[v7,07/24] IB/umem: use get_user_pages_fast() to pin DMA pages
diff mbox series

Message ID 20191121071354.456618-8-jhubbard@nvidia.com
State New
Headers show
Series
  • mm/gup: track dma-pinned pages: FOLL_PIN
Related show

Commit Message

John Hubbard Nov. 21, 2019, 7:13 a.m. UTC
And get rid of the mmap_sem calls, as part of that. Note
that get_user_pages_fast() will, if necessary, fall back to
__gup_longterm_unlocked(), which takes the mmap_sem as needed.

Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
---
 drivers/infiniband/core/umem.c | 17 ++++++-----------
 1 file changed, 6 insertions(+), 11 deletions(-)

Comments

Christoph Hellwig Nov. 21, 2019, 8:07 a.m. UTC | #1
On Wed, Nov 20, 2019 at 11:13:37PM -0800, John Hubbard wrote:
> And get rid of the mmap_sem calls, as part of that. Note
> that get_user_pages_fast() will, if necessary, fall back to
> __gup_longterm_unlocked(), which takes the mmap_sem as needed.
> 
> Reviewed-by: Jan Kara <jack@suse.cz>
> Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
> Reviewed-by: Ira Weiny <ira.weiny@intel.com>
> Signed-off-by: John Hubbard <jhubbard@nvidia.com>

Looks fine,

Reviewed-by: Christoph Hellwig <hch@lst.de>

Jason, can you queue this up for 5.5 to reduce this patch stack a bit?
Jason Gunthorpe Nov. 21, 2019, 2:36 p.m. UTC | #2
On Thu, Nov 21, 2019 at 12:07:46AM -0800, Christoph Hellwig wrote:
> On Wed, Nov 20, 2019 at 11:13:37PM -0800, John Hubbard wrote:
> > And get rid of the mmap_sem calls, as part of that. Note
> > that get_user_pages_fast() will, if necessary, fall back to
> > __gup_longterm_unlocked(), which takes the mmap_sem as needed.
> > 
> > Reviewed-by: Jan Kara <jack@suse.cz>
> > Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
> > Reviewed-by: Ira Weiny <ira.weiny@intel.com>
> > Signed-off-by: John Hubbard <jhubbard@nvidia.com>
> 
> Looks fine,
> 
> Reviewed-by: Christoph Hellwig <hch@lst.de>
> 
> Jason, can you queue this up for 5.5 to reduce this patch stack a bit?

Yes, I said I'd do this in an earlier revision. Now that it is clear this
won't go through Andrew's tree, applied to rdma for-next

Thanks,
Jason
John Hubbard Nov. 24, 2019, 6:14 a.m. UTC | #3
On 11/21/19 6:36 AM, Jason Gunthorpe wrote:
> On Thu, Nov 21, 2019 at 12:07:46AM -0800, Christoph Hellwig wrote:
>> On Wed, Nov 20, 2019 at 11:13:37PM -0800, John Hubbard wrote:
>>> And get rid of the mmap_sem calls, as part of that. Note
>>> that get_user_pages_fast() will, if necessary, fall back to
>>> __gup_longterm_unlocked(), which takes the mmap_sem as needed.
>>>
>>> Reviewed-by: Jan Kara <jack@suse.cz>
>>> Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
>>> Reviewed-by: Ira Weiny <ira.weiny@intel.com>
>>> Signed-off-by: John Hubbard <jhubbard@nvidia.com>
>>
>> Looks fine,
>>
>> Reviewed-by: Christoph Hellwig <hch@lst.de>
>>
>> Jason, can you queue this up for 5.5 to reduce this patch stack a bit?
> 
> Yes, I said I'd do this in an earlier revision. Now that it is clear this
> won't go through Andrew's tree, applied to rdma for-next
> 

Great, I'll plan on it going up through that tree. To be clear, is it headed 
for:

    git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma.git for-next

?


thanks,
Leon Romanovsky Nov. 24, 2019, 10:07 a.m. UTC | #4
On Thu, Nov 21, 2019 at 10:36:43AM -0400, Jason Gunthorpe wrote:
> On Thu, Nov 21, 2019 at 12:07:46AM -0800, Christoph Hellwig wrote:
> > On Wed, Nov 20, 2019 at 11:13:37PM -0800, John Hubbard wrote:
> > > And get rid of the mmap_sem calls, as part of that. Note
> > > that get_user_pages_fast() will, if necessary, fall back to
> > > __gup_longterm_unlocked(), which takes the mmap_sem as needed.
> > >
> > > Reviewed-by: Jan Kara <jack@suse.cz>
> > > Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
> > > Reviewed-by: Ira Weiny <ira.weiny@intel.com>
> > > Signed-off-by: John Hubbard <jhubbard@nvidia.com>
> >
> > Looks fine,
> >
> > Reviewed-by: Christoph Hellwig <hch@lst.de>
> >
> > Jason, can you queue this up for 5.5 to reduce this patch stack a bit?
>
> Yes, I said I'd do this in an earlier revision. Now that it is clear this
> won't go through Andrew's tree, applied to rdma for-next

Jason,

This patch broke RDMA completely.
Change from get_user_pages() to get_user_pages_fast() causes to endless
amount of splats due to combination of the following code:

189 struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
190                             size_t size, int access)
...
263         if (!umem->writable)
264                 gup_flags |= FOLL_FORCE;
265

and

2398 int get_user_pages_fast(unsigned long start, int nr_pages,
2399                         unsigned int gup_flags, struct page **pages)
2400 {
2401         unsigned long addr, len, end;
2402         int nr = 0, ret = 0;
2403
2404         if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM)))
2405                 return -EINVAL;


[   72.623266] ------------[ cut here ]------------
[   72.624143] WARNING: CPU: 1 PID: 2557 at mm/gup.c:2404 get_user_pages_fast+0x115/0x180
[   72.625426] Modules linked in: xt_MASQUERADE iptable_nat nf_nat
   br_netfilter overlay ib_srp scsi_transport_srp rpcrdma rdma_ucm
   ib_iser libiscsi scsi_transport_iscsi rdma_cm ib_umad iw_cm ib_ipoib
   ib_cm mlx5_ib ib_uverbs ib_core mlx5_core mlxfw ptp pps_core
[   72.629024] CPU: 1 PID: 2557 Comm: python3 Not tainted 5.4.0-rc5-J6120-Geeb831ec5b80 #1
[   72.630435] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
[   72.631973] RIP: 0010:get_user_pages_fast+0x115/0x180
[   72.632873] Code: 8d 0c 10 85 c0 89 d0 0f 49 c1 eb 97 fa 4c 8d
       44 24 0c 4c 89 e9 44 89 e2 48 89 df e8 25 dd ff ff fb 8b 44 24 0c
       e9 75 ff ff ff <0f> 0b b8 ea ff ff ff e9 6d ff ff ff 65 4c 8b 2c
       25 00 5d 01 00 49
[   72.635967] RSP: 0018:ffffc90000edf930 EFLAGS: 00010202
[   72.636896] RAX: 0000000000000000 RBX: 00007f931c392000 RCX: ffff8883f4909000
[   72.638117] RDX: 0000000000010011 RSI: 0000000000000001 RDI: 00007f931c392000
[   72.639353] RBP: 00007f931c392000 R08: 0000000000000020 R09: ffff8883f4909000
[   72.640602] R10: 0000000000000000 R11: ffff8884068d6760 R12: ffff888409e45ba0
[   72.641858] R13: ffff8884068d6760 R14: 0000160000000000 R15: ffff8883f4909000
[   72.643115] FS:  00007f9323754700(0000) GS:ffff88842fa80000(0000) knlGS:0000000000000000
[   72.644586] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   72.645628] CR2: 00007f931c3ca000 CR3: 00000003f83d6006 CR4: 00000000007606a0
[   72.646875] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   72.648120] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   72.649374] PKRU: 55555554
[   72.649935] Call Trace:
[   72.650477]  ib_umem_get+0x298/0x550 [ib_uverbs]
[   72.651347]  mlx5_ib_db_map_user+0xad/0x130 [mlx5_ib]
[   72.652279]  mlx5_ib_create_cq+0x1e8/0xaa0 [mlx5_ib]
[   72.653207]  create_cq+0x1c8/0x2d0 [ib_uverbs]
[   72.654050]  ib_uverbs_create_cq+0x70/0xa0 [ib_uverbs]
[   72.654988]  ib_uverbs_handler_UVERBS_METHOD_INVOKE_WRITE+0xc2/0xf0 [ib_uverbs]
[   72.656331]  ib_uverbs_cmd_verbs.isra.6+0x5be/0xbe0 [ib_uverbs]
[   72.657398]  ? uverbs_disassociate_api+0xd0/0xd0 [ib_uverbs]
[   72.658423]  ? kvm_clock_get_cycles+0xd/0x10
[   72.659229]  ? kmem_cache_alloc+0x176/0x1c0
[   72.660025]  ? filemap_map_pages+0x18c/0x350
[   72.660838]  ib_uverbs_ioctl+0xc0/0x120 [ib_uverbs]
[   72.661756]  do_vfs_ioctl+0xa1/0x610
[   72.662458]  ksys_ioctl+0x70/0x80
[   72.663119]	__x64_sys_ioctl+0x16/0x20
[   72.663850]  do_syscall_64+0x42/0x110
[   72.664562]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
[   72.665492] RIP: 0033:0x7f9332958267
[   72.666185] Code: b3 66 90 48
	8b 05 19 3c 2c 00 64 c7 00 26 00
	00 00 48 c7 c0 ff ff ff ff c3 66
	2e 0f 1f 84 00 00 00 00 00 b8 10
	00 00 00 0f 05 <48> 3d 01 f0 ff
	ff 73 01 c3 48 8b 0d e9 3b 2c 00
	f7 d8 64 89 01 48
[   72.669347] RSP: 002b:00007f9323751928 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
[   72.670706] RAX: ffffffffffffffda RBX: 00007f93237519b8 RCX:	00007f9332958267
[   72.671937] RDX: 00007f93237519a0 RSI: 00000000c0181b01 RDI: 0000000000000008
[   72.673176] RBP: 00007f9323751980 R08: 0000000000000005 R09: 00007f931c3ffef0
[   72.674415] R10: 0000000000001000 R11: 0000000000000246 R12: 00007f931c400030
[   72.675653] R13: 00007f9323751980 R14: 00007f9323751ea8 R15: 00000000000000ff
[   72.676885] ---[ end trace 033465440517865c ]---

>
> Thanks,
> Jason
>
John Hubbard Nov. 25, 2019, 12:05 a.m. UTC | #5
On 11/24/19 2:07 AM, Leon Romanovsky wrote:
> On Thu, Nov 21, 2019 at 10:36:43AM -0400, Jason Gunthorpe wrote:
>> On Thu, Nov 21, 2019 at 12:07:46AM -0800, Christoph Hellwig wrote:
>>> On Wed, Nov 20, 2019 at 11:13:37PM -0800, John Hubbard wrote:
>>>> And get rid of the mmap_sem calls, as part of that. Note
>>>> that get_user_pages_fast() will, if necessary, fall back to
>>>> __gup_longterm_unlocked(), which takes the mmap_sem as needed.
>>>>
>>>> Reviewed-by: Jan Kara <jack@suse.cz>
>>>> Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
>>>> Reviewed-by: Ira Weiny <ira.weiny@intel.com>
>>>> Signed-off-by: John Hubbard <jhubbard@nvidia.com>
>>>
>>> Looks fine,
>>>
>>> Reviewed-by: Christoph Hellwig <hch@lst.de>
>>>
>>> Jason, can you queue this up for 5.5 to reduce this patch stack a bit?
>>
>> Yes, I said I'd do this in an earlier revision. Now that it is clear this
>> won't go through Andrew's tree, applied to rdma for-next
> 
> Jason,
> 
> This patch broke RDMA completely.
> Change from get_user_pages() to get_user_pages_fast() causes to endless
> amount of splats due to combination of the following code:
> 
> 189 struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
> 190                             size_t size, int access)
> ...
> 263         if (!umem->writable)
> 264                 gup_flags |= FOLL_FORCE;
> 265
> 
> and
> 
> 2398 int get_user_pages_fast(unsigned long start, int nr_pages,
> 2399                         unsigned int gup_flags, struct page **pages)
> 2400 {
> 2401         unsigned long addr, len, end;
> 2402         int nr = 0, ret = 0;
> 2403
> 2404         if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM)))
> 2405                 return -EINVAL;
> 

Hi Leon,

I looked into this, and I believe that the problem is in gup.c. There appears to
have been an oversight, in commit 817be129e6f2 ("mm: validate get_user_pages_fast
flags"), in filtering out FOLL_FORCE. There is nothing in the _fast() implementation
that requires that we avoid writing to the pages.

So I intend, to post a two-patch series that includes this fix, first, but maybe this
means that it should go in via -mm. I'm not sure what's the best tree.

diff --git a/mm/gup.c b/mm/gup.c
index 8f236a335ae9..745b4036cdfd 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -2401,7 +2401,8 @@ int get_user_pages_fast(unsigned long start, int nr_pages,
 	unsigned long addr, len, end;
 	int nr = 0, ret = 0;
 
-	if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM)))
+	if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM |
+				       FOLL_FORCE)))
 		return -EINVAL;
 
 	start = untagged_addr(start) & PAGE_MASK;


thanks,
Jason Gunthorpe Nov. 25, 2019, 12:53 a.m. UTC | #6
On Sun, Nov 24, 2019 at 04:05:16PM -0800, John Hubbard wrote:
 
> I looked into this, and I believe that the problem is in gup.c. There appears to
> have been an oversight, in commit 817be129e6f2 ("mm: validate get_user_pages_fast
> flags"), in filtering out FOLL_FORCE. There is nothing in the _fast() implementation
> that requires that we avoid writing to the pages.

I think it is too late to be doing these kinds of changes, I will
revert the patch and this will miss this merge window.

Jason

From ec6cb45292d21d1af9b9d95997b8cf204bbe854c Mon Sep 17 00:00:00 2001
From: Jason Gunthorpe <jgg@mellanox.com>
Date: Sun, 24 Nov 2019 20:47:59 -0400
Subject: [PATCH] Revert "IB/umem: use get_user_pages_fast() to pin DMA pages"

This reverts commit c9a7a2ed837c563f9f89743a6db732591cb4035b.

This was merged before enough testing was done, and it triggers a WARN_ON()
in get_user_pages_fast():

  WARNING: CPU: 1 PID: 2557 at mm/gup.c:2404 get_user_pages_fast+0x115/0x180
  Call Trace:
   ib_umem_get+0x298/0x550 [ib_uverbs]
   mlx5_ib_db_map_user+0xad/0x130 [mlx5_ib]
   mlx5_ib_create_cq+0x1e8/0xaa0 [mlx5_ib]
   create_cq+0x1c8/0x2d0 [ib_uverbs]
   ib_uverbs_create_cq+0x70/0xa0 [ib_uverbs]
   ib_uverbs_handler_UVERBS_METHOD_INVOKE_WRITE+0xc2/0xf0 [ib_uverbs]
   ib_uverbs_cmd_verbs.isra.6+0x5be/0xbe0 [ib_uverbs]
   ? uverbs_disassociate_api+0xd0/0xd0 [ib_uverbs]
   ? kvm_clock_get_cycles+0xd/0x10
   ? kmem_cache_alloc+0x176/0x1c0
   ? filemap_map_pages+0x18c/0x350
   ib_uverbs_ioctl+0xc0/0x120 [ib_uverbs]
   do_vfs_ioctl+0xa1/0x610
   ksys_ioctl+0x70/0x80
   __x64_sys_ioctl+0x16/0x20
   do_syscall_64+0x42/0x110
   entry_SYSCALL_64_after_hwframe+0x44/0xa9

2404         if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM)))
2405                 return -EINVAL;

While we think this WARN_ON is probably bogus, resolving this will have to
wait.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
---
 drivers/infiniband/core/umem.c | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)

diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
index 214e87aa609d6e..7a3b99597eada1 100644
--- a/drivers/infiniband/core/umem.c
+++ b/drivers/infiniband/core/umem.c
@@ -266,13 +266,16 @@ struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
 	sg = umem->sg_head.sgl;
 
 	while (npages) {
-		ret = get_user_pages_fast(cur_base,
-					  min_t(unsigned long, npages,
-						PAGE_SIZE /
-						sizeof(struct page *)),
-					  gup_flags | FOLL_LONGTERM, page_list);
-		if (ret < 0)
+		down_read(&mm->mmap_sem);
+		ret = get_user_pages(cur_base,
+				     min_t(unsigned long, npages,
+					   PAGE_SIZE / sizeof (struct page *)),
+				     gup_flags | FOLL_LONGTERM,
+				     page_list, NULL);
+		if (ret < 0) {
+			up_read(&mm->mmap_sem);
 			goto umem_release;
+		}
 
 		cur_base += ret * PAGE_SIZE;
 		npages   -= ret;
@@ -280,6 +283,8 @@ struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
 		sg = ib_umem_add_sg_table(sg, page_list, ret,
 			dma_get_max_seg_size(context->device->dma_device),
 			&umem->sg_nents);
+
+		up_read(&mm->mmap_sem);
 	}
 
 	sg_mark_end(sg);

Patch
diff mbox series

diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
index 24244a2f68cc..3d664a2539eb 100644
--- a/drivers/infiniband/core/umem.c
+++ b/drivers/infiniband/core/umem.c
@@ -271,16 +271,13 @@  struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
 	sg = umem->sg_head.sgl;
 
 	while (npages) {
-		down_read(&mm->mmap_sem);
-		ret = get_user_pages(cur_base,
-				     min_t(unsigned long, npages,
-					   PAGE_SIZE / sizeof (struct page *)),
-				     gup_flags | FOLL_LONGTERM,
-				     page_list, NULL);
-		if (ret < 0) {
-			up_read(&mm->mmap_sem);
+		ret = get_user_pages_fast(cur_base,
+					  min_t(unsigned long, npages,
+						PAGE_SIZE /
+						sizeof(struct page *)),
+					  gup_flags | FOLL_LONGTERM, page_list);
+		if (ret < 0)
 			goto umem_release;
-		}
 
 		cur_base += ret * PAGE_SIZE;
 		npages   -= ret;
@@ -288,8 +285,6 @@  struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
 		sg = ib_umem_add_sg_table(sg, page_list, ret,
 			dma_get_max_seg_size(context->device->dma_device),
 			&umem->sg_nents);
-
-		up_read(&mm->mmap_sem);
 	}
 
 	sg_mark_end(sg);