[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <1375257e-38c3-f9e3-049e-d1a602d084c9@linux.ibm.com>
Date: Thu, 13 Sep 2018 11:12:09 +0200
From: Halil Pasic <pasic@...ux.ibm.com>
To: Tony Krowiak <akrowiak@...ux.ibm.com>,
Cornelia Huck <cohuck@...hat.com>,
Tony Krowiak <akrowiak@...ux.vnet.ibm.com>
Cc: linux-s390@...r.kernel.org, linux-kernel@...r.kernel.org,
kvm@...r.kernel.org, freude@...ibm.com, schwidefsky@...ibm.com,
heiko.carstens@...ibm.com, borntraeger@...ibm.com,
kwankhede@...dia.com, bjsdjshi@...ux.vnet.ibm.com,
pbonzini@...hat.com, alex.williamson@...hat.com,
pmorel@...ux.vnet.ibm.com, alifm@...ux.vnet.ibm.com,
mjrosato@...ux.vnet.ibm.com, jjherne@...ux.vnet.ibm.com,
thuth@...hat.com, pasic@...ux.vnet.ibm.com, berrange@...hat.com,
fiuczy@...ux.vnet.ibm.com, buendgen@...ibm.com,
frankja@...ux.ibm.com
Subject: Re: [PATCH v9 13/22] s390: vfio-ap: sysfs interface to view matrix
mdev matrix
On 09/12/2018 07:01 PM, Tony Krowiak wrote:
> On 08/20/2018 10:08 AM, Cornelia Huck wrote:
>> I find this output to be a bit confusing; but OTOH, I'm probably not
>> the person to parse it :) Still, some comments.
>>
>> From previous discussions, ISTR that this is mainly supposed to be a
>> debugging/administration aid. Of course, this generates some questions:
>> - Should this be in sysfs (sysfs attributes are supposed to follow the
>> "one value per file" rule, at least for the most part), or would
>> debugfs be a better fit?
>> - Should userspace code be able to introspect the current
>> configuration? If yes, it might be better to have some
>> not-so-nice-but-easily-parsable output, possibly one attribute for
>> the assigned adapters and one for the assigned domains, and a tool
>> which distills that into a nice "matrix" with labeled rows and
>> columns for human consumption.
>>
>> That said, I don't really have major objections to that interface.
>
> For now, I will leave it in, but we could conceivably simplify configuration
> of the matrix and adhere to sysfs rules by modeling this after the AP bus
> apmask and aqmask sysfs interfaces. If we did that, there would need to be only
> three RW sysfs interfaces:
>
> apm or apmask or ap_mask or adapter_mask or apid_mask
> aqm or aqmask or aq_mask or domains or usage_domain_mask or apqi_mask
> adm or admask or ad_mask or control_domains or control_domain_mask or domain_mask
>
> To assign a device, either a bitmask or a comma separated list of IDs prepended
> with a '+' could be passed in. To unassign a device, either a bitmask in or a
> comma-separated list of IDs prepended with a '-' could be passed in.
>
> Reading an attribute would return the mask. As you suggested, tools could be
> provided to parse the output and display it in a human-readable format.
>
I would prefer an interface where one can set at least one whole mask in one
go over the current one that let us manipulate just one bit at a time.
Hallil
>>
>>> Signed-off-by: Tony Krowiak <akrowiak@...ux.ibm.com>
>>> Reviewed-by: Halil Pasic <pasic@...ux.ibm.com>
>>> Tested-by: Michael Mueller <mimu@...ux.ibm.com>
>>> Tested-by: Farhan Ali <alifm@...ux.ibm.com>
>>> Tested-by: Pierre Morel <pmorel@...ux.ibm.com>
>>> Signed-off-by: Christian Borntraeger <borntraeger@...ibm.com>
>>> ---
>>> drivers/s390/crypto/vfio_ap_ops.c | 52 +++++++++++++++++++++++++++++++++++++
>>> 1 files changed, 52 insertions(+), 0 deletions(-)
>
>
Powered by blists - more mailing lists