November 2015 - kernelnewbies@kernelnewbies.org

How to handle Hotplug with UIO userspace driver
by Mandeep Sandhu Sept. 28, 2017

Sept. 28, 2017

Hi All, We're working on a custom PCIe based hardware, which can be hotplugged into the system. We're using the UIO approach of having a minimal kernel driver which basically handles the interrupt and provides a mmap of the device memory to userspace and userspace driver which mmap's the provided memory and manages the device. We're using systemd and udev rules to spawn off a user process whenever the hardware is plugged into the system. Now my question is, how can we gracefully inform the userspace process if someone pulls out the hardware (sitting on a PCI slot). Even with full hotplug support, there can always be a case of "surprise removal", where someone yanks the board without going through the whole "attention button" routine. While the kernel driver has no problem with hot-removal, I see that if I "unregister" our UIO driver in response to the removal, then the mapping for the hardware's physical memory would become "invalid" (done by UIO on unregsiter). This could possibly cause the user-space process to crash due to illegal memory access. Is there a "standard" way to handle this scenario, i.e for hotpluggable hardware using UIO? Basically, we'd want to "unregister" the UIO device only after the userspace process is informed of the removal so that it could unmap/close the device and exit gracefully. We could try doing it using some systemd trick, where we put a dependency on one of the UIO created files (/dev/uioX), so that when the file is destroyed, systemd could kill/shutdown the process. Although, there's no guarantee that the userspace process won't try to access the mmaped space during that interval, so this isn't a reliable solution either. Any hints, thoughts? Thanks., -mandeep

3 7

eudyptula challenge, stuck at task 06 - no reply for four weeks
by Alexander Ivanov Dec. 31, 2015

Dec. 31, 2015

Hi there, Is there anybody stuck at task 06 too? --ami

6 6

Why a “barrier()” is enough for disabling or enabling the preemption?
by Nan Xiao Nov. 30, 2015

Nov. 30, 2015

Hi all, Below is the preemption disable/enable code: #ifdef CONFIG_PREEMPT_COUNT #define preempt_disable() \ do { \ inc_preempt_count(); \ barrier(); \ } while (0) ...... #define preempt_enable() \ do { \ preempt_enable_no_resched(); \ barrier(); \ preempt_check_resched(); \ } while (0) #else /* !CONFIG_PREEMPT_COUNT */ /* * Even if we don't have any preemption, we need preempt disable/enable * to be barriers, so that we don't have things like get_user/put_user * that can cause faults and scheduling migrate into our preempt-protected * region. */ #define preempt_disable() barrier() ...... #define preempt_enable() barrier() ...... #endif /* CONFIG_PREEMPT_COUNT */ And I have a question about it:If the CONFIG_PREEMPT_COUNT is OFF, while CONFIG_PREEMPT is ON, why only a barrier()can disable/enable preemption? What is the magic behind it? Or On SMP, the CONFIG_PREEMPT_COUNT should always be ON? BTW, I also post this issue on SO(http://stackoverflow.com/questions/33864903/why-a-barrier-is-enough-for-…, but can't receive clear answers. Thanks in advance! Best Regards Nan Xiao

2 1

git tags (RC kernels ) and kernel versions (www.kernel.org)
by Kevin Wilson Nov. 26, 2015

Nov. 26, 2015

I want to be able to work with git commands with the Linux kernel versions according to what I see in www.kernel.org, and I encounter a difficulty. The following example demonstrates the difficulty I encounter. I see under: https://www.kernel.org/pub/linux/kernel/v4.x/ 13 sub version of the 4.1 kernel: 4.1.1 4.1.2 ..... .... 4.1.13 For each one there is a correspoinding ChangeLog ( ChangeLog-4.1.3, ChangeLog-4.1.4, and so on) I know of course that I can find the commits for each such subversion by looking in the corresponding ChangeLog. However, I want to be able to work with the different subversions of 4.1 (for example) with git. This means to navigate with git commands from, let's say, 4.1.13 to 4.1.5, and also go back and forth with git commends to 4.0 and earlier versions. And also of course see the commits by "git show", see the history by "git log", and in general run all git commands. However, when I run in Linus tree (https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git) the following: git tag | grep 4.1 I get: v4.1 v4.1-rc1 v4.1-rc2 v4.1-rc3 v4.1-rc4 v4.1-rc5 v4.1-rc6 v4.1-rc7 v4.1-rc8 So these 8 rc subversions of 4.1 do not seem to correspond the the 1..13 subversions that I see in https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git as mentioned below. Any ideas? Regards, Kevin

4 3

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