Re: [PATCH 0/4] do not use sg if not properly supported by usb controller

[Stanislaw Gruszka <sgruszka@redhat.com>]

On Wed, Feb 13, 2019 at 12:00:42PM +0100, Lorenzo Bianconi wrote:
> On Feb 13, Stanislaw Gruszka wrote:
> > On Tue, Feb 12, 2019 at 11:09:03PM +0100, Lorenzo Bianconi wrote:
> > > > > On Tue, Feb 12, 2019 at 04:08:37PM +0100, Lorenzo Bianconi wrote:
> > > > > > > > > What unnecessary operation ?
> > > > > > > > 
> > > > > > > > the ones in mt76u_fill_bulk_urb()
> > > > > > > 
> > > > > > > Your patches also add extra operations on hotpath due to urb->num_sgs and
> > > > > > > dev->usb.sg_en checks.
> > > > > > 
> > > > > > here I guess you should use mt76u_check_sg() instead of
> > > > > > udev->bus->sg_tablesize > 0 so I think you need to precompute it since it is in
> > > > > > the hotpath
> > > > > 
> > > > > It's not necessary. If udev->bus->sg_tablesize > 0 is true and rest
> > > > > of mt76u_check_sg() is not, we will submit urb with urb->num_sgs = 1 to
> > > > > usb host driver, what is fine. 
> > > > > 
> > > > > > Moreover the RFC series has been tested by multiple users and on multiple
> > > > > > devices
> > > > > 
> > > > > I know. Perhaps you could test my patch on rpi ?
> > > > 
> > > > sure, I will do it later
> > > 
> > > I confirm that even with Stanislaw's patch the usb dongle is properly working
> > > on rpi3+
> > 
> > Thanks for testing. Would be ok to you to post my patch against
> > wireless-drivers tree and cc stable as fix for non-SG usb hosts,
> > drop your set for -next and work for fix for SG issue on AMD IOMMU?
> 
> I agree that your patch works (since it does not use SG I/O :)) but

It still use SG in mt76usb driver (urb->sg, sg_set_page(),
sg_init_marker(), etc ...) for all usb host controllers. It just
not submit urb->num_sgs > 1 to USB host driver if it does not
support SG.

> I think it is more clear (and manageable) to have two separated
> routines for memory allocation.

Hmm, I disagree on that and don't understand how you consider clearer
and manageable design with two separate buffer allocation methods
and bunch of extra ->num_sgs and ->sg_en checks, compered to solution
with one allocation method and without those checks.

Also some additional cleanups/simplification can be done after
applying my patch in mt76usb not possible after applying your set.

> Moreover I think that this check has to be done in the control plane instead
> of the data plane, so I would like to spend some more time in order to see if it is
> possible to remove some checks in the hot-path

I'm not sure what you mean by data-pane and control-plane in this context.

I considered to do mt76u_fill_buk_urb:

urb->num_sgs = (buf->num_sgs > 1) ? buf->num_sgs : 0;

Instead of usb->bus->sg_tablesize check. It should work, but it's not
what usb_sg_init() does and as pointed by Alan that function always
do the right thing, so I just copied code from there.

Beside that I don't think this check in mt76u_fill_buk_urb()
affects performance whatsoever.

Stanislaw