Skip to contentSkip to conversation

Correction of the TLB handling of the OpenRISC target

PATCH In review Sebastian Macke 3 participants 5 messages Oct 2, 2013 – Oct 3, 2013

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Correction of the TLB handling of the OpenRISC target

Hi,

this patch corrects two problems for the OpenRISC Target in QEMU. The first one corrects one obvious bug
concerning the handling of page faults while reading from a page. The second part removes a non-conforming behavior for the first page of the memory.

I have tested this patch with the newest Linux kernel and compared the output with or1ksim.

Sebastian

Quoted 1 line

From 4491bae7109e2b4de5a8de8a7e4b08d1f19ac70e Mon Sep 17 00:00:00 2001

From: Sebastian Macke <[email protected]>
Date: Tue, 1 Oct 2013 21:39:38 -0700
Subject: [PATCH] Correction of the TLB handling of the OpenRISC target

This patch correct two problems. The first one corrects one obvious bug concerning the handling of page faults while reading from a page. The second part removes a non-conforming behavior for the first page of the memory.

I have tested this patch with the newest Linux kernel and compared the output with or1ksim.

---
 target-openrisc/mmu.c |    9 +--------
 1 files changed, 1 insertions(+), 8 deletions(-)
target-openrisc/mmu.c +1 -8
index 57f5616..22d7cbe 100644
@@ -102,7 +102,7 @@ int cpu_openrisc_get_phys_data(OpenRISCCPU *cpu,
         }
     }
 
-    if ((rw & 0) && ((right & PAGE_READ) == 0)) {
+    if (!(rw & 1) && ((right & PAGE_READ) == 0)) {
         return TLBRET_BADADDR;
     }
     if ((rw & 1) && ((right & PAGE_WRITE) == 0)) {
@@ -122,13 +122,6 @@ static int cpu_openrisc_get_phys_addr(OpenRISCCPU *cpu,
 {
     int ret = TLBRET_MATCH;
 
-    /* [0x0000--0x2000]: unmapped */
-    if (address < 0x2000 && (cpu->env.sr & SR_SM)) {
-        *physical = address;
-        *prot = PAGE_READ | PAGE_WRITE;
-        return ret;
-    }
-
     if (rw == 2) {    /* ITLB */
        *physical = 0;
         ret = cpu->env.tlb->cpu_openrisc_map_address_code(cpu, physical,
Signature · 1 line
1.7.9
#1 lore raw

Hi Sebastian,

Quoted 5 lines · Sebastian Macke

Hi,

this patch corrects two problems for the OpenRISC Target in QEMU. The first one corrects one obvious bug
concerning the handling of page faults while reading from a page.

@@ -102,7 +102,7 @@ int cpu_openrisc_get_phys_data(OpenRISCCPU *cpu,

         }
     }

- if ((rw & 0) && ((right & PAGE_READ) == 0)) {

+    if (!(rw & 1) && ((right & PAGE_READ) == 0)) {
         return TLBRET_BADADDR;
     }
     if ((rw & 1) && ((right & PAGE_WRITE) == 0)) {

They are just two type of one code...

Quoted 2 lines

The second
part removes a non-conforming behavior for the first page of the memory.

@@ -122,13 +122,6 @@ static int cpu_openrisc_get_phys_addr(OpenRISCCPU *cpu,

 {
     int ret = TLBRET_MATCH;

- /* [0x0000--0x2000]: unmapped */
- if (address < 0x2000 && (cpu->env.sr & SR_SM)) {

-        *physical = address;
-        *prot = PAGE_READ | PAGE_WRITE;
-        return ret;

- }
-

May you please explain more about why the first page is non-conforming? The Arch manual told me 0x0000--0x2000 is unmapped.

Quoted 3 lines

I have tested this patch with the newest Linux kernel and compared the output with or1ksim.

May you please upload a newest Linux kernel to somewhere?

Quoted 8 lines

Sebastian

_______________________________________________
OpenRISC mailing list
[email protected]
http://lists.openrisc.net/listinfo/openrisc

Regards,
Jia

#2 lore raw
Quoted 21 lines · Jia Liu

Hi Sebastian,

Quoted 3 lines · Sebastian Macke

Hi,

this patch corrects two problems for the OpenRISC Target in QEMU. The

first

Quoted 2 lines

one corrects one obvious bug
concerning the handling of page faults while reading from a page.

target-openrisc/mmu.c +1 -1
@@ -102,7 +102,7 @@ int cpu_openrisc_get_phys_data(OpenRISCCPU *cpu,
         }
     }
 
-    if ((rw & 0) && ((right & PAGE_READ) == 0)) {
+    if (!(rw & 1) && ((right & PAGE_READ) == 0)) {
         return TLBRET_BADADDR;
     }
     if ((rw & 1) && ((right & PAGE_WRITE) == 0)) {
Quoted 2 lines

They are just two type of one code...

No, (rw & 0) always evaluates to 0.

Quoted 17 lines
Quoted 3 lines

The second
part removes a non-conforming behavior for the first page of the memory.

target-openrisc/mmu.c +0 -7
@@ -122,13 +122,6 @@ static int cpu_openrisc_get_phys_addr(OpenRISCCPU
 *cpu,
 {
     int ret = TLBRET_MATCH;
 
-    /* [0x0000--0x2000]: unmapped */
-    if (address < 0x2000 && (cpu->env.sr & SR_SM)) {
-        *physical = address;
-        *prot = PAGE_READ | PAGE_WRITE;
-        return ret;
-    }
-
 
Quoted 3 lines

May you please explain more about why the first page is non-conforming? The Arch manual told me 0x0000--0x2000 is unmapped.

It shows an example where *software* leaves 0x0000 - 0x2000 unmapped, the hardware should still allow for this area to be mapped.

Stefan

Reviewed-by
#3 lore raw
Quoted 46 lines · Stefan Kristiansson
Quoted 22 lines · Jia Liu

Hi Sebastian,

Quoted 6 lines · Sebastian Macke

Hi,

this patch corrects two problems for the OpenRISC Target in QEMU. The first
one corrects one obvious bug
concerning the handling of page faults while reading from a page.

target-openrisc/mmu.c +1 -1
@@ -102,7 +102,7 @@ int cpu_openrisc_get_phys_data(OpenRISCCPU *cpu,
         }
     }
 
-    if ((rw & 0) && ((right & PAGE_READ) == 0)) {
+    if (!(rw & 1) && ((right & PAGE_READ) == 0)) {
         return TLBRET_BADADDR;
     }
     if ((rw & 1) && ((right & PAGE_WRITE) == 0)) {
Quoted 1 line

They are just two type of one code...

No, (rw & 0) always evaluates to 0.

Quoted 18 lines
Quoted 4 lines

The second
part removes a non-conforming behavior for the first page of the memory.

target-openrisc/mmu.c +0 -7
@@ -122,13 +122,6 @@ static int cpu_openrisc_get_phys_addr(OpenRISCCPU
 *cpu,
 {
     int ret = TLBRET_MATCH;
 
-    /* [0x0000--0x2000]: unmapped */
-    if (address < 0x2000 && (cpu->env.sr & SR_SM)) {
-        *physical = address;
-        *prot = PAGE_READ | PAGE_WRITE;
-        return ret;
-    }
-
 
Quoted 6 lines
Quoted 2 lines

May you please explain more about why the first page is non-conforming? The Arch manual told me 0x0000--0x2000 is unmapped.

It shows an example where *software* leaves 0x0000 - 0x2000 unmapped, the hardware should still allow for this area to be mapped.

OK, thank you. I will send a PULL Request soon.

Quoted 2 lines

Stefan

Regards,
Jia

#4 lore raw
In reply to Jia Liu

Hi Jia,

Quoted 18 lines · Jia Liu

Hi Sebastian,

Quoted 5 lines · Sebastian Macke

Hi,

this patch corrects two problems for the OpenRISC Target in QEMU. The first one corrects one obvious bug
concerning the handling of page faults while reading from a page.

target-openrisc/mmu.c +1 -1
@@ -102,7 +102,7 @@ int cpu_openrisc_get_phys_data(OpenRISCCPU *cpu,
          }
      }
 
-    if ((rw & 0) && ((right & PAGE_READ) == 0)) {
+    if (!(rw & 1) && ((right & PAGE_READ) == 0)) {
          return TLBRET_BADADDR;
      }
      if ((rw & 1) && ((right & PAGE_WRITE) == 0)) {
Quoted 1 line

They are just two type of one code...

No the result of (rw&0) is always zero and therefore a logic false. The whole comparison will therefore never be executed.

Quoted 14 lines
Quoted 2 lines

The second
part removes a non-conforming behavior for the first page of the memory.

target-openrisc/mmu.c +0 -7
@@ -122,13 +122,6 @@ static int cpu_openrisc_get_phys_addr(OpenRISCCPU *cpu,
  {
      int ret = TLBRET_MATCH;
 
-    /* [0x0000--0x2000]: unmapped */
-    if (address < 0x2000 && (cpu->env.sr & SR_SM)) {
-        *physical = address;
-        *prot = PAGE_READ | PAGE_WRITE;
-        return ret;
-    }
-
 
Quoted 2 lines

May you please explain more about why the first page is non-conforming? The Arch manual told me 0x0000--0x2000 is unmapped.

There is a good reason why this page should be a mapped one. Think of an accidental

int *a = NULL;
....
*a = 0;

In the current implementation this would work in the supervisor mode and could never be catched. The Linux kernel handles this page as a mapped one without write access which is correct.

The specification is not clear about this mapping region because it is not consistent with the rest of the specification and of the implementation in or1ksim and the Linux kernel.

Quoted 3 lines
Quoted 2 lines

I have tested this patch with the newest Linux kernel and compared the output with or1ksim.

May you please upload a newest Linux kernel to somewhere?

Try http://www.simulationcorner.net/vmlinux
start it with
or32-system-qemu -m 64 -kernel vmlinux
In / there are two example executables which accidently dereferences a null pointer in the Linux kernel. The correct behavior should be a kernel oops.

Sebastian

Quoted 9 lines
Quoted 7 lines

Sebastian

_______________________________________________
OpenRISC mailing list
[email protected]
http://lists.openrisc.net/listinfo/openrisc

Regards,
Jia