Skip to contentSkip to conversation

mm: memblock: always inline memblock_alloc

PATCH In review Faiyaz Mohammed 2 participants 4 messages Nov 10, 2020 – Nov 11, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm: memblock: always inline memblock_alloc

Since memblock_alloc is not getting inlined, memblock_reserve owner info is lost. Below information is not enough for memory accounting. for example:

[    0.000000] memblock_alloc_try_nid: 1490 bytes align=0x40 nid=-1 from=0x0000000000000000 max_addr=0x0000000000000000 memblock_alloc+0x20/0x2c
[    0.000000] memblock_reserve: [0x000000023f09a3c0-0x000000023f09a991] memblock_alloc_range_nid+0xc0/0x188

Add "__always_inline" to make sure it get inlined and to get the exact owner of the memblock_reserve.
After adding __always_inline:

[    0.000000] memblock_alloc_try_nid: 1490 bytes align=0x40 nid=-1 from=0x0000000000000000 max_addr=0x0000000000000000 start_kernel+0xa4/0x568
[    0.000000] memblock_reserve: [0x000000023f09a3c0-0x000000023f09a991] memblock_alloc_range_nid+0xc0/0x188
---
 include/linux/memblock.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/memblock.h +1 -1
index ef13125..54f9544 100644
@@ -404,7 +404,7 @@ void *memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align,
 			     phys_addr_t min_addr, phys_addr_t max_addr,
 			     int nid);
 
-static inline void * __init memblock_alloc(phys_addr_t size,  phys_addr_t align)
+static __always_inline void * __init memblock_alloc(phys_addr_t size,  phys_addr_t align)
 {
 	return memblock_alloc_try_nid(size, align, MEMBLOCK_LOW_LIMIT,
 				      MEMBLOCK_ALLOC_ACCESSIBLE, NUMA_NO_NODE);
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of the Code Aurora Forum, hosted by The Linux Foundation
maintainer
#1 lore raw

Hi,

Quoted 11 lines · Faiyaz Mohammed

Since memblock_alloc is not getting inlined, memblock_reserve owner info is lost. Below information is not enough for memory accounting. for example:

[    0.000000] memblock_alloc_try_nid: 1490 bytes align=0x40 nid=-1 from=0x0000000000000000 max_addr=0x0000000000000000 memblock_alloc+0x20/0x2c
[    0.000000] memblock_reserve: [0x000000023f09a3c0-0x000000023f09a991] memblock_alloc_range_nid+0xc0/0x188

Add "__always_inline" to make sure it get inlined and to get the exact owner of the memblock_reserve.
After adding __always_inline:

[    0.000000] memblock_alloc_try_nid: 1490 bytes align=0x40 nid=-1 from=0x0000000000000000 max_addr=0x0000000000000000 start_kernel+0xa4/0x568
[    0.000000] memblock_reserve: [0x000000023f09a3c0-0x000000023f09a991] memblock_alloc_range_nid+0xc0/0x188

I agree that making memblock_alloc() inline as well as other similar wrappers would improve the debugability.
Still, it has nothing to do with memory accounting and owner tracking. Please update the patch description to better explain what it actually improves.

Quoted 10 lines
---
 include/linux/memblock.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/memblock.h +0 -0
index ef13125..54f9544 100644

include/linux/memblock.h void *memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, · line 407

Quoted 5 lines
include/linux/memblock.h +1 -1
@@ -404,7 +404,7 @@ void *memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align,
 			     phys_addr_t min_addr, phys_addr_t max_addr,
 			     int nid);
 
-static inline void * __init memblock_alloc(phys_addr_t size,  phys_addr_t align)
+static __always_inline void * __init memblock_alloc(phys_addr_t size,  phys_addr_t align)

I think simply dropping __init here will make memblock_alloc() inline. There are also several more convenience wrappers marked __init, do you mind removing the __init annotation for them as well?

Quoted 3 lines
include/linux/memblock.h +0 -0
 {
 	return memblock_alloc_try_nid(size, align, MEMBLOCK_LOW_LIMIT,
 				      MEMBLOCK_ALLOC_ACCESSIBLE, NUMA_NO_NODE);
Signature · 2 lines
Sincerely yours,
Mike.
#2 lore raw
Quoted 15 lines · Mike Rapoport

Hi,

Quoted 11 lines · Faiyaz Mohammed

Since memblock_alloc is not getting inlined, memblock_reserve owner info is lost. Below information is not enough for memory accounting. for example:

[    0.000000] memblock_alloc_try_nid: 1490 bytes align=0x40 nid=-1 from=0x0000000000000000 max_addr=0x0000000000000000 memblock_alloc+0x20/0x2c
[    0.000000] memblock_reserve: [0x000000023f09a3c0-0x000000023f09a991] memblock_alloc_range_nid+0xc0/0x188

Add "__always_inline" to make sure it get inlined and to get the exact owner of the memblock_reserve.
After adding __always_inline:

[    0.000000] memblock_alloc_try_nid: 1490 bytes align=0x40 nid=-1 from=0x0000000000000000 max_addr=0x0000000000000000 start_kernel+0xa4/0x568
[    0.000000] memblock_reserve: [0x000000023f09a3c0-0x000000023f09a991] memblock_alloc_range_nid+0xc0/0x188
Quoted 5 lines

I agree that making memblock_alloc() inline as well as other similar wrappers would improve the debugability.
Still, it has nothing to do with memory accounting and owner tracking. Please update the patch description to better explain what it actually improves.

As describe in other thread, do memblock reserved accounting to track owners

to know size of memory allocated by different drivers/owners through

memblock_reserve, which help in comparing different kernel version and in

optimizations.

Quoted 14 lines
Quoted 14 lines
---
  include/linux/memblock.h | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/memblock.h +0 -0
index ef13125..54f9544 100644
@@ -404,7 +404,7 @@ void *memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align,
  			     phys_addr_t min_addr, phys_addr_t max_addr,
  			     int nid);
 
Quoted 5 lines

include/linux/memblock.h void *memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, · line 407

Quoted 2 lines
include/linux/memblock.h +1 -1
-static inline void * __init memblock_alloc(phys_addr_t size,  phys_addr_t align)
+static __always_inline void * __init memblock_alloc(phys_addr_t size,  phys_addr_t align)

I think simply dropping __init here will make memblock_alloc() inline. There are also several more convenience wrappers marked __init, do you mind removing the __init annotation for them as well?

Yes, if we drop __init, memblock_alloc will get inline but would it not increase

kernel footprint as the function will no more be released after kernel init?

Quoted 3 lines
Quoted 3 lines
include/linux/memblock.h +0 -0
  {
  	return memblock_alloc_try_nid(size, align, MEMBLOCK_LOW_LIMIT,
  				      MEMBLOCK_ALLOC_ACCESSIBLE, NUMA_NO_NODE);
maintainer
#3 lore raw
Quoted 48 lines · Faiyaz Mohammed
Quoted 14 lines · Mike Rapoport

Hi,

Quoted 11 lines · Faiyaz Mohammed

Since memblock_alloc is not getting inlined, memblock_reserve owner info is lost. Below information is not enough for memory accounting. for example:

[    0.000000] memblock_alloc_try_nid: 1490 bytes align=0x40 nid=-1 from=0x0000000000000000 max_addr=0x0000000000000000 memblock_alloc+0x20/0x2c
[    0.000000] memblock_reserve: [0x000000023f09a3c0-0x000000023f09a991] memblock_alloc_range_nid+0xc0/0x188

Add "__always_inline" to make sure it get inlined and to get the exact owner of the memblock_reserve.
After adding __always_inline:

[    0.000000] memblock_alloc_try_nid: 1490 bytes align=0x40 nid=-1 from=0x0000000000000000 max_addr=0x0000000000000000 start_kernel+0xa4/0x568
[    0.000000] memblock_reserve: [0x000000023f09a3c0-0x000000023f09a991] memblock_alloc_range_nid+0xc0/0x188
Quoted 5 lines

I agree that making memblock_alloc() inline as well as other similar wrappers would improve the debugability.
Still, it has nothing to do with memory accounting and owner tracking. Please update the patch description to better explain what it actually improves.

As describe in other thread, do memblock reserved accounting to track owners

to know size of memory allocated by different drivers/owners through

memblock_reserve, which help in comparing different kernel version and in

optimizations.

Quoted 6 lines
Quoted 6 lines
---
  include/linux/memblock.h | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 12 lines

include/linux/memblock.h void *memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, · line 406

Quoted 9 lines
include/linux/memblock.h +1 -1
index ef13125..54f9544 100644
@@ -404,7 +404,7 @@ void *memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align,
  			     phys_addr_t min_addr, phys_addr_t max_addr,
  			     int nid);
-static inline void * __init memblock_alloc(phys_addr_t size,  phys_addr_t align)
+static __always_inline void * __init memblock_alloc(phys_addr_t size,  phys_addr_t align)

I think simply dropping __init here will make memblock_alloc() inline. There are also several more convenience wrappers marked __init, do you mind removing the __init annotation for them as well?

Quoted 3 lines

Yes, if we drop __init, memblock_alloc will get inline but would it not increase
kernel footprint as the function will no more be released after kernel init?

If memblock_alloc(), or any other function for that matter, is inlined, it does not have an entry in the object file, but it is rather inserted at the call sites.

The functions that use memblock_alloc() may change, but they all should be __init anyway so I won't expect increase in the kernel post-init memory footprint.

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
include/linux/memblock.h +0 -0
  {
  	return memblock_alloc_try_nid(size, align, MEMBLOCK_LOW_LIMIT,
  				      MEMBLOCK_ALLOC_ACCESSIBLE, NUMA_NO_NODE);
Signature · 2 lines
Sincerely yours,
Mike.