Skip to contentSkip to conversation

initrd: move initrd_start calculate within linear mapping range check

PATCH In review pierre Kuo 4 participants 10 messages Mar 14, 2019 – Apr 8, 2019 arm64

lore
Share & tools
lore

pierre Kuo

author
#0 lore raw

PATCH v2 1/1 initrd: move initrd_start calculate within linear mapping range check

in the previous case, initrd_start and initrd_end can be successfully returned either (base < memblock_start_of_DRAM()) or (base + size > memblock_start_of_DRAM() + linear_region_size).

That means even linear mapping range check fail for initrd_start and initrd_end, it still can get virtual address. Here we put initrd_start/initrd_end to be calculated only when linear mapping check pass.

Fixes c756c592e442 ("arm64: Utilize phys_initrd_start/phys_initrd_size")

---
Changes in v2:
- add Fixes tag

 arch/arm64/mm/init.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/arm64/mm/init.c +3 -5
index 7205a9085b4d..1adf418de685 100644
@@ -425,6 +425,9 @@ void __init arm64_memblock_init(void)
 			memblock_remove(base, size); /* clear MEMBLOCK_ flags */
 			memblock_add(base, size);
 			memblock_reserve(base, size);
+			/* the generic initrd code expects virtual addresses */
+			initrd_start = __phys_to_virt(phys_initrd_start);
+			initrd_end = initrd_start + phys_initrd_size;
 		}
 	}
 
@@ -450,11 +453,6 @@ void __init arm64_memblock_init(void)
 	 * pagetables with memblock.
 	 */
 	memblock_reserve(__pa_symbol(_text), _end - _text);
-	if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
-		/* the generic initrd code expects virtual addresses */
-		initrd_start = __phys_to_virt(phys_initrd_start);
-		initrd_end = initrd_start + phys_initrd_size;
-	}
 
 	early_init_fdt_scan_reserved_mem();
 
Signature · 1 line
2.17.1
author
#1 lore raw

Hi Steven, Catalin and Will:

Quoted 10 lines

in the previous case, initrd_start and initrd_end can be successfully returned either (base < memblock_start_of_DRAM()) or (base + size > memblock_start_of_DRAM() + linear_region_size).

That means even linear mapping range check fail for initrd_start and initrd_end, it still can get virtual address. Here we put initrd_start/initrd_end to be calculated only when linear mapping check pass.

Quoted 7 lines
Fixes c756c592e442 ("arm64: Utilize phys_initrd_start/phys_initrd_size")

---
Changes in v2:
- add Fixes tag

Would you mind to give some comment and suggestion for this v2 patch? If there is anything that are not noticed, please let me know.

Sincerely appreciate ur kind help.

maintainer
#2 lore raw
In reply to pierre Kuo
Quoted 10 lines · pierre Kuo

in the previous case, initrd_start and initrd_end can be successfully returned either (base < memblock_start_of_DRAM()) or (base + size > memblock_start_of_DRAM() + linear_region_size).

That means even linear mapping range check fail for initrd_start and initrd_end, it still can get virtual address. Here we put initrd_start/initrd_end to be calculated only when linear mapping check pass.

Fixes c756c592e442 ("arm64: Utilize phys_initrd_start/phys_initrd_size")

For future versions, please also cc the author of the original commit you are fixing.

Quoted 24 lines

---
Changes in v2:
- add Fixes tag

 arch/arm64/mm/init.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/arm64/mm/init.c +3 -0
index 7205a9085b4d..1adf418de685 100644
@@ -425,6 +425,9 @@ void __init arm64_memblock_init(void)
 			memblock_remove(base, size); /* clear MEMBLOCK_ flags */
 			memblock_add(base, size);
 			memblock_reserve(base, size);
+			/* the generic initrd code expects virtual addresses */
+			initrd_start = __phys_to_virt(phys_initrd_start);
+			initrd_end = initrd_start + phys_initrd_size;
 		}
 	}
 

arch/arm64/mm/init.c void __init arm64_memblock_init(void) · line 457

Quoted 8 lines
arch/arm64/mm/init.c +0 -5
@@ -450,11 +453,6 @@ void __init arm64_memblock_init(void)
 	 * pagetables with memblock.
 	 */
 	memblock_reserve(__pa_symbol(_text), _end - _text);
-	if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
-		/* the generic initrd code expects virtual addresses */
-		initrd_start = __phys_to_virt(phys_initrd_start);
-		initrd_end = initrd_start + phys_initrd_size;
-	}

With CONFIG_RANDOMIZE_BASE we can get a further change to memstart_addr after the place where you moved the initrd_{start,end} setting, which would result in a different value for __phys_to_virt(phys_initrd_start).

Signature · 1 line
Catalin
author
#3 lore raw

hi Catalin:

Quoted 10 lines · pierre Kuo
Quoted 9 lines

in the previous case, initrd_start and initrd_end can be successfully returned either (base < memblock_start_of_DRAM()) or (base + size > memblock_start_of_DRAM() + linear_region_size).

That means even linear mapping range check fail for initrd_start and initrd_end, it still can get virtual address. Here we put initrd_start/initrd_end to be calculated only when linear mapping check pass.

Quoted 4 lines
Quoted 1 line
Fixes c756c592e442 ("arm64: Utilize phys_initrd_start/phys_initrd_size")

For future versions, please also cc the author of the original commit you are fixing.

Got it and thanks for ur warm reminder ^^

Quoted 18 lines
Quoted 18 lines
 arch/arm64/mm/init.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/arm64/mm/init.c +3 -0
index 7205a9085b4d..1adf418de685 100644
@@ -425,6 +425,9 @@ void __init arm64_memblock_init(void)
                      memblock_remove(base, size); /* clear MEMBLOCK_ flags */
                      memblock_add(base, size);
                      memblock_reserve(base, size);
+                     /* the generic initrd code expects virtual addresses */
+                     initrd_start = __phys_to_virt(phys_initrd_start);
+                     initrd_end = initrd_start + phys_initrd_size;
              }
      }
 
Quoted 13 lines

arch/arm64/mm/init.c void __init arm64_memblock_init(void) · line 457

Quoted 9 lines
arch/arm64/mm/init.c +0 -5
@@ -450,11 +453,6 @@ void __init arm64_memblock_init(void)
       * pagetables with memblock.
       */
      memblock_reserve(__pa_symbol(_text), _end - _text);
-     if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
-             /* the generic initrd code expects virtual addresses */
-             initrd_start = __phys_to_virt(phys_initrd_start);
-             initrd_end = initrd_start + phys_initrd_size;
-     }

With CONFIG_RANDOMIZE_BASE we can get a further change to memstart_addr after the place where you moved the initrd_{start,end} setting, which would result in a different value for __phys_to_virt(phys_initrd_start).

I found what you mean, since __phys_to_virt will use PHYS_OFFSET (memstart_addr) for calculating.
How about moving CONFIG_RANDOMIZE_BASE part of code ahead of CONFIG_BLK_DEV_INITRD checking?

That means below (d) moving ahead of (c)
prvious:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)

now:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) { ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ----------------(b)
if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} --------------(d) if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

Appreciate your kind advice.

author
#4 lore raw

hi Catalin:

Quoted 16 lines
Quoted 3 lines

With CONFIG_RANDOMIZE_BASE we can get a further change to memstart_addr after the place where you moved the initrd_{start,end} setting, which would result in a different value for __phys_to_virt(phys_initrd_start).

I found what you mean, since __phys_to_virt will use PHYS_OFFSET (memstart_addr) for calculating.
How about moving CONFIG_RANDOMIZE_BASE part of code ahead of CONFIG_BLK_DEV_INITRD checking?

That means below (d) moving ahead of (c)
prvious:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)

Quoted 6 lines

now:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) { ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ----------------(b)
if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} --------------(d) if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

After tracing the kernel code,
is it even possible to move CONFIG_RANDOMIZE_BASE part of code ahead of memory_limit?

that mean the flow may look like below:
now2:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

in (b), the result of __pa_symbol(_text), memory_limit, memblock_mem_limit_remove_map and memblock_add
are not depended on memsart_addr.
So the now2 flow can grouping modification of memstart_address, put
(a) and (d) together.

Sincerely Appreciate your comment.

#5 lore raw

[+Ard in case I'm missing something]

Quoted 34 lines · pierre kuo
Quoted 16 lines
Quoted 3 lines

With CONFIG_RANDOMIZE_BASE we can get a further change to memstart_addr after the place where you moved the initrd_{start,end} setting, which would result in a different value for __phys_to_virt(phys_initrd_start).

I found what you mean, since __phys_to_virt will use PHYS_OFFSET (memstart_addr) for calculating.
How about moving CONFIG_RANDOMIZE_BASE part of code ahead of CONFIG_BLK_DEV_INITRD checking?

That means below (d) moving ahead of (c)
prvious:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)

Quoted 6 lines

now:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) { ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ----------------(b)
if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} --------------(d) if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

After tracing the kernel code,
is it even possible to move CONFIG_RANDOMIZE_BASE part of code ahead of memory_limit?

that mean the flow may look like below:
now2:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

Quoted 5 lines

in (b), the result of __pa_symbol(_text), memory_limit, memblock_mem_limit_remove_map and memblock_add
are not depended on memsart_addr.
So the now2 flow can grouping modification of memstart_address, put
(a) and (d) together.

I'm afraid that you've lost me with this. Why isn't it just as simple as the diff below?

Will

--->8

arch/arm64/mm/init.c +1 -1
index c29b17b520cd..ec3487c94b10 100644
@@ -377,7 +377,7 @@ void __init arm64_memblock_init(void)
 			 base + size > memblock_start_of_DRAM() +
 				       linear_region_size,
 			"initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
-			initrd_start = 0;
+			phys_initrd_size = 0;
 		} else {
 			memblock_remove(base, size); /* clear MEMBLOCK_ flags */
 			memblock_add(base, size);
author
#6 lore raw

hi Will:

Quoted 38 lines

[+Ard in case I'm missing something]

Quoted 34 lines · pierre kuo
Quoted 16 lines
Quoted 3 lines

With CONFIG_RANDOMIZE_BASE we can get a further change to memstart_addr after the place where you moved the initrd_{start,end} setting, which would result in a different value for __phys_to_virt(phys_initrd_start).

I found what you mean, since __phys_to_virt will use PHYS_OFFSET (memstart_addr) for calculating.
How about moving CONFIG_RANDOMIZE_BASE part of code ahead of CONFIG_BLK_DEV_INITRD checking?

That means below (d) moving ahead of (c)
prvious:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)

Quoted 6 lines

now:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) { ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ----------------(b)
if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} --------------(d) if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

After tracing the kernel code,
is it even possible to move CONFIG_RANDOMIZE_BASE part of code ahead of memory_limit?

that mean the flow may look like below:
now2:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

Quoted 7 lines
Quoted 5 lines

in (b), the result of __pa_symbol(_text), memory_limit, memblock_mem_limit_remove_map and memblock_add
are not depended on memsart_addr.
So the now2 flow can grouping modification of memstart_address, put
(a) and (d) together.

I'm afraid that you've lost me with this.

welcome for ur kind suggestion ^^

Quoted 7 lines

Why isn't it just as simple as
the diff below?

arch/arm64/mm/init.c +0 -0
index c29b17b520cd..ec3487c94b10 100644

arch/arm64/mm/init.c void __init arm64_memblock_init(void) · line 383

Quoted 8 lines
arch/arm64/mm/init.c +1 -1
@@ -377,7 +377,7 @@ void __init arm64_memblock_init(void)
                         base + size > memblock_start_of_DRAM() +
                                       linear_region_size,
                        "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
-                       initrd_start = 0;
+                       phys_initrd_size = 0;
                } else {
                        memblock_remove(base, size); /* clear MEMBLOCK_ flags */
                        memblock_add(base, size);

This patch will also fix the issue, but it still needs 2 "if comparisions" for getting initrd_start/initrd_end.
By possible grouping modification of memstart_address, and put initrd_start/initrd_end to be calculated only when linear mapping check pass. Maybe (just if) can let the code be more concise.

Sincerely appreciate all of yours great comment.

#7 lore raw
Quoted 57 lines · pierre kuo
Quoted 35 lines · pierre kuo
Quoted 34 lines
Quoted 16 lines
Quoted 3 lines

With CONFIG_RANDOMIZE_BASE we can get a further change to memstart_addr after the place where you moved the initrd_{start,end} setting, which would result in a different value for __phys_to_virt(phys_initrd_start).

I found what you mean, since __phys_to_virt will use PHYS_OFFSET (memstart_addr) for calculating.
How about moving CONFIG_RANDOMIZE_BASE part of code ahead of CONFIG_BLK_DEV_INITRD checking?

That means below (d) moving ahead of (c)
prvious:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)

Quoted 6 lines

now:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) { ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ----------------(b)
if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} --------------(d) if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

After tracing the kernel code,
is it even possible to move CONFIG_RANDOMIZE_BASE part of code ahead of memory_limit?

that mean the flow may look like below:
now2:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

Quoted 7 lines
Quoted 5 lines

in (b), the result of __pa_symbol(_text), memory_limit, memblock_mem_limit_remove_map and memblock_add
are not depended on memsart_addr.
So the now2 flow can grouping modification of memstart_address, put
(a) and (d) together.

I'm afraid that you've lost me with this.

welcome for ur kind suggestion ^^

Quoted 13 lines

Why isn't it just as simple as
the diff below?

arch/arm64/mm/init.c +1 -1
index c29b17b520cd..ec3487c94b10 100644
@@ -377,7 +377,7 @@ void __init arm64_memblock_init(void)
                         base + size > memblock_start_of_DRAM() +
                                       linear_region_size,
                        "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
-                       initrd_start = 0;
+                       phys_initrd_size = 0;
                } else {
Quoted 8 lines

arch/arm64/mm/init.c void __init arm64_memblock_init(void) · line 383

Quoted 2 lines
arch/arm64/mm/init.c +0 -0
                        memblock_remove(base, size); /* clear MEMBLOCK_ flags */
                        memblock_add(base, size);

This patch will also fix the issue, but it still needs 2 "if comparisions" for getting initrd_start/initrd_end.
By possible grouping modification of memstart_address, and put initrd_start/initrd_end to be calculated only when linear mapping check pass. Maybe (just if) can let the code be more concise.

Maybe, but I don't think we've seen a patch which accomplishes that. I think I'll go ahead and commit the basic one-liner, then we can always improve it afterwards if somebody sends a patch. It's not like this is a fastpath.

Will

#8 lore raw
Quoted 67 lines · Will Deacon
Quoted 57 lines · pierre kuo
Quoted 35 lines · pierre kuo
Quoted 34 lines
Quoted 16 lines
Quoted 3 lines

With CONFIG_RANDOMIZE_BASE we can get a further change to memstart_addr after the place where you moved the initrd_{start,end} setting, which would result in a different value for __phys_to_virt(phys_initrd_start).

I found what you mean, since __phys_to_virt will use PHYS_OFFSET (memstart_addr) for calculating.
How about moving CONFIG_RANDOMIZE_BASE part of code ahead of CONFIG_BLK_DEV_INITRD checking?

That means below (d) moving ahead of (c)
prvious:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)

Quoted 6 lines

now:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) { ---(a)
if (memory_limit != PHYS_ADDR_MAX) {} ----------------(b)
if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} --------------(d) if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

After tracing the kernel code,
is it even possible to move CONFIG_RANDOMIZE_BASE part of code ahead of memory_limit?

that mean the flow may look like below:
now2:
if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {} ---(a) if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {} ---(d)
if (memory_limit != PHYS_ADDR_MAX) {} ---(b)
if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {} ---(c)

Quoted 7 lines
Quoted 5 lines

in (b), the result of __pa_symbol(_text), memory_limit, memblock_mem_limit_remove_map and memblock_add
are not depended on memsart_addr.
So the now2 flow can grouping modification of memstart_address, put
(a) and (d) together.

I'm afraid that you've lost me with this.

welcome for ur kind suggestion ^^

Quoted 13 lines

Why isn't it just as simple as
the diff below?

arch/arm64/mm/init.c +1 -1
index c29b17b520cd..ec3487c94b10 100644
@@ -377,7 +377,7 @@ void __init arm64_memblock_init(void)
                         base + size > memblock_start_of_DRAM() +
                                       linear_region_size,
                        "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
-                       initrd_start = 0;
+                       phys_initrd_size = 0;
                } else {
Quoted 8 lines

arch/arm64/mm/init.c void __init arm64_memblock_init(void) · line 383

Quoted 2 lines
arch/arm64/mm/init.c +0 -0
                        memblock_remove(base, size); /* clear MEMBLOCK_ flags */
                        memblock_add(base, size);

This patch will also fix the issue, but it still needs 2 "if comparisions" for getting initrd_start/initrd_end.
By possible grouping modification of memstart_address, and put initrd_start/initrd_end to be calculated only when linear mapping check pass. Maybe (just if) can let the code be more concise.

Quoted 3 lines

Maybe, but I don't think we've seen a patch which accomplishes that. I think I'll go ahead and commit the basic one-liner, then we can always improve it afterwards if somebody sends a patch. It's not like this is a fastpath.

Sorry for the slow response and introducing the bug in the first place, yes, I agree here, an one-liner is a better way to get that fixed:

Signature · 1 line
Florian
author
#9 lore raw
In reply to Will Deacon

hi Will:

Quoted 4 lines

Maybe, but I don't think we've seen a patch which accomplishes that. I think I'll go ahead and commit the basic one-liner, then we can always improve it afterwards if somebody sends a patch. It's not like this is a fastpath.

Sorry for not showing the patches I try to explain to sir. I will attach v3 patches for ur reference.

Thanks for ur kind help,