January 2024 - kernelnewbies@kernelnewbies.org

Source code organization
by Billie Alsup (balsup) April 23, 2024

April 23, 2024

I have a set of drivers that I would like to upstream. These are primarily MFD style drivers supporting Cisco-8000 FPGAs. The devices can be instantiated through multiple top level drivers, which provide the access mechanism for the MFD driver. For example, an FPGA can be accessed via PCI, or via i2c, or via SLPC, or via P2PM (a point-to-point interface). We currently build these drivers out of tree, under a single directory drivers/cisco. I note that existing drivers are spread out across the kernel tree, sometimes by bus (pci, i2c), sometimes by function (gpio, net/mdio, spi), and sometimes under the generic mfd. I would like to start the upstream review process for our drivers, but first want to get recommendations on the source code layout. Is it permissible to keep a top level directory such as drivers/cisco to organize our code? It is not only the source code that is affected, but also provides a central place for Kconfig entries. Our FPGAs have multiple logical blocks, each of which is handled by a different MFD driver, e.g. i2c controllers, gpio controllers, spi controllers, mdio controllers. There can be multiple instances of each block as well (so multiple MFD devices are instantiated for each driver). And of course, there can be multiple FPGAs in a system, each with different combinations of logical blocks. The drivers themselves are pretty specific to our FPGAs, thus it makes sense to use Kconfig to select a hardware platform to automatically select the set of MFD drivers (and top level bus drivers) that would apply. Would a source layout putting all the code under drivers/cisco be acceptable in this case, or do I need to move things around and spread out the Kconfig entries across directories? I note that a single drivers/cisco would simplify any related modifications to MAINTAINERS as well.

2 5

Completion wait timeout error thrown by i2c-ismt.c driver
by Rahul Gore (Nokia) Jan. 23, 2024

Jan. 23, 2024

Hi, I have come across the following problem: Error: After pasting ~130kb text in console, multiple "ismt_smbus 0000:00:12.0: completion wait timed out" messages are seen in dmesg. Configuration: 1. Custom Debian-based operating system based on Linux kernel 6.1.25-13-amd64. 2. Sysrq is set to 16. 3. Following peripherals are present: a. Intel Corporation Atom Processor C3000 Series SMBus Contoller (Driver i2c_ismt). b. Serial controller: Intel Corporation Atom Processor C3000 Series HSUART Controller (Driver 8250_mid) 4. The original baudrate of the serial controller is 115200, however, we configure its baudrate to 57600. 5. Model name: Intel(R) Atom(TM) CPU C3558 @ 2.20GHz; Number of cores = 4; RAM = 16GB. Additional observations: 1. The paste operation takes around 61 minutes. 2. Number of interrupts on serial controller during this time is around 175000. 3. Number of interrupts on SMBus controller during this time is around 346000. 4. When sysrq is set to 0, the problem does not occur. 5. The interrupts on the IRQ lines for serial driver and SMBus driver are serviced by the same core. 6. If I ensure that the interrupts on the serial and SMBus driver are managed by different cores, the problem does not occur. 7. During the paste process, the CPU usage of the core (that handles both interrupts) goes to 100%, then suddenly drops to 0%, before returning to normal values (around 90%). 8. If the baudrate is set to 115200 (original value), the problem does not occur. I studied the source code for the serial driver. I could not find anything which may have caused other interrupts to get blocked. I would appreciate any direction and/or explanation for this behavior. Please let me know if any additional information is needed from my end. Regards, Rahul

1 0

Generic USB LED HID driver
by Kirill Yatsenko Jan. 19, 2024

Jan. 19, 2024

Hello, I'm trying to find the Linux kernel driver in the tree that implements the HID LED usage table. The only driver that I've found is the hid-led.c. However, it seems to support only specific devices. The device with which I'm trying to interface is the custom-made board with atxmega and some LEDS connected to it. Currently, the communication is done through the vendor-specific HID protocol. The custom USB HID kernel module is implemented to communicate with the device. The idea is to move away from the custom protocol to the generic implementation so that the custom kernel module can dropped. Best regards, Kirill Yatsenko

2 4

Lost memory
by Andrea Tomassetti Jan. 18, 2024

Jan. 18, 2024

Hi everyone, I've been facing this problem for the past few weeks and I'm out of ideas of how to analyze it, maybe someone here has some thoughts on how to address it. Basically, if I compare the output of `free` and the total RSS reported by `ps` there are 200 GiB of difference: $ free total used free shared buff/cache available Mem: 521933440 371541084 19246028 4100 131146328 117933472 Swap: 20971516 15208556 5762960 $ ps -eo rss | awk '{ rss += $0 } END { print rss}' 176765072 total used: 386749640 total RSS: 176765072 difference: 209984568 (~200.3 GiB) I have some programs running that make intense use of networking with thousands of sockets opened, so I checked sockstat but with no luck: $ cat /proc/net/sockstat sockets: used 90319 TCP: inuse 40798 orphan 0 tw 56 alloc 89689 mem 558949 # *4 = 2235796 KiB = ~2.13 GiB UDP: inuse 31 mem 2 UDPLITE: inuse 0 RAW: inuse 0 FRAG: inuse 0 memory 0 Even /proc/meminfo is quite useless [1]. I started to kill one process at a time till I found the culprit: a Java process with high network usage. But inspecting this process' memory consumption didn't give me any conclusion because its memory consumption appears to be low: $ sudo pmap -x 268169 | head -n2 | tail -n1; sudo pmap -x 268169 | tail -n 1 Address Kbytes RSS Dirty Mode Mapping total kB 171403084 23775460 23697900 ~ 22.7 GiB But this is the process that is helding the resources because, once killed, the memory difference/leak disappears. How is it possible that no utility or kernel information is capable of outlining this memory leak? What can I possibly do to investigate it further? For sure there's a problem releasing resources in the Java code, but I would like to see it at a kernel level. Any help is very appreciated. Thank you in advance, Andrea [1] $ cat /proc/meminfo MemTotal: 521933440 kB MemFree: 20600964 kB MemAvailable: 118570008 kB Buffers: 18092 kB Cached: 125086816 kB SwapCached: 1792868 kB Active: 115786784 kB Inactive: 187218964 kB Active(anon): 79144440 kB Inactive(anon): 98768932 kB Active(file): 36642344 kB Inactive(file): 88450032 kB Unevictable: 20556 kB Mlocked: 20556 kB SwapTotal: 20971516 kB SwapFree: 5762448 kB Dirty: 114392 kB Writeback: 0 kB AnonPages: 176129992 kB Mapped: 315908 kB Shmem: 4100 kB KReclaimable: 5322088 kB Slab: 13095480 kB SReclaimable: 5322088 kB SUnreclaim: 7773392 kB KernelStack: 135600 kB PageTables: 496504 kB NFS_Unstable: 0 kB Bounce: 0 kB WritebackTmp: 0 kB CommitLimit: 542904956 kB Committed_AS: 210582988 kB VmallocTotal: 133143592960 kB VmallocUsed: 256200 kB VmallocChunk: 0 kB Percpu: 62976 kB HardwareCorrupted: 0 kB AnonHugePages: 0 kB ShmemHugePages: 0 kB ShmemPmdMapped: 0 kB FileHugePages: 0 kB FilePmdMapped: 0 kB CmaTotal: 32768 kB CmaFree: 0 kB HugePages_Total: 0 HugePages_Free: 0 HugePages_Rsvd: 0 HugePages_Surp: 0 Hugepagesize: 2048 kB Hugetlb: 0 kB

1 0

Boot stucks at "Loading initial ramdisk"
by Dileep Sankhla Jan. 5, 2024

Jan. 5, 2024

Hello, I have cloned and built the latest rc kernel using the default config but on boot, it stucks at "Loading initial ramdisk". I have tried booting after removing the "quiet" option from the kernel parameters just before selecting the kernel from the grub menu. I have also tried adding "dis_ucode_ldr" parameter to the initial parameters but all in vain. How to look for the underlying issue and solve it? Regards, Dileep

5 12

Information about interrupt in userpace
by Patryk Jan. 5, 2024

Jan. 5, 2024

Hi, I have a device that is connected through the SPI to my CPU. Additionally, the device exposes an interrupt pin which I can also connect to my processor. The interrupt is triggered each time the device has some data available. This information should be propagated up to the userspace process which uses my driver to communicate with the device. How should I implement this? Should I e.g. implement a poll function in the driver which will be called from the userspace process and which will inform about triggered interrupt? What are the other ways to deal with it? BR Patryk

1 0

Backtracing to find the assignment of a null pointer.
by Dileep Sankhla Jan. 5, 2024

Jan. 5, 2024

Hello everyone, I am currently addressing a power management issue. Upon waking up from suspend, the login screen briefly appears for a couple of seconds, followed by the screen going blank for around 7 seconds. Afterward, the login screen is displayed again. I've identified an ACPI error in the journalctl entries. In the source code, I located the line corresponding to the error message, which indicates an error if the handler is a null pointer. This handler is one of the function parameters, and I am interested in determining which function assigns its value and where, how the value was set to NULL initially. Initially, I attempted manual debugging, but after a few backtraces, I couldn't pinpoint from where the function was called. Since I have some experience with gdb, I decided to try it and followed both the kernel's and gdb's documentation. To reproduce the bug, I needed a real machine. I ran `gdbserver` by attaching it to pid 1 (I think ACPI driver is initialized by init (pid 1)), and in another terminal tab, I executed `gdb vmlinux`. I then set a hardware breakpoint, but when I used `step`, I encountered an "ignoring packet error." Here is the output: $ cd ~/src/linux-next $ gdb vmlinux Reading symbols from vmlinux... (gdb) tar remote :1234 warning: Build ID mismatch between current exec-file /home/dileep/src/linux-next/vmlinux and automatically determined exec-file /usr/lib/systemd/systemd exec-file-mismatch handling is currently "ask" Load new symbol table from "/usr/lib/systemd/systemd"? (y or n) n (gdb) hbreak drivers/acpi/acpica/evregion.c:131 Hardware assisted breakpoint 1 at 0xffffffff8197e240: drivers/acpi/acpica/evregion.c:131. (2 locations) (gdb) step warning: Remote failure reply: E01 Ignoring packet error, continuing... Ignoring packet error, continuing... Ignoring packet error, continuing... Ignoring packet error, continuing... Ignoring packet error, continuing... Ignoring packet error, continuing... And for gdbserver: $ sudo gdbserver --attach :1234 1 Attached; pid = 1 Listening on port 1234 Remote debugging from host 127.0.0.1, port 52364 gdbserver: Couldn't write debug register: Invalid argument. Listening on port 1234 input_interrupt, count = 1 c = 36 ('$') Firstly, the breakpoint 1 should be set at only one location. Secondly, I'm puzzled by the error occurring with the `step` command in both gdb and gdbserver. Is this the correct approach for remote kernel debugging on the same machine? If not, what would be the proper method? I also have a MacBook Pro as a second machine. If necessary, I can use it as a host to run gdb. Regards, Dileep

1 0

Identifying the initial bug within the x86 architecture subsystem
by Dileep Sankhla Jan. 3, 2024

Jan. 3, 2024

Hello, I am new to Linux kernel development and have a keen interest in the x86 architecture subsystem, particularly in the areas of boot and feature flags. Before diving into adding or changing features, I would like to address a few existing bugs. I am good in C and have some knowledge of assembly language. Last night, I dedicated time to go through bugs on Bugzilla (see [1]), considering their priorities but I could not figure out which one to pick. While I found only a couple of bugs with the latest modification date, I lack the same hardware as the original poster (OP) to reproduce and fix those bugs. How can I identify the first bug in this subsystem? Given the relatively low number of bugs, should I consider reaching out to the maintainer of `arch/x86/boot` if I cannot find a suitable one to fix? P.S. I have one commit in the kernel from 2018 (see [2]) , but it was merely addressing a `checkpatch.pl` warning. Regards, Dileep [1]: https://bugzilla.kernel.org/buglist.cgi?bug_severity=blocking&bug_severity=… [2]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?…

2 4