On Fri, 2 Aug 2019 11:03:48 +1000
Daniel Black <dan...@linux.ibm.com> wrote:
> On Thu, 1 Aug 2019 12:41:59 +0200
> Greg Kurz <gr...@kaod.org> wrote:
>
> > On Thu, 1 Aug 2019 13:38:19 +1000
> > Daniel Black <dan...@linux.ibm.com> wrote:
> >
> > > Its not immediately obvious how cap-X=Y setting need to be applied
> > > to the command line so, for spapr capability error messages, this
> > > has been clarified to:
> > >
> ...
> > > index bbb001f84a..1c0222a081 100644
> > > --- a/hw/ppc/spapr_caps.c
> > > +++ b/hw/ppc/spapr_caps.c
> > > @@ -37,6 +37,8 @@
> > >
> > > #include "hw/ppc/spapr.h"
> > >
> > > +#define CAPABILITY_ERROR(X) "appending -machine " X
> >
> > I would make that:
> >
> > #define CAPABILITY_HINT() "try appending -machine " X
> >
> > because it is really an hint for the user, not an
> > error,
>
> Works for me. At the lowest layer it is a hint.
>
> > and all original strings have "try",
>
> True.
>
> > except...
>
>
> > > @@ -249,11 +255,13 @@ static void
> > > cap_safe_cache_apply(SpaprMachineState *spapr, uint8_t val, if
> > > (tcg_enabled() && val) { /* TCG only supports broken, allow other
> > > values and print a warning */ error_setg(&local_err,
> > > - "TCG doesn't support requested feature,
> > > cap-cfpc=%s",
> > > + "TCG doesn't support requested feature, "
> > > + CAPABILITY_ERROR("cap-cfpc=%s"),
> >
> > ... this one, but it doesn't look like a hint to me. It just tells
> > which is the unsupported cap.
>
> This is one of 3 that local_error (commit
> 006e9d3618698eeef2f3e07628d22cb6f5c2a039) - intentionally just a
> warning and to TLDR the commit/Suraj conversation; defaults apply
> to all machine types; hardware security measures don't make sense in
> TCG; hence warning.
>
Sure. What I meant is that the warning for "cap-cfpc" should be left alone,
otherwise we get:
$ ppc64-softmmu/qemu-system-ppc64 -nodefaults -nographic -machine
pseries,accel=tcg
qemu-system-ppc64: warning: TCG doesn't support requested feature, appending
-machine cap-cfpc=workaround
qemu-system-ppc64: warning: TCG doesn't support requested feature,
cap-sbbc=workaround
qemu-system-ppc64: warning: TCG doesn't support requested feature,
cap-ibs=workaround
and even weirder:
$ ppc64-softmmu/qemu-system-ppc64 -nodefaults -nographic -machine
pseries,accel=tcg -machine cap-cfpc=workaround
qemu-system-ppc64: warning: TCG doesn't support requested feature, appending
-machine cap-cfpc=workaround
qemu-system-ppc64: warning: TCG doesn't support requested feature,
cap-sbbc=workaround
qemu-system-ppc64: warning: TCG doesn't support requested feature,
cap-ibs=workaround
> For every function with CAPABILITY_[ERROR|HINT] its called by
> spapr_caps_apply, has its errp as &error_fatal (intentionally - spoke
> to Suraj - migrations to machines without capabilities need to fail and
> defaults (kvm) should be secure unless explicitly disabled).
>
> > > cap_cfpc_possible.vals[val]);
> > > } else if (kvm_enabled() && (val > kvm_val)) {
> > > error_setg(errp,
> > > -"Requested safe cache capability level not supported by kvm, try
> > > cap-cfpc=%s", +"Requested safe cache capability level not supported
> > > by kvm, try "
> > > + CAPABILITY_ERROR("cap-cfpc=%s"),
> > > cap_cfpc_possible.vals[kvm_val]);
> >
> > Also, we have a dedicated API for hints, which are only printed under
> > the monitor but ignored under QMP.
>
> Ok.
>
> > Not sure why it isn't used here but it should be something like:
>
> If error_append_hint should be used for fatal errors (all that use
> errp), then this patten should be applied further to
> CAPABILITY_[HINT|ERROR] functions.
>
Hmm... looking again at error_append_hint(), it shouldn't be passed
&error_fatal or &error_abort. This calls for an extra local_err and
error_propagate() dance... Definitely not in the scope of this patch.
> If error_append_hint needs to apply to warnings
> cap_[cfpc/sbbc/ibs]_apply functions need to use it.
>
The current warnings from commit 006e9d3618698eeef2f3e07628d22cb6f5c2a039
don't contain hints. We could potentially add one that says "try appending
-machine cap-blah=broken" but it doesn't bring much...
> Would I be right in I'm assuming that the below pattern needs to apply
> to both of these cases?
>
> > error_setg(errp,
> > "Requested safe cache capability level not
> > supported by kvm");
> > error_append_hint(errp,
> > CAPABILITY_HINT("cap-cfpc=%s") "\n", cap_cfpc_possible.vals[kvm_val]);
>
> This is going a little beyond the scope of fixing a message, ok, but
> lets not extend the scope too much more.
>
Yes, I agree. Your patch should only be about fixing the message.
I'll have a look at the error_append_hint() story when I'm back
from holidays :)
So, to sum up:
- s/CAPABILITY_ERROR/CAPABILITY_HINT and move "try" there
- drop the unwanted change in the "cap-cfpc" warning
With these fixed:
Reviewed-by: Greg Kurz <gr...@kaod.org>
Cheers,
--
Greg
