if_msk.c link negotiation / packet drops

[Arnaud Lacombe]

Hi,

On Thu, Oct 13, 2011 at 4:47 PM, YongHyeon PYUN <pyunyh at gmail.com> wrote:
> On Thu, Oct 13, 2011 at 12:59:35AM -0700, perryh at pluto.rain.com wrote:
>> YongHyeon PYUN <pyunyh at gmail.com> wrote:
>> > On Wed, Oct 12, 2011 at 10:07:02AM -0400, Karim wrote:
>> > > ... why are we ORing the same call twice isn't the same thing
>> > > as calling it once:
>> > >
>> > > bmsr = PHY_READ(sc, E1000_SR) | PHY_READ(sc, E1000_SR);
>> >
>> > The E1000_SR_LINK_STATUS bit is latched low so it should be read
>> > twice.
>>
>> It might not be a bad idea to check the generated code to be sure
>> that the read _is_ being done twice.  An optimizer might well come
>> to the same conclusion as Karim, and discard the "redundant" second
>> instance (unless there's a "volatile" declaration somewhere in the
>> expansion of PHY_READ, to explicitly indicate that it has side
>> effects).
>
> Last time I checked it, compiler generated correct code.
> Tried again on amd64 and I can still see the code is there.
>
What about other architecture (especially i386) ? which optimization
level did you use ? which compiler version ?

About the last question, I know for sure that there has been change in
FreeBSD's gcc between 7-STABLE, and FreeBSD -CURRENT.

I agree with perryh@ than such hardware requirement _requires_ being
explicit in the code, ie proper `volatile' marking.

 - Arnaud