

we dont know how many times we should skip to find the pci_dev which number

is *pos.

Best Regards,

Guixin Liu