[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5d0b26b1-ffc5-94dd-ced2-52cfef3842a4@linux.ibm.com>
Date: Thu, 19 Jan 2023 11:12:11 -0500
From: Matthew Rosato <mjrosato@...ux.ibm.com>
To: Niklas Schnelle <schnelle@...ux.ibm.com>,
Joerg Roedel <joro@...tes.org>, Will Deacon <will@...nel.org>,
Robin Murphy <robin.murphy@....com>,
Jason Gunthorpe <jgg@...dia.com>,
Wenjia Zhang <wenjia@...ux.ibm.com>
Cc: Gerd Bayer <gbayer@...ux.ibm.com>,
Pierre Morel <pmorel@...ux.ibm.com>, iommu@...ts.linux.dev,
linux-s390@...r.kernel.org, borntraeger@...ux.ibm.com,
hca@...ux.ibm.com, gor@...ux.ibm.com,
gerald.schaefer@...ux.ibm.com, agordeev@...ux.ibm.com,
svens@...ux.ibm.com, linux-kernel@...r.kernel.org,
Julian Ruess <julianr@...ux.ibm.com>
Subject: Re: [PATCH v4 4/7] s390/pci: Use dma-iommu layer
On 1/19/23 11:04 AM, Niklas Schnelle wrote:
> On Thu, 2023-01-19 at 10:59 -0500, Matthew Rosato wrote:
>> On 1/19/23 6:03 AM, Niklas Schnelle wrote:
>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>> static char *pci_sw_names[] = {
>>>>>>>>>>>>>>>>>>> - "Allocated pages",
>>>>>>>>>>>>>>>>>>> +/* TODO "Allocated pages", */
>>>>>>>>>>>
>>>>>>>>>>> ? Forgot to finish this?
>>>
>>> Definitely forgot to remove the TODO. I think my latest plan was to
>>> just remove this counter. With the DMA API conversion the
>>> dma_map_ops.alloc and dma_map_ops.free move to common code and I don't
>>> see how we could differentiate these from map/unmap on our side. I'm
>>> not sure how helpful this counter really is either. If you're
>>> interested in how many pages are mapped long term I think it makes more
>>> sense to look at the difference between mapped and unmapped pages. What
>>> do you think?
>>>>>>>>>>
>>
>> Sounds reasonable to me, but I also note that without this series, when viewing statistics for a device, mapped - unmapped != allocated. Maybe allocated pages was already broken, or is it taking into account something else that mapped - unmapped would not (maybe mapping the same page multiple times)?
>>
>>
>
> Allocated Pages only counts the memory allocated via dma_map_ops.alloc
> so it would not count long term mappings of memory the driver allocated
> differently and then mapped for long term use.
Oh, right, I see it now.
Seems to me then that mapped-unmapped is more indicative of the actual footprint anyway so in the absence of an obvious analogue I'm fine with just getting rid of it.
Powered by blists - more mailing lists