July 2019 - kernelnewbies@kernelnewbies.org - Kernelnewbies Mailing Lists

of_parse_phandle in device_node
by Nuno Sá July 29, 2019

July 29, 2019

Hi guys, So I'm working on a driver and on dt parsing i need to get a reference to device_node using of_parse_phandle. I was trying to see if there is any equivalent on the fwnode_* api's and i could find nothing... I use device_for_each_child_node(dev, child) { ... } and then some of my fwnode's need to reference some other ones. Now, i'm using fwnode's api and using to_of_node() and of_fwnode_handle() macros to use of_parse_phandle() but i don't really want to mix device_node with of_node so if there's nothing i guess i will just go with of_node api's. So, my question, Is there any equivalent to of_parse_phandle() in the device_node api's? Note that i don't have any strong requirement to use device_node instead of of_node but I now i just got curious... Regards, Nuno Sá

1 0

internal / external include files
by Martin Kaiser July 25, 2019

July 25, 2019

Dear all, is it correct that the files under include/uapi/ can be included by a user space application? I just ran into an error where my application ended up including include/uapi/linux/stddef.h (with a number of intermediate steps). This fails because the kernel's stddef.h includes include/linux/compiler_types.h and this file is internal to the kernel. What is the correct way to solve this? Should I fix my include path to make sure that my application picks the stddef.h in the compiler's sysroot rather than the kernel's stddef.h? Or should include/uapi/linux/stddef.h guard the kernel-internal parts using #ifdef __KERNEL__? Thanks, Martin

3 3

Problem with kernel major than 4.15.x
by Gilberto Nunes July 24, 2019

July 24, 2019

Hi there I am usging Linux mint here 19.1. My laptop has the following configuration: CPU - AMD A12-9720P RADEON R7, 12 COMPUTE CORES 4C+8G GPU - Wani [Radeon R5/R6/R7 Graphics] (amdgpu) Network Interface card: 01:00.1 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL8111/8168/8411 PCI Express Gigabit Ethernet Controller (rev 12) When I install kernel 4.18.x or 5.x, the network doesn't work anymore... I can loaded the modulo r8169 andr8168. I saw enp1s0f1 as well but there's no link at all! Even when I fixed the IP none link! I cannot ping the network gateway or any other IP inside LAN! Is there somebody with the same problem???? Thanks a lot. --- Gilberto Nunes Ferreira (47) 3025-5907 (47) 99676-7530 - Whatsapp / Telegram Skype: gilberto.nunes36

2 4

bitops api for set_bit function
by Wes Edens July 22, 2019

July 22, 2019

Hi all, I have been looking at the bitops api for various architectures. I am curious why some architectures define the 'nr' argument as a signed int and some architectures define it as an unsigned int. there are even some architectures (mips, sparc) that mix the prototypes between atomic and non-atomic functions (set_bit and __set_bit). The x86 architecture uses a long as its 'nr' argument. the atomic operations documentation gives a prototype of the bit operations using unsigned int for 'nr'. (https://elixir.bootlin.com/linux/latest/source/Documentation/core-api/atomi…) i've looked at the commit where the sparc architecture changes from using signed to unsigned, but the commit message didn't have an explanation on that part of it. (https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?…) i've dug through the uses of this function and most drivers/kernel subsystems use this to set bits in a status bitfield where the 'nr' argument is a #define. it doesn't seem to matter which type the argument is, but i'm curious why there are different implementations. asm-generic implementation: static inline void __set_bit(int nr, volatile unsigned long *addr) { unsigned long mask = BIT_MASK(nr); unsigned long *p = ((unsigned long *)addr) + BIT_WORD(nr); *p |= mask; } sparc implementation: static inline void set_bit(unsigned long nr, volatile unsigned long *addr) { unsigned long *ADDR, mask; ADDR = ((unsigned long *) addr) + (nr >> 5); mask = 1 << (nr & 31); (void) ___set_bit(ADDR, mask); } thank you, wes edens

1 0

Are MSRs saved and restored on context switches?
by Josué Feliu July 12, 2019

July 12, 2019

Hi all, I was curious about whether a context switch in Linux saves and restores the MSR registers of the tasks involved in the context switch. I searched in the source code and could not find this functionality. I'm not sure if all MSRs need or can be associated to tasks but I understand that, for example, MSRs related to performance counters should be saved and restored on context switches to correctly monitor events on a per PID basis. Thanks a lot, Josué

1 0

Re: [IMX] [DRM]: suspend/resume support
by Pintu Agarwal July 11, 2019

July 11, 2019

On Tue, Jun 18, 2019 at 7:32 PM Fabio Estevam <festevam(a)gmail.com> wrote: > > Hi Pintu, > > On Mon, Jun 17, 2019 at 10:50 AM Pintu Agarwal <pintu.ping(a)gmail.com> wrote: > > > > Hi All, > > > > I am trying to implement suspend/resume functionality to to imx-hdmi driver: > > https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/drive… > > Please test a recent kernel, such as 5.1.11. I recall that last time > you reported an issue with the dw hdmi you were running kernel 3.10 > with backported patches from 4.9. > > Just tried suspend/resume on a imx6q board and HDMI behaves correctly. Thank you so much for your reply. Yes, I already checked the latest mainline kernel changes until 5.2-**. I see that there are many cleanup related changes, but still I see that suspend/resume functions are missing in dw_hdmi-imx driver. So, I am trying to implement myself. Also, I see that suspend/resume is missing even in : drivers/gpu/drm/imx/ipuv3-crtc.c I am trying this also (not sure if we need this as well). Can you point out some specific patches which can be helpful ? Currently, what I observed is that hdmi bind() functions are not getting called. Not sure who is responsible for triggering this bind function. Or, irq remains disabled after resume, and we need to enable it back again (but how ?) Or, we may need to enable vblank events and thus irqs can be enabled. So, I am exploring many possibilities. If you have any suggestions please let me know. Thanks, Pintu

4 15

Guide to building the Linux kernel tools update
by Rob Coulson (bcoulson) July 10, 2019

July 10, 2019

Good afternoon, I recently had to reconstruct my Linux Kernel Ubuntu VM development environment and went through this wiki as a quick reference: https://kernelnewbies.org/KernelBuild Specifically, the ‘What tools do I need?’ Ubuntu section: while all of the tools listed installed correctly, there seems to be some missing tools that are required for completing a linux-5.2.y kernel build: flex, bison, and libelf-dev tools. bison, flex: the defconfig failed w/o these pkgs. libelf-dev: kernel make failed w/o this pkg. Since I do not have the privilege to update this wiki page, would someone on this list please do so? Cheers, RobC.

1 0

CONFIG_CC_STACKPROTECTOR_STRONG
by Muni Sekhar July 9, 2019

July 9, 2019

Hi All, My Kernel is built with the following options: CONFIG_CC_STACKPROTECTOR=y CONFIG_CC_STACKPROTECTOR_STRONG=y We use out-of-tree kernel modules in our project and I need to measure the performance of it by using a bit older gcc version 4.8.5. I can build the modules with gcc above 5 version, but when I try to use gcc version 4.8.5 I’m getting the following error: Makefile:693: Cannot use CONFIG_CC_STACKPROTECTOR_STRONG: -fstack-protector-strong not supported by compiler So it looks like gcc 4.8.5 does not support CONFIG_CC_STACKPROTECTOR_STRONG option, but how to instruct gcc to ignore this one? In runtime, is it possible to instruct kernel to ignore “CONFIG_CC_STACKPROTECTOR_STRONG=y” (without building the kernel) by using sysctl or any other method? Any help is appreciated. -- Thanks, Sekhar

3 5

how tasklet_kill() worked?
by kinglt July 6, 2019

July 6, 2019

Dear mentors, After tired digging, I'm wholly stuck in tasklet_kill(). Hope your help. >From the source tree (4.9.168), I got: 1. TASKLET_STATE_SCHED flag is only set by tasklet_schedule(). (also in dccp_tasklet_schedule()). 2. tasklet_schedule() immediately put the tasklet to queue tasklet_vec. 3. softirq handler tasklet_action() run the queue and clear their SCHED flag. 4. Unlike tasklet_kill_immediate, tasklet_kill() doesn't touch tasklet_vec. 5. tasklet_kill() just yield to waiting the tasklet complete and clear its SCHED flag. Then how to say the tasklet been killed? If a tasklet SCHED flag been set, it must be in tasklet_vec queue. At this time, tasklet_kill does not kick it out of the queue but waiting it to run over. Is my understanding right or I miss something? Thanks, laokz

1 0

questions about kfifo
by laokz July 1, 2019

July 1, 2019

Hello, all I have a few questions about kfifo, wishing your opinions. 1. Why __kfifo_init() roundup_pow_of_two(size) nor down? __kfifo_init() is called by kfifo_init() to initialize a fifo using a preallocated buffer. If the buffer size is not pow_of_two, it will be rounded up to the next power of two. But __kfifo_init() only adjusts the fifo's internal argument 'mask' to reflect the new size, DOES NOT adjust the preallocated buffer! If so, there would be a potential 'memory collapse'. Fortunately, I haven't caught any complaint about this. Maybe as kfifo_init() is the only caller and no much driver use it. >From http://linux-kernel.2935.n7.nabble.com/template/NamlServlet.jtp?macro=searc… , I know, before 3.9 its algorithm was round down. After a big discussion, __kfifo_alloc and __kfifo_init all adopted 'to alloc at least the requested number of elements'. For __kfifo_alloc there is no problem, but for __kfifo_init it is wrong I think, because __kfifo_alloc's buffer is managed by kernel itself, __kfifo_init's buffer by user. My question is, though few application scenarios and 'pow_of_two' rule should be obey, is it more robust back to round down? Then, 'if a user do the wrong thing, then the user will get also a wrong result', but NOT a wrong system. 2. Is it better to optimize __kfifo_in_r()? __kfifo_in_r() is an enqueue function, first it calls __kfifo_poke_n() to write the number of elements, then kfifo_copy_in() for actual elements. In kfifo_copy_in(), it adjusts 'buffer offset' by multiply 'element size' if esize != 1 before writing, but __kfifo_poke_n() doesn't -- that is, in special situation __kfifo_in_r() may use inconsistent offset unit to write this 2 kinds of data. Dequeue function __kfifo_out_r() also has the issue. I learn that the FIFO model with 'sizeof(datatype)>1 && recsize>0' is actually not supported by kernel. But to make a more 'generic kernel FIFO implementation', is it better to do a little fix of 2 helpers __KFIFO_PEEK and __KFIFO_POKE? Thanks, laokz

1 0