lkml.org 
[lkml]   [2019]   [Apr]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH V2 1/2] trace,x86: Add external_interrupts to the irq_vectors class
From
Date
On 4/5/19 11:15 PM, Thomas Gleixner wrote:
> On Mon, 1 Apr 2019, Daniel Bristot de Oliveira wrote:
>
>> Currently, the irq_vectors is showing the entry and exit events for
>> the interrupts of the architecture, but not for external interrupts.
>
> Those are covered by the irq tracepoints. Is there a really good reason why
> we need both?

The irq_handler_* tracepoints might give the "imprecise" idea that more than one
interrupts were raised when we have shared handlers. For instance:

------------ %< ----------------
f-892 [000] d.h. 790.617251: external_interrupt_entry: vector=37
f-892 [000] d.h. 790.617257: irq_handler_entry: irq=11 name=uhci_hcd:usb3
f-892 [000] d.h. 790.617343: irq_handler_exit: irq=11 ret=handled
f-892 [000] d.h. 790.617343: irq_handler_entry: irq=11 name=uhci_hcd:usb4
f-892 [000] d.h. 790.617349: irq_handler_exit: irq=11 ret=unhandled
f-892 [000] d.h. 790.617350: irq_handler_entry: irq=11 name=qxl
f-892 [000] d.h. 790.617360: irq_handler_exit: irq=11 ret=handled
f-892 [000] d.h. 790.617387: external_interrupt_exit: vector=37
------------ >% ----------------

In this case, a single interrupt occurrence (vector 37) caused two handlers to
handle their interrupt.

From a latency analysis perspective, the external_interrupt_* tracepoints turn
clearer that a single interrupt interfered in the thread execution happened, not
two separated executions of the same vector.

The outer-most tracepoints also help to have a more precise accounting of the
interference:

Using the sum of the irq_handler_* tracepoints we have:
(790.617343−790.617257)+(790.617349−790.617343)+(790.617360−790.617350) = 102 us

While using the irq_vector based one:
790.617387-790.617251 = 136 us

So, the proposed tracepoints help clarify the logical sequence of the interrupt
handling, while increasing the precision of the measurements.

Thoughts?

-- Daniel
> Thanks,
>
> tglx
>

\
 
 \ /
  Last update: 2019-04-08 11:49    [W:0.052 / U:7.212 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site