RE: DMA from high memory regions

RE: DMA from high memory regions

[Imran Badr]

Many thanks for the reply. It sure works and I noticed that code in
pci_map_page() api. Previously I was using kernel 2.4.2 fron RedHat7.1 which
has no such kernel function.
Now, if I need to access that page in my device driver, how can I get kernel
virtual address without using kmap?

Thanks,
Imran.

-----Original Message-----
From: Larry Sendlosky [mailto:Larry.Sendlosky@storigen.com]
Sent: Wednesday, June 26, 2002 6:33 AM
To: imran.badr@cavium.com
Subject: RE: DMA from high memory regions


There is no need to kmap.

1) call map_user_kiobuf to make sure buffer is mapped and pages
   are not swapped.

2) call lock_kiovec to lock the pages for I/O

3) use the struct page pointers in the kiobuf maplist
   to compute physical page address of each page.
   eg  ((struct_page_pointer - mem_map) << PAGE_SHIFT)   is physical
   address of the page described by 'struct page *struct_page_pointer'.

   You now have the physical addresses needed to give to the device
   for DMA.


-----Original Message-----
From: Imran Badr [mailto:imran.badr@cavium.com]
Sent: Tuesday, June 25, 2002 7:30 PM
To: 'lkml'
Subject: DMA from high memory regions


Hi,

I am trying to setup DMA to/from user space. I get a user pointer in my
device driver, from which I build up kiobuf and call map_user_kiobuf. After
this, I call kmap to map pages to kernel virtual space and then virt_to_bus
to get bus address. Now if I define CONFIG_HIGHMEM and that page happens to
be in the memory region near 1GB then DMA never happens and I donot see any
data in result pointer. If CONFIG_HIGHMEM is not defined, then everything
works perfectly. Please suggest any solution.

Thanks,
Imran.



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

RE: DMA from high memory regions, GFP_KERNEL and virt_to_bus

[Imran Badr]

I really appreciate your useful suggestions. I have another question about
kmalloc. Whenever kmalloc() is called with GFP_KERNEL flag, does it always
returns mapped memory pointer?  Is it guaranteed that virt_to_bus() is going
to return correct bus address for that memory pointer ?

Thanks,
Imran.


-----Original Message-----
From: linux-kernel-owner@vger.kernel.org
[mailto:linux-kernel-owner@vger.kernel.org]On Behalf Of Imran Badr
Sent: Wednesday, June 26, 2002 9:43 AM
To: 'Larry Sendlosky'; 'lkml'
Subject: RE: DMA from high memory regions



Many thanks for the reply. It sure works and I noticed that code in
pci_map_page() api. Previously I was using kernel 2.4.2 fron RedHat7.1 which
has no such kernel function.
Now, if I need to access that page in my device driver, how can I get kernel
virtual address without using kmap?

Thanks,
Imran.

-----Original Message-----
From: Larry Sendlosky [mailto:Larry.Sendlosky@storigen.com]
Sent: Wednesday, June 26, 2002 6:33 AM
To: imran.badr@cavium.com
Subject: RE: DMA from high memory regions


There is no need to kmap.

1) call map_user_kiobuf to make sure buffer is mapped and pages
   are not swapped.

2) call lock_kiovec to lock the pages for I/O

3) use the struct page pointers in the kiobuf maplist
   to compute physical page address of each page.
   eg  ((struct_page_pointer - mem_map) << PAGE_SHIFT)   is physical
   address of the page described by 'struct page *struct_page_pointer'.

   You now have the physical addresses needed to give to the device
   for DMA.


-----Original Message-----
From: Imran Badr [mailto:imran.badr@cavium.com]
Sent: Tuesday, June 25, 2002 7:30 PM
To: 'lkml'
Subject: DMA from high memory regions


Hi,

I am trying to setup DMA to/from user space. I get a user pointer in my
device driver, from which I build up kiobuf and call map_user_kiobuf. After
this, I call kmap to map pages to kernel virtual space and then virt_to_bus
to get bus address. Now if I define CONFIG_HIGHMEM and that page happens to
be in the memory region near 1GB then DMA never happens and I donot see any
data in result pointer. If CONFIG_HIGHMEM is not defined, then everything
works perfectly. Please suggest any solution.

Thanks,
Imran.



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Re: Linux 2.4.19-rc1

[Simon Kirby]

Hello,

This fix did not yet make it in to the tg3 driver as of -rc1.  This patch
fixes an occasional crash problem with several of our boxes and the tg3
driver (fix from Jes).

--

Hi Dave

Here's another one, tg3_recycle_rx() needs to declare dest_idx_unmasked
as unsigned int or it will do funny stuff on modular division once we go
above 31 bit values.

I don't think this is the bug that Scott was hitting, but it is of
course possible. It takes about 120 minutes for me with a packet
generator, spewing out 64 byte packets, to trigger it.

Cheers,
Jes

--- ../orig/drivers/net/tg3.c	Tue May 14 07:30:52 2002
+++ drivers/net/tg3.c	Fri Jun 14 11:20:48 2002
@@ -1689,7 +1669,7 @@
  * tg3_alloc_rx_skb for full details.
  */
 static void tg3_recycle_rx(struct tg3 *tp, u32 opaque_key,
-			   int src_idx, int dest_idx_unmasked)
+			   int src_idx, u32 dest_idx_unmasked)
 {
 	struct tg3_rx_buffer_desc *src_desc, *dest_desc;
 	struct ring_info *src_map, *dest_map;

--

Simon-

[  Stormix Technologies Inc.  ][  NetNation Communications Inc. ]
[       sim@stormix.com       ][       sim@netnation.com        ]
[ Opinions expressed are not necessarily those of my employers. ]

DMA from high memory regions

[Imran Badr]

Hi,

I am trying to setup DMA to/from user space. I get a user pointer in my
device driver, from which I build up kiobuf and call map_user_kiobuf. After
this, I call kmap to map pages to kernel virtual space and then virt_to_bus
to get bus address. Now if I define CONFIG_HIGHMEM and that page happens to
be in the memory region near 1GB then DMA never happens and I donot see any
data in result pointer. If CONFIG_HIGHMEM is not defined, then everything
works perfectly. Please suggest any solution.

Thanks,
Imran.