Skip to contentSkip to conversation

memblock: exclude MEMBLOCK_NOMAP regions from kmemleak

PATCH In review Mike Rapoport 5 participants 9 messages Oct 21, 2021 – Oct 22, 2021

Subsystems (4)
lore
Share & tools
lore

Mike Rapoport

author maintainer
#0 lore raw

v2 cover memblock: exclude MEMBLOCK_NOMAP regions from kmemleak

From: Mike Rapoport <[email protected]>

Hi,

This is take 2 to fix interaction between MEMBLOCK_NOMAP and kmemleak.

The previous version caused boot failures Qian Cai reported here:

https://lore.kernel.org/all/[email protected]

The failures happened because calling kmemleak_free_part_phys() (or any kmemleak phys APIs for that matter) too early means it cannot use __va() on arm64.

This late in the cycle I can only think of reverting kmemleak wavier from memblock_mark_nomap() and putting it in the only two callers that set MEMBLOCK_NOMAP to an allocated chunk rather than marking NOMAP "unusable" memory reported by firmware.

The first patch here is the revert of v1 and the second patch is actual v2 implementation.

Vladimir and Qian, I'd appreciate if you could verify that v2 works for you.

v2:
* move kmemleak waiver from memblock_mark_nomap() to callers that need it
* use kmemleak_ignore_phys() rather than kmemleak_free_part_phys() as Catalin suggested.

v1: https://lore.kernel.org/all/[email protected]

Mike Rapoport (2):
Revert "memblock: exclude NOMAP regions from kmemleak"
memblock: exclude MEMBLOCK_NOMAP regions from kmemleak

 drivers/acpi/tables.c        |  3 +++
 drivers/of/of_reserved_mem.c |  2 ++
 mm/memblock.c                | 10 ++++------
 3 files changed, 9 insertions(+), 6 deletions(-)

base-commit: 519d81956ee277b4419c723adfb154603c2565ba

Signature · 1 line
2.28.0
author maintainer
#1 lore raw

PATCH v2 1/2 Revert "memblock: exclude NOMAP regions from kmemleak"

From: Mike Rapoport <[email protected]>

Commit 6e44bd6d34d6 ("memblock: exclude NOMAP regions from kmemleak") breaks boot on EFI systems with kmemleak and VM_DEBUG enabled:

efi: Processing EFI memory map:
efi:   0x000090000000-0x000091ffffff [Conventional|   |  |  |  |  |  |  |  |  |   |WB|WT|WC|UC]
efi:   0x000092000000-0x0000928fffff [Runtime Data|RUN|  |  |  |  |  |  |  |  |   |WB|WT|WC|UC]
------------[ cut here ]------------
kernel BUG at mm/kmemleak.c:1140!
Internal error: Oops - BUG: 0 [#1] SMP
Modules linked in:
CPU: 0 PID: 0 Comm: swapper Not tainted 5.15.0-rc6-next-20211019+ #104
pstate: 600000c5 (nZCv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : kmemleak_free_part_phys+0x64/0x8c
lr : kmemleak_free_part_phys+0x38/0x8c

sp : ffff800011eafbc0
x29: ffff800011eafbc0 x28: 1fffff7fffb41c0d x27: fffffbfffda0e068
x26: 0000000092000000 x25: 1ffff000023d5f94 x24: ffff800011ed84d0
x23: ffff800011ed84c0 x22: ffff800011ed83d8 x21: 0000000000900000
x20: ffff800011782000 x19: 0000000092000000 x18: ffff800011ee0730
x17: 0000000000000000 x16: 0000000000000000 x15: 1ffff0000233252c
x14: ffff800019a905a0 x13: 0000000000000001 x12: ffff7000023d5ed7
x11: 1ffff000023d5ed6 x10: ffff7000023d5ed6 x9 : dfff800000000000 x8 : ffff800011eaf6b7 x7 : 0000000000000001 x6 : ffff800011eaf6b0 x5 : 00008ffffdc2a12a x4 : ffff7000023d5ed7 x3 : 1ffff000023dbf99 x2 : 1ffff000022f0463 x1 : 0000000000000000 x0 : ffffffffffffffff

Call trace:
 kmemleak_free_part_phys+0x64/0x8c
 memblock_mark_nomap+0x5c/0x78
 reserve_regions+0x294/0x33c
 efi_init+0x2d0/0x490
 setup_arch+0x80/0x138
 start_kernel+0xa0/0x3ec
 __primary_switched+0xc0/0xc8
Code: 34000041 97d526e7 f9418e80 36000040 (d4210000)
random: get_random_bytes called from print_oops_end_marker+0x34/0x80 with crng_init=0
---[ end trace 0000000000000000 ]---

The crash happens because kmemleak_free_part_phys() tries to use __va() before memstart_addr is initialized and this triggers a VM_BUG_ON() in arch/arm64/include/asm/memory.h:

Revert 6e44bd6d34d6 ("memblock: exclude NOMAP regions from kmemleak"), the issue it is fixing will be fixed differently.

---
 mm/memblock.c | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
mm/memblock.c +1 -6
index 5c3503c98b2f..184dcd2e5d99 100644
@@ -936,12 +936,7 @@ int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
  */
 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
 {
-	int ret = memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
-
-	if (!ret)
-		kmemleak_free_part_phys(base, size);
-
-	return ret;
+	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
 }
 
 /**
Signature · 1 line
2.28.0
Acked-by on v2 1/2
#2 lore raw

Re: [PATCH v2 1/2] Revert "memblock: exclude NOMAP regions from kmemleak"

Quoted 43 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Commit 6e44bd6d34d6 ("memblock: exclude NOMAP regions from kmemleak") breaks boot on EFI systems with kmemleak and VM_DEBUG enabled:

efi: Processing EFI memory map:
efi:   0x000090000000-0x000091ffffff [Conventional|   |  |  |  |  |  |  |  |  |   |WB|WT|WC|UC]
efi:   0x000092000000-0x0000928fffff [Runtime Data|RUN|  |  |  |  |  |  |  |  |   |WB|WT|WC|UC]
------------[ cut here ]------------
kernel BUG at mm/kmemleak.c:1140!
Internal error: Oops - BUG: 0 [#1] SMP
Modules linked in:
CPU: 0 PID: 0 Comm: swapper Not tainted 5.15.0-rc6-next-20211019+ #104
pstate: 600000c5 (nZCv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : kmemleak_free_part_phys+0x64/0x8c
lr : kmemleak_free_part_phys+0x38/0x8c

sp : ffff800011eafbc0
x29: ffff800011eafbc0 x28: 1fffff7fffb41c0d x27: fffffbfffda0e068
x26: 0000000092000000 x25: 1ffff000023d5f94 x24: ffff800011ed84d0
x23: ffff800011ed84c0 x22: ffff800011ed83d8 x21: 0000000000900000
x20: ffff800011782000 x19: 0000000092000000 x18: ffff800011ee0730
x17: 0000000000000000 x16: 0000000000000000 x15: 1ffff0000233252c
x14: ffff800019a905a0 x13: 0000000000000001 x12: ffff7000023d5ed7
x11: 1ffff000023d5ed6 x10: ffff7000023d5ed6 x9 : dfff800000000000 x8 : ffff800011eaf6b7 x7 : 0000000000000001 x6 : ffff800011eaf6b0 x5 : 00008ffffdc2a12a x4 : ffff7000023d5ed7 x3 : 1ffff000023dbf99 x2 : 1ffff000022f0463 x1 : 0000000000000000 x0 : ffffffffffffffff

Call trace:
 kmemleak_free_part_phys+0x64/0x8c
 memblock_mark_nomap+0x5c/0x78
 reserve_regions+0x294/0x33c
 efi_init+0x2d0/0x490
 setup_arch+0x80/0x138
 start_kernel+0xa0/0x3ec
 __primary_switched+0xc0/0xc8
Code: 34000041 97d526e7 f9418e80 36000040 (d4210000)
random: get_random_bytes called from print_oops_end_marker+0x34/0x80 with crng_init=0
---[ end trace 0000000000000000 ]---

The crash happens because kmemleak_free_part_phys() tries to use __va() before memstart_addr is initialized and this triggers a VM_BUG_ON() in arch/arm64/include/asm/memory.h:

Quoted 5 lines

Revert 6e44bd6d34d6 ("memblock: exclude NOMAP regions from kmemleak"), the issue it is fixing will be fixed differently.

Reviewed-by on v2 1/2
#3 lore raw

Re: [PATCH v2 1/2] Revert "memblock: exclude NOMAP regions from kmemleak"

In reply to Mike Rapoport
Quoted 61 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Commit 6e44bd6d34d6 ("memblock: exclude NOMAP regions from kmemleak") breaks boot on EFI systems with kmemleak and VM_DEBUG enabled:

efi: Processing EFI memory map:
efi:   0x000090000000-0x000091ffffff [Conventional|   |  |  |  |  |  |  |  |  |   |WB|WT|WC|UC]
efi:   0x000092000000-0x0000928fffff [Runtime Data|RUN|  |  |  |  |  |  |  |  |   |WB|WT|WC|UC]
------------[ cut here ]------------
kernel BUG at mm/kmemleak.c:1140!
Internal error: Oops - BUG: 0 [#1] SMP
Modules linked in:
CPU: 0 PID: 0 Comm: swapper Not tainted 5.15.0-rc6-next-20211019+ #104
pstate: 600000c5 (nZCv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : kmemleak_free_part_phys+0x64/0x8c
lr : kmemleak_free_part_phys+0x38/0x8c

sp : ffff800011eafbc0
x29: ffff800011eafbc0 x28: 1fffff7fffb41c0d x27: fffffbfffda0e068
x26: 0000000092000000 x25: 1ffff000023d5f94 x24: ffff800011ed84d0
x23: ffff800011ed84c0 x22: ffff800011ed83d8 x21: 0000000000900000
x20: ffff800011782000 x19: 0000000092000000 x18: ffff800011ee0730
x17: 0000000000000000 x16: 0000000000000000 x15: 1ffff0000233252c
x14: ffff800019a905a0 x13: 0000000000000001 x12: ffff7000023d5ed7
x11: 1ffff000023d5ed6 x10: ffff7000023d5ed6 x9 : dfff800000000000 x8 : ffff800011eaf6b7 x7 : 0000000000000001 x6 : ffff800011eaf6b0 x5 : 00008ffffdc2a12a x4 : ffff7000023d5ed7 x3 : 1ffff000023dbf99 x2 : 1ffff000022f0463 x1 : 0000000000000000 x0 : ffffffffffffffff

Call trace:
 kmemleak_free_part_phys+0x64/0x8c
 memblock_mark_nomap+0x5c/0x78
 reserve_regions+0x294/0x33c
 efi_init+0x2d0/0x490
 setup_arch+0x80/0x138
 start_kernel+0xa0/0x3ec
 __primary_switched+0xc0/0xc8
Code: 34000041 97d526e7 f9418e80 36000040 (d4210000)
random: get_random_bytes called from print_oops_end_marker+0x34/0x80 with crng_init=0
---[ end trace 0000000000000000 ]---

The crash happens because kmemleak_free_part_phys() tries to use __va() before memstart_addr is initialized and this triggers a VM_BUG_ON() in arch/arm64/include/asm/memory.h:

Revert 6e44bd6d34d6 ("memblock: exclude NOMAP regions from kmemleak"), the issue it is fixing will be fixed differently.

---
 mm/memblock.c | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
mm/memblock.c +0 -0
index 5c3503c98b2f..184dcd2e5d99 100644
@@ -936,12 +936,7 @@ int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
  */
 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
 {

mm/memblock.c int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) · line 942

Quoted 10 lines
mm/memblock.c +1 -6
-	int ret = memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
-
-	if (!ret)
-		kmemleak_free_part_phys(base, size);
-
-	return ret;
+	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
 }
 
 /**
Signature · 3 lines
Thanks,

David / dhildenb
author maintainer
#4 lore raw

PATCH v2 2/2 memblock: exclude MEMBLOCK_NOMAP regions from kmemleak

In reply to Mike Rapoport

From: Mike Rapoport <[email protected]>

Vladimir Zapolskiy reports:

commit a7259df76702 ("memblock: make memblock_find_in_range method private") invokes a kernel panic while running kmemleak on OF platforms with nomaped regions:

  Unable to handle kernel paging request at virtual address fff000021e00000
  [...]
    scan_block+0x64/0x170
    scan_gray_list+0xe8/0x17c
    kmemleak_scan+0x270/0x514
    kmemleak_write+0x34c/0x4ac

The memory allocated from memblock is registered with kmemleak, but if it is marked MEMBLOCK_NOMAP it won't have linear map entries so an attempt to scan such areas will fault.

Ideally, memblock_mark_nomap() would inform kmemleak to ignore MEMBLOCK_NOMAP memory, but it can be called before kmemleak interfaces operating on physical addresses can use __va() conversion.

Make sure that functions that mark allocated memory as MEMBLOCK_NOMAP take care of informing kmemleak to ignore such memory.

---
 drivers/acpi/tables.c        | 3 +++
 drivers/of/of_reserved_mem.c | 2 ++
 mm/memblock.c                | 3 +++
 3 files changed, 8 insertions(+)
drivers/acpi/tables.c +3 -0
index f9383736fa0f..71419eb16e09 100644
@@ -21,6 +21,7 @@
 #include <linux/earlycpio.h>
 #include <linux/initrd.h>
 #include <linux/security.h>
+#include <linux/kmemleak.h>
 #include "internal.h"
 
 #ifdef CONFIG_ACPI_CUSTOM_DSDT
@@ -601,6 +602,8 @@ void __init acpi_table_upgrade(void)
 	 */
 	arch_reserve_mem_area(acpi_tables_addr, all_tables_size);
 
+	kmemleak_ignore_phys(acpi_tables_addr);
+
 	/*
 	 * early_ioremap only can remap 256k one time. If we map all
 	 * tables one time, we will hit the limit. Need to map chunks
drivers/of/of_reserved_mem.c +2 -0
index 59c1390cdf42..9da8835ba5a5 100644
@@ -21,6 +21,7 @@
 #include <linux/sort.h>
 #include <linux/slab.h>
 #include <linux/memblock.h>
+#include <linux/kmemleak.h>
 
 #include "of_private.h"
 
@@ -46,6 +47,7 @@ static int __init early_init_dt_alloc_reserved_memory_arch(phys_addr_t size,
 		err = memblock_mark_nomap(base, size);
 		if (err)
 			memblock_free(base, size);
+		kmemleak_ignore_phys(base);
 	}
 
 	return err;
mm/memblock.c +3 -0
index 184dcd2e5d99..dab804b09d62 100644
@@ -932,6 +932,9 @@ int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
  * covered by the memory map. The struct page representing NOMAP memory
  * frames in the memory map will be PageReserved()
  *
+ * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
+ * memblock, the caller must inform kmemleak to ignore that memory
+ *
  * Return: 0 on success, -errno on failure.
  */
 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
Signature · 1 line
2.28.0
Tested-by on v2 2/2
#5 lore raw

Hi Mike,

Quoted 23 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Vladimir Zapolskiy reports:

commit a7259df76702 ("memblock: make memblock_find_in_range method private") invokes a kernel panic while running kmemleak on OF platforms with nomaped regions:

   Unable to handle kernel paging request at virtual address fff000021e00000
   [...]
     scan_block+0x64/0x170
     scan_gray_list+0xe8/0x17c
     kmemleak_scan+0x270/0x514
     kmemleak_write+0x34c/0x4ac

The memory allocated from memblock is registered with kmemleak, but if it is marked MEMBLOCK_NOMAP it won't have linear map entries so an attempt to scan such areas will fault.

Ideally, memblock_mark_nomap() would inform kmemleak to ignore MEMBLOCK_NOMAP memory, but it can be called before kmemleak interfaces operating on physical addresses can use __va() conversion.

Quoted 8 lines

Make sure that functions that mark allocated memory as MEMBLOCK_NOMAP take care of informing kmemleak to ignore such memory.

this change variant also solves the reported problem, thank you.

--
Best wishes,
Vladimir

Reviewed-by on v2 2/2
#6 lore raw
In reply to Mike Rapoport
Quoted 34 lines · Mike Rapoport
drivers/acpi/tables.c +3 -0
index f9383736fa0f..71419eb16e09 100644
@@ -21,6 +21,7 @@
 #include <linux/earlycpio.h>
 #include <linux/initrd.h>
 #include <linux/security.h>
+#include <linux/kmemleak.h>
 #include "internal.h"
 
 #ifdef CONFIG_ACPI_CUSTOM_DSDT
@@ -601,6 +602,8 @@ void __init acpi_table_upgrade(void)
 	 */
 	arch_reserve_mem_area(acpi_tables_addr, all_tables_size);
 
+	kmemleak_ignore_phys(acpi_tables_addr);
+
 	/*
 	 * early_ioremap only can remap 256k one time. If we map all
 	 * tables one time, we will hit the limit. Need to map chunks
drivers/of/of_reserved_mem.c +1 -0
index 59c1390cdf42..9da8835ba5a5 100644
@@ -21,6 +21,7 @@
 #include <linux/sort.h>
 #include <linux/slab.h>
 #include <linux/memblock.h>
+#include <linux/kmemleak.h>
 
 #include "of_private.h"
 

drivers/of/of_reserved_mem.c static int __init early_init_dt_alloc_reserved_memory_arch(phys_addr_t size, · line 53

Quoted 7 lines
drivers/of/of_reserved_mem.c +1 -0
@@ -46,6 +47,7 @@ static int __init early_init_dt_alloc_reserved_memory_arch(phys_addr_t size,
 		err = memblock_mark_nomap(base, size);
 		if (err)
 			memblock_free(base, size);
+		kmemleak_ignore_phys(base);
 	}
 
 	return err;

More of a nitpick as there's no kmemleak scanning to race with during early boot: I'd normally call kmemleak_ignore_phys() before marking it nomap. Either way:

Reviewed-by on v2 2/2
#7 lore raw
In reply to Mike Rapoport
Quoted 84 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Vladimir Zapolskiy reports:

commit a7259df76702 ("memblock: make memblock_find_in_range method private") invokes a kernel panic while running kmemleak on OF platforms with nomaped regions:

  Unable to handle kernel paging request at virtual address fff000021e00000
  [...]
    scan_block+0x64/0x170
    scan_gray_list+0xe8/0x17c
    kmemleak_scan+0x270/0x514
    kmemleak_write+0x34c/0x4ac

The memory allocated from memblock is registered with kmemleak, but if it is marked MEMBLOCK_NOMAP it won't have linear map entries so an attempt to scan such areas will fault.

Ideally, memblock_mark_nomap() would inform kmemleak to ignore MEMBLOCK_NOMAP memory, but it can be called before kmemleak interfaces operating on physical addresses can use __va() conversion.

Make sure that functions that mark allocated memory as MEMBLOCK_NOMAP take care of informing kmemleak to ignore such memory.

---
 drivers/acpi/tables.c        | 3 +++
 drivers/of/of_reserved_mem.c | 2 ++
 mm/memblock.c                | 3 +++
 3 files changed, 8 insertions(+)
drivers/acpi/tables.c +3 -0
index f9383736fa0f..71419eb16e09 100644
@@ -21,6 +21,7 @@
 #include <linux/earlycpio.h>
 #include <linux/initrd.h>
 #include <linux/security.h>
+#include <linux/kmemleak.h>
 #include "internal.h"
 
 #ifdef CONFIG_ACPI_CUSTOM_DSDT
@@ -601,6 +602,8 @@ void __init acpi_table_upgrade(void)
 	 */
 	arch_reserve_mem_area(acpi_tables_addr, all_tables_size);
 
+	kmemleak_ignore_phys(acpi_tables_addr);
+
 	/*
 	 * early_ioremap only can remap 256k one time. If we map all
 	 * tables one time, we will hit the limit. Need to map chunks
drivers/of/of_reserved_mem.c +2 -0
index 59c1390cdf42..9da8835ba5a5 100644
@@ -21,6 +21,7 @@
 #include <linux/sort.h>
 #include <linux/slab.h>
 #include <linux/memblock.h>
+#include <linux/kmemleak.h>
 
 #include "of_private.h"
 
@@ -46,6 +47,7 @@ static int __init early_init_dt_alloc_reserved_memory_arch(phys_addr_t size,
 		err = memblock_mark_nomap(base, size);
 		if (err)
 			memblock_free(base, size);
+		kmemleak_ignore_phys(base);
 	}
 
 	return err;
mm/memblock.c +0 -0
index 184dcd2e5d99..dab804b09d62 100644

mm/memblock.c int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) · line 940

Quoted 9 lines
mm/memblock.c +3 -0
@@ -932,6 +932,9 @@ int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
  * covered by the memory map. The struct page representing NOMAP memory
  * frames in the memory map will be PageReserved()
  *
+ * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
+ * memblock, the caller must inform kmemleak to ignore that memory
+ *
  * Return: 0 on success, -errno on failure.
  */
 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
Signature · 3 lines
Thanks,

David / dhildenb
#8 lore raw
In reply to Mike Rapoport
Quoted 19 lines · Mike Rapoport

From: Mike Rapoport <[email protected]>

Hi,

This is take 2 to fix interaction between MEMBLOCK_NOMAP and kmemleak.

The previous version caused boot failures Qian Cai reported here:

https://lore.kernel.org/all/[email protected]

The failures happened because calling kmemleak_free_part_phys() (or any kmemleak phys APIs for that matter) too early means it cannot use __va() on arm64.

This late in the cycle I can only think of reverting kmemleak wavier from memblock_mark_nomap() and putting it in the only two callers that set MEMBLOCK_NOMAP to an allocated chunk rather than marking NOMAP "unusable" memory reported by firmware.

Quoted 5 lines

The first patch here is the revert of v1 and the second patch is actual v2 implementation.

Vladimir and Qian, I'd appreciate if you could verify that v2 works for you.

FYI, this passed the regression tests here.

Quoted 20 lines

v2:
* move kmemleak waiver from memblock_mark_nomap() to callers that need it
* use kmemleak_ignore_phys() rather than kmemleak_free_part_phys() as Catalin suggested.

v1: https://lore.kernel.org/all/[email protected]

Mike Rapoport (2):
Revert "memblock: exclude NOMAP regions from kmemleak"
memblock: exclude MEMBLOCK_NOMAP regions from kmemleak

 drivers/acpi/tables.c        |  3 +++
 drivers/of/of_reserved_mem.c |  2 ++
 mm/memblock.c                | 10 ++++------
 3 files changed, 9 insertions(+), 6 deletions(-)

base-commit: 519d81956ee277b4419c723adfb154603c2565ba