Hi Bob, > I think that's the normal case to reuse that page after its context have been > transmitted. > I'm looking at the block driver since it's a iSCSI block device , the > sequence is > like this: > 1. A user call get_page() and then call submit_bio() 2. Then the page data is > going to be transmitted through iSCSI and ixgbe. Fix me if I was wrong, I'm > not clear how ixgbe was involved. > 3. When the bio is completed, bio-> bi_end_io() function will be triggered > and call put_page(). I'm not sure about your implementation but there are plenty of other iSCSI components (targets and initiators) that work with our drivers today. So what you are doing in the application is doing something different. You may want to look for some of these other iSCSI components and look to see what they are doing regarding transmits. As I've said there is no way for the driver to indicate back up the stack to indicate that a transmit of any given packet has happened. Also, If you think about it how would the driver know that a packet is part of a certain page (or iSCSI frame) to indicate that information back up the stack? The base drivers have no mechanisms to do this. Since other iSCSI agents are implemented and running in the kernel without such a feature are all working this seems to be something in your implementation.
Please look around at other Linux iSCSI code to see what they are doing. Thanks. Cheers, John > -----Original Message----- > From: Bob Liu [mailto:lliu...@gmail.com] > Sent: Tuesday, March 10, 2015 7:26 AM > To: Ronciak, John > Cc: Kirsher, Jeffrey T; e1000-devel; Tantilov, Emil S > Subject: Re: [E1000-devel] ixgbe: kernel BUG at include/linux/mm.h > > On Tue, Mar 10, 2015 at 9:58 PM, Ronciak, John <john.ronc...@intel.com> > wrote: > >> -----Original Message----- > >> From: Jeff Kirsher [mailto:jeffrey.t.kirs...@intel.com] > >> Sent: Tuesday, March 10, 2015 6:38 AM > >> To: Bob Liu > >> Cc: e1000-devel; Tantilov, Emil S > >> Subject: Re: [E1000-devel] ixgbe: kernel BUG at include/linux/mm.h > >> > >> On Tue, 2015-03-10 at 21:29 +0800, Bob Liu wrote: > >> > > Would you mind sending your actual test cases to prove that the > >> > > page is free when calling _skb_frag_unref > >> > > >> > I'm afraid not, it's a complex testing program and difficult to run > >> > on external environment because of many dependencies. > >> > > >> > > from the ixgbe code. This seems like a viable area this would be > >> > going > >> > > wrong and we need synchronization for packets calling > >> > > dev_kfree_skb_any in ixgbe_clean_tx_irq if this > >> > is > >> > > happening. > >> > > >> > Since it's a iSCSI device, I'd like to know whether ixgbe driver > >> > will notifiy the upper layer "transmit package complete" before > >> > dev_kfree_skb_any()? > >> > > >> > The situation I'm worry about is like this: > >> > > >> > block driver: net driver: > >> > submit_bio() > >> > > >> > ixgbe_xmit_frame_ring() > >> > notify block driver > >> > tx done > >> > > >> > bio_done() > >> > { > >> > free_page() > >> > } > >> > > >> > dev_kfree_skb_any() > >> > ^^^ Here the page > >> > already freed by block driver. > >> > >> Bob- > >> I am sorry you had to deal with Nick. He is not an Intel developer, > >> even though he talks as if he was. He has now been banned from our > >> mailing list (this has been coming for some time now). > >> > >> In a few hours, most of the ixgbe developers will be in the office > >> and can assist you with your questions/inquiries. > >> > >> Cheers, > >> Jeff > > Hi Bob, > > > > Thanks for your reply, John. > > > There is really no way for a base Ethernet driver to do this. There is no > kernel interface for it. It's the drivers responsibility to free the skb and > any > associated buffers on transmit. So I have to ask, why are you freeing a > buffer > (page) that you have sent off to be transmitted? > > > > I think that's the normal case to reuse that page after its context have been > transmitted. > I'm looking at the block driver since it's a iSCSI block device , the > sequence is > like this: > 1. A user call get_page() and then call submit_bio() 2. Then the page data is > going to be transmitted through iSCSI and ixgbe. Fix me if I was wrong, I'm > not clear how ixgbe was involved. > 3. When the bio is completed, bio-> bi_end_io() function will be triggered > and call put_page(). > > -- > Regards, > --Bob ------------------------------------------------------------------------------ Dive into the World of Parallel Programming The Go Parallel Website, sponsored by Intel and developed in partnership with Slashdot Media, is your hub for all things parallel software development, from weekly thought leadership blogs to news, videos, case studies, tutorials and more. Take a look and join the conversation now. http://goparallel.sourceforge.net/ _______________________________________________ E1000-devel mailing list E1000-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/e1000-devel To learn more about Intel® Ethernet, visit http://communities.intel.com/community/wired
