- kernelnewbies@kernelnewbies.org - Kernelnewbies Mailing Lists

Raw Sockets + Userland TCP Kernel Resetting Packet
by Bradley Falzon Nov. 25, 2015

Nov. 25, 2015

Hi, I'm writing a program to assist in learning/debugging tcp stacks, it creates a raw socket and replicates an exceedingly simple tcp stack. By which I mean, it sends and receives packets, setting correct flags, payloads etc. One difficulty I've come across, after I've created a raw socket and sent my syn packet, when the syn+ack packet arrives from the peer, my program receives it fine - but the kernel also resets the connection. Example tcpdump dump where 192.168.1.12 is the program and 192.168.1.1 is the remote peer I'm trying to establish a connection with: 08:33:36.660321 IP 192.168.1.12.47775 > 192.168.1.1.80: Flags [S], seq 1023, win 0, length 0 08:33:36.660665 IP 192.168.1.1.80 > 192.168.1.12.47775: Flags [S.], seq 640280152, ack 1024, win 14600, options [mss 1460], length 0 08:33:36.660682 IP 192.168.1.12.47775 > 192.168.1.1.80: Flags [R], seq 1024, win 0, length 0 In previous iterations of this, I've simply created a iptables rules entry to drop this packet, but I feel there might be a better way. Others have mentioned to also bind to the socket, but this doesn't appear to be working. I've been trying to trace the path through the kernel to find out if there's any support for what I'm trying to do (effectively discard the packet). The code itself is written in Go, but the following are the syscalls I'm using (output from systrace): socket(PF_INET, SOCK_RAW|SOCK_CLOEXEC|SOCK_NONBLOCK, IPPROTO_TCP) = 5 setsockopt(5, SOL_SOCKET, SO_BROADCAST, [1], 4) = 0 bind(5, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("192.168.1.12")}, 16) = 0 I don't believe the bind call is actually required, and I've used hping utility which uses socket(PF_PACKET, SOCK_RAW, 768) and exhibits the same behaviour (syn+ack is reset). Also, I'm not debating the behaviour of the Kernel, just whether I can do anything to change it. Is there another option, or if this is too far off topic another place I could ask? Thanks -- Bradley Falzon brad(a)teambrad.net

2 1

dumping large buffers to file
by Ramon Fried Nov. 24, 2015

Nov. 24, 2015

Hi. I'm currently debugging a ISP DMA issue in a kernel module. For debugging, I would like to dump buffers to a file on run time. The buffers are quite big, 1MB at least. As saving a file directly from kernel is discouraged, is there any available framework for doing such a thing ? I can think of numerous ways of transferring the buffer to user space and saving it by a dedicated user process (Netlink, sysfs, char device, etc.) I just don't want to invent the wheel. How do you do it ? Thanks. Ramon

3 2

Accessing novel DMAd hardware from userspace
by Henry Gomersall Nov. 21, 2015

Nov. 21, 2015

I asked this question on IRC, but my experience has been that IRC is variable in whether the right person sees the message, so I apologise for potentially breaking etiquette. I've recently been working with a Xilinx Zynq FPGA/ARM combined device, in which I wish to use DMA to transfer data from the FPGA side to the ARM side. Now, this is well supported at the kernel level. Xilinx have a driver that uses with the DMA engine subsystem In this email thread: http://www.spinics.net/lists/dmaengin/msg04599.html there was discussion about writing a userspace facing driver for presenting the DMAd data and the transactions. There wasn't much discussion, but one criticism seemed to be that there was no place for userspace facing DMA access and that there should be a driver that encapsulates each device that uses the relevant subsystem. The problem is, there isn't always a subsystem that fits every weird and wonderful new bit of hardware that someone might create, which is a pretty huge space of things when using FPGAs. What is the preferred way to solve this problem? It was suggested to me that the uio subsystem is the way to go, and I'm aware of this approach, but I don't fully understand how to use this. Does the uio handle the DMA through the DMA engine subsystem, or do I need to handle the DMA from userspace? If so, can I add this to the uio driver (to gain the benefits of the DMA engine). Is there a better way to present entirely new bits of hardware (and specifically hardware accessed through DMA) to userspace? Thanks, Henry

1 0

Accessing novel DMAd hardware from userspace
by Henry Gomersall Nov. 21, 2015

Nov. 21, 2015

I asked this question on IRC, but my experience has been that IRC is variable in its reliability, so I apologise for potentially breaking etiquette. I've recently been working with a Xilinx Zynq FPGA/ARM combined device, in which I wish to use DMA to transfer data from the FPGA side to the ARM side. Now, this is well supported at the kernel level. Xilinx have a driver that uses with the DMA engine subsystem In this email thread: http://www.spinics.net/lists/dmaengin/msg04599.html there was discussion about writing a userspace facing driver for presenting the DMAd data and the transactions. There wasn't much discussion, but one criticism seemed to be that there was no place for userspace facing DMA access and that there should be a driver that encapsulates each device that uses the relevant subsystem. The problem is, there isn't always a subsystem that fits every weird and wonderful new bit of hardware that someone might create, which is a pretty huge space of things when using FPGAs. What is the preferred way to solve this problem? It was suggested to me that the uio subsystem is the way to go, and I'm aware of this approach, but I don't fully understand how to use this. Does the uio handle the DMA through the DMA engine subsystem, or do I need to handle the DMA from userspace? If so, can I add this to the uio driver (to gain the benefits of the DMA engine). Is there a better way to present entirely new bits of hardware (and specifically hardware accessed through DMA) to userspace? Thanks, Henry

1 0

viability of supporting interrupts in user space driver?
by Robert P. J. Day Nov. 20, 2015

Nov. 20, 2015

some colleagues are considering implementing some user space drivers that need to recognize and process interrupts. i'm aware of the possibilities -- UIO, for example -- but other than the technical possibility of doing that, can anyone provide feedback on the viability or benchmarking of that? is there an example someone can point at that demonstrates the goodness or not of such an approach? has anyone out there done this, and lived to regret it in terms of performance? thanks. rday -- ======================================================================== Robert P. J. Day Ottawa, Ontario, CANADA http://crashcourse.ca Twitter: http://twitter.com/rpjday LinkedIn: http://ca.linkedin.com/in/rpjday ========================================================================

2 2

Correctly locking a Block Device Request Handler
by Marcel Müller Nov. 19, 2015

Nov. 19, 2015

Hello everyone, I'm currently writing a block device driver and got stuck at trying to understand how to correctly handle the locking in the reqfn one passes to `blk_init_queue`. My code looks like this: static DEFINE_SPINLOCK(rblk_lock); /* Code */ static void rblk_request_handler(struct request_queue *q) __releases(q->queue_lock) __acquires(q->queue_lock) { struct request* req; unsigned long flags = 0; printk(KERN_INFO "rblk: Got request(s) \n"); while ((req = blk_fetch_request(q)) != NULL) { printk(KERN_INFO "rblk: Handling request \n"); // <- Gets printed spin_lock_irqsave(q->queue_lock, flags); blk_end_request_all(req, -ENOTTY); spin_unlock_irqrestore(q->queue_lock, flags); printk(KERN_INFO "rblk: Handled request \n"); // <- Does not get printed } } static in rblk_init() { /* Get major number, allocate devices */ for (i = 0; i < rblk_cnt; i++) { // For each device /* alloc_disk, check for allocation fail */ disk->queue = blk_init_queue(rblk_request_handler, &rblk_lock); } } This didn't work, and it was obvious to me that it was hanging in the spinlock. So I tried removing the locking, (this SO answer says that the queue is already locked: https://stackoverflow.com/questions/19418979/proper-way-to-lock-the-queue-i…) however, I still get the problem that the queue locks up and the second message never hits the message queue. What /does/ work is if I invert the order of locking. As in, unlock first, end_request and then lock again. However that doesn't seem to be the correct way. What am I doing completely wrong, what did I misunderstand? Full code: https://gist.github.com/TheNeikos/8798788defa1a9f316e6 Thanks

2 2

[PATCH v4 4/4] devicetree: update documentation for fw_cfg ARM bindings
by Gabriel L. Somlo Nov. 18, 2015

Nov. 18, 2015

From: Gabriel Somlo <somlo(a)cmu.edu> Remove redundant details from Documentation/devicetree/bindings/arm/fw-cfg.txt, and replace them with a pointer to the more comprehensive fw_cfg documentation privided by Documentation/ABI/testing/sysfs-firmware-qemu_fw_cfg, leaving the specific ARM DTB node description in place. Signed-off-by: Gabriel Somlo <somlo(a)cmu.edu> Cc: Laszlo Ersek <lersek(a)redhat.com> --- Documentation/devicetree/bindings/arm/fw-cfg.txt | 37 ++---------------------- 1 file changed, 2 insertions(+), 35 deletions(-) diff --git a/Documentation/devicetree/bindings/arm/fw-cfg.txt b/Documentation/devicetree/bindings/arm/fw-cfg.txt index 953fb64..7aeb48a 100644 --- a/Documentation/devicetree/bindings/arm/fw-cfg.txt +++ b/Documentation/devicetree/bindings/arm/fw-cfg.txt @@ -11,43 +11,10 @@ QEMU exposes the control and data register to ARM guests as memory mapped registers; their location is communicated to the guest's UEFI firmware in the DTB that QEMU places at the bottom of the guest's DRAM. -The guest writes a selector value (a key) to the selector register, and then -can read the corresponding data (produced by QEMU) via the data register. If -the selected entry is writable, the guest can rewrite it through the data -register. -The selector register takes keys in big endian byte order. +For a comprehensive description of the behavior of fw_cfg, please see +Documentation/ABI/testing/sysfs-firmware-qemu_fw_cfg. -The data register allows accesses with 8, 16, 32 and 64-bit width (only at -offset 0 of the register). Accesses larger than a byte are interpreted as -arrays, bundled together only for better performance. The bytes constituting -such a word, in increasing address order, correspond to the bytes that would -have been transferred by byte-wide accesses in chronological order. - -The interface allows guest firmware to download various parameters and blobs -that affect how the firmware works and what tables it installs for the guest -OS. For example, boot order of devices, ACPI tables, SMBIOS tables, kernel and -initrd images for direct kernel booting, virtual machine UUID, SMP information, -virtual NUMA topology, and so on. - -The authoritative registry of the valid selector values and their meanings is -the QEMU source code; the structure of the data blobs corresponding to the -individual key values is also defined in the QEMU source code. - -The presence of the registers can be verified by selecting the "signature" blob -with key 0x0000, and reading four bytes from the data register. The returned -signature is "QEMU". - -The outermost protocol (involving the write / read sequences of the control and -data registers) is expected to be versioned, and/or described by feature bits. -The interface revision / feature bitmap can be retrieved with key 0x0001. The -blob to be read from the data register has size 4, and it is to be interpreted -as a uint32_t value in little endian byte order. The current value -(corresponding to the above outer protocol) is zero. - -The guest kernel is not expected to use these registers (although it is -certainly allowed to); the device tree bindings are documented here because -this is where device tree bindings reside in general. Required properties: -- 2.4.3

4 7

depmod: ERROR: Found 2 modules in dependency cycles!
by Lucas Tanure Nov. 18, 2015

Nov. 18, 2015

Hi, I'm trying to compile and run linux-next tree, but no matter what I do I got : depmod: ERROR: Found 2 modules in dependency cycles! depmod: ERROR: Cycle detected: lnet -> libcfs -> lnet Makefile:1139: recipe for target '_modinst_post' failed make: *** [_modinst_post] Error 1 Can anyone give a way to fix this ? I already re-cloned the tree, tried different .config files, tried with the stable branch. But nothing works. Thanks -- Lucas Tanure +55 (19) 988176559

2 1

hi, i want to use a script in kernel
by Naver Nov. 17, 2015

Nov. 17, 2015

hi, i want to use a script in kernel, but i can't use this. in my terminal. yoon@pc:~/project/aosp/kernel/linux/scripts$ ./checkpatch.pl sortextable.c Must be run from the top-level dir. of a kernel tree Please let me know how to use this script. Thanks.

7 6

looking for decent, current online discussion of user-space drivers
by Robert P. J. Day Nov. 17, 2015

Nov. 17, 2015

a colleague asks me about the implications of porting a pile of code from a non-linux embedded OS to linux, and simply rewriting most of it as user-space drivers, as opposed to a total rewrite to kernel code. i'm in no way an authority on user-space drivers, so i'm looking for any decent, up-to-date discussions, not so much on the technical aspects, but more the *implications* of doing this. i'm aware of mmap(), and i'm reading up on UIO, and i've found a few online pieces that look informative: http://www.embedded.com/design/operating-systems/4401769/Device-drivers-in-… http://www.enea.com/Documents/Resources/Whitepapers/Enea-User-Space-Drivers… so is there something that elaborates on both the pros and cons of doing this? in particular, are there any nice examples of this that can be downloaded, built and played with? thanks muchly. rday -- ======================================================================== Robert P. J. Day Ottawa, Ontario, CANADA http://crashcourse.ca Twitter: http://twitter.com/rpjday LinkedIn: http://ca.linkedin.com/in/rpjday ========================================================================

2 2