linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: "Pankaj Dubey" <pankaj.dubey@samsung.com>
To: "'Andrew Murray'" <andrew.murray@arm.com>
Cc: <linux-pci@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
	"'Jingoo Han'" <jingoohan1@gmail.com>,
	<gustavo.pimentel@synopsys.com>, <lorenzo.pieralisi@arm.com>,
	"'Bjorn Helgaas'" <bhelgaas@google.com>,
	"'Anvesh Salveru'" <anvesh.s@samsung.com>
Subject: RE: [PATCH 1/2] PCI: dwc: Add support to disable GEN3 equalization
Date: Thu, 12 Sep 2019 17:09:41 +0530	[thread overview]
Message-ID: <23a601d5695e$c54f2b80$4fed8280$@samsung.com> (raw)
In-Reply-To: <20190911092310.GN9720@e119886-lin.cambridge.arm.com>



> From: Andrew Murray <andrew.murray@arm.com>
> 
> On Tue, Sep 10, 2019 at 09:46:28PM +0530, Pankaj Dubey wrote:
> > On Tue, 10 Sep 2019 at 19:56, Andrew Murray <andrew.murray@arm.com>
> wrote:
> > >
> > > On Tue, Sep 10, 2019 at 05:55:01PM +0530, Pankaj Dubey wrote:
> > > > From: Anvesh Salveru <anvesh.s@samsung.com>
> > > >
> > > > In some platforms, PCIe PHY may have issues which will prevent
> > > > linkup to happen in GEN3 or high speed. In case equalization
> > > > fails, link will fallback to GEN1.
> > >
> > > When you refer to "high speed", do you mean "higher speeds" as in
> > > GEN3, GEN4, etc?
> > >
> >
> > Yes. Will reword the commit message as "higher speeds"
> >
> > > >
> > > > Designware controller has support for disabling GEN3 equalization
> > > > if required. This patch enables the designware driver to disable
> > > > the PCIe
> > > > GEN3 equalization by writing into PCIE_PORT_GEN3_RELATED.
> > >
> > > Thus limiting to GEN2 speeds max, right?
> > >
> > > Is the purpose of PORT_LOGIC_GEN3_EQ_DISABLE to disable GEN3 and
> > > above even though we advertise GEN3 and above speeds? I.e. the IP
> > > advertises
> > > GEN3 but the phy can't handle it, we can't change what the IP
> > > advertises and so we disable equalization to limit to GEN2?
> > >
> > > I notice many of the other dwc drivers (dra7xx, keystone, tegra194,
> > > imx6) seem to use the device tree to specify a max-link-speed and
> > > then impose that limit by changing the value in PCI_EXP_LNKCAP. Is
> > > your PORT_LOGIC_GEN3_EQ_DISABLE approach and alternative to the
> > > PCI_EXP_LNKCAP approach, or does your approach add something else?
> > >
> >
> > No, max speed will be still as per advertised by link or it will be
> > equal to the limited speed as per DT property if any.
> > This register will prohibit to perform all phases of equalization and
> > thus allowing link to happen in maximum supported/advertised speed.
> >
> > This is not to limit max link speed, this register helps link to
> > happen in higher speeds (GEN3/4) without going through equalization
> > phases. It is intended to use only if at all link fails to latch up in
> > GEN3/4 due to failure in equalization phases.
> 
> I thought that for GEN3 and beyond equalization was *required* - with only
> phases 2 and 3 being optional. Therefore I'm suprised to see that if
equalization
> does fail we continue to train the link anyway. Have I understood this
correctly?
> 

AFAIK, equalization is not mandatory for GEN3 and higher speed. I mean in
case there is some H/W issue we can still go ahead and disable complete
equalization.

> Also are there any plans to provide patches to use this quirk on any
drivers?
> 

Yes, we have plans and soon the user driver of this feature will land for
review. But I believe this should not be blocker to consider this patch, as
it is a feature provided in Designware H/W and so we can add S/W support for
the same.

> >
> > > >
> > > > Platform drivers can disable equalization by setting the
> > > > dwc_pci_quirk flag DWC_EQUALIZATION_DISABLE.
> > > >
> > > > Signed-off-by: Anvesh Salveru <anvesh.s@samsung.com>
> > > > Signed-off-by: Pankaj Dubey <pankaj.dubey@samsung.com>
> > > > ---
> > > >  drivers/pci/controller/dwc/pcie-designware.c | 7 +++++++
> > > > drivers/pci/controller/dwc/pcie-designware.h | 7 +++++++
> > > >  2 files changed, 14 insertions(+)
> > > >
> > > > diff --git a/drivers/pci/controller/dwc/pcie-designware.c
> > > > b/drivers/pci/controller/dwc/pcie-designware.c
> > > > index 7d25102..bf82091 100644
> > > > --- a/drivers/pci/controller/dwc/pcie-designware.c
> > > > +++ b/drivers/pci/controller/dwc/pcie-designware.c
> > > > @@ -466,4 +466,11 @@ void dw_pcie_setup(struct dw_pcie *pci)
> > > >               break;
> > > >       }
> > > >       dw_pcie_writel_dbi(pci, PCIE_LINK_WIDTH_SPEED_CONTROL, val);
> > > > +
> > > > +     val = dw_pcie_readl_dbi(pci, PCIE_PORT_GEN3_RELATED);
> > > > +
> > > > +     if (pci->dwc_pci_quirk & DWC_EQUALIZATION_DISABLE)
> > > > +             val |= PORT_LOGIC_GEN3_EQ_DISABLE;
> > > > +
> > > > +     dw_pcie_writel_dbi(pci, PCIE_PORT_GEN3_RELATED, val);
> > >
> > > The problem here is that even when DWC_EQUALIZATION_DISABLE is not
> > > set the driver will read and write PCIE_PORT_GEN3_RELATED when it is
> > > not needed. How about something like:
> > >
> > > > +
> > > > +     if (pci->dwc_pci_quirk & DWC_EQUALIZATION_DISABLE) {
> > > > +             val = dw_pcie_readl_dbi(pci, PCIE_PORT_GEN3_RELATED);
> > > > +             val |= PORT_LOGIC_GEN3_EQ_DISABLE;
> > > > +             dw_pcie_writel_dbi(pci, PCIE_PORT_GEN3_RELATED, val);
> > > > +     }
> > >
> >
> > Yes, before posting we taught about it, but then next patchset is
> > adding one more quirk and in that case we need to repeat read and
> > write under each if condition. I hope that repetition should be fine.
> 
> I understand. I think the repetition is prefered over needlessly reading
and
> writing registers.
> 

OK, will handle this in v2.

> Given these quirks are so similar, I wouldn't have a problem with them
being in
> the same patch.
> 

I still prefer to keep these in two separate patches, unless until it is
something mandatory.

> Thanks,
> 
> Andrew Murray
> 
> >
> > > >  }
> > > > diff --git a/drivers/pci/controller/dwc/pcie-designware.h
> > > > b/drivers/pci/controller/dwc/pcie-designware.h
> > > > index ffed084..a1453c5 100644
> > > > --- a/drivers/pci/controller/dwc/pcie-designware.h
> > > > +++ b/drivers/pci/controller/dwc/pcie-designware.h
> > > > @@ -29,6 +29,9 @@
> > > >  #define LINK_WAIT_MAX_IATU_RETRIES   5
> > > >  #define LINK_WAIT_IATU                       9
> > > >
> > > > +/* Parameters for PCIe Quirks */
> > > > +#define DWC_EQUALIZATION_DISABLE     0x1
> > >
> > > How about using BIT(1) instead? Thus implying that you can combine
> > > quirks.
> > >
> >
> > Agreed.
> >
> > > Thanks,
> > >
> > > Andrew Murray
> > >
> > > > +
> > > >  /* Synopsys-specific PCIe configuration registers */
> > > >  #define PCIE_PORT_LINK_CONTROL               0x710
> > > >  #define PORT_LINK_MODE_MASK          GENMASK(21, 16)
> > > > @@ -60,6 +63,9 @@
> > > >  #define PCIE_MSI_INTR0_MASK          0x82C
> > > >  #define PCIE_MSI_INTR0_STATUS                0x830
> > > >
> > > > +#define PCIE_PORT_GEN3_RELATED               0x890
> > > > +#define PORT_LOGIC_GEN3_EQ_DISABLE   BIT(16)
> > > > +
> > > >  #define PCIE_ATU_VIEWPORT            0x900
> > > >  #define PCIE_ATU_REGION_INBOUND              BIT(31)
> > > >  #define PCIE_ATU_REGION_OUTBOUND     0
> > > > @@ -244,6 +250,7 @@ struct dw_pcie {
> > > >       struct dw_pcie_ep       ep;
> > > >       const struct dw_pcie_ops *ops;
> > > >       unsigned int            version;
> > > > +     unsigned int            dwc_pci_quirk;
> > > >  };
> > > >
> > > >  #define to_dw_pcie_from_pp(port) container_of((port), struct
> > > > dw_pcie, pp)
> > > > --
> > > > 2.7.4
> > > >


  reply	other threads:[~2019-09-12 11:39 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <CGME20190910122514epcas5p4f00c0f999333dd7707c0a353fd06b57f@epcas5p4.samsung.com>
2019-09-10 12:25 ` Pankaj Dubey
     [not found]   ` <CGME20190910122520epcas5p1faeb16f7c38ee057ce93783a637e6bf4@epcas5p1.samsung.com>
2019-09-10 12:25     ` [PATCH 2/2] PCI: dwc: Add support to disable equalization phase 2 and 3 Pankaj Dubey
2019-09-10 14:05       ` Andrew Murray
2019-09-10 16:21         ` Pankaj Dubey
2019-09-11  9:27           ` Andrew Murray
2019-09-12 11:44             ` Pankaj Dubey
2019-09-10 13:58   ` [PATCH 1/2] PCI: dwc: Add support to disable GEN3 equalization Andrew Murray
2019-09-10 16:16     ` Pankaj Dubey
2019-09-11  9:23       ` Andrew Murray
2019-09-12 11:39         ` Pankaj Dubey [this message]
2019-09-12 12:42           ` Andrew Murray
2019-09-12 14:21   ` Gustavo Pimentel
2019-09-13 10:53     ` Pankaj Dubey

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='23a601d5695e$c54f2b80$4fed8280$@samsung.com' \
    --to=pankaj.dubey@samsung.com \
    --cc=andrew.murray@arm.com \
    --cc=anvesh.s@samsung.com \
    --cc=bhelgaas@google.com \
    --cc=gustavo.pimentel@synopsys.com \
    --cc=jingoohan1@gmail.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pci@vger.kernel.org \
    --cc=lorenzo.pieralisi@arm.com \
    --subject='RE: [PATCH 1/2] PCI: dwc: Add support to disable GEN3 equalization' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).