Skip to contentSkip to conversation

i386/monitor: Fix page table walking issue for LA57 enabled guest

PATCH Superseded Yuan Yao 3 participants 5 messages Jun 9, 2022 – Jun 24, 2022

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Yuan Yao

author
#0 lore raw

PATCH 1/1 i386/monitor: Fix page table walking issue for LA57 enabled guest

Don't skip next leve page table for pdpe/pde when the PG_PRESENT_MASK is set.

This fixs the issue that no mapping information was collected from "info mem" for guest with LA57 enabled.

---
 target/i386/monitor.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
target/i386/monitor.c +2 -2
index 8e4b4d600c..3339550bbe 100644
@@ -489,7 +489,7 @@ static void mem_info_la57(Monitor *mon, CPUArchState *env)
                 cpu_physical_memory_read(pdp_addr + l2 * 8, &pdpe, 8);
                 pdpe = le64_to_cpu(pdpe);
                 end = (l0 << 48) + (l1 << 39) + (l2 << 30);
-                if (pdpe & PG_PRESENT_MASK) {
+                if (!(pdpe & PG_PRESENT_MASK)) {
                     prot = 0;
                     mem_print(mon, env, &start, &last_prot, end, prot);
                     continue;
@@ -508,7 +508,7 @@ static void mem_info_la57(Monitor *mon, CPUArchState *env)
                     cpu_physical_memory_read(pd_addr + l3 * 8, &pde, 8);
                     pde = le64_to_cpu(pde);
                     end = (l0 << 48) + (l1 << 39) + (l2 << 30) + (l3 << 21);
-                    if (pde & PG_PRESENT_MASK) {
+                    if (!(pde & PG_PRESENT_MASK)) {
                         prot = 0;
                         mem_print(mon, env, &start, &last_prot, end, prot);
                         continue;

base-commit: 6d940eff4734bcb40b1a25f62d7cec5a396f994a

Signature · 1 line
2.27.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 14 lines

-----Original Message-----
From: Qemu-devel <qemu-devel- [email protected]> On Behalf Of Yuan Yao
Sent: Thursday, June 9, 2022 4:35 PM
To: Paolo Bonzini <[email protected]>; Philippe Mathieu-Daudé <[email protected]>; Dr. David Alan Gilbert <[email protected]>; Zhong, Yang <[email protected]>; Connor Kuehl <[email protected]>
Cc: [email protected]; Yao, Yuan <[email protected]>; Yamahata, Isaku <[email protected]>
Subject: [PATCH 1/1] i386/monitor: Fix page table walking issue for LA57 enabled guest

Quoted 1 line

Don't skip next leve page table for pdpe/pde when the PG_PRESENT_MASK

S/leve/level

Quoted 6 lines

is set.

This fixs the issue that no mapping information was collected from "info mem" for guest with LA57 enabled.

LGTM.
It should same with the excp_helper.c/mmu_translate() la57 implementation.

Add Eric and Markus for double check.

Thanks
Chen

Quoted 33 lines
---
 target/i386/monitor.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
target/i386/monitor.c index renamed from target/i386/monitor.c +0 -0

8e4b4d600c..3339550bbe 100644

target/i386/monitor.c +1 -1
@@ -489,7 +489,7 @@ static void mem_info_la57(Monitor *mon,
 CPUArchState *env)
                 cpu_physical_memory_read(pdp_addr + l2 * 8, &pdpe, 8);
                 pdpe = le64_to_cpu(pdpe);
                 end = (l0 << 48) + (l1 << 39) + (l2 << 30);
-                if (pdpe & PG_PRESENT_MASK) {
+                if (!(pdpe & PG_PRESENT_MASK)) {
                     prot = 0;
                     mem_print(mon, env, &start, &last_prot, end, prot);
                     continue;
target/i386/monitor.c +1 -1
@@ -508,7 +508,7 @@ static void mem_info_la57(Monitor *mon,
 CPUArchState *env)
                     cpu_physical_memory_read(pd_addr + l3 * 8, &pde, 8);
                     pde = le64_to_cpu(pde);
                     end = (l0 << 48) + (l1 << 39) + (l2 << 30) + (l3 << 21);
-                    if (pde & PG_PRESENT_MASK) {
+                    if (!(pde & PG_PRESENT_MASK)) {
                         prot = 0;
                         mem_print(mon, env, &start, &last_prot, end, prot);
                         continue;

base-commit: 6d940eff4734bcb40b1a25f62d7cec5a396f994a --
2.27.0

author
#2 lore raw
Quoted 23 lines

-----Original Message-----
From: Zhang, Chen <[email protected]>
Sent: Friday, June 24, 2022 10:04
To: Yao, Yuan <[email protected]>; Paolo Bonzini <[email protected]>; Philippe Mathieu-Daudé <[email protected]>; Dr. David Alan Gilbert <[email protected]>; Zhong, Yang <[email protected]>; Connor Kuehl <[email protected]>; Eric Blake <[email protected]>; Markus Armbruster <[email protected]>
Cc: [email protected]; Yao, Yuan <[email protected]>; Yamahata, Isaku <[email protected]>
Subject: RE: [PATCH 1/1] i386/monitor: Fix page table walking issue for LA57 enabled guest

Quoted 12 lines

-----Original Message-----
From: Qemu-devel <qemu-devel- [email protected]> On Behalf Of Yuan Yao
Sent: Thursday, June 9, 2022 4:35 PM
To: Paolo Bonzini <[email protected]>; Philippe Mathieu-Daudé <[email protected]>; Dr. David Alan Gilbert <[email protected]>; Zhong, Yang <[email protected]>; Connor Kuehl <[email protected]>
Cc: [email protected]; Yao, Yuan <[email protected]>; Yamahata, Isaku <[email protected]>
Subject: [PATCH 1/1] i386/monitor: Fix page table walking issue for LA57 enabled guest

Quoted 3 lines
Quoted 1 line

Don't skip next leve page table for pdpe/pde when the PG_PRESENT_MASK

S/leve/level

Sorry for the typo.

Quoted 6 lines
Quoted 6 lines

is set.

This fixs the issue that no mapping information was collected from "info mem" for guest with LA57 enabled.

Quoted 7 lines
Quoted 1 line

LGTM.
It should same with the excp_helper.c/mmu_translate() la57 implementation.

Add Eric and Markus for double check.

Thanks for your comments.

Quoted 37 lines

Thanks
Chen

Quoted 33 lines
---
 target/i386/monitor.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
target/i386/monitor.c index renamed from target/i386/monitor.c +0 -0

8e4b4d600c..3339550bbe 100644

target/i386/monitor.c +1 -1
@@ -489,7 +489,7 @@ static void mem_info_la57(Monitor *mon,
 CPUArchState *env)
                 cpu_physical_memory_read(pdp_addr + l2 * 8, &pdpe, 8);
                 pdpe = le64_to_cpu(pdpe);
                 end = (l0 << 48) + (l1 << 39) + (l2 << 30);
-                if (pdpe & PG_PRESENT_MASK) {
+                if (!(pdpe & PG_PRESENT_MASK)) {
                     prot = 0;
                     mem_print(mon, env, &start, &last_prot, end, prot);
                     continue;
target/i386/monitor.c +1 -1
@@ -508,7 +508,7 @@ static void mem_info_la57(Monitor *mon,
 CPUArchState *env)
                     cpu_physical_memory_read(pd_addr + l3 * 8, &pde, 8);
                     pde = le64_to_cpu(pde);
                     end = (l0 << 48) + (l1 << 39) + (l2 << 30) + (l3 << 21);
-                    if (pde & PG_PRESENT_MASK) {
+                    if (!(pde & PG_PRESENT_MASK)) {
                         prot = 0;
                         mem_print(mon, env, &start, &last_prot, end, prot);
                         continue;

base-commit: 6d940eff4734bcb40b1a25f62d7cec5a396f994a --
2.27.0

#3 lore raw
In reply to Yuan Yao
Quoted 1 line · Yuan Yao

Don't skip next leve page table for pdpe/pde when the

level

Quoted 3 lines

PG_PRESENT_MASK is set.

This fixs the issue that no mapping information was

fixes

Quoted 3 lines

collected from "info mem" for guest with LA57 enabled.

Should we add

Fixes 6c7c3c21f95dd9af8a0691c0dd29b07247984122

?

Quoted 27 lines
---
 target/i386/monitor.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
target/i386/monitor.c +2 -2
index 8e4b4d600c..3339550bbe 100644
@@ -489,7 +489,7 @@ static void mem_info_la57(Monitor *mon, CPUArchState *env)
                 cpu_physical_memory_read(pdp_addr + l2 * 8, &pdpe, 8);
                 pdpe = le64_to_cpu(pdpe);
                 end = (l0 << 48) + (l1 << 39) + (l2 << 30);
-                if (pdpe & PG_PRESENT_MASK) {
+                if (!(pdpe & PG_PRESENT_MASK)) {
                     prot = 0;
                     mem_print(mon, env, &start, &last_prot, end, prot);
                     continue;
@@ -508,7 +508,7 @@ static void mem_info_la57(Monitor *mon, CPUArchState *env)
                     cpu_physical_memory_read(pd_addr + l3 * 8, &pde, 8);
                     pde = le64_to_cpu(pde);
                     end = (l0 << 48) + (l1 << 39) + (l2 << 30) + (l3 << 21);
-                    if (pde & PG_PRESENT_MASK) {
+                    if (!(pde & PG_PRESENT_MASK)) {
                         prot = 0;
                         mem_print(mon, env, &start, &last_prot, end, prot);
                         continue;
Quoted 1 line

base-commit: 6d940eff4734bcb40b1a25f62d7cec5a396f994a

The commit message talks about not skipping something when the flag is set. However, the patch *flips* the sense of conditions, which means were *also* changing behavior when the flag is unset. How?

author
#4 lore raw
Quoted 9 lines

-----Original Message-----
From: Markus Armbruster <[email protected]>
Sent: Friday, June 24, 2022 13:35
To: Yao, Yuan <[email protected]>
Cc: Paolo Bonzini <[email protected]>; Philippe Mathieu-Daudé <[email protected]>; Dr. David Alan Gilbert <[email protected]>; Zhong, Yang <[email protected]>; Connor Kuehl <[email protected]>; [email protected]; Yamahata, Isaku <[email protected]>
Subject: Re: [PATCH 1/1] i386/monitor: Fix page table walking issue for LA57 enabled guest

Quoted 5 lines
Quoted 1 line · Yuan Yao

Don't skip next leve page table for pdpe/pde when the

level

Sorry for my typo.

Quoted 7 lines
Quoted 4 lines

PG_PRESENT_MASK is set.

This fixs the issue that no mapping information was

fixes

Quoted 7 lines
Quoted 3 lines

collected from "info mem" for guest with LA57 enabled.

Should we add

Fixes 6c7c3c21f95dd9af8a0691c0dd29b07247984122

Yes, I will add this next time, thanks.

Quoted 30 lines

?

Quoted 27 lines
---
 target/i386/monitor.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
target/i386/monitor.c +2 -2
index 8e4b4d600c..3339550bbe 100644
@@ -489,7 +489,7 @@ static void mem_info_la57(Monitor *mon, CPUArchState *env)
                 cpu_physical_memory_read(pdp_addr + l2 * 8, &pdpe, 8);
                 pdpe = le64_to_cpu(pdpe);
                 end = (l0 << 48) + (l1 << 39) + (l2 << 30);
-                if (pdpe & PG_PRESENT_MASK) {
+                if (!(pdpe & PG_PRESENT_MASK)) {
                     prot = 0;
                     mem_print(mon, env, &start, &last_prot, end, prot);
                     continue;
@@ -508,7 +508,7 @@ static void mem_info_la57(Monitor *mon, CPUArchState *env)
                     cpu_physical_memory_read(pd_addr + l3 * 8, &pde, 8);
                     pde = le64_to_cpu(pde);
                     end = (l0 << 48) + (l1 << 39) + (l2 << 30) + (l3 << 21);
-                    if (pde & PG_PRESENT_MASK) {
+                    if (!(pde & PG_PRESENT_MASK)) {
                         prot = 0;
                         mem_print(mon, env, &start, &last_prot, end, prot);
                         continue;
Quoted 5 lines
Quoted 1 line

base-commit: 6d940eff4734bcb40b1a25f62d7cec5a396f994a

The commit message talks about not skipping something when the flag is set. However, the patch *flips* the sense of conditions, which means were *also* changing behavior when the flag is unset. How?

Yes, this also changes the behavior when the flag is unset, because the original code does wrong for both set and unset , flips the checking condition bring all of them to right behavior. I think I can add these to the commit message in v2.