Re: [Xen-devel] [PATCH V2] Decouple SnadyBridge quirk form VTd timeout

2014-11-17 Thread Dugger, Donald D
Well, Allen (the original VTd maintainer and the author of the IGD quirk code) agrees that the original IGD quirk timeout code just used the VTd timeout as a convenience. He's the one who identified 670 msec as the theoretical max latency for IGD access but, due to certification issues, suggest

Re: [Xen-devel] [PATCH V2] Decouple SnadyBridge quirk form VTd timeout

2014-11-17 Thread Jan Beulich
>>> On 17.11.14 at 15:51, wrote: > My primary goal is to decouple the IGD quirk code from the VTd timeout value, > the two are unrelated and shouldn't be using the same define. In the process > we can clean up the IGD code so that, if a user wants, the user can specify a > more appropriate tim

Re: [Xen-devel] [PATCH V2] Decouple SnadyBridge quirk form VTd timeout

2014-11-17 Thread Dugger, Donald D
My primary goal is to decouple the IGD quirk code from the VTd timeout value, the two are unrelated and shouldn't be using the same define. In the process we can clean up the IGD code so that, if a user wants, the user can specify a more appropriate timeout value for the quirk code. There's no

Re: [Xen-devel] [PATCH V2] Decouple SnadyBridge quirk form VTd timeout

2014-11-17 Thread Jan Beulich
>>> On 17.11.14 at 15:27, wrote: > I'm a big believer in backward compatibility, especially in not doing things > that change current defined behavior. The current `snb_igd_quirk' flag > enables the quirk code with a 1sec timeout. Even though that value is > excessive silently changing the me

Re: [Xen-devel] [PATCH V2] Decouple SnadyBridge quirk form VTd timeout

2014-11-17 Thread Dugger, Donald D
I'm a big believer in backward compatibility, especially in not doing things that change current defined behavior. The current `snb_igd_quirk' flag enables the quirk code with a 1sec timeout. Even though that value is excessive silently changing the meaning of the parameter just seems wrong.

Re: [Xen-devel] [PATCH V2] Decouple SnadyBridge quirk form VTd timeout

2014-11-17 Thread Jan Beulich
>>> On 17.11.14 at 14:32, wrote: > Hmm, good ideas. How about I change the `snb_igd_quirk' parameter to be: > > snb_igd_quirk => current behavior, enable > quirk with 1 sec timeout > snb_igd_quirk=default => enable quirk with theoretical max > t

Re: [Xen-devel] [PATCH V2] Decouple SnadyBridge quirk form VTd timeout

2014-11-17 Thread Dugger, Donald D
Hmm, good ideas. How about I change the `snb_igd_quirk' parameter to be: snb_igd_quirk => current behavior, enable quirk with 1 sec timeout snb_igd_quirk=default => enable quirk with theoretical max timeout of 670 msec snb_igd_qui

Re: [Xen-devel] [PATCH V2] Decouple SnadyBridge quirk form VTd timeout

2014-11-17 Thread Jan Beulich
>>> On 17.11.14 at 05:11, wrote: > @@ -237,6 +254,18 @@ > } > } > > +static void __init parse_snb_timeout(const char *s) > +{ > + > + if (*s == '\0') > + snb_igd_timeout = SNB_IGD_TIMEOUT; > + else > + snb_igd_timeout = MILLISECS(simple_strtoul(s, &s, 0)); >