thr3ads.net - Nouveau - [Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation [Mar 2013]

If this information is useful, please help other people find it:
Share via:

Marcin Slusarz

2013-Mar-05 18:57 UTC

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

Page tables on nv50 take 48kB, which can be hard to allocate in one piece.
Let's use vmalloc.

Signed-off-by: Marcin Slusarz <marcin.slusarz at gmail.com>
Cc: stable at vger.kernel.org [3.7+]
---
 drivers/gpu/drm/nouveau/core/subdev/vm/base.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
index 77c67fc..e66fb77 100644
--- a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
+++ b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
@@ -362,7 +362,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64
length,
 	vm->fpde = offset >> (vmm->pgt_bits + 12);
 	vm->lpde = (offset + length - 1) >> (vmm->pgt_bits + 12);
 
-	vm->pgt  = kcalloc(vm->lpde - vm->fpde + 1, sizeof(*vm->pgt),
GFP_KERNEL);
+	vm->pgt  = vzalloc((vm->lpde - vm->fpde + 1) * sizeof(*vm->pgt));
 	if (!vm->pgt) {
 		kfree(vm);
 		return -ENOMEM;
@@ -371,7 +371,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64
length,
 	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length
>> 12,
 			      block >> 12);
 	if (ret) {
-		kfree(vm->pgt);
+		vfree(vm->pgt);
 		kfree(vm);
 		return ret;
 	}
@@ -446,7 +446,7 @@ nouveau_vm_del(struct nouveau_vm *vm)
 	}
 
 	nouveau_mm_fini(&vm->mm);
-	kfree(vm->pgt);
+	vfree(vm->pgt);
 	kfree(vm);
 }
 
-- 
1.8.1.4

Marcin Slusarz

2013-Mar-19 18:18 UTC

head link

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

On Tue, Mar 05, 2013 at 07:57:08PM +0100, Marcin Slusarz
wrote:> Page tables on nv50 take 48kB, which can be hard to allocate in one piece.
> Let's use vmalloc.
> 
> Signed-off-by: Marcin Slusarz <marcin.slusarz at gmail.com>
> Cc: stable at vger.kernel.org [3.7+]
> ---
>  drivers/gpu/drm/nouveau/core/subdev/vm/base.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
> index 77c67fc..e66fb77 100644
> --- a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
> +++ b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
> @@ -362,7 +362,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64
offset, u64 length,
>  	vm->fpde = offset >> (vmm->pgt_bits + 12);
>  	vm->lpde = (offset + length - 1) >> (vmm->pgt_bits + 12);
>  
> -	vm->pgt  = kcalloc(vm->lpde - vm->fpde + 1, sizeof(*vm->pgt),
GFP_KERNEL);
> +	vm->pgt  = vzalloc((vm->lpde - vm->fpde + 1) *
sizeof(*vm->pgt));
>  	if (!vm->pgt) {
>  		kfree(vm);
>  		return -ENOMEM;
> @@ -371,7 +371,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64
offset, u64 length,
>  	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length
>> 12,
>  			      block >> 12);
>  	if (ret) {
> -		kfree(vm->pgt);
> +		vfree(vm->pgt);
>  		kfree(vm);
>  		return ret;
>  	}
> @@ -446,7 +446,7 @@ nouveau_vm_del(struct nouveau_vm *vm)
>  	}
>  
>  	nouveau_mm_fini(&vm->mm);
> -	kfree(vm->pgt);
> +	vfree(vm->pgt);
>  	kfree(vm);
>  }
>  
> -- 
What's up with this patch?

Marcin

Maarten Lankhorst

2013-Jun-11 08:58 UTC

head link

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

Op 05-03-13 19:57, Marcin Slusarz schreef:> Page tables on nv50 take 48kB, which can be hard to allocate in one piece.
> Let's use vmalloc.
>
> Signed-off-by: Marcin Slusarz <marcin.slusarz at gmail.com>
> Cc: stable at vger.kernel.org [3.7+]
> ---
>  drivers/gpu/drm/nouveau/core/subdev/vm/base.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
> index 77c67fc..e66fb77 100644
> --- a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
> +++ b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
> @@ -362,7 +362,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64
offset, u64 length,
>  	vm->fpde = offset >> (vmm->pgt_bits + 12);
>  	vm->lpde = (offset + length - 1) >> (vmm->pgt_bits + 12);
>  
> -	vm->pgt  = kcalloc(vm->lpde - vm->fpde + 1, sizeof(*vm->pgt),
GFP_KERNEL);
> +	vm->pgt  = vzalloc((vm->lpde - vm->fpde + 1) *
sizeof(*vm->pgt));
>  	if (!vm->pgt) {
>  		kfree(vm);
>  		return -ENOMEM;
> @@ -371,7 +371,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64
offset, u64 length,
>  	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length
>> 12,
>  			      block >> 12);
>  	if (ret) {
> -		kfree(vm->pgt);
> +		vfree(vm->pgt);
>  		kfree(vm);
>  		return ret;
>  	}
> @@ -446,7 +446,7 @@ nouveau_vm_del(struct nouveau_vm *vm)
>  	}
>  
>  	nouveau_mm_fini(&vm->mm);
> -	kfree(vm->pgt);
> +	vfree(vm->pgt);
>  	kfree(vm);
>  }Could this patch be upstreamed?

I was hitting the same allocation failure on my fermi after keeping my system
running for a
a week, and doing plenty of suspend/resume cycles (16+) in between it was
failing an order 6
allocation, which I assume means it was attempting to allocate 256 kilobyte of
physical memory
contiguously. My system has 16 gigabyte of ram, and had about 4 gb free
afterwards when I
checked, so it wasn't like there was not enough memory. It was probably just
too fragmented
to allocate with kmalloc, while vmalloc doesn't require the memory to be
physicallly contiguous.

~Maarten

Zlatko Calusic

2013-Jun-11 09:51 UTC

head link

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

On 11.06.2013 10:58, Maarten Lankhorst wrote:> Op 05-03-13 19:57, Marcin Slusarz schreef:
>> Page tables on nv50 take 48kB, which can be hard to allocate in one
piece.
>> Let's use vmalloc.
>>
>> Signed-off-by: Marcin Slusarz <marcin.slusarz at gmail.com>
>> Cc: stable at vger.kernel.org [3.7+]
>> ---
>>   drivers/gpu/drm/nouveau/core/subdev/vm/base.c | 6 +++---
>>   1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
>> index 77c67fc..e66fb77 100644
>> --- a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
>> +++ b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
>> @@ -362,7 +362,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64
offset, u64 length,
>>   	vm->fpde = offset >> (vmm->pgt_bits + 12);
>>   	vm->lpde = (offset + length - 1) >> (vmm->pgt_bits +
12);
>>   
>> -	vm->pgt  = kcalloc(vm->lpde - vm->fpde + 1,
sizeof(*vm->pgt), GFP_KERNEL);
>> +	vm->pgt  = vzalloc((vm->lpde - vm->fpde + 1) *
sizeof(*vm->pgt));
>>   	if (!vm->pgt) {
>>   		kfree(vm);
>>   		return -ENOMEM;
>> @@ -371,7 +371,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64
offset, u64 length,
>>   	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12,
mm_length >> 12,
>>   			      block >> 12);
>>   	if (ret) {
>> -		kfree(vm->pgt);
>> +		vfree(vm->pgt);
>>   		kfree(vm);
>>   		return ret;
>>   	}
>> @@ -446,7 +446,7 @@ nouveau_vm_del(struct nouveau_vm *vm)
>>   	}
>>   
>>   	nouveau_mm_fini(&vm->mm);
>> -	kfree(vm->pgt);
>> +	vfree(vm->pgt);
>>   	kfree(vm);
>>   }
> Could this patch be upstreamed?
> 
> I was hitting the same allocation failure on my fermi after keeping my
system running for a
> a week, and doing plenty of suspend/resume cycles (16+) in between it was
failing an order 6
> allocation, which I assume means it was attempting to allocate 256 kilobyte
of physical memory
> contiguously. My system has 16 gigabyte of ram, and had about 4 gb free
afterwards when I
> checked, so it wasn't like there was not enough memory. It was probably
just too fragmented
> to allocate with kmalloc, while vmalloc doesn't require the memory to
be physicallly contiguous.
> 
Same thing here. Except I don't do suspend/resume. Occasionally I get this:

Jun  8 22:29:15 atlas kernel: iceweasel: page allocation failure: order:4,
mode:0x10c0d0
Jun  8 22:29:15 atlas kernel: CPU: 1 PID: 15976 Comm: iceweasel Not tainted
3.10.0-rc4+ #1
Jun  8 22:29:15 atlas kernel: Hardware name: System manufacturer System Product
Name/P5Q PRO TURBO, BIOS 0602    08/04/2009
Jun  8 22:29:15 atlas kernel: 0000000000000000 ffff8800190a5920 ffffffff81574002
ffff8800190a59a8
Jun  8 22:29:15 atlas kernel: ffffffff810d29b0 0000000000000380 0000000000000010
0000000000000010
Jun  8 22:29:15 atlas kernel: ffff8800190a5978 00000000000003c0 0000001000000001
0000000000000002
Jun  8 22:29:15 atlas kernel: Call Trace:
Jun  8 22:29:15 atlas kernel: [<ffffffff81574002>] dump_stack+0x19/0x1b
Jun  8 22:29:15 atlas kernel: [<ffffffff810d29b0>]
warn_alloc_failed+0xf0/0x140
Jun  8 22:29:15 atlas kernel: [<ffffffff810d4a1a>]
__alloc_pages_nodemask+0x17a/0x6a0
Jun  8 22:29:15 atlas kernel: [<ffffffff810d4f52>]
__get_free_pages+0x12/0x50
Jun  8 22:29:15 atlas kernel: [<ffffffff810fd8ed>] __kmalloc+0xed/0x140
Jun  8 22:29:15 atlas kernel: [<ffffffff8130a9ce>]
nouveau_vm_create+0xae/0x140
Jun  8 22:29:15 atlas kernel: [<ffffffff8130c1db>]
nv50_vm_create+0x2b/0x30
Jun  8 22:29:15 atlas kernel: [<ffffffff8130aa8a>]
nouveau_vm_new+0x2a/0x30
Jun  8 22:29:15 atlas kernel: [<ffffffff81363769>]
nouveau_drm_open+0xc9/0x150
Jun  8 22:29:15 atlas kernel: [<ffffffff81070719>] ? ns_capable+0x29/0x50
Jun  8 22:29:15 atlas kernel: [<ffffffff812c7173>] drm_open+0x283/0x6e0
Jun  8 22:29:15 atlas kernel: [<ffffffff812c76d8>]
drm_stub_open+0x108/0x1a0
Jun  8 22:29:15 atlas kernel: [<ffffffff81104f46>] chrdev_open+0x96/0x1c0
Jun  8 22:29:15 atlas kernel: [<ffffffff8110b2f2>] ?
generic_permission+0xe2/0x100
Jun  8 22:29:15 atlas kernel: [<ffffffff81104eb0>] ? cdev_put+0x20/0x20
Jun  8 22:29:15 atlas kernel: [<ffffffff810feffe>]
do_dentry_open.isra.17+0x1ee/0x280
Jun  8 22:29:15 atlas kernel: [<ffffffff810ff179>] finish_open+0x19/0x30
Jun  8 22:29:15 atlas kernel: [<ffffffff8110e1be>]
do_last.isra.65+0x26e/0xc30
Jun  8 22:29:15 atlas kernel: [<ffffffff8110b503>] ?
inode_permission+0x13/0x50
Jun  8 22:29:15 atlas kernel: [<ffffffff8110b808>] ?
link_path_walk+0x68/0x8a0
Jun  8 22:29:15 atlas kernel: [<ffffffff8110ec2e>]
path_openat.isra.66+0xae/0x480
Jun  8 22:29:15 atlas kernel: [<ffffffff8110f34c>] do_filp_open+0x3c/0x90
Jun  8 22:29:15 atlas kernel: [<ffffffff8111b52b>] ? __alloc_fd+0xcb/0x120
Jun  8 22:29:15 atlas kernel: [<ffffffff8110025f>] do_sys_open+0xef/0x1d0
Jun  8 22:29:15 atlas kernel: [<ffffffff8110035d>] SyS_open+0x1d/0x20
Jun  8 22:29:15 atlas kernel: [<ffffffff81579492>]
system_call_fastpath+0x16/0x1b
Jun  8 22:29:15 atlas kernel: Mem-Info:
Jun  8 22:29:15 atlas kernel: DMA per-cpu:
Jun  8 22:29:15 atlas kernel: CPU    0: hi:    0, btch:   1 usd:   0
Jun  8 22:29:15 atlas kernel: CPU    1: hi:    0, btch:   1 usd:   0
Jun  8 22:29:15 atlas kernel: DMA32 per-cpu:
Jun  8 22:29:15 atlas kernel: CPU    0: hi:  186, btch:  31 usd: 108
Jun  8 22:29:15 atlas kernel: CPU    1: hi:  186, btch:  31 usd:   0
Jun  8 22:29:15 atlas kernel: Normal per-cpu:
Jun  8 22:29:15 atlas kernel: CPU    0: hi:  186, btch:  31 usd:  96
Jun  8 22:29:15 atlas kernel: CPU    1: hi:  186, btch:  31 usd:   0
Jun  8 22:29:15 atlas kernel: active_anon:884379 inactive_anon:7116
isolated_anon:0
Jun  8 22:29:15 atlas kernel: active_file:444168 inactive_file:483902
isolated_file:32
Jun  8 22:29:15 atlas kernel: unevictable:10 dirty:3173 writeback:0 unstable:0
Jun  8 22:29:15 atlas kernel: free:103104 slab_reclaimable:41561
slab_unreclaimable:8207
Jun  8 22:29:15 atlas kernel: mapped:41417 shmem:11137 pagetables:8576 bounce:0
Jun  8 22:29:15 atlas kernel: free_cma:0
Jun  8 22:29:15 atlas kernel: DMA free:15888kB min:20kB low:24kB high:28kB
active_anon:0kB inactive_anon:0kB active_file:0kB inactive_file:0kB
unevictable:0kB isolated(anon):0kB isolated(file):0kB present:15980kB
managed:15896kB mlocked:0kB dirty:0kB writeback:0kB mapped:0kB shmem:0kB
slab_reclaimable:0kB slab_unreclaimable:8kB kernel_stack:0kB pagetables:0kB
unstable:0kB bounce:0kB free_cma:0kB writeback_tmp:0kB pages_scanned:0
all_unreclaimable? yes
Jun  8 22:29:15 atlas kernel: lowmem_reserve[]: 0 3237 7960 7960
Jun  8 22:29:15 atlas kernel: DMA32 free:387096kB min:4636kB low:5792kB
high:6952kB active_anon:1002804kB inactive_anon:12248kB active_file:823744kB
inactive_file:886900kB unevictable:8kB isolated(anon):0kB isolated(file):0kB
present:3390912kB managed:3315368kB mlocked:8kB dirty:5264kB writeback:0kB
mapped:31200kB shmem:20504kB slab_reclaimable:92620kB slab_unreclaimable:8764kB
kernel_stack:352kB pagetables:4704kB unstable:0kB bounce:0kB free_cma:0kB
writeback_tmp:0kB pages_scanned:108 all_unreclaimable? no
Jun  8 22:29:15 atlas kernel: lowmem_reserve[]: 0 0 4723 4723
Jun  8 22:29:15 atlas kernel: Normal free:9432kB min:6764kB low:8452kB
high:10144kB active_anon:2534712kB inactive_anon:16216kB active_file:952928kB
inactive_file:1048580kB unevictable:32kB isolated(anon):0kB isolated(file):128kB
present:4980736kB managed:4836504kB mlocked:32kB dirty:7428kB writeback:0kB
mapped:134468kB shmem:24044kB slab_reclaimable:73624kB
slab_unreclaimable:24056kB kernel_stack:3584kB pagetables:29600kB unstable:0kB
bounce:0kB free_cma:0kB writeback_tmp:0kB pages_scanned:128 all_unreclaimable?
no
Jun  8 22:29:15 atlas kernel: lowmem_reserve[]: 0 0 0 0
Jun  8 22:29:15 atlas kernel: DMA: 0*4kB 0*8kB 1*16kB (U) 0*32kB 2*64kB (U)
1*128kB (U) 1*256kB (U) 0*512kB 1*1024kB (U) 1*2048kB (U) 3*4096kB (MR) =
15888kB
Jun  8 22:29:15 atlas kernel: DMA32: 7294*4kB (EM) 34492*8kB (UEM) 5096*16kB
(UEM) 1*32kB (M) 1*64kB (R) 0*128kB 0*256kB 1*512kB (R) 0*1024kB 0*2048kB
0*4096kB = 387256kB
Jun  8 22:29:15 atlas kernel: Normal: 1787*4kB (UEM) 60*8kB (UEM) 13*16kB (UM)
1*32kB (R) 5*64kB (R) 3*128kB (R) 0*256kB 0*512kB 1*1024kB (R) 0*2048kB 0*4096kB
= 9596kB
Jun  8 22:29:15 atlas kernel: 939160 total pagecache pages
Jun  8 22:29:15 atlas kernel: 2097151 pages RAM
Jun  8 22:29:15 atlas kernel: 54796 pages reserved
Jun  8 22:29:15 atlas kernel: 1246398 pages shared
Jun  8 22:29:15 atlas kernel: 1323236 pages non-shared

-- 
Zlatko

Marcin Ślusarz

2013-Jun-12 20:15 UTC

head link

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

2013/6/11 Maarten Lankhorst <maarten.lankhorst at
canonical.com>:> Op 05-03-13 19:57, Marcin Slusarz schreef:
>> Page tables on nv50 take 48kB, which can be hard to allocate in one
piece.
>> Let's use vmalloc.
>>
>> Signed-off-by: Marcin Slusarz <marcin.slusarz at gmail.com>
>> Cc: stable at vger.kernel.org [3.7+]
>> ---
>>  drivers/gpu/drm/nouveau/core/subdev/vm/base.c | 6 +++---
>>  1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
>> index 77c67fc..e66fb77 100644
>> --- a/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
>> +++ b/drivers/gpu/drm/nouveau/core/subdev/vm/base.c
>> @@ -362,7 +362,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64
offset, u64 length,
>>       vm->fpde = offset >> (vmm->pgt_bits + 12);
>>       vm->lpde = (offset + length - 1) >> (vmm->pgt_bits +
12);
>>
>> -     vm->pgt  = kcalloc(vm->lpde - vm->fpde + 1,
sizeof(*vm->pgt), GFP_KERNEL);
>> +     vm->pgt  = vzalloc((vm->lpde - vm->fpde + 1) *
sizeof(*vm->pgt));
>>       if (!vm->pgt) {
>>               kfree(vm);
>>               return -ENOMEM;
>> @@ -371,7 +371,7 @@ nouveau_vm_create(struct nouveau_vmmgr *vmm, u64
offset, u64 length,
>>       ret = nouveau_mm_init(&vm->mm, mm_offset >> 12,
mm_length >> 12,
>>                             block >> 12);
>>       if (ret) {
>> -             kfree(vm->pgt);
>> +             vfree(vm->pgt);
>>               kfree(vm);
>>               return ret;
>>       }
>> @@ -446,7 +446,7 @@ nouveau_vm_del(struct nouveau_vm *vm)
>>       }
>>
>>       nouveau_mm_fini(&vm->mm);
>> -     kfree(vm->pgt);
>> +     vfree(vm->pgt);
>>       kfree(vm);
>>  }
> Could this patch be upstreamed?
Why do you ask me? I sent it to the Nouveau mailing list more than 3 months ago,
pinged Ben after 2 weeks, and as USUAL didn't get any response.
Have fun.

Marcin

Possibly Parallel Threads

Search for more apparently analagous threads

Nouveau - Mar 2013 - [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

[Nouveau] [RFC PATCH] drm/nouveau: use vmalloc for pgt allocation

Possibly Parallel Threads