lkml.org 
[lkml]   [2021]   [Jul]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH v5 0/3] Add Qualcomm PCIe Endpoint driver support
On Mon, Jul 12, 2021 at 1:53 AM Manivannan Sadhasivam
<manivannan.sadhasivam@linaro.org> wrote:
>
> On Thu, Jul 01, 2021 at 09:25:01AM -0600, Rob Herring wrote:
> > On Tue, Jun 29, 2021 at 9:47 PM Manivannan Sadhasivam
> > <manivannan.sadhasivam@linaro.org> wrote:
> > >
> > > Hello,
> > >
> > > This series adds support for Qualcomm PCIe Endpoint controller found
> > > in platforms like SDX55. The Endpoint controller is based on the designware
> > > core with additional Qualcomm wrappers around the core.
> > >
> > > The driver is added separately unlike other Designware based drivers that
> > > combine RC and EP in a single driver. This is done to avoid complexity and
> > > to maintain this driver autonomously.
> > >
> > > The driver has been validated with an out of tree MHI function driver on
> > > SDX55 based Telit FN980 EVB connected to x86 host machine over PCIe.
> > >
> > > Thanks,
> > > Mani
> > >
> > > Changes in v5:
> > >
> > > * Removed the DBI register settings that are not needed
> > > * Used the standard definitions available in pci_regs.h
> > > * Added defines for all the register fields
> > > * Removed the left over code from previous iteration
> > >
> > > Changes in v4:
> > >
> > > * Removed the active_config settings needed for IPA integration
> > > * Switched to writel for couple of relaxed versions that sneaked in
> >
> > I thought we resolved this discussion. Use _relaxed variants unless
> > you need the stronger ones.
> >
>
> I thought the discussion was resolved in favor of using read/writel. Here
> is the last reply from Bjorn:
>
> "I think we came to the conclusion that writel() was better
> than incorrect use of writel_relaxed() followed by wmb(). And in this
> particular case it's definitely not happening in a hot code path..."

Certainly if you're needing wmb(), then you shouldn't be using
_relaxed() variants.

> IMO, it is safer to use readl/writel calls than the relaxed variants.

Sure, and it's safer to have one big lock than it is to have no locks
or lots of small locks.

> And so far the un-written rule I assumed is, only consider using the
> relaxed variants if the code is in hot path (but somehow I used the
> relaxed version in v1 :P )

If you want any real conclusion, then best to fix the 'un-written' part.

But what I say for every PCI review, is use the _relaxed() variants.

Rob

\
 
 \ /
  Last update: 2021-07-12 21:57    [W:0.071 / U:0.168 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site