On Tue, Apr 06, 2021 at 08:45:50PM +0800, Yicong Yang wrote:
> HiSilicon PCIe tune and trace device(PTT) is a PCIe Root Complex
> integrated Endpoint(RCiEP) device, providing the capability
> to dynamically monitor and tune the PCIe traffic(tune),
> and trace the TLP headers(trace). The driver exposes the user
> interface through debugfs, so no need for extra user space tools.
> The usage is described in the document.
Why use debugfs and not the existing perf tools for debugging?
thanks,
greg k-h
On 2021/4/6 21:49, Greg KH wrote:
> On Tue, Apr 06, 2021 at 08:45:50PM +0800, Yicong Yang wrote:
>> HiSilicon PCIe tune and trace device(PTT) is a PCIe Root Complex
>> integrated Endpoint(RCiEP) device, providing the capability
>> to dynamically monitor and tune the PCIe traffic(tune),
>> and trace the TLP headers(trace). The driver exposes the user
>> interface through debugfs, so no need for extra user space tools.
>> The usage is described in the document.
>
> Why use debugfs and not the existing perf tools for debugging?
>
The perf doesn't match our device as we've analyzed.
For the tune function it doesn't do the sampling at all.
User specifys one link parameter and reads its current value or set
the desired one. The process is static. We didn't find a
way to adapt to perf.
For the trace function, we may barely adapt to the perf framework
but it doesn't seems like a better choice. We have our own format
of data and don't need perf doing the parsing, and we'll get extra
information added by perf as well. The settings through perf tools
won't satisfy our needs, we cannot present available settings
(filter BDF number, TLP types, buffer controls) to
the user and user cannot set in a friendly way. For example,
we cannot count on perf to decode the usual format BDF number like
<domain>:<bus>:<dev>.<fn>, which user can use filter the TLP
headers.
So we intended to make the operation of this driver a bit like
ftrace. user sets the control settings through control files
and get the result through files as well. No additional tools
is necessay. A user space tool is necessary if we use a character
device or misc device for implementing this. The trace data maybe
hundreds of megabytes, and debugfs file can just satisfy it.
Thanks,
Yicong
On Wed, Apr 07, 2021 at 06:03:11PM +0800, Yicong Yang wrote:
> On 2021/4/6 21:49, Greg KH wrote:
> > On Tue, Apr 06, 2021 at 08:45:50PM +0800, Yicong Yang wrote:
> >> HiSilicon PCIe tune and trace device(PTT) is a PCIe Root Complex
> >> integrated Endpoint(RCiEP) device, providing the capability
> >> to dynamically monitor and tune the PCIe traffic(tune),
> >> and trace the TLP headers(trace). The driver exposes the user
> >> interface through debugfs, so no need for extra user space tools.
> >> The usage is described in the document.
> >
> > Why use debugfs and not the existing perf tools for debugging?
> >
>
> The perf doesn't match our device as we've analyzed.
>
> For the tune function it doesn't do the sampling at all.
> User specifys one link parameter and reads its current value or set
> the desired one. The process is static. We didn't find a
> way to adapt to perf.
>
> For the trace function, we may barely adapt to the perf framework
> but it doesn't seems like a better choice. We have our own format
> of data and don't need perf doing the parsing, and we'll get extra
> information added by perf as well. The settings through perf tools
> won't satisfy our needs, we cannot present available settings
> (filter BDF number, TLP types, buffer controls) to
> the user and user cannot set in a friendly way. For example,
> we cannot count on perf to decode the usual format BDF number like
> <domain>:<bus>:<dev>.<fn>, which user can use filter the TLP
> headers.
Please work with the perf developers to come up with a solution. I find
it hard to believe that your hardware is so different than all the other
hardware that perf currently supports. I would need their agreement
that you can not use perf before accepting this patchset.
thanks,
greg k-h
On 2021/4/7 18:25, Greg KH wrote:
> On Wed, Apr 07, 2021 at 06:03:11PM +0800, Yicong Yang wrote:
>> On 2021/4/6 21:49, Greg KH wrote:
>>> On Tue, Apr 06, 2021 at 08:45:50PM +0800, Yicong Yang wrote:
>>>> HiSilicon PCIe tune and trace device(PTT) is a PCIe Root Complex
>>>> integrated Endpoint(RCiEP) device, providing the capability
>>>> to dynamically monitor and tune the PCIe traffic(tune),
>>>> and trace the TLP headers(trace). The driver exposes the user
>>>> interface through debugfs, so no need for extra user space tools.
>>>> The usage is described in the document.
>>>
>>> Why use debugfs and not the existing perf tools for debugging?
>>>
>>
>> The perf doesn't match our device as we've analyzed.
>>
>> For the tune function it doesn't do the sampling at all.
>> User specifys one link parameter and reads its current value or set
>> the desired one. The process is static. We didn't find a
>> way to adapt to perf.
>>
>> For the trace function, we may barely adapt to the perf framework
>> but it doesn't seems like a better choice. We have our own format
>> of data and don't need perf doing the parsing, and we'll get extra
>> information added by perf as well. The settings through perf tools
>> won't satisfy our needs, we cannot present available settings
>> (filter BDF number, TLP types, buffer controls) to
>> the user and user cannot set in a friendly way. For example,
>> we cannot count on perf to decode the usual format BDF number like
>> <domain>:<bus>:<dev>.<fn>, which user can use filter the TLP
>> headers.
>
> Please work with the perf developers to come up with a solution. I find
> it hard to believe that your hardware is so different than all the other
> hardware that perf currently supports. I would need their agreement
> that you can not use perf before accepting this patchset.
>
Sure. I'll resend this series with more detailed information and with perf list
and developers cc'ed to collect more suggestions on this device and driver.
Thanks,
Yicong