Re: [RFC PATCH] usb: xhci-mtk: handle bandwidth table rollover

[Chunfeng Yun (云春峰)]

On Wed, 2021-08-18 at 10:43 +0800, Ikjoon Jang wrote:
> Hi Chunfeng,
> 
> On Thu, Aug 12, 2021 at 7:49 PM Chunfeng Yun (云春峰)
> <Chunfeng.Yun@xxxxxxxxxxxx> wrote:
> > 
> > On Thu, 2021-08-12 at 17:31 +0800, Ikjoon Jang wrote:
> > > HI,
> > > 
> > > On Wed, Aug 11, 2021 at 5:02 PM Chunfeng Yun (云春峰)
> > > <Chunfeng.Yun@xxxxxxxxxxxx> wrote:
> > > > 
> > > > On Mon, 2021-08-09 at 17:42 +0800, Ikjoon Jang wrote:
> > > > > On Mon, Aug 9, 2021 at 5:11 PM Greg Kroah-Hartman
> > > > > <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > > > > > 
> > > > > > On Mon, Aug 09, 2021 at 04:59:29PM +0800, Ikjoon Jang
> > > > > > wrote:
> > > > > > > xhci-mtk has 64 slots for periodic bandwidth calculations
> > > > > > > and
> > > > > > > each
> > > > > > > slot represents byte budgets on a microframe. When an
> > > > > > > endpoint's
> > > > > > > allocation sits on the boundary of the table, byte
> > > > > > > budgets'
> > > > > > > slot
> > > > > > > should be rolled over but the current implementation
> > > > > > > doesn't.
> > > > > > > 
> > > > > > > This patch applies a 6 bits mask to the microframe index
> > > > > > > to
> > > > > > > handle
> > > > > > > its rollover 64 slots and prevent out-of-bounds array
> > > > > > > access.
> > > > > > > 
> > > > > > > Signed-off-by: Ikjoon Jang <ikjn@xxxxxxxxxxxx>
> > > > > > > ---
> > > > > > > 
> > > > > > >  drivers/usb/host/xhci-mtk-sch.c | 79 +++++++++--------
> > > > > > > ----
> > > > > > > ----
> > > > > > > --------
> > > > > > >  drivers/usb/host/xhci-mtk.h     |  1 +
> > > > > > >  2 files changed, 23 insertions(+), 57 deletions(-)
> > > > > > 
> > > > > > Why is this "RFC"?  What needs to be addressed in this
> > > > > > change
> > > > > > before it
> > > > > > can be accepted?
> > > > > 
> > > > > sorry, I had to mention why this is RFC:
> > > > > 
> > > > > I simply don't know about the details of the xhci-mtk
> > > > > internals.
> > > > > It was okay from my tests with mt8173 and I think this will
> > > > > be
> > > > > harmless
> > > > > as this is "better than before".
> > > > > 
> > > > > But when I removed get_esit_boundary(), I really have no idea
> > > > > why
> > > > > it was there. I'm wondering if there was another reason of
> > > > > that
> > > > > function
> > > > > other than just preventing out-of-bounds. Maybe chunfeng can
> > > > > answer
> > > > > this?
> > > > 
> > > > We use @esit to prevent out-of-bounds array access. it's not a
> > > > ring,
> > > > can't insert out-of-bounds value into head slot.
> > > 
> > > Thanks, so that function was only for out-of-bounds array access.
> > > then I think we just can remove that function and use it as a
> > > ring.
> > > Can you tell me _why_ it can't be used as a ring?
> > 
> > Treat it as a period, roll over slot equals to put it into the next
> > period.
> > 
> > > 
> > > I think a transaction (e.g. esit_boundary = 7) can start its
> > > first
> > > SSPLIT
> > > from Y_7 (offset = 7). But will that allocation be matched with
> > > this?
> > > 
> > > -               if ((offset + sch_ep->num_budget_microframes) >
> > > esit_boundary)
> > > -                       break;
> > > 
> > > I mean I'm not sure why this is needed.
> > 
> > Prevent out-of-bounds.
> 
> If it was for preventing drivers from out-of-bound array access,
> I couldn't find any reasons why we cannot remove the above lines.
> So can I know if it was just for preventing xhci-mtk drivers from
> out-of-bounds array access?
Due to it use an array to calculate bandwidth, if use ring, can remove
it.

> 
> If xhci-mtk HC itself can continue the transaction from Y_7 to
> (Y+1)_n;
> including the case of Y==63, I think it's just okay to rollover to
> (Y+1).
> 
> If it's prohibited by xhci-mtk hw, or if you think this patch is not
> allowed by any other reasons, can you please  tell me what
> kinds of problems can happen with this patch?
Seems sw limitation, or avoid repeated calculation;
I'll change it as a ring, and do some tests.
> 
> Otherwise, please consider minimizing the bw constraints from
> xhci-mtk-sch on your side. Note that we're still having other usb
> audio headsets which cannot be configured with xhci-mtk
> even with this patch.
Ok, try my best to do it, thanks a lot

> 
> Thanks.
> 
> > 
> > > 
> > > Until now, I couldn't find a way to accept the USB audio headset
> > > with a configuration of { INT-IN 64 + ISOC-OUT 384 + ISOC-IN 192
> > > }
> > > without this patch.
> > 
> > what is the interval value of each endpoint?
> > 
> > > 
> > > > 
> > > > > 
> > > > > Thanks!
> > > > > 
> > > > > > 
> > > > > > thanks,
> > > > > > 
> > > > > > greg k-h