Skip to contentSkip to conversation

acpi: fix suspend with Xen PV

PATCH In review Juergen Gross 3 participants 6 messages Jan 17, 2023 – Jan 18, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 acpi: fix suspend with Xen PV

Commit f1e525009493 ("x86/boot: Skip realmode init code when running as Xen PV guest") missed one code path accessing real_mode_header, leading to dereferencing NULL when suspending the system under Xen:

    [  348.284004] PM: suspend entry (deep)
    [  348.289532] Filesystems sync: 0.005 seconds
    [  348.291545] Freezing user space processes ... (elapsed 0.000 seconds) done.
    [  348.292457] OOM killer disabled.
    [  348.292462] Freezing remaining freezable tasks ... (elapsed 0.104 seconds) done.
    [  348.396612] printk: Suspending console(s) (use no_console_suspend to debug)
    [  348.749228] PM: suspend devices took 0.352 seconds
    [  348.769713] ACPI: EC: interrupt blocked
    [  348.816077] BUG: kernel NULL pointer dereference, address: 000000000000001c
    [  348.816080] #PF: supervisor read access in kernel mode
    [  348.816081] #PF: error_code(0x0000) - not-present page
    [  348.816083] PGD 0 P4D 0
    [  348.816086] Oops: 0000 [#1] PREEMPT SMP NOPTI
    [  348.816089] CPU: 0 PID: 6764 Comm: systemd-sleep Not tainted 6.1.3-1.fc32.qubes.x86_64 #1
    [  348.816092] Hardware name: Star Labs StarBook/StarBook, BIOS 8.01 07/03/2022
    [  348.816093] RIP: e030:acpi_get_wakeup_address+0xc/0x20

Fix that by adding an optional acpi callback allowing to skip setting the wakeup address, as in the Xen PV case this will be handled by the hypervisor anyway.

Fixes f1e525009493 ("x86/boot: Skip realmode init code when running as Xen PV guest")

---
V2:
- new approach, avoid calling acpi_get_wakeup_address()

---
 arch/x86/include/asm/acpi.h | 8 ++++++++
 drivers/acpi/sleep.c        | 6 +++++-
 2 files changed, 13 insertions(+), 1 deletion(-)
arch/x86/include/asm/acpi.h +8 -0
index 65064d9f7fa6..8eb74cf386db 100644
@@ -14,6 +14,7 @@
 #include <asm/mmu.h>
 #include <asm/mpspec.h>
 #include <asm/x86_init.h>
+#include <asm/cpufeature.h>
 
 #ifdef CONFIG_ACPI_APEI
 # include <asm/pgtable_types.h>
@@ -63,6 +64,13 @@ extern int (*acpi_suspend_lowlevel)(void);
 /* Physical address to resume after wakeup */
 unsigned long acpi_get_wakeup_address(void);
 
+static inline bool acpi_skip_set_wakeup_address(void)
+{
+	return cpu_feature_enabled(X86_FEATURE_XENPV);
+}
+
+#define acpi_skip_set_wakeup_address acpi_skip_set_wakeup_address
+
 /*
  * Check if the CPU can handle C2 and deeper
  */
drivers/acpi/sleep.c +5 -1
index 0b557c0d405e..4ca667251272 100644
@@ -60,13 +60,17 @@ static struct notifier_block tts_notifier = {
 	.priority	= 0,
 };
 
+#ifndef acpi_skip_set_wakeup_address
+#define acpi_skip_set_wakeup_address() false
+#endif
+
 static int acpi_sleep_prepare(u32 acpi_state)
 {
 #ifdef CONFIG_ACPI_SLEEP
 	unsigned long acpi_wakeup_address;
 
 	/* do we have a wakeup address for S2 and S3? */
-	if (acpi_state == ACPI_STATE_S3) {
+	if (acpi_state == ACPI_STATE_S3 && !acpi_skip_set_wakeup_address()) {
 		acpi_wakeup_address = acpi_get_wakeup_address();
 		if (!acpi_wakeup_address)
 			return -EFAULT;
Signature · 1 line
2.35.3
#1 lore raw
Quoted 26 lines · Juergen Gross

Commit f1e525009493 ("x86/boot: Skip realmode init code when running as Xen PV guest") missed one code path accessing real_mode_header, leading to dereferencing NULL when suspending the system under Xen:

    [  348.284004] PM: suspend entry (deep)
    [  348.289532] Filesystems sync: 0.005 seconds
    [  348.291545] Freezing user space processes ... (elapsed 0.000 seconds) done.
    [  348.292457] OOM killer disabled.
    [  348.292462] Freezing remaining freezable tasks ... (elapsed 0.104 seconds) done.
    [  348.396612] printk: Suspending console(s) (use no_console_suspend to debug)
    [  348.749228] PM: suspend devices took 0.352 seconds
    [  348.769713] ACPI: EC: interrupt blocked
    [  348.816077] BUG: kernel NULL pointer dereference, address: 000000000000001c
    [  348.816080] #PF: supervisor read access in kernel mode
    [  348.816081] #PF: error_code(0x0000) - not-present page
    [  348.816083] PGD 0 P4D 0
    [  348.816086] Oops: 0000 [#1] PREEMPT SMP NOPTI
    [  348.816089] CPU: 0 PID: 6764 Comm: systemd-sleep Not tainted 6.1.3-1.fc32.qubes.x86_64 #1
    [  348.816092] Hardware name: Star Labs StarBook/StarBook, BIOS 8.01 07/03/2022
    [  348.816093] RIP: e030:acpi_get_wakeup_address+0xc/0x20

Fix that by adding an optional acpi callback allowing to skip setting the wakeup address, as in the Xen PV case this will be handled by the hypervisor anyway.

Quoted 6 lines
Fixes f1e525009493 ("x86/boot: Skip realmode init code when running as Xen PV guest")

---
V2:
- new approach, avoid calling acpi_get_wakeup_address()

I'll queue this up for 6.3 if the x86 people don't object.

Thanks!

Quoted 55 lines
---
 arch/x86/include/asm/acpi.h | 8 ++++++++
 drivers/acpi/sleep.c        | 6 +++++-
 2 files changed, 13 insertions(+), 1 deletion(-)
arch/x86/include/asm/acpi.h +8 -0
index 65064d9f7fa6..8eb74cf386db 100644
@@ -14,6 +14,7 @@
 #include <asm/mmu.h>
 #include <asm/mpspec.h>
 #include <asm/x86_init.h>
+#include <asm/cpufeature.h>
 
 #ifdef CONFIG_ACPI_APEI
 # include <asm/pgtable_types.h>
@@ -63,6 +64,13 @@ extern int (*acpi_suspend_lowlevel)(void);
 /* Physical address to resume after wakeup */
 unsigned long acpi_get_wakeup_address(void);
 
+static inline bool acpi_skip_set_wakeup_address(void)
+{
+       return cpu_feature_enabled(X86_FEATURE_XENPV);
+}
+
+#define acpi_skip_set_wakeup_address acpi_skip_set_wakeup_address
+
 /*
  * Check if the CPU can handle C2 and deeper
  */
drivers/acpi/sleep.c +5 -2
index 0b557c0d405e..4ca667251272 100644
@@ -60,13 +60,17 @@ static struct notifier_block tts_notifier = {
        .priority       = 0,
 };
 
+#ifndef acpi_skip_set_wakeup_address
+#define acpi_skip_set_wakeup_address() false
+#endif
+
 static int acpi_sleep_prepare(u32 acpi_state)
 {
 #ifdef CONFIG_ACPI_SLEEP
        unsigned long acpi_wakeup_address;
 
        /* do we have a wakeup address for S2 and S3? */
-       if (acpi_state == ACPI_STATE_S3) {
+       if (acpi_state == ACPI_STATE_S3 && !acpi_skip_set_wakeup_address()) {
                acpi_wakeup_address = acpi_get_wakeup_address();
                if (!acpi_wakeup_address)
                        return -EFAULT;
--
#2 lore raw
Quoted 7 lines · Rafael J. Wysocki
Quoted 6 lines
Fixes f1e525009493 ("x86/boot: Skip realmode init code when running as Xen PV guest")

---
V2:
- new approach, avoid calling acpi_get_wakeup_address()

I'll queue this up for 6.3 if the x86 people don't object.

That 'Fixes:' patch looked to me like it was applied during the 6.2 merge window. I would have expected a fix to go up for 6.2 final.

#3 lore raw
Quoted 8 lines · Dave Hansen
Quoted 6 lines · Rafael J. Wysocki
Quoted 6 lines
Fixes f1e525009493 ("x86/boot: Skip realmode init code when running as Xen PV guest")

---
V2:
- new approach, avoid calling acpi_get_wakeup_address()

Quoted 4 lines
Quoted 1 line

I'll queue this up for 6.3 if the x86 people don't object.

That 'Fixes:' patch looked to me like it was applied during the 6.2 merge window. I would have expected a fix to go up for 6.2 final.

Right, sorry. Definitely, 6.2-rc material.

#4 lore raw
Quoted 8 lines · Rafael J. Wysocki
Quoted 7 lines · Dave Hansen
Quoted 6 lines · Rafael J. Wysocki
Quoted 6 lines
Fixes f1e525009493 ("x86/boot: Skip realmode init code when running as Xen PV guest")

---
V2:
- new approach, avoid calling acpi_get_wakeup_address()

Quoted 4 lines
Quoted 3 lines
Quoted 1 line

I'll queue this up for 6.3 if the x86 people don't object.

That 'Fixes:' patch looked to me like it was applied during the 6.2 merge window. I would have expected a fix to go up for 6.2 final.

Right, sorry. Definitely, 6.2-rc material.

I think the original issue went up in x86/boot, so how about I queue this in x86/urgent? Rafael, should I add an ack from you on it?

maintainer Acked-by
#5 lore raw
Quoted 10 lines · Dave Hansen
Quoted 8 lines · Rafael J. Wysocki
Quoted 7 lines · Dave Hansen
Quoted 6 lines · Rafael J. Wysocki
Quoted 6 lines
Fixes f1e525009493 ("x86/boot: Skip realmode init code when running as Xen PV guest")

---
V2:
- new approach, avoid calling acpi_get_wakeup_address()

Quoted 7 lines
Quoted 4 lines
Quoted 3 lines
Quoted 1 line

I'll queue this up for 6.3 if the x86 people don't object.

That 'Fixes:' patch looked to me like it was applied during the 6.2 merge window. I would have expected a fix to go up for 6.2 final.

Right, sorry. Definitely, 6.2-rc material.

I think the original issue went up in x86/boot, so how about I queue this in x86/urgent? Rafael, should I add an ack from you on it?

Yes, please.