Unified Kernel

[zephyrbot]

Reported by Gajinder Vij:

As a Zephyr developer, I would like a unified and more complete kernel API with optimized performance, so that I don't have to choose between the nanokernel or the microkernel when developing applications.

The Zephyr dual-kernel model is confusing for some, and is somewhat harder to use than traditional cookie-cutter RTOS kernels by even experienced developers. Also, while the nanokernel performances for both footprint and speed of execution (context-switch time, latency, etc.), the microkernel could be improved on both fronts.

Problems
- Non-intuitive nature of the nanokernel/microkernel split
- Double-context switch affecting kernel performance (speed and footprint)
- Duplication of object types between the nanokernel and microkernel
- System initialization running in the idle task

Proposed Improvements:
- Make the nanokernel 'pre-emptible thread' aware
- Unify fibers and tasks as one type of threads by dropping the Microkernel server
- Allow cooperative threads to operate on all types of objects
- Clarify duplicated object types
- Create a new, more streamlined API, without any loss of functionality

• Update "Changes from Version 1 Kernel" to include a "How-To Port Apps" section #3034: Update "Changes from Version 1 Kernel" to include a "How-To Port Apps" section
• Update Architecture Porting Guide #2799: Update Architecture Porting Guide
• Configuration talks about >32 thread prios, but the kernel does not support it #2782: Configuration talks about >32 thread prios, but the kernel does not support it
• Optimize interrupt return code on ARC. #2716: Optimize interrupt return code on ARC.
• Fix all test failures from TCF #2606: Fix all test failures from TCF
• cortex_m_systick programmed cycles are 1.7 tick less than expected, for tickless idle #2575: cortex_m_systick programmed cycles are 1.7 tick less than expected, for tickless idle
• Have the kernel give the leftover memory to the IP stack #2568: Have the kernel give the leftover memory to the IP stack
• Create a tool for finding out stack sizes automatically. #2566: Create a tool for finding out stack sizes automatically.
• shrink k_block struct size #2533: shrink k_block struct size
• workq in Kernel primer for unified kernel #2528: workq in Kernel primer for unified kernel
• Cache next ready thread instead of finding out the long way #2494: Cache next ready thread instead of finding out the long way
• Add doxygen documentation to both include/kernel.h and include/legacy.h #2486: Add doxygen documentation to both include/kernel.h and include/legacy.h
• Add a (hidden?) Kconfig option for disabling legacy API #2469: Add a (hidden?) Kconfig option for disabling legacy API
• Create example sample for new unified kernel API usage #2462: Create example sample for new unified kernel API usage
• Revisit the timeslicing algorithm #2455: Revisit the timeslicing algorithm
• Adapt networking to unified kernel #2444: Adapt networking to unified kernel
• Adapt drivers to unified kernel #2443: Adapt drivers to unified kernel
• Kernel logger support (validation) #2442: Kernel logger support (validation)
• NIOS_II port #2441: NIOS_II port
• Unified kernel ARC port #2440: Unified kernel ARC port
• Adapt kernel sample & test projects #2439: Adapt kernel sample & test projects
• Finalize kernel file naming & locations #2438: Finalize kernel file naming & locations
• Cutover to unified kernel #2437: Cutover to unified kernel
• Verify the preempt-thread-only and coop-thread-only configurations #2436: Verify the preempt-thread-only and coop-thread-only configurations
• API changes to event handling #2435: API changes to event handling
• API changes to memory maps #2434: API changes to memory maps
• API changes to memory pools #2433: API changes to memory pools
• API changes to message queues #2432: API changes to message queues
• Revise documentation for Interrupts #2431: Revise documentation for Interrupts
• Revise documentation for Timing #2430: Revise documentation for Timing
• Revise documentation for Kernel Event Logger #2429: Revise documentation for Kernel Event Logger
• Miscellaneous documentation work #2428: Miscellaneous documentation work
• Investigate malloc/free support #2427: Investigate malloc/free support
• Purge obsolete microkernel & nanokernel code #2426: Purge obsolete microkernel & nanokernel code
• Add ring buffer support #2425: Add ring buffer support
• Add abort handler support #2424: Add abort handler support
• Eliminate kernel object API anomalies #2423: Eliminate kernel object API anomalies
• O(1) pend queue support #2422: O(1) pend queue support
• Improving locking algorithms in kernel objects #2421: Improving locking algorithms in kernel objects
• Place thread stacks in their own linker section #2420: Place thread stacks in their own linker section
• Finish renaming kernel object types #2419: Finish renaming kernel object types
• Refine thread priorities & locking #2418: Refine thread priorities & locking
• Adapt tickless idle for x86 #2417: Adapt tickless idle for x86
• Adapt tickless idle + power management for ARM #2416: Adapt tickless idle + power management for ARM
• Add task offload to fiber support #2415: Add task offload to fiber support
• Test memory pool support (with mailboxes) #2414: Test memory pool support (with mailboxes)
• [unified] Add scheduler time slicing support #2413: [unified] Add scheduler time slicing support
• clearly document internal kernel interfaces #2144: clearly document internal kernel interfaces
• clean up how internal APIs are used #1655: clean up how internal APIs are used

(Imported from Jira ZEP-334)

[zephyrbot]

by Benjamin Walsh:

FYI: the tasks/users stories for this are tracked internally at WR since we're doing at least the initial push for this. If/when needed, we can transition them to the public Jira. I've added Al so that he can handle this if needed during my vacation.

[zephyrbot]

by Benjamin Walsh:

RFC available here:

https://gerrit.zephyrproject.org/r/#/c/2255/2/incoming/unified_kernel.rst

and discussed on the devel mailing list.

[zephyrbot]

by Benjamin Walsh:

Extra info from porting Feature from WR's internal tracking:

DESCRIPTION

As a Zephyr user (i.e. an application developer using Zephyr or an RTOS developer modifying Zephyr) I don't want to deal with the kernel's layered Microkernel and Nanokernel model, as it makes the kernel difficult to understand and can increase development overhead.

This feature merges the Microkernel and Nanokernel into a single Kernel entity. Where feasible, duplicated services are combined and undesirable service gaps are filled. These changes are made transparent to users by providing "legacy API" versions of any API made obsolete by the unified kernel.

DOCUMENTATION

1. Update public Zephyr project documentation to describe new unified kernel APIs, and to mark obsolete APIs as deprecated.
2. Update public Zephyr project documentation to remove references to nanokernel & microkernel layers.

RATIONALE

1. A unified kernel is easier for users to understand, and provides a unified approach that eliminates some existing restrictions of the kernel. This makes Zephyr more likely to meet the expectations of experienced RTOS developers.
2. A unified kernel is easier for Wind River developers (and others) to support and extend. It eliminates the need for duplication of effort when supporting both Microkernel and Nanokernel configurations.
3. A unified kernel provides significant opportunities for both footprint and performance improvements, particularly in Microkernel configurations.
4. Potential customers/partners could potentially merge all of their development into Zephyr. Easier to do with a well-known model.

ACCEPTANCE CRITERIA

1. All user stories accepted.
2. All test cases passed.
3. Before/After comparison of benchmark test cases on Galileo Gen 2

[zephyrbot]

Blocks GH-2144