[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200313141416.00002e89@Huawei.com>
Date: Fri, 13 Mar 2020 14:14:16 +0000
From: Jonathan Cameron <Jonathan.Cameron@...wei.com>
To: Robin Murphy <robin.murphy@....com>
CC: Qi Liu <liuqi115@...wei.com>, <will@...nel.org>,
<mark.rutland@....com>, <bhelgaas@...gle.com>,
<linux-pci@...r.kernel.org>, <linuxarm@...wei.com>,
<linux-arm-kernel@...ts.infradead.org>,
<linux-kernel@...r.kernel.org>
Subject: Re: [PATCH RFC] perf:Add driver for HiSilicon PCIe PMU
On Fri, 13 Mar 2020 13:23:53 +0000
Robin Murphy <robin.murphy@....com> wrote:
> On 2020-03-12 12:06 pm, Qi Liu wrote:
> > From: Qi liu <liuqi115@...wei.com>
> >
> > PCIe PMU Root Complex Integrate End Point(IEP) device is
> > supported to sample bandwidth, latency, buffer occupation,
> > bandwidth utilization etc.
> > Each PMU IEP device monitors multiple root ports, and each
> > IEP is registered as a pmu in /sys/bus/event_source/devices,
> > so users can select the target IEP, and use filter to select
> > root port, function and event.
> > Filtering options are:
> > event: - select the event.
> > subevent: - select the subevent.
> > port: - select target root port.
> > func: - select target EP device under the port.
> >
> > Example: hisi_pcie_00_14_00/event=0x08,subevent=0x04, \
> > port=0x05,func=0x00/ -I 1000
> >
> > PMU IEP device is described by its bus, device and function,
> > target root port is 0x05 and target EP under it is function
> > 0x00. Subevent 0x04 of event 0x08 is sampled.
> >
> > Note that in this RFC:
> > 1. PCIe PMU IEP hardware is still in development.
> > 2. Perf common event list is undetermined, and name of these
> > events still need to be discussed.
> > 3. port filter could only select one port each time.
> > 4. There are two possible schemes of pmu registration, one is
> > register each root port as a pmu, it is easier for users to
> > select target port. The other one is register each IEP as pmu,
> > for counters are per IEP, not per root port, the second scheme
> > describes hardware PMC much more reasonable, but need to add
> > "port" filter option to select port. We use the second one in
> > this RFC.
> >
Whilst it's great to have detailed feedback, just to make it clear...
This is an RFC for the reasons above. They include that the hardware
is still in development - i.e. we can't test it because they've not finished
implementation yet.
The intention of posting so early is to get some feedback on the general
approach and the specific points in 2,3 and 4 above.
The key fiddly point with this is that it is a shared PMU across a set
of PCIe Root Ports (there are several of these devices on each SoC in
the system covering a set of ports each).
That makes for a somewhat fiddly interface, hence the RFC.
Thanks to everyone who has reviewed though as definitely some stuff for liuqi to
fix!
Jonathan
Powered by blists - more mailing lists