Re: [RFC PATCH 2/5] entry: Add calls for save/restore auxiliary pt_regs
From: Ira Weiny
Date: Tue Aug 09 2022 - 18:33:22 EST
On Tue, Aug 09, 2022 at 11:38:48PM +0200, Borislav Petkov wrote:
> On Tue, Aug 09, 2022 at 11:14:19PM +0200, Thomas Gleixner wrote:
> > Ira is right. If we want it for everything, then the generic code is the
> > right place.
>
> But what is "everything"? Currently, and AFAIU, it is for the PKS use
> case on x86 only.
>
> I'm not saying it should not be done this way eventually - all I'm
> saying is we should not design "preemptively" before it is really needed
> for other arches.
>
> Unless putting it in generic code makes it all simpler and cleaner to
> do, that is.
For the cpu use case we could limit the number of call sites. However, for PKS
the patch would have required changing x86 code in approximately 9 places for
the enter code.
$ git grep 'irqentry_enter(regs)' arch/x86 | wc -l
9
How about we drop patch 1 (I'll rework it to be less churn and submit it for
clean up separately because it will no longer be needed). Keep patch 3 as is.
Then combine 2 and 5 as below. The saving of the CPU can be lifted later if
needed.
Ira
commit 4c1d646888dd7471ae71a24109d587901a00f87d
Author: Ira Weiny <ira.weiny@xxxxxxxxx>
Date: Mon Jan 10 15:06:07 2022 -0800
x86/entry: Store CPU info on exception entry
x86 has auxiliary pt_regs space available to store information on the
stack during exceptions. This information is easier to obtain and store
within C code.
The CPU information of a page fault is useful in determining where bad
CPUs are in a large data center.
Define aux_pt_regs_save_cpu() and set ARCH_HAS_PTREGS_AUXILIARY default
to yes.
Store the CPU on page fault entry and use it later.
Cc: Rik van Riel <riel@xxxxxxxxxxx>
Suggested-by: Borislav Petkov <bp@xxxxxxxxx>
Suggested-by: Dave Hansen <dave.hansen@xxxxxxxxx>
Suggested-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Signed-off-by: Ira Weiny <ira.weiny@xxxxxxxxx>
---
Changes from RFC:
New patch combining 2 and 5 from original series and modified.
Boris/Thomas - eliminate generic calls to save the cpu and call
only from exc_page_fault
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index b35f6a472e09..707650a6ecb2 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1876,7 +1876,7 @@ config X86_INTEL_MEMORY_PROTECTION_KEYS
config ARCH_HAS_PTREGS_AUXILIARY
depends on X86_64
- bool
+ def_bool y
choice
prompt "TSX enable mode"
diff --git a/arch/x86/include/asm/ptrace.h b/arch/x86/include/asm/ptrace.h
index 5a9c85893459..b403b469996f 100644
--- a/arch/x86/include/asm/ptrace.h
+++ b/arch/x86/include/asm/ptrace.h
@@ -97,6 +97,7 @@ struct pt_regs {
* ARCH_HAS_PTREGS_AUXILIARY. Failure to do so will result in a build failure.
*/
struct pt_regs_auxiliary {
+ int cpu;
};
struct pt_regs_extended {
diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index 82cf23975aa1..b9b8344b69ad 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -768,9 +768,9 @@ static inline void
show_signal_msg(struct pt_regs *regs, unsigned long error_code,
unsigned long address, struct task_struct *tsk)
{
+ struct pt_regs_auxiliary *aux_pt_regs = &to_extended_pt_regs(regs)->aux;
const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG;
- /* This is a racy snapshot, but it's better than nothing. */
- int cpu = raw_smp_processor_id();
+ int cpu = aux_pt_regs->cpu;
if (!unhandled_signal(tsk, SIGSEGV))
return;
@@ -1503,6 +1503,13 @@ handle_page_fault(struct pt_regs *regs, unsigned long error_code,
}
}
+static void aux_pt_regs_save_cpu(struct pt_regs *regs)
+{
+ struct pt_regs_auxiliary *aux_pt_regs = &to_extended_pt_regs(regs)->aux;
+
+ aux_pt_regs->cpu = raw_smp_processor_id();
+}
+
DEFINE_IDTENTRY_RAW_ERRORCODE(exc_page_fault)
{
unsigned long address = read_cr2();
@@ -1546,6 +1553,7 @@ DEFINE_IDTENTRY_RAW_ERRORCODE(exc_page_fault)
*/
state = irqentry_enter(regs);
+ aux_pt_regs_save_cpu(regs);
instrumentation_begin();
handle_page_fault(regs, error_code, address);
instrumentation_end();