This is an old revision of the document!
Kernel headers should be copied into /usr/include to be used by libraries and compiled programs. These should be the headers glibc has been compiled against, and should be kept independent of the ones in the kernel source tree. New packages are compiled against these headers, otherwise, if packages are compiled against headers that do not match the headers glibc was compiled against, segfaults or other strange problems may appear. /usr/include headers should be what matches the library object files. By copying the headers instead of symlink them to /usr/src/<linux>/include, we make sure that they remain available if we upgrade our kernel.
Process State
- TASK_RUNNING, either running or in a runqueue waiting to run. Only possible state for a process executing in user-space.
- TASK_INTERRUPTIBLE, process blocked, waiting for some condition to exist.
- TASK_UNINTERRUPTIBLE, like TASK_INTERRUPTIBLE but without responding to signals.
- TASK_TRACED
- TASK_STOPPED
Process Context
A program running in user-space executes a system call or triggers an exception, making the kernel to run in process context.
There is also interrupt context, mostly applied to when the system is executing an interrupt handler (no process is tied to interrupt handlers). Other examples are the execution of a softirq as well as initializing or running the scheduler.
The Kernel may also run kernel threads, standard processes that exist only in kernel-space, like management threads that maintain the buffer and page caches. These processes are scheduled like normal processes and are allowed to sleep.
Concurrency and shared resources
- Interrupts
- Kernel preemption, causing the kernel to switch from the execution of one system call to another
- Multiprocessing
We need to keep as much local state as possible and use locking for the shared resources.
Locks
- Mutex, the main kernel locking primitive. Usable in contexts where sleeping is allowed.
- Spinlocks, used with code that is not allowed to sleep (interrupt handlers) or that doesn't want to sleep (critical sections).
- Alternatives: Lock-free algorithms like Read Copy Update (RCU) and atomic operations
Debugging
ftp://82.96.64.7/pub/software/scm/git/docs/v1.5.0/howto/isolate-bugs-with-bisect.txt
http://www.reactivated.net/weblog/archives/2006/01/using-git-bisect-to-find-buggy-kernel-patches/
git bisect start – [path]
to restrict bisection to commits that touch [path].
Porting the kernel to ARM devices
ARM specific files
arch/arm/Makefile has macros to detect machine name
def-configs includes the default config file
kernel/debug.s assembly level debugging
lib
boot/Makefile lists start address from where kernel image has to be decompressed
tools/mach-types machine ID
mm/proc-[CPU_TYPE]
include/asm-arm set up functions for a specific CPU
asm/arch/uncompress.h output kernel decompression msgs to UART
hardware.h IO map of HW architecture
io.h HW specific macros
irq.h base address of diferent IRQs
irqs.h IRQ numbers
time.h time interrupt handlers and macros