Messages in this thread | | | Subject | Re: [PATCH v2 2/6] x86/mm/pat: Add pat_disable() interface | From | Toshi Kani <> | Date | Tue, 22 Mar 2016 15:40:45 -0600 |
| |
On Tue, 2016-03-22 at 17:59 +0100, Borislav Petkov wrote: > On Wed, Mar 16, 2016 at 06:46:55PM -0600, Toshi Kani wrote: > > In preparation to fix a regression caused by 'commit 9cd25aac1f44 > > ("x86/mm/pat: Emulate PAT when it is disabled")', PAT needs to > > provide an interface that disables the OS to initialize PAT MSR. > > prevents the OS from initializing the PAT MSR.
Right. Will do.
> > > > PAT MSR initialization must be done on all CPUs with the specific > > s/with/using/
Ditto.
> > sequence of operations defined in Intel SDM. This requires MTRR > ^ > the > > s/MTRR/MTRRs/
Ditto.
> > to be enabled since pat_init() is called as part of MTRR init > > from mtrr_rendezvous_handler(). > > > > Change pat_disable() as the interface to disable the OS to initialize > > PAT MSR, and set PAT table with pat_keep_handoff_state(). This > > interface can be called when PAT initialization may not be performed. > > This paragraph reads funky and I can't really parse what it is trying to > say.
Sorry... Here is a retry:
Make pat_disable() as the interface that prevents the OS from initializing the PAT MSR. MTRR will call this interface when it cannot provide the SDM- defined sequence to initialize PAT.
> > This also assures that pat_disable() called from pat_bsp_init() > > to set PAT table properly when CPU does not support PAT. > > : > > > > -static inline void pat_disable(const char *reason) > > +/** > > + * pat_disable() - Disable the OS to initialize PAT MSR > > ^^^^ > > Err, what? The function name can't be more clear.
Will change to "Prevent the OS from initializing the PAT MSR".
I wanted to clarify that "disable" does not mean to disable PAT MSR.
> > + * > > + * This function disables the OS to initialize PAT MSR, and calls > > "prevents the OS from initializing the PAT MSR..."
Will do.
> > + * pat_keep_handoff_state() to set PAT table to the handoff state. > > We can see what is calls. You're explaining *what* the code does instead > of *why* again.
Right...
> > + */ > > +void pat_disable(const char *reason) > > { > > Why aren't you checking __pat_enabled here? > > if (!__pat_enabled) > return;
pat_keep_handoff_state() is a no-op after the initial call, but I agree that having this check is better. Will do.
> You can save yourself the other guards in that function, especially that > pr_err() below.
The pr_err() below is for a difference case -- PAT is enabled, but a call is made to disable it after pat_init() is called. We cannot allow this case.
> > + if (boot_cpu_done) { > > + pr_err("x86/PAT: PAT cannot be disabled after > > initialization " > > + "(attempting: %s)\n", reason); > > Please integrate checkpatch.pl into your patch creation workflow as it > sometimes has valid complaints: > > WARNING: quoted string split across lines > #79: FILE: arch/x86/mm/pat.c:55: > + pr_err("x86/PAT: PAT cannot be disabled after > initialization " > + "(attempting: %s)\n", reason);
I've run checkpatch.pl and thought it was OK to have this warning (instead of a >80 warning) since the error message part was not split. The "attempting" part is for debugging and its string is passed from the caller.
> More to the point: why do we need that pr_err() call? What is that > supposed to tell the user? > > I think it is more for the programmer to catch wrong use of > pat_disable() and then it should be WARN_ONCE() or so...
Yes, this case is for the programmer to catch wrong use. I will change it to use WARN_ONCE() and remove the "(attempting: %s)\n" part of the message.
> > + return; > > + } > > + > > __pat_enabled = 0; > > pr_info("x86/PAT: %s\n", reason); > > + > > + pat_keep_handoff_state(); > > } > > > > static int __init nopat(char *str) > > @@ -202,7 +217,7 @@ static void pat_bsp_init(u64 pat) > > { > > u64 tmp_pat; > > > > - if (!cpu_has_pat) { > > + if (!boot_cpu_has(X86_FEATURE_PAT)) { > > pat_disable("PAT not supported by CPU."); > > return; > > } > > @@ -220,7 +235,7 @@ static void pat_bsp_init(u64 pat) > > > > static void pat_ap_init(u64 pat) > > { > > - if (!cpu_has_pat) { > > + if (!boot_cpu_has(X86_FEATURE_PAT)) { > > /* > > * If this happens we are on a secondary CPU, but > > switched to > > * PAT on the boot CPU. We have no way to undo PAT. > > Those last two hunks are unrelated changes and should be a separate > patch.
Will do.
Thanks, -Toshi
| |