On Thu, Jan 30, 2025 at 02:07:46PM -0500, Steven Rostedt wrote: > On Thu, 30 Jan 2025 15:14:38 -0300 > Wander Lairson Costa <[email protected]> wrote: > > > The irq_handler_entry tracepoint includes a "name" field, which allows > > for easy identification and filtering of specific IRQs. However, the > > irq_handler_exit tracepoint lacks this field, making it difficult to > > pair corresponding entry and exit events for analysis. > > > > Add the "name" field to irq_handler_exit to enable consistent tracking > > and correlation of IRQ entry and exit events. > > I'm not sure what you mean by this. >
What I meant was looking at the name is easier when you are analysing several interrupts overlapping on different cores, in special if they are shared... > > # trace-cmd record -e irq_handl* sleep 5 > # trace-cmd report > > <idle>-0 [000] d.h1. 4690.628641: irq_handler_entry: irq=55 > name=virtio5-output.0 > <idle>-0 [000] d.h1. 4690.628650: irq_handler_exit: irq=55 > ret=handled > <idle>-0 [007] d.H1. 4690.628895: irq_handler_entry: irq=54 > name=virtio5-input.0 > <idle>-0 [007] d.H1. 4690.628900: irq_handler_exit: irq=54 > ret=handled > <idle>-0 [000] d.h1. 4690.802977: irq_handler_entry: irq=41 > name=virtio2-req.0 > <idle>-0 [000] dNh1. 4690.803046: irq_handler_exit: irq=41 > ret=handled > <idle>-0 [000] d.h1. 4690.813958: irq_handler_entry: irq=41 > name=virtio2-req.0 > <idle>-0 [000] dNh1. 4690.814008: irq_handler_exit: irq=41 > ret=handled > <idle>-0 [000] d.H1. 4690.814257: irq_handler_entry: irq=41 > name=virtio2-req.0 > <idle>-0 [000] dNH1. 4690.814301: irq_handler_exit: irq=41 > ret=handled > <idle>-0 [000] d.H1. 4690.815374: irq_handler_entry: irq=41 > name=virtio2-req.0 > > # cat /proc/interrupts |grep -e 55 -e 54 -e 41 > 41: 462 0 0 0 0 0 > 0 0 PCI-MSIX-0000:04:00.0 1-edge virtio2-req.0 > 54: 0 0 0 0 0 0 > 0 4539 PCI-MSIX-0000:07:00.0 1-edge virtio5-input.0 > 55: 4544 0 0 0 0 0 > 0 0 PCI-MSIX-0000:07:00.0 2-edge virtio5-output.0 > RES: 1293 1411 1357 1307 1343 1345 > 1330 1315 Rescheduling interrupts > > We have the names right there. Heck, I could probably write a plugin for > trace-cmd to record the interrupt names in the trace.dat file and print > them out with the events. Any analysis tool could also record that. > > That's better than wasting ring buffer space for every event. > ...But you have a point I was not considering before (ring buffer space). > -- Steve >
