If CONFIG_ARCH_DMA_ADDR_T_64BIT enabled for x86 systems and physical
memory is more than 4GB, dma_map_page may return a valid memory
address which greater than 0xffffffff. As a result, the mlx5 device page
allocator RB tree will be initialized with valid addresses greater than
0xfffffff.
However, (addr & PAGE_MASK) set the high four bytes to zeros. So, it's
impossible for the function, free_4k, to release the pages whose
addresses greater than 4GB. Memory leaks. And mlx5_ib module can't
release the pages when user try to remove the module, as a result,
system hang.
[root@rdma05 root]# dmesg | grep addr | head
addr = 3fe384000
addr & PAGE_MASK = fe384000
[root@rdma05 root]# rmmod mlx5_ib <---- hang on
---------------------- cosnole log -----------------
mlx5_ib 0000:04:00.0: irq 138 for MSI/MSI-X
alloc irq_desc for 139 on node -1
alloc kstat_irqs on node -1
mlx5_ib 0000:04:00.0: irq 139 for MSI/MSI-X
0000:04:00.0:free_4k:221:(pid 1519): page not found
0000:04:00.0:free_4k:221:(pid 1519): page not found
0000:04:00.0:free_4k:221:(pid 1519): page not found
0000:04:00.0:free_4k:221:(pid 1519): page not found
---------------------- cosnole log -----------------
Signed-off-by: Honggang Li <[email protected]>
---
drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c b/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c
index df22383..27c72da 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c
@@ -211,12 +211,14 @@ static int alloc_4k(struct mlx5_core_dev *dev, u64 *addr)
return 0;
}
+#define MLX5_U64_4K_PAGE_MASK ((~(u64)0U) << PAGE_SHIFT)
+
static void free_4k(struct mlx5_core_dev *dev, u64 addr)
{
struct fw_page *fwp;
int n;
- fwp = find_fw_page(dev, addr & PAGE_MASK);
+ fwp = find_fw_page(dev, addr & MLX5_U64_4K_PAGE_MASK);
if (!fwp) {
mlx5_core_warn(dev, "page not found\n");
return;
@@ -241,7 +243,7 @@ static void free_4k(struct mlx5_core_dev *dev, u64 addr)
static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id)
{
struct page *page;
- u64 addr;
+ u64 addr = 0;
int err;
int nid = dev_to_node(&dev->pdev->dev);
--
1.8.3.1
Good catch, thanks!
There are more places in this file where PAGE_MASK is wrongly used. Need to fix them as well.
Also, see below [Eli]
-----Original Message-----
From: Honggang Li [mailto:[email protected]]
Sent: Sunday, April 12, 2015 2:06 PM
To: Eli Cohen; [email protected]; [email protected]; [email protected]
Cc: Honggang Li
Subject: [linux-next PATCH] mlx5: wrong page mask if CONFIG_ARCH_DMA_ADDR_T_64BIT enabled for 32Bit architectures
If CONFIG_ARCH_DMA_ADDR_T_64BIT enabled for x86 systems and physical memory is more than 4GB, dma_map_page may return a valid memory address which greater than 0xffffffff. As a result, the mlx5 device page allocator RB tree will be initialized with valid addresses greater than 0xfffffff.
However, (addr & PAGE_MASK) set the high four bytes to zeros. So, it's impossible for the function, free_4k, to release the pages whose addresses greater than 4GB. Memory leaks. And mlx5_ib module can't release the pages when user try to remove the module, as a result, system hang.
[root@rdma05 root]# dmesg | grep addr | head
addr = 3fe384000
addr & PAGE_MASK = fe384000
[root@rdma05 root]# rmmod mlx5_ib <---- hang on
---------------------- cosnole log ----------------- mlx5_ib 0000:04:00.0: irq 138 for MSI/MSI-X
alloc irq_desc for 139 on node -1
alloc kstat_irqs on node -1
mlx5_ib 0000:04:00.0: irq 139 for MSI/MSI-X 0000:04:00.0:free_4k:221:(pid 1519): page not found 0000:04:00.0:free_4k:221:(pid 1519): page not found 0000:04:00.0:free_4k:221:(pid 1519): page not found 0000:04:00.0:free_4k:221:(pid 1519): page not found
---------------------- cosnole log -----------------
Signed-off-by: Honggang Li <[email protected]>
---
drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c b/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c
index df22383..27c72da 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c
@@ -211,12 +211,14 @@ static int alloc_4k(struct mlx5_core_dev *dev, u64 *addr)
return 0;
}
+#define MLX5_U64_4K_PAGE_MASK ((~(u64)0U) << PAGE_SHIFT)
+
static void free_4k(struct mlx5_core_dev *dev, u64 addr) {
struct fw_page *fwp;
int n;
- fwp = find_fw_page(dev, addr & PAGE_MASK);
+ fwp = find_fw_page(dev, addr & MLX5_U64_4K_PAGE_MASK);
if (!fwp) {
mlx5_core_warn(dev, "page not found\n");
return;
@@ -241,7 +243,7 @@ static void free_4k(struct mlx5_core_dev *dev, u64 addr) static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id) {
struct page *page;
- u64 addr;
+ u64 addr = 0; [Eli] Why is this required?
int err;
int nid = dev_to_node(&dev->pdev->dev);
--
1.8.3.1
On Sun, Apr 12, 2015 at 03:04:16PM +0000, Eli Cohen wrote:
> Good catch, thanks!
>
> There are more places in this file where PAGE_MASK is wrongly used. Need to fix them as well.
>
Yes, replaced all of the PAGE_MASK in the file. Please see attched new
patch.
> Also, see below [Eli]
>
> @@ -241,7 +243,7 @@ static void free_4k(struct mlx5_core_dev *dev, u64 addr) static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id) {
> struct page *page;
> - u64 addr;
> + u64 addr = 0; [Eli] Why is this required?
For 32bit architectures, if CONFIG_ARCH_DMA_ADDR_T_64BIT disabled and
physical memory is less than 4GB, dma_map_page will return an u32 integer
less than 0xffff_ffff. And if addr had been initialized with random
rubbish in the stack of alloc_system_page, the high four bytes is
unpredictable. The new mask, MLX5_NUM_4K_IN_PAGE, will reserve the high
four bytes. So, free_4k/find_fw_page will randomly failed.
thanks
> int err;
> int nid = dev_to_node(&dev->pdev->dev);
>
> --
> 1.8.3.1
>
On Mon, Apr 13, 2015 at 08:33:33AM +0800, Honggang LI wrote:
>
> Yes, replaced all of the PAGE_MASK in the file. Please see attched new
> patch.
>
I think you need to send the new patch inline in your email.
> >
> > @@ -241,7 +243,7 @@ static void free_4k(struct mlx5_core_dev *dev, u64 addr) static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id) {
> > struct page *page;
> > - u64 addr;
> > + u64 addr = 0; [Eli] Why is this required?
>
> For 32bit architectures, if CONFIG_ARCH_DMA_ADDR_T_64BIT disabled and
> physical memory is less than 4GB, dma_map_page will return an u32 integer
> less than 0xffff_ffff. And if addr had been initialized with random
> rubbish in the stack of alloc_system_page, the high four bytes is
> unpredictable. The new mask, MLX5_NUM_4K_IN_PAGE, will reserve the high
> four bytes. So, free_4k/find_fw_page will randomly failed.
>
Sorry, I still don't understand the issue here. MLX5_NUM_4K_IN_PAGE is
not a mask and will always get the correct value which is fairly
small.
On Mon, Apr 13, 2015 at 06:47:11AM +0000, Eli Cohen wrote:
> On Mon, Apr 13, 2015 at 08:33:33AM +0800, Honggang LI wrote:
> >
> > Yes, replaced all of the PAGE_MASK in the file. Please see attched new
> > patch.
> >
>
> I think you need to send the new patch inline in your email.
>
I will resent the patch.
> > >
> > > @@ -241,7 +243,7 @@ static void free_4k(struct mlx5_core_dev *dev, u64 addr) static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id) {
> > > struct page *page;
> > > - u64 addr;
> > > + u64 addr = 0; [Eli] Why is this required?
> >
> > For 32bit architectures, if CONFIG_ARCH_DMA_ADDR_T_64BIT disabled and
> > physical memory is less than 4GB, dma_map_page will return an u32 integer
> > less than 0xffff_ffff. And if addr had been initialized with random
> > rubbish in the stack of alloc_system_page, the high four bytes is
> > unpredictable. The new mask, MLX5_NUM_4K_IN_PAGE, will reserve the high
> > four bytes. So, free_4k/find_fw_page will randomly failed.
> >
>
> Sorry, I still don't understand the issue here. MLX5_NUM_4K_IN_PAGE is
> not a mask and will always get the correct value which is fairly
> small.
Sorry. It was a typo. It should be MLX5_U64_4K_PAGE_MASK.