July 2015 - kernelnewbies@kernelnewbies.org - Kernelnewbies Mailing Lists

CPU STALL ISSUE
by Amit Gupta July 23, 2015

July 23, 2015

Hi All, I am facing one issue with Linux kernel 4.0.4. *CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.0.4+ #19* This issue i am facing with below mentioned two scenario First Scenario: At the time when I start using PCIe based Atheros wifi card along with my Ethernet controller. If I am using both standalone then I am not facing this issue. (I think this is not hardware related issue as with the same combination of h/w but on different linux kernel version (3.4), I am not facing this issue.) Second Scenario: This issue also repeated if I keep on executing below mentioned command: # hostapd -B /etc/hostapd.conf My main focus is on first scenario. I googled it this issue and found that few people also faced this issue previously on Linux kernel 3.1/3.2/3.3 . Some people are saying that disabling IPv6 support from kernel configuration will resolve this issue but I need IPv6 support in my kernel due to project specific task. By digging out into Linux kernel and observing dump stack log(at the time CPU STALL), I found culprit function 'ieee80211_wake_queues_by_reason'. I go through the series of functions calling by this function, and stuck at one point. I found there are two 'rcu_read_lock();' in sequence. * So my question is can we call rcu_read_lock() function is sequence(two times) and later call rcu_read_unlock() one by one;??* Please look into the pictorial representation of my question: ieee80211_wake_queues_by_reason -- |----> __ieee80211_wake_queue ---> if (skb_queue_empty(&local->pending[queue])) { rcu_read_lock(); ieee80211_propagate_queue_wake(local, queue); rcu_read_unlock(); } ieee80211_propagate_queue_wake(local, queue) | --> netif_wake_subqueue(sdata->dev, ac) | --> rcu_read_lock(); q = rcu_dereference(txq->qdisc); __netif_schedule(q); rcu_read_unlock(); My Approach: I increased RCU stall time from 21 jiffies to 60 jiffies. -- CONFIG_RCU_CPU_STALL_TIMEOUT=21 ++ CONFIG_RCU_CPU_STALL_TIMEOUT=60 But this does not solve my problem. Please suggest me any approach to solve this problem. Thanks, Amit Gupta

1 0

Fuzzing scheduler CFS
by Meyer Lansky July 23, 2015

July 23, 2015

Thank you very much! I will learn to look for bottlenecks in the program. You sent me to the right path!

1 0

how to delete kernels cleanly
by Ahmed Soliman July 22, 2015

July 22, 2015

I have many kernels on my machine and I want to delete some of them what can I do some of these kernels are shipped with the OS ubuntu 14.04 and others I built by make oldconfig make all make modules sudo make modules_install sudo make install now how can I these kernels

4 4

Is it possible to reserve a huge page beginning at a fixed physical address
by Yunzhao li July 22, 2015

July 22, 2015

Hello, All The hugetlb feature helps to reserve a large amount of memory. I am wondering is there any method that could reserve huge pages beginning at a fixed physical memory? Or, for reserved physical memory via "memmap" boot parameter, is it possible to use hugetlb feature? Thanks! Yunzhao

1 0

Fuzzing scheduler CFS
by Meyer Lansky July 22, 2015

July 22, 2015

I am very interested in finding bugs in such code. In such large projects which method you recommend to seek effective error? I do not necessarily insist on the method of fuzzing, it is one of the options. the answer to your question: The scheduler should work very quickly, and this demonstrates the quality and level of the algorithm. I would like to improve the code, but has not found a practical method of finding errors. Thank you very much for your reply!

2 1

Delegating printk work to UART interrupt
by Alexander July 22, 2015

July 22, 2015

Hi! When i checked how kernel printing works, i mentioned that it takes messages from log_buffer in console_unlock and gives it to call_console_drivers -> ... -> some uart bsp function. Basically, as i see this BSP realization tries to flush all message chars in busyloop ... so it waits until FIFO_NOT_FULL bit will be dropped by UART and it will be able to push the next byte. Basically, as i see userspace printing do something different. It puts N_FIFO_BYTES and exits, next, when FIFO will be freed - interrupt will be generated, and other characters will be put into UART FIFO. Can we do something similar for kernel printing? i.e. do not busyloop sending char after char, but put N_FIFO chars and flush other in interrupt. When panic will occur we can do busyloop printing again. Is it reliable? Suppose we have several cores. Thank you. -- Alexander <alexhoppus111(a)gmail.com>

2 3

Fuzzing scheduler CFS
by Meyer Lansky July 22, 2015

July 22, 2015

Hello everyone! I decided to take fuzzing task Scheduler CFS http://lxr.free-electrons.com/source/kernel/sched/fair.c Question: I just recently learning fuzzing and it is at all possible fuzzy scheduler? If so then why will advise you start? Excuse me for my english.

2 1

Obfuscate code GPL 2 - The source uses kernel structs and GPL2 API
by Lucas Tanure July 21, 2015

July 21, 2015

Hi, This company released a obfuscated kernel module in GPL 2. http://www.incentivespro.com/downloads.html So, they didn't release the code at all. This is ok ? This against the law ? If you download the source for linux you will se this code: /* * * Copyright (C) 2007-2015 SimplyCore, LLC * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as * published by the Free Software Foundation. * * */ #include "usbd.h" #ifdef _USBD_DEBUG_MEMORY_ atomic_t llIlllll=ATOMIC_INIT((0x575+1935-0xd04));atomic_t lIIlIIIll=ATOMIC_INIT ((0xa1d+3263-0x16dc));atomic_t IllIlIlI=ATOMIC_INIT((0x128f+4169-0x22d8)); atomic_t IlIllllI=ATOMIC_INIT((0xf24+561-0x1155)); #ifdef ATOMIC64_INIT atomic64_t llIlIlll=ATOMIC64_INIT((0xeb9+1200-0x1369)); #endif atomic_t IllllIlI=ATOMIC_INIT((0x95+5058-0x1457));atomic_t lIIllllI=ATOMIC_INIT( (0x54+4166-0x109a));atomic_t llIllIlI=ATOMIC_INIT((0x90+8277-0x20e5)); #endif struct lllIl*IIIIlll=NULL; #ifdef _USBD_ENABLE_STUB_ spinlock_t IIIIlII;struct list_head IIIIlIlI; #endif static int lIIIlIII=(0x1b5b+1617-0x21ac);module_param(lIIIlIII,int, ... many lines .... #endif IIIll("\x75\x73\x62\x64\x5f\x65\x78\x69\x74" "\n");}module_init(IlIIllIIl); module_exit(usbd_exit);MODULE_LICENSE("\x47\x50\x4c"); -- Lucas Tanure +55 (19) 988176559

6 7

what is the use of #ifndefs
by Ahmed Soliman July 21, 2015

July 21, 2015

currently I started reading through the linux kernel and I started reading liunx/include/linux/list.h> I understood some of the functions but still I dont know what does these lines of code do #ifndef _LINUX_LIST_H #define _LINUX_LIST_H which exist at the very beginning of the file I also noticed that there is many similar ifndefs in almost any .h file in the kernel note that I understand wnat does ifndef do bu I dont understand what goal is it supposed to achieve at the beginning of the headerfile

10 10

Allocate write combined memory for BAR of a PCIe device and transfer data in bursts
by Ankit Pradhan July 21, 2015

July 21, 2015

Hello everyone I want to be able to send 1-2KB of data from host, using BARs of a PCIe device, with very low latency. I am working on an FPGA project in which I have to send packet data from Host to FPGA over PCIe 3.0 interface. This is a latency critical application and therefore I want to use BARs to send data in bursts and not use DMA because it has a lot of overhead. I started out with the device driver provided to me by my FPGA board vendor. That driver allocates 'uncached' memory for the BAR via mmap and provides its virtual pointer to the user application for use. When I initiate a transfer by dereferencing that pointer in my application and observe the write signals that arrive at my FPGA using a logic analyzer. I see all write requests arriving after constant time period and the burst count signal is always 1. I have concluded that these transfers are not in bursts. I modified the driver by using the ioremap_wc() on the physical address of BAR and then passed the virtual pointer it returned to set_memory_wc() (apparently I have to use both). I was able to get 'write combined' memory for the physical BAR address. When I do writes using this pointer and observe from logic analyzer I can see burst transfers. That is, I see 4 bursts of 5-6 write requests sent within a short time period. The problem is that the pointer to this memory is a kernel virtual pointer. I want to be able to do the above from user space. How can I do this? Thanks Ankit

1 0