On 12/9/2024 3:52 PM, Nathan Hartman wrote:
Hi all, Unfortunately I missed the PR before it was merged, but PR-15073 has broken High Priority, Zero Latency Interrupts! Fortunately I caught it now. It was merged 17 hours ago. This PR removed a very important Kconfig: config ARMV7M_USEBASEPRI, and much of the logic related to it. This config is *required* when using CONFIG_ARCH_HIPRI_INTERRUPT. High Priority, Zero Latency Interrupts are documented here: https://nuttx.apache.org/docs/latest/guides/zerolatencyinterrupts.html I depend strongly on Zero Latency Interrupts. It is *required* for critical handling in my firmware. I cannot use NuttX at all if NuttX does not have Zero Latency Interrupts anymore. PR-15073 was meant to fix a regression in PR-14881. PR-14881 attempted to optimize handling of g_current_regs. The commit log of PR-15073 explains this. [1] We have to find another way to fix it, without destroying the Zero Latency Interrupts feature. Otherwise, it is a very serious problem. I suggest to revert PR-15073, and possibly revert PR-14881, and find another way to implement PR-14881 that doesn't cause this problem. References: [1] PR-15073: https://github.com/apache/nuttx/commit/0e1b432dd068880d353ae91c01e610f85b4a16ea#diff-24754055e0d9f2d702b8503e7f5b520394cb59affb5d24457fe0641c1d7678ceL27 [2] PR-14881: https://github.com/apache/nuttx/pull/14881 [3] High Priority Zero Latency Interrupts: https://nuttx.apache.org/docs/latest/guides/zerolatencyinterrupts.html Thanks, Nathan
Using the basepri register would also be important if you ever want to implement nested interrupts. I purist would say that nested, priority interrupts is necessary for true real-time performance. More info: https://cwiki.apache.org/confluence/display/NUTTX/Nested+Interrupts
Samsung has a working implementation of this in TizenRT: The are about 3 or 4 commits in May of 2019 that provides the full implementation. It should drop into those architectures pretty simply.
