Hi Robin,

> -----Original Message-----
> From: Robin Murphy <robin.mur...@arm.com>
> Sent: Friday, February 28, 2020 8:32 PM
> 
> [ +Laurentiu ]
> 
> Hi Russell,
> 
> Thanks for sharing a log, now I properly understand what's up... further
> comments at the end (for context).
> 
> On 28/02/2020 10:06 am, Russell King - ARM Linux admin wrote:
> > On Fri, Feb 28, 2020 at 09:33:40AM +0000, John Garry wrote:
> >> On 28/02/2020 02:16, Lu Baolu wrote:
> >>> Hi,
> >>>
> >>> On 2020/2/27 19:57, Russell King wrote:
> >>>> On the LX2160A, there are lots (about 160) of IOMMU messages produced
> >>>> during boot; this is excessive.  Reduce the severity of these
> messages
> >>>> to debug level.
> >>>>
> >>>> Signed-off-by: Russell King <rmk+ker...@armlinux.org.uk>
> >>>> ---
> >>>>    drivers/iommu/iommu.c | 4 ++--
> >>>>    1 file changed, 2 insertions(+), 2 deletions(-)
> >>>>
> >>>> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> >>>> index 3ead597e1c57..304281ec623b 100644
> >>>> --- a/drivers/iommu/iommu.c
> >>>> +++ b/drivers/iommu/iommu.c
> >>>> @@ -741,7 +741,7 @@ int iommu_group_add_device(struct iommu_group
> >>>> *group, struct device *dev)
> >>>>        trace_add_device_to_group(group->id, dev);
> >>>> -    dev_info(dev, "Adding to iommu group %d\n", group->id);
> >>>> +    dev_dbg(dev, "Adding to iommu group %d\n", group->id);
> >>>
> >>> I'm not strongly against this. But to me this message seems to be a
> good
> >>> indicator that a device was probed successfully by the iommu subsystem.
> >>> Keeping it in the default kernel message always helps to the kernel
> >>> debugging.
> >>>
> >>
> >> I would tend to agree.
> >

[snip]

> >
> > # dmesg |grep 'Adding to iommu' | wc -l
> > 164
> > # dmesg |grep -v 'Adding to iommu' | wc -l
> > 551
> >
> > So, 23% of the kernel messages on this platform are "Adding to iommu",
> > which is excessive.
> 
> Indeed, however I would note that on most platforms bringing up a
> network interface involves hot-adding 0 devices, so hot-adding 19
> devices as full-blown DMA masters is arguably the root of "excessive"
> already. Per the concern I initially raised, each of those messages
> represents a whole bunch of internal allocation and bookkeeping going
> on, which if it isn't necessary would be far better avoided altogether,
> than simply done more quietly.
> 
> Laurentiu, I guess at the moment the nature of the of_dma_configure()
> integration means we end up treating all DPAA2 objects identically, but
> do you think we have scope to be a bit cleverer in that regard?
> Presumably not every type of object that shows up on the fsl_mc bus is
> really an independent DMA master, so if we could skip doing the full
> DMA/IOMMU/MSI setup for the ones that don't need it, it would work out
> nicer all round. In fact your .dma_configure proposal (which I'll try to
> take a proper look at next week) couldn't have come at a better time for
> that argument :)
 
Thanks! That's a very good point - I'll check on which devices we actually use 
dma apis and filter the rest out. Will keep in mind for the next spin of the 
patches.

---
Best Regards, Laurentiu

_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Reply via email to