[+cc Tony]
On Wed, Feb 25, 2015 at 10:54 AM, Borislav Petkov <bp@xxxxxxx> wrote:
> On Tue, Feb 24, 2015 at 11:50:46PM +0100, Rasmus Villemoes wrote:
>> Commit fab4c256a58b ("PCI/AER: Add a TLP header print helper")
>> introduced the helper function __print_tlp_header, but contrary to the
>> intention, the behaviour did change: Since we're taking the address of
>
> Whoops, good catch.
>
>> the parameter t, the first 4 or 8 bytes printed will be the value of
>> the pointer t itself, and the remaining 12 or 8 bytes will be
>> who-knows-what (something from the stack).
>>
>> We want to treat the four members of the struct aer_header_log_regs as
>> little-endian 32 bit numbers and print those. That can be done without
>> ugly and confusing casts.
>>
>> Fixes: fab4c256a58b ("PCI/AER: Add a TLP header print helper")
>> Signed-off-by: Rasmus Villemoes <linux@xxxxxxxxxxxxxxxxxx>
>> ---
>> drivers/pci/pcie/aer/aerdrv_errprint.c | 13 +++----------
>> 1 file changed, 3 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/pci/pcie/aer/aerdrv_errprint.c b/drivers/pci/pcie/aer/aerdrv_errprint.c
>> index c6849d9e86ce..e328978038c1 100644
>> --- a/drivers/pci/pcie/aer/aerdrv_errprint.c
>> +++ b/drivers/pci/pcie/aer/aerdrv_errprint.c
>> @@ -132,16 +132,9 @@ static const char *aer_agent_string[] = {
>> static void __print_tlp_header(struct pci_dev *dev,
>> struct aer_header_log_regs *t)
>> {
>> - unsigned char *tlp = (unsigned char *)&t;
>> -
>> - dev_err(&dev->dev, " TLP Header:"
>> - " %02x%02x%02x%02x %02x%02x%02x%02x"
>> - " %02x%02x%02x%02x %02x%02x%02x%02x\n",
>> - *(tlp + 3), *(tlp + 2), *(tlp + 1), *tlp,
>> - *(tlp + 7), *(tlp + 6), *(tlp + 5), *(tlp + 4),
>> - *(tlp + 11), *(tlp + 10), *(tlp + 9),
>> - *(tlp + 8), *(tlp + 15), *(tlp + 14),
>> - *(tlp + 13), *(tlp + 12));
>> + dev_err(&dev->dev, " TLP Header: %08x %08x %08x %08x\n",
>> + le32_to_cpu(t->dw0), le32_to_cpu(t->dw1),
>> + le32_to_cpu(t->dw2), le32_to_cpu(t->dw3));
>
> I'm not sure about this: I think the original intention was to dump the
> dwords MS-bit to LS-bit like this here:
>
> http://www.fpga4fun.com/PCI-Express4.html
>
> Now, if this runs on a big endian machine, converting to CPU order would
> be wrong IMHO. You'd rather want do do cpu_to_le32() for consistency.
> But I don't know whether big endian machines are even sporting PCIE
> AER...
I think we should expect AER to be used on big-endian machines. I'm
pretty sure it's used on Itanium in big-endian mode.
Why are we worrying about byte order here at all? I'd think we could
just print t->dw0 directly with %08x.
Any byte order issues should be handled when we fill in the struct
aer_header_log_regs. For normal AER (non-APEI), we use
pci_read_config_dword() to directly fill in info->tlp.dw0, etc. in
get_device_error_info(), so I don't think there's a problem there.
For APEI, it looks like it would happen somewhere in
ghes_read_estatus(). I didn't follow the whole path here, but I would
argue that by the time we put data in t->dw0, it should be in CPU
order so a mask like 0x80000000 would work the same on LE and BE
boxes.
Bjorn
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
