Re: [PATCH] PCI/VPD: Use unaligned access helper in pci_vpd_write

[Heiner Kallweit <hkallweit1@gmail.com>]

On 30.04.2021 22:29, Bjorn Helgaas wrote:
> On Sun, Apr 18, 2021 at 11:19:29AM +0200, Heiner Kallweit wrote:
>> Use helper get_unaligned_le32() to simplify the code.
>>
>> Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
>> ---
>>  drivers/pci/vpd.c | 15 +++++----------
>>  1 file changed, 5 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/pci/vpd.c b/drivers/pci/vpd.c
>> index 60573f27a..2888bb300 100644
>> --- a/drivers/pci/vpd.c
>> +++ b/drivers/pci/vpd.c
>> @@ -9,6 +9,7 @@
>>  #include <linux/delay.h>
>>  #include <linux/export.h>
>>  #include <linux/sched/signal.h>
>> +#include <asm/unaligned.h>
>>  #include "pci.h"
>>  
>>  /* VPD access through PCI 2.2+ VPD capability */
>> @@ -235,10 +236,9 @@ static ssize_t pci_vpd_read(struct pci_dev *dev, loff_t pos, size_t count,
>>  }
>>  
>>  static ssize_t pci_vpd_write(struct pci_dev *dev, loff_t pos, size_t count,
>> -			     const void *arg)
>> +			     const void *buf)
>>  {
>>  	struct pci_vpd *vpd = dev->vpd;
>> -	const u8 *buf = arg;
>>  	loff_t end = pos + count;
>>  	int ret = 0;
>>  
>> @@ -264,14 +264,8 @@ static ssize_t pci_vpd_write(struct pci_dev *dev, loff_t pos, size_t count,
>>  		goto out;
>>  
>>  	while (pos < end) {
>> -		u32 val;
>> -
>> -		val = *buf++;
>> -		val |= *buf++ << 8;
>> -		val |= *buf++ << 16;
>> -		val |= *buf++ << 24;
>> -
>> -		ret = pci_user_write_config_dword(dev, vpd->cap + PCI_VPD_DATA, val);
>> +		ret = pci_user_write_config_dword(dev, vpd->cap + PCI_VPD_DATA,
>> +						  get_unaligned_le32(buf));
> 
> If I understand correctly, get_unaligned_le32() is equivalent to the
> open-coded "val = *buf++; ..."  In other words, this doesn't fix a bug
> on any platforms, and it doesn't change the bits written to VPD.
> Right?
> 
Right.

> If so, this makes sense.  I'd like it better if we could also make
> pci_vpd_read() look more symmetric, since it has similar byte
> order-related code.  The read side is a little more complicated since
> the size need not be 4-byte aligned.  But maybe there's a way?  Seems
> like there should be other examples of this sort of thing, but my
> quick look didn't find one.
> 
I also find the read side a little bit hard to understand.
Let me check whether I find some way to improve the code.


>>  		if (ret < 0)
>>  			break;
>>  		ret = pci_user_write_config_word(dev, vpd->cap + PCI_VPD_ADDR,
>> @@ -286,6 +280,7 @@ static ssize_t pci_vpd_write(struct pci_dev *dev, loff_t pos, size_t count,
>>  			break;
>>  
>>  		pos += sizeof(u32);
>> +		buf += sizeof(u32);
>>  	}
>>  out:
>>  	mutex_unlock(&vpd->lock);
>> -- 
>> 2.31.1
>>