Skip to contentSkip to conversation

hw/intc: Fix incorrect calculation of core in liointc_read() and liointc_write()

PATCH Applied AlexChen 4 participants 10 messages Nov 3, 2020 – Nov 5, 2020

lore
Share & tools
lore

AlexChen

author
#0 lore raw

PATCH 1/1 hw/intc: Fix incorrect calculation of core in liointc_read() and liointc_write()

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

---
 hw/intc/loongson_liointc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
hw/intc/loongson_liointc.c +2 -2
index 30fb375b72..fbbfb57ee9 100644
@@ -130,7 +130,7 @@ liointc_read(void *opaque, hwaddr addr, unsigned int size)
 
     if (addr >= R_PERCORE_ISR(0) &&
         addr < R_PERCORE_ISR(NUM_CORES)) {
-        int core = (addr - R_PERCORE_ISR(0)) / 4;
+        int core = (addr - R_PERCORE_ISR(0)) / 8;
         r = p->per_core_isr[core];
         goto out;
     }
@@ -173,7 +173,7 @@ liointc_write(void *opaque, hwaddr addr,
 
     if (addr >= R_PERCORE_ISR(0) &&
         addr < R_PERCORE_ISR(NUM_CORES)) {
-        int core = (addr - R_PERCORE_ISR(0)) / 4;
+        int core = (addr - R_PERCORE_ISR(0)) / 8;
         p->per_core_isr[core] = value;
         goto out;
     }
Signature · 1 line
2.19.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 5 lines · AlexChen

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

Hi Alex

Thanks!

That was my fault.. Per Core ISA is rarely used by kernel..

Quoted 1 line

Btw:
How can you discover this by robot?
Huawei owns real artifical intelligence technology lol :-)

- Jiaxun

Quoted 27 lines
---
  hw/intc/loongson_liointc.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
hw/intc/loongson_liointc.c +2 -2
index 30fb375b72..fbbfb57ee9 100644
@@ -130,7 +130,7 @@ liointc_read(void *opaque, hwaddr addr, unsigned int size)
 
      if (addr >= R_PERCORE_ISR(0) &&
          addr < R_PERCORE_ISR(NUM_CORES)) {
-        int core = (addr - R_PERCORE_ISR(0)) / 4;
+        int core = (addr - R_PERCORE_ISR(0)) / 8;
          r = p->per_core_isr[core];
          goto out;
      }
@@ -173,7 +173,7 @@ liointc_write(void *opaque, hwaddr addr,
 
      if (addr >= R_PERCORE_ISR(0) &&
          addr < R_PERCORE_ISR(NUM_CORES)) {
-        int core = (addr - R_PERCORE_ISR(0)) / 4;
+        int core = (addr - R_PERCORE_ISR(0)) / 8;
          p->per_core_isr[core] = value;
          goto out;
      }
author
#2 lore raw
Quoted 12 lines · Jiaxun Yang
Quoted 5 lines · AlexChen

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

Hi Alex

Thanks!

Quoted 9 lines

That was my fault.. Per Core ISA is rarely used by kernel..

Quoted 1 line

Btw:
How can you discover this by robot?
Huawei owns real artifical intelligence technology lol :-)

Thanks for your review.
EulerRobot is a virtualization software quality automation project that integrates some tools and test suites such as gcc/clang make test, qemu ut, qtest, coccinelle scripts and avocado-vt.
The code checking tool found there was a potential array out of bounds at 'r = p->per_core_isr[core]', since 'core' may be 7 which is bigger than 'per_core_isr' array size 3.
So we found this bug.

Thanks,
Alex

Quoted 30 lines

- Jiaxun

Quoted 19 lines
---
  hw/intc/loongson_liointc.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
hw/intc/loongson_liointc.c +1 -1
index 30fb375b72..fbbfb57ee9 100644
@@ -130,7 +130,7 @@ liointc_read(void *opaque, hwaddr addr, unsigned int size)
 
      if (addr >= R_PERCORE_ISR(0) &&
          addr < R_PERCORE_ISR(NUM_CORES)) {
-        int core = (addr - R_PERCORE_ISR(0)) / 4;
+        int core = (addr - R_PERCORE_ISR(0)) / 8;
          r = p->per_core_isr[core];
          goto out;
      }

hw/intc/loongson_liointc.c liointc_write(void *opaque, hwaddr addr, · line 179

Quoted 8 lines
hw/intc/loongson_liointc.c +1 -1
@@ -173,7 +173,7 @@ liointc_write(void *opaque, hwaddr addr,
 
      if (addr >= R_PERCORE_ISR(0) &&
          addr < R_PERCORE_ISR(NUM_CORES)) {
-        int core = (addr - R_PERCORE_ISR(0)) / 4;
+        int core = (addr - R_PERCORE_ISR(0)) / 8;
          p->per_core_isr[core] = value;
          goto out;
      }

.

#3 lore raw
Quoted 9 lines · AlexChen
Quoted 8 lines · Jiaxun Yang

閸︼拷 2020/11/3 17:32, AlexChen 閸愭瑩浜�:

Quoted 5 lines

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

Quoted 5 lines
Quoted 5 lines

Hi Alex

Thanks!

That was my fault.. Per Core ISA is rarely used by kernel..

No board in QEMU use this device. So we are safe =)

Quoted 53 lines
Quoted 8 lines
Quoted 1 line

Btw:
How can you discover this by robot?
Huawei owns real artifical intelligence technology lol :-閿涳拷

Thanks for your review.
EulerRobot is a virtualization software quality automation project that integrates some tools and test suites such as gcc/clang make test, qemu ut, qtest, coccinelle scripts and avocado-vt.
The code checking tool found there was a potential array out of bounds at 'r = p->per_core_isr[core]', since 'core' may be 7 which is bigger than 'per_core_isr' array size 3.
So we found this bug.

Thanks,
Alex

Quoted 30 lines

- Jiaxun

Quoted 19 lines
---
  hw/intc/loongson_liointc.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
hw/intc/loongson_liointc.c +1 -1
index 30fb375b72..fbbfb57ee9 100644
@@ -130,7 +130,7 @@ liointc_read(void *opaque, hwaddr addr, unsigned int size)
 
      if (addr >= R_PERCORE_ISR(0) &&
          addr < R_PERCORE_ISR(NUM_CORES)) {
-        int core = (addr - R_PERCORE_ISR(0)) / 4;
+        int core = (addr - R_PERCORE_ISR(0)) / 8;
          r = p->per_core_isr[core];
          goto out;
      }

hw/intc/loongson_liointc.c liointc_write(void *opaque, hwaddr addr, · line 179

Quoted 8 lines
hw/intc/loongson_liointc.c +1 -1
@@ -173,7 +173,7 @@ liointc_write(void *opaque, hwaddr addr,
 
      if (addr >= R_PERCORE_ISR(0) &&
          addr < R_PERCORE_ISR(NUM_CORES)) {
-        int core = (addr - R_PERCORE_ISR(0)) / 4;
+        int core = (addr - R_PERCORE_ISR(0)) / 8;
          p->per_core_isr[core] = value;
          goto out;
      }

.

Reviewed-by on v1 1/1
#4 lore raw
In reply to AlexChen
Quoted 11 lines · AlexChen

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

---
 hw/intc/loongson_liointc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

For a model added in 2020, its code style is a bit disappointing (leading to that kind of hidden bugs). I'm even surprised it passed the review process.

Thanks for the fix.

#5 lore raw

于 2020年11月3日 GMT+08:00 下午8:28:27, "Philippe Mathieu-Daudé" <[email protected]> 写到:

Quoted 13 lines · AlexChen
Quoted 11 lines

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

---
 hw/intc/loongson_liointc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 7 lines

For a model added in 2020, its code style is a bit disappointing (leading to that kind of hidden bugs). I'm even surprised it passed the review process.

Thanks for the fix.

It was my proof of concept code.

Any suggestions on enhancement?
I'll have some free time afterwards so probably can do something.

Thanks.

-Jiaxun

#6 lore raw
Quoted 18 lines · Jiaxun Yang

于 2020年11月3日 GMT+08:00 下午8:28:27, "Philippe Mathieu-Daudé" <[email protected]> 写到:

Quoted 17 lines · AlexChen
Quoted 11 lines

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

---
 hw/intc/loongson_liointc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

For a model added in 2020, its code style is a bit disappointing (leading to that kind of hidden bugs). I'm even surprised it passed the review process.

Quoted 5 lines
Quoted 3 lines

Thanks for the fix.

It was my proof of concept code.

Don't worry Jiaxun, this comment is not to you, but to the QEMU community as a whole. We should have asked improvements during review, and explain what could be improve, what is not the best style but acceptable, and what is good.

Quoted 2 lines

Any suggestions on enhancement?
I'll have some free time afterwards so probably can do something.

It is a bit awkward to not comment on a patch (diff). Comment I'd have made:

- Add definition for 0x8 magic value in R_PERCORE_ISR()
- Replace R_PERCORE_ISR() macro my function
- Replace dead D() code by trace events
- Use a simple 32-bit implementation (pic_ops.impl.min/max = 4) to let the generic memory code deal with the 8-bit accesses to mapper[].

If you have time, today what would be more useful is to have tests for the Loongson-3 board.

You can see some examples with the Malta board in the repository:

$ git-grep malta tests/acceptance/

Regards,

Phil.

#7 lore raw

Hi, Philippe and Jiaxun,

Quoted 45 lines · Philippe Mathieu-Daudé
Quoted 18 lines · Jiaxun Yang

于 2020年11月3日 GMT+08:00 下午8:28:27, "Philippe Mathieu-Daudé" <[email protected]> 写到:

Quoted 17 lines · AlexChen
Quoted 11 lines

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

---
 hw/intc/loongson_liointc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

For a model added in 2020, its code style is a bit disappointing (leading to that kind of hidden bugs). I'm even surprised it passed the review process.

Quoted 5 lines
Quoted 3 lines

Thanks for the fix.

It was my proof of concept code.

Don't worry Jiaxun, this comment is not to you, but to the QEMU community as a whole. We should have asked improvements during review, and explain what could be improve, what is not the best style but acceptable, and what is good.

Quoted 2 lines

Any suggestions on enhancement?
I'll have some free time afterwards so probably can do something.

It is a bit awkward to not comment on a patch (diff). Comment I'd have made:

- Add definition for 0x8 magic value in R_PERCORE_ISR()
- Replace R_PERCORE_ISR() macro my function
- Replace dead D() code by trace events
- Use a simple 32-bit implementation (pic_ops.impl.min/max = 4) to let the generic memory code deal with the 8-bit accesses to mapper[].

Quoted 2 lines

If you have time, today what would be more useful is to have tests for the Loongson-3 board.

As you suggested, LOONGSON_LIOINTC will be used in loongson3_virt.c and it is defined in a .c file now, so this is a chance for me to rework liointc.

Huacai

Quoted 9 lines

You can see some examples with the Malta board in the repository:

$ git-grep malta tests/acceptance/

Regards,

Phil.

Signature · 1 line
Huacai Chen
#8 lore raw

Hi, Philippe,

Quoted 48 lines · chen huacai

Hi, Philippe and Jiaxun,

Quoted 44 lines · Philippe Mathieu-Daudé
Quoted 18 lines · Jiaxun Yang

于 2020年11月3日 GMT+08:00 下午8:28:27, "Philippe Mathieu-Daudé" <[email protected]> 写到:

Quoted 17 lines · AlexChen
Quoted 11 lines

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

---
 hw/intc/loongson_liointc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

For a model added in 2020, its code style is a bit disappointing (leading to that kind of hidden bugs). I'm even surprised it passed the review process.

Quoted 5 lines
Quoted 3 lines

Thanks for the fix.

It was my proof of concept code.

Don't worry Jiaxun, this comment is not to you, but to the QEMU community as a whole. We should have asked improvements during review, and explain what could be improve, what is not the best style but acceptable, and what is good.

Quoted 2 lines

Any suggestions on enhancement?
I'll have some free time afterwards so probably can do something.

It is a bit awkward to not comment on a patch (diff). Comment I'd have made:

- Add definition for 0x8 magic value in R_PERCORE_ISR()
- Replace R_PERCORE_ISR() macro my function
- Replace dead D() code by trace events
- Use a simple 32-bit implementation (pic_ops.impl.min/max = 4) to let the generic memory code deal with the 8-bit accesses to mapper[].

I have reworked, but I haven't take the last suggestion (Use a simple 32-bit implementation). Because the mapper[] are naturely accessed in 8-bit, if use 32-bit implementation, the data type of mapper[] should also be changed, which looks very strange.

Huacai

Quoted 21 lines
Quoted 3 lines

If you have time, today what would be more useful is to have tests for the Loongson-3 board.

As you suggested, LOONGSON_LIOINTC will be used in loongson3_virt.c and it is defined in a .c file now, so this is a chance for me to rework liointc.

Huacai

Quoted 9 lines

You can see some examples with the Malta board in the repository:

$ git-grep malta tests/acceptance/

Regards,

Phil.

--
Huacai Chen

Signature · 1 line
Huacai Chen
#9 lore raw
In reply to AlexChen
Quoted 11 lines · AlexChen

According to the loongson spec
(http://www.loongson.cn/uploadfile/cpu/3B1500/Loongson_3B1500_cpu_user_1.pdf)
and the macro definition(#define R_PERCORE_ISR(x) (0x40 + 0x8 * x)), we know that the ISR size of per CORE is 8, so here we need to divide (addr - R_PERCORE_ISR(0)) by 8, not 4.

---
 hw/intc/loongson_liointc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Thanks, applied to mips-next.