ARM: memblock: set allocation limit for MEMBLOCK_ALLOC_ANYWHERE
PATCH 1/1 ARM: memblock: set allocation limit for MEMBLOCK_ALLOC_ANYWHERE
From: Will Deacon <[email protected]>
Commit e63075a3 removed the explicit MEMBLOCK_REAL_LIMIT #define and introduced the requirement that arch code calls memblock_set_current_limit to ensure that the __va macro can be used on physical addresses returned from memblock_alloc.
This patch ensures that memblock allocates from lowmem on ARM by setting the allocation limit to lowmem_end_addr.
--- arch/arm/include/asm/memblock.h | 3 --- arch/arm/mm/mmu.c | 1 + 2 files changed, 1 insertions(+), 3 deletions(-)
arch/arm/include/asm/memblock.h +0 -3
@@ -3,9 +3,6 @@#ifdef CONFIG_MMU extern phys_addr_t lowmem_end_addr; -#define MEMBLOCK_REAL_LIMIT lowmem_end_addr -#else -#define MEMBLOCK_REAL_LIMIT 0#endif struct meminfo ;
arch/arm/mm/mmu.c +1 -0
@@ -753,6 +753,7 @@ static void __init sanity_check_meminfo(void)int i, j, highmem =0 ; lowmem_end_addr =__pa (vmalloc_min -1 ) +1 ; +memblock_set_current_limit (lowmem_end_addr);for (i =0 , j =0 ; i < meminfo.nr_banks; i++) {struct membank *bank = &meminfo.bank[j];
PATCH 1/1 ARM: memblock: set allocation limit for MEMBLOCK_ALLOC_ANYWHERE
Quoted 22 lines · Catalin Marinas
From: Will Deacon <[email protected]>
Commit
e63075a3removed the explicitMEMBLOCK_REAL_LIMIT#define and introduced the requirement that arch code callsmemblock_set_current_limitto ensure that the__vamacro can be used on physical addresses returned frommemblock_alloc.This patch ensures that memblock allocates from lowmem on ARM by setting the allocation limit to
lowmem_end_addr.Signed-off-by Will Deacon <[email protected]>Signed-off-by Catalin Marinas <[email protected]>--- arch/arm/include/asm/memblock.h | 3 --- arch/arm/mm/mmu.c | 1 + 2 files changed, 1 insertions(+), 3 deletions(-)arch/arm/include/asm/memblock.h +0
-0
arch/arm/include/asm/memblock.h line 5
Quoted 3 lines
arch/arm/include/asm/memblock.h +0
-0@@ -3,9 +3,6 @@#ifdef CONFIG_MMU extern phys_addr_t lowmem_end_addr;
This doesn't need to be exported anymore.
Quoted 11 lines
arch/arm/include/asm/memblock.h +0
-3-#define MEMBLOCK_REAL_LIMIT lowmem_end_addr -#else -#define MEMBLOCK_REAL_LIMIT 0#endif struct meminfo ;arch/arm/mm/mmu.c +0
-0
arch/arm/mm/mmu.c static void __init sanity_check_meminfo(void) · line 756
Quoted 4 lines
arch/arm/mm/mmu.c +1
-0@@ -753,6 +753,7 @@ static void __init sanity_check_meminfo(void)int i, j, highmem =0 ; lowmem_end_addr =__pa (vmalloc_min -1 ) +1 ; +memblock_set_current_limit (lowmem_end_addr);
So this might as well become:
memblock_set_current_limit(__pa(vmalloc_min - 1) + 1);
PATCH 1/1 ARM: memblock: set allocation limit for MEMBLOCK_ALLOC_ANYWHERE
Hi Russell,
Quoted 7 lines · Russell King - ARM Linux
Quoted 6 lines · Catalin Marinas
From: Will Deacon <[email protected]>
Commit
e63075a3removed the explicitMEMBLOCK_REAL_LIMIT#define and introduced the requirement that arch code callsmemblock_set_current_limitto ensure that the__vamacro can be used on physical addresses returned frommemblock_alloc.
[...]
Quoted 7 lines
Review
Quoted 1 line
(quoted diff) +0
-0
b/arch/arm/include/asm/memblock.hQuoted 5 lines
index fdbc43b..655b4c4 100644
arch/arm/include/asm/memblock.h +0
-0@@ -3,9 +3,6 @@
Quoted 5 lines
arch/arm/include/asm/memblock.h line 5
Quoted 2 lines
arch/arm/include/asm/memblock.h +0
-0#ifdef CONFIG_MMU extern phys_addr_t lowmem_end_addr;This doesn't need to be exported anymore.
Thanks for confirming this. I'll remove it from the header file.
Quoted 7 lines
Quoted 7 lines
arch/arm/mm/mmu.c +0
-0@@ -753,6 +753,7 @@ static void __init sanity_check_meminfo(void)int i, j, highmem =0 ;
Quoted 5 lines
arch/arm/mm/mmu.c static void __init sanity_check_meminfo(void) · line 756
Quoted 2 lines
arch/arm/mm/mmu.c +1
-0lowmem_end_addr =__pa (vmalloc_min -1 ) +1 ; +memblock_set_current_limit (lowmem_end_addr);So this might as well become:
memblock_set_current_limit(__pa(vmalloc_min - 1) + 1);
Right. I could also move this call out of sanity_check_meminfo and into arm_memblock_init now that we don't need lowmem_end_addr to be initialised. If you're happy with me making that change then I'll post a revised version to the patch system.
Thanks,
Will
PATCH 1/1 ARM: memblock: set allocation limit for MEMBLOCK_ALLOC_ANYWHERE
Quoted 4 lines · Will Deacon
Right. I could also move this call out of
sanity_check_meminfoand intoarm_memblock_initnow that we don't needlowmem_end_addrto be initialised. If you're happy with me making that change then I'll post a revised version to the patch system.
It looks like we're doing a few things out of order. Yes, it makes sense to move that initialization to arm_memblock_init(), but as we're not running sanity_check_meminfo(), it could mean there's problems with out of order membanks and similar.
I'll look at that when I'm back.
PATCH 1/1 ARM: memblock: set allocation limit for MEMBLOCK_ALLOC_ANYWHERE
Russell,
Quoted 11 lines · Russell King - ARM Linux
Quoted 4 lines · Will Deacon
Right. I could also move this call out of
sanity_check_meminfoand intoarm_memblock_initnow that we don't needlowmem_end_addrto be initialised. If you're happy with me making that change then I'll post a revised version to the patch system.It looks like we're doing a few things out of order. Yes, it makes sense to move that initialization to
arm_memblock_init(), but as we're not runningsanity_check_meminfo(), it could mean there's problems with out of order membanks and similar.
We could move the arm_memblock_init() call into paging_init, immediately after the call to sanity_check_meminfo. I think everything should be safe then.
Quoted 1 line
I'll look at that when I'm back.
Great, thanks.
Will
PATCH 1/1 ARM: memblock: set allocation limit for MEMBLOCK_ALLOC_ANYWHERE
Quoted 9 lines · Will Deacon
Russell,
Quoted 6 lines · Russell King - ARM Linux
Quoted 4 lines · Will Deacon
Right. I could also move this call out of
sanity_check_meminfoand intoarm_memblock_initnow that we don't needlowmem_end_addrto be initialised. If you're happy with me making that change then I'll post a revised version to the patch system.
Quoted 8 lines
Quoted 5 lines
It looks like we're doing a few things out of order. Yes, it makes sense to move that initialization to
arm_memblock_init(), but as we're not runningsanity_check_meminfo(), it could mean there's problems with out of order membanks and similar.We could move the
arm_memblock_init()call intopaging_init, immediately after the call tosanity_check_meminfo. I think everything should be safe then.
Well, there's actually some things which the mmu-ful stuff does (such as sorting) which the nommu stuff should also do for safety sake, so I think a little bit of a cleanup in this area would be no bad thing, rather than trying to move arm_memblock_init() into both the MMU and noMMU code.