Skip to contentSkip to conversation

x86/fred: Fix int80 emulation for FRED

PATCH Superseded Xin Li (Intel) 3 participants 9 messages Apr 12, 2024 – Apr 16, 2024

Subsystems (2)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/fred: Fix int80 emulation for FRED

Commit 55617fb991df added a bunch of tests to the int $0x80 path, however they are unnecessary and event wrong in fact under FRED.

First FRED distinguishes external interrupts from software interrupts, thus int80_emulation() should NEVER be called for handling an external interrupt, and then int80_is_external() should be skipped under FRED.

Second, the FRED kernel entry handler NEVER dispatches INTx, which is of event type EVENT_TYPE_SWINT, so the user mode checking in do_int80_emulation() is redundant, and should be skipped.

It might be even better to strip down do_int80_emulation() to a lean fred_int80_emulation(), not to mention int80_emulation() does a CLEAR_BRANCH_HISTORY.

---
 arch/x86/entry/common.c | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)
arch/x86/entry/common.c +20 -2
index 6de50b80702e..aed745fc8333 100644
@@ -174,6 +174,18 @@ static __always_inline bool int80_is_external(void)
 	const unsigned int offs = (0x80 / 32) * 0x10;
 	const u32 bit = BIT(0x80 % 32);
 
+	/*
+	 * FRED distinguishes external interrupts from software interrupts
+	 * with different event types, EVENT_TYPE_EXTINT v.s. EVENT_TYPE_SWINT,
+	 * thus we should NEVER get here when FRED is enabled, and then the
+	 * following APIC ISR check makes no sense.
+	 *
+	 * Furthermore, the external interrupt vector 0x80 is available as
+	 * a hardware interrupt under FRED.
+	 */
+	if (cpu_feature_enabled(X86_FEATURE_FRED))
+		return false;
+
 	/* The local APIC on XENPV guests is fake */
 	if (cpu_feature_enabled(X86_FEATURE_XENPV))
 		return false;
@@ -211,8 +223,14 @@ __visible noinstr void do_int80_emulation(struct pt_regs *regs)
 {
 	int nr;
 
-	/* Kernel does not use INT $0x80! */
-	if (unlikely(!user_mode(regs))) {
+	/*
+	 * Kernel does not use INT $0x80!
+	 *
+	 * The FRED kernel entry handler NEVER dispatches INTx (panic or
+	 * oops otherwise), so it is redundant to check user mode here.
+	 */
+	if (!cpu_feature_enabled(X86_FEATURE_FRED) &&
+	    unlikely(!user_mode(regs))) {
 		irqentry_enter(regs);
 		instrumentation_begin();
 		panic("Unexpected external interrupt 0x80\n");

base-commit: 86d1b22a75cffa50160e4621da00311e6f6f48de

Signature · 1 line
2.44.0
#1 lore raw
Quoted 15 lines · Xin Li (Intel)

Commit 55617fb991df added a bunch of tests to the int $0x80 path, however they are unnecessary and event wrong in fact under FRED.

First FRED distinguishes external interrupts from software interrupts, thus int80_emulation() should NEVER be called for handling an external interrupt, and then int80_is_external() should be skipped under FRED.

Second, the FRED kernel entry handler NEVER dispatches INTx, which is of event type EVENT_TYPE_SWINT, so the user mode checking in do_int80_emulation() is redundant, and should be skipped.

It might be even better to strip down do_int80_emulation() to a lean fred_int80_emulation(), not to mention int80_emulation() does a CLEAR_BRANCH_HISTORY.

Quoted 5 lines
---
  arch/x86/entry/common.c | 22 ++++++++++++++++++++--
  1 file changed, 20 insertions(+), 2 deletions(-)

Note: this is the minimal bug fix versions and belongs in x86/urgent.

-hpa

maintainer
#2 lore raw
In reply to Xin Li (Intel)
Quoted 1 line · Xin Li (Intel)

Commit 55617fb991df

Use the full commit abbreviation when mentioning commits:

"Commit

55617fb991df ("x86/entry: Do not allow external 0x80 interrupts")

..."

Quoted 1 line

added a bunch of tests to the int $0x80 path,

Added a bunch of tests?

What does that even mean?

Quoted 1 line

however they are unnecessary and event wrong in fact under FRED.

Are the bunch of tests wrong or is do_int80_emulation() simply the wrong handler to use on a FRED?

Quoted 11 lines

First FRED distinguishes external interrupts from software interrupts, thus int80_emulation() should NEVER be called for handling an external interrupt, and then int80_is_external() should be skipped under FRED.

Second, the FRED kernel entry handler NEVER dispatches INTx, which is of event type EVENT_TYPE_SWINT, so the user mode checking in do_int80_emulation() is redundant, and should be skipped.

It might be even better to strip down do_int80_emulation() to a lean fred_int80_emulation(), not to mention int80_emulation() does a CLEAR_BRANCH_HISTORY.

Yah, how about you do a FRED-specific INT80 handler instead of sprinkling moar tests around? fred_intx() looks like the right place to stuff it in...

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
author
#3 lore raw
Quoted 10 lines · Borislav Petkov
Quoted 1 line · Xin Li (Intel)

Commit 55617fb991df

Use the full commit abbreviation when mentioning commits:

"Commit

55617fb991df ("x86/entry: Do not allow external 0x80 interrupts")

.."

oh, I should have done it!

Quoted 5 lines
Quoted 2 lines

added a bunch of tests to the int $0x80 path,

Added a bunch of tests?

'checks' might be a better term to use?

Quoted 6 lines

What does that even mean?

Quoted 1 line

however they are unnecessary and event wrong in fact under FRED.

Are the bunch of tests wrong or is do_int80_emulation() simply the wrong handler to use on a FRED?

I think the explanations are below, and you saw it a bit later.

BTW, comments are added around the code changes.

Quoted 9 lines
Quoted 9 lines

First FRED distinguishes external interrupts from software interrupts, thus int80_emulation() should NEVER be called for handling an external interrupt, and then int80_is_external() should be skipped under FRED.

Second, the FRED kernel entry handler NEVER dispatches INTx, which is of event type EVENT_TYPE_SWINT, so the user mode checking in do_int80_emulation() is redundant, and should be skipped.

Quoted 8 lines
Quoted 3 lines

It might be even better to strip down do_int80_emulation() to a lean fred_int80_emulation(), not to mention int80_emulation() does a CLEAR_BRANCH_HISTORY.

Yah, how about you do a FRED-specific INT80 handler instead of sprinkling moar tests around? fred_intx() looks like the right place to stuff it in...

I will add fred_int80_emulation(), which duplicates a big part of do_int80_emulation(), however this seems more readable.

Thanks!
Xin

#4 lore raw
In reply to Borislav Petkov
Quoted 30 lines · Borislav Petkov
Quoted 1 line · Xin Li (Intel)

Commit 55617fb991df

Use the full commit abbreviation when mentioning commits:

"Commit

55617fb991df ("x86/entry: Do not allow external 0x80 interrupts")

..."

Quoted 1 line

added a bunch of tests to the int $0x80 path,

Added a bunch of tests?

What does that even mean?

Quoted 1 line

however they are unnecessary and event wrong in fact under FRED.

Are the bunch of tests wrong or is do_int80_emulation() simply the wrong handler to use on a FRED?

Quoted 8 lines

First FRED distinguishes external interrupts from software interrupts, thus int80_emulation() should NEVER be called for handling an external interrupt, and then int80_is_external() should be skipped under FRED.

Second, the FRED kernel entry handler NEVER dispatches INTx, which is of event type EVENT_TYPE_SWINT, so the user mode checking in do_int80_emulation() is redundant, and should be skipped.

Quoted 8 lines
Quoted 3 lines

It might be even better to strip down do_int80_emulation() to a lean fred_int80_emulation(), not to mention int80_emulation() does a CLEAR_BRANCH_HISTORY.

Yah, how about you do a FRED-specific INT80 handler instead of sprinkling moar tests around? fred_intx() looks like the right place to stuff it in...

The question was if you wanted a quick fix for x86/urgent. It's pretty obvious that a FRED fork of the int80 code is called for.

maintainer
#5 lore raw
Quoted 2 lines · H. Peter Anvin

The question was if you wanted a quick fix for x86/urgent. It's pretty obvious that a FRED fork of the int80 code is called for.

Yeah, I'm thinking the FRED fork would be even easier to send to Linus now since no existing hw will break.

Btw, I presume that patch needs to have

Fixes 55617fb991df ("x86/entry: Do not allow external 0x80 interrupts")

?

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#6 lore raw
Quoted 6 lines · Borislav Petkov
Quoted 2 lines · H. Peter Anvin

The question was if you wanted a quick fix for x86/urgent. It's pretty obvious that a FRED fork of the int80 code is called for.

Yeah, I'm thinking the FRED fork would be even easier to send to Linus now since no existing hw will break.

Looking at the resulting code (Xin's v2 path), I would agree with you. The code is downright trivial.

Quoted 4 lines

Btw, I presume that patch needs to have

Fixes 55617fb991df ("x86/entry: Do not allow external 0x80 interrupts")

Yes. My fault, I gave Xin incorrect information w.r.t. Fixes: tags.

-hpa

maintainer
#7 lore raw
Quoted 1 line · H. Peter Anvin

Yes. My fault, I gave Xin incorrect information w.r.t. Fixes: tags.

No worries, I'll fix that up before applying.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#8 lore raw
Quoted 4 lines · Borislav Petkov
Quoted 1 line · H. Peter Anvin

Yes. My fault, I gave Xin incorrect information w.r.t. Fixes: tags.

No worries, I'll fix that up before applying.

Let me submit my comments (about comments) first :)

-hpa