Skip to contentSkip to conversation

Fix PIC interrupt handling of x86 CPUs if APIC is globally disabled

PATCH Superseded Bernhard Beschow 2 participants 6 messages Jan 3, 2024

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover Fix PIC interrupt handling of x86 CPUs if APIC is globally disabled

This two-patch series is part of my work emulating the VIA Apollo Pro 133T chipset in QEMU [1] and testing it by running real-world BIOSes on it. The first patch fixes an issue regarding PIC interrupt handling, the second one just fixes a typo in a comment.

During testing, I've found that the boot process gets stuck for some BIOSes that disable the LAPIC globally (by disabling the enable bit in the base address register). QEMU seems to emulate PIC interrupt handling only if a CPU doesn't have a LAPIC, and always emulates LAPIC interrupt handling if one is present. According to the Intel documentation, a CPU should resort to PIC interrupt handling if its LAPIC is globally didabled. This series fixes this corner case which makes the boot process succeed. More details can be found in the commit message.

Testing done:
* make check
* make check-avocado

[1] https://github.com/shentok/qemu/tree/via-apollo-pro-133t

Bernhard Beschow (2):
hw/i386/x86: Fix PIC interrupt handling if APIC globally disabled target/i386/cpu: Fix small typo in comment

 hw/i386/x86.c     | 10 +++++-----
 target/i386/cpu.c |  2 +-
 2 files changed, 6 insertions(+), 6 deletions(-)
Signature · 1 line
2.43.0
#1 lore raw

PATCH 1/2 hw/i386/x86: Fix PIC interrupt handling if APIC globally disabled

QEMU populates the apic_state attribute of x86 CPUs if supported by real hardware. Even when the APIC is globally disabled by a guest, this attribute stays populated. This means that the APIC code paths are still used in this case. However, chapter 10.4.3 of [1] requires that:

When IA32_APIC_BASE[11] is 0, the processor is functionally equivalent to an IA-32 processor without an on-chip APIC. The CPUID feature flag for the APIC [...] is also set to 0.

Fix this by checking the APIC feature flag rather than apic_state when deciding whether PIC or APIC behavior is required. This fixes some real-world BIOSes.

Notice that presence of the CPUID_APIC flag implies that apic_state is non-NULL.

[1] Intel 64 and IA-32 Architectures Software Developer's Manual, Vol. 3A:
System Programming Guide, Part 1

---
 hw/i386/x86.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
hw/i386/x86.c +5 -5
index 2b6291ad8d..a753d1aeca 100644
@@ -516,10 +516,10 @@ static void x86_nmi(NMIState *n, int cpu_index, Error **errp)
     CPU_FOREACH(cs) {
         X86CPU *cpu = X86_CPU(cs);
 
-        if (!cpu->apic_state) {
-            cpu_interrupt(cs, CPU_INTERRUPT_NMI);
-        } else {
+        if (cpu->env.features[FEAT_1_EDX] & CPUID_APIC) {
             apic_deliver_nmi(cpu->apic_state);
+        } else {
+            cpu_interrupt(cs, CPU_INTERRUPT_NMI);
         }
     }
 }
@@ -551,8 +551,8 @@ static void pic_irq_request(void *opaque, int irq, int level)
     X86CPU *cpu = X86_CPU(cs);
 
     trace_x86_pic_interrupt(irq, level);
-    if (cpu->apic_state && !kvm_irqchip_in_kernel() &&
-        !whpx_apic_in_platform()) {
+    if ((cpu->env.features[FEAT_1_EDX] & CPUID_APIC) &&
+        !kvm_irqchip_in_kernel() && !whpx_apic_in_platform()) {
         CPU_FOREACH(cs) {
             cpu = X86_CPU(cs);
             if (apic_accept_pic_intr(cpu->apic_state)) {
Signature · 1 line
2.43.0
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] hw/i386/x86: Fix PIC interrupt handling if APIC globally disabled

Quoted 27 lines · Bernhard Beschow

QEMU populates the apic_state attribute of x86 CPUs if supported by real hardware. Even when the APIC is globally disabled by a guest, this attribute stays populated. This means that the APIC code paths are still used in this case. However, chapter 10.4.3 of [1] requires that:

When IA32_APIC_BASE[11] is 0, the processor is functionally equivalent to an IA-32 processor without an on-chip APIC. The CPUID feature flag for the APIC [...] is also set to 0.

Fix this by checking the APIC feature flag rather than apic_state when deciding whether PIC or APIC behavior is required. This fixes some real-world BIOSes.

Notice that presence of the CPUID_APIC flag implies that apic_state is non-NULL.

[1] Intel 64 and IA-32 Architectures Software Developer's Manual, Vol. 3A:
System Programming Guide, Part 1

---
 hw/i386/x86.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
hw/i386/x86.c +0 -0
index 2b6291ad8d..a753d1aeca 100644

hw/i386/x86.c static void x86_nmi(NMIState *n, int cpu_index, Error **errp) · line 519

Quoted 7 lines
hw/i386/x86.c +1 -3
@@ -516,10 +516,10 @@ static void x86_nmi(NMIState *n, int cpu_index, Error **errp)
     CPU_FOREACH(cs) {
         X86CPU *cpu = X86_CPU(cs);
 
-        if (!cpu->apic_state) {
-            cpu_interrupt(cs, CPU_INTERRUPT_NMI);
-        } else {
+        if (cpu->env.features[FEAT_1_EDX] & CPUID_APIC) {

You could assert the relationship between the feature and ->apic_state with:

g_assert(cpu->apic_state)

But probably unnecessary in the grand scheme of things. Anyway:

Signature · 2 lines
Alex Bennée
Virtualisation Tech Lead @ Linaro
#3 lore raw

Re: [PATCH 1/2] hw/i386/x86: Fix PIC interrupt handling if APIC globally disabled

Am 3. Januar 2024 09:12:24 UTC schrieb "Alex Bennée" <[email protected]>:

Quoted 39 lines
Quoted 27 lines · Bernhard Beschow

QEMU populates the apic_state attribute of x86 CPUs if supported by real hardware. Even when the APIC is globally disabled by a guest, this attribute stays populated. This means that the APIC code paths are still used in this case. However, chapter 10.4.3 of [1] requires that:

When IA32_APIC_BASE[11] is 0, the processor is functionally equivalent to an IA-32 processor without an on-chip APIC. The CPUID feature flag for the APIC [...] is also set to 0.

Fix this by checking the APIC feature flag rather than apic_state when deciding whether PIC or APIC behavior is required. This fixes some real-world BIOSes.

Notice that presence of the CPUID_APIC flag implies that apic_state is non-NULL.

[1] Intel 64 and IA-32 Architectures Software Developer's Manual, Vol. 3A:
System Programming Guide, Part 1

---
 hw/i386/x86.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
hw/i386/x86.c +0 -0
index 2b6291ad8d..a753d1aeca 100644

hw/i386/x86.c static void x86_nmi(NMIState *n, int cpu_index, Error **errp) · line 519

Quoted 7 lines
hw/i386/x86.c +1 -3
@@ -516,10 +516,10 @@ static void x86_nmi(NMIState *n, int cpu_index, Error **errp)
     CPU_FOREACH(cs) {
         X86CPU *cpu = X86_CPU(cs);
 
-        if (!cpu->apic_state) {
-            cpu_interrupt(cs, CPU_INTERRUPT_NMI);
-        } else {
+        if (cpu->env.features[FEAT_1_EDX] & CPUID_APIC) {

You could assert the relationship between the feature and ->apic_state with:

Quoted 3 lines

g_assert(cpu->apic_state)

But probably unnecessary in the grand scheme of things.

I like the idea so I'll respin.

Thanks,
Bernhard

Quoted 4 lines

Anyway:

#4 lore raw

PATCH 2/2 target/i386/cpu: Fix small typo in comment

In reply to Bernhard Beschow
---
 target/i386/cpu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
target/i386/cpu.c +1 -1
index 95d5f16cd5..2ae271e9a1 100644
@@ -2179,7 +2179,7 @@ static const CPUCaches epyc_genoa_cache_info = {
  *  Conceal VM entries from PT
  *  Enable ENCLS exiting
  *  Mode-based execute control (XS/XU)
- s  TSC scaling (Skylake Server and newer)
+ *  TSC scaling (Skylake Server and newer)
  *  GPA translation for PT (IceLake and newer)
  *  User wait and pause
  *  ENCLV exiting
Signature · 1 line
2.43.0
Reviewed-by on v1 2/2
#5 lore raw

Re: [PATCH 2/2] target/i386/cpu: Fix small typo in comment

Quoted 10 lines · Bernhard Beschow
---
 target/i386/cpu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
target/i386/cpu.c +0 -0
index 95d5f16cd5..2ae271e9a1 100644

target/i386/cpu.c static const CPUCaches epyc_genoa_cache_info = { · line 2185

Quoted 8 lines
target/i386/cpu.c +1 -1
@@ -2179,7 +2179,7 @@ static const CPUCaches epyc_genoa_cache_info = {
  *  Conceal VM entries from PT
  *  Enable ENCLS exiting
  *  Mode-based execute control (XS/XU)
- s  TSC scaling (Skylake Server and newer)
+ *  TSC scaling (Skylake Server and newer)
  *  GPA translation for PT (IceLake and newer)
  *  User wait and pause
  *  ENCLV exiting
Signature · 2 lines
Alex Bennée
Virtualisation Tech Lead @ Linaro