diff mbox series

PCI: amlogic: meson: Don't use FAST_LINK_MODE to set up link

Message ID 20200429164230.309922-1-maz@kernel.org
State New
Headers show
Series PCI: amlogic: meson: Don't use FAST_LINK_MODE to set up link | expand

Commit Message

Marc Zyngier April 29, 2020, 4:42 p.m. UTC 
My vim3l board stubbornly refuses to play ball with a bog
standard PCIe switch (ASM1184e), spitting all kind of errors
ranging from link never coming up to crazy things like downstream
ports falling off the face of the planet.

Upon investigating how the PCIe RC is configured, I found the
following nugget: the Sysnopsys DWC PCIe Reference Manual, in the
section dedicated to the PLCR register, describes bit 7 (FAST_LINK_MODE)
as:

"Sets all internal timers to fast mode for simulation purposes."

I completely understand the need for setting this bit from a simulation
perspective, but what I have on my desk is actual silicon, which
expects timers to have a nominal value (and I expect this is the
case for most people).

Making sure the FAST_LINK_MODE bit is cleared when configuring the RC
solves this problem.

Fixes: 9c0ef6d34fdb ("PCI: amlogic: Add the Amlogic Meson PCIe controller driver")
Signed-off-by: Marc Zyngier <maz@kernel.org>
---
 drivers/pci/controller/dwc/pci-meson.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Comments

Marc Zyngier May 6, 2020, 10:43 a.m. UTC | #1 
On Wed, 29 Apr 2020 17:42:30 +0100
Marc Zyngier <maz@kernel.org> wrote:

> My vim3l board stubbornly refuses to play ball with a bog
> standard PCIe switch (ASM1184e), spitting all kind of errors
> ranging from link never coming up to crazy things like downstream
> ports falling off the face of the planet.
> 
> Upon investigating how the PCIe RC is configured, I found the
> following nugget: the Sysnopsys DWC PCIe Reference Manual, in the
> section dedicated to the PLCR register, describes bit 7 (FAST_LINK_MODE)
> as:
> 
> "Sets all internal timers to fast mode for simulation purposes."
> 
> I completely understand the need for setting this bit from a simulation
> perspective, but what I have on my desk is actual silicon, which
> expects timers to have a nominal value (and I expect this is the
> case for most people).
> 
> Making sure the FAST_LINK_MODE bit is cleared when configuring the RC
> solves this problem.
> 
> Fixes: 9c0ef6d34fdb ("PCI: amlogic: Add the Amlogic Meson PCIe controller driver")
> Signed-off-by: Marc Zyngier <maz@kernel.org>
> ---
>  drivers/pci/controller/dwc/pci-meson.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/pci/controller/dwc/pci-meson.c b/drivers/pci/controller/dwc/pci-meson.c
> index 3715dceca1bf..ca59ba9e0ecd 100644
> --- a/drivers/pci/controller/dwc/pci-meson.c
> +++ b/drivers/pci/controller/dwc/pci-meson.c
> @@ -289,11 +289,11 @@ static void meson_pcie_init_dw(struct meson_pcie *mp)
>  	meson_cfg_writel(mp, val, PCIE_CFG0);
>  
>  	val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
> -	val &= ~LINK_CAPABLE_MASK;
> +	val &= ~(LINK_CAPABLE_MASK | FAST_LINK_MODE);
>  	meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
>  
>  	val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
> -	val |= LINK_CAPABLE_X1 | FAST_LINK_MODE;
> +	val |= LINK_CAPABLE_X1;
>  	meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
>  
>  	val = meson_elb_readl(mp, PCIE_GEN2_CTRL_OFF);

Yue, Kevin: any comment on this?

I found that the issue is reproducible even without a PCIe switch,
depending on the single device I plug in this machine (an Intel SSD
works fine, while a Marvell Ethernet adapter never shows up) as the
LTSSM times out much earlier than it really should (HW timers running
too quickly). Applying this patch makes every single device I have
lying around work fine.

Thanks,

	M.
Rob Herring May 7, 2020, 9:03 p.m. UTC | #2 
On Wed, 29 Apr 2020 17:42:30 +0100, Marc Zyngier wrote:
> My vim3l board stubbornly refuses to play ball with a bog
> standard PCIe switch (ASM1184e), spitting all kind of errors
> ranging from link never coming up to crazy things like downstream
> ports falling off the face of the planet.
> 
> Upon investigating how the PCIe RC is configured, I found the
> following nugget: the Sysnopsys DWC PCIe Reference Manual, in the
> section dedicated to the PLCR register, describes bit 7 (FAST_LINK_MODE)
> as:
> 
> "Sets all internal timers to fast mode for simulation purposes."
> 
> I completely understand the need for setting this bit from a simulation
> perspective, but what I have on my desk is actual silicon, which
> expects timers to have a nominal value (and I expect this is the
> case for most people).
> 
> Making sure the FAST_LINK_MODE bit is cleared when configuring the RC
> solves this problem.
> 
> Fixes: 9c0ef6d34fdb ("PCI: amlogic: Add the Amlogic Meson PCIe controller driver")
> Signed-off-by: Marc Zyngier <maz@kernel.org>
> ---
>  drivers/pci/controller/dwc/pci-meson.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 

Acked-by: Rob Herring <robh@kernel.org>
Marc Zyngier May 9, 2020, 1:07 p.m. UTC | #3 
Hi Yue,

On Thu, 07 May 2020 02:43:31 +0100,
"yue.wang@amlogic.com" <yue.wang@amlogic.com> wrote:
> 
> [1  <text/plain; utf-8 (base64)>]
> Marc,
> 
> This patch looks all right. I tested in my meson board and pcie
> EP(QCA9888) worked well.
> 
> Fast link mode is enabled for simulation purposes, it should be
> disabled in the real hardware.

Thanks for confirming my reading of the manual and having tested it.
Can I take this as an "Acked-by:" and a "Tested-by:"?

Cheers,

	M.

>
> yue.wang@amlogic.com
>  
> From: Marc Zyngier
> Date: 2020-05-06 18:43
> To: linux-pci; linux-amlogic; linux-arm-kernel; linux-kernel; Kevin Hilman; Yue Wang
> CC: Bjorn Helgaas; Rob Herring; Lorenzo Pieralisi
> Subject: Re: [PATCH] PCI: amlogic: meson: Don't use FAST_LINK_MODE to set up link
> On Wed, 29 Apr 2020 17:42:30 +0100
> Marc Zyngier <maz@kernel.org> wrote:
>  
> > My vim3l board stubbornly refuses to play ball with a bog
> > standard PCIe switch (ASM1184e), spitting all kind of errors
> > ranging from link never coming up to crazy things like downstream
> > ports falling off the face of the planet.
> > 
> > Upon investigating how the PCIe RC is configured, I found the
> > following nugget: the Sysnopsys DWC PCIe Reference Manual, in the
> > section dedicated to the PLCR register, describes bit 7 (FAST_LINK_MODE)
> > as:
> > 
> > "Sets all internal timers to fast mode for simulation purposes."
> > 
> > I completely understand the need for setting this bit from a simulation
> > perspective, but what I have on my desk is actual silicon, which
> > expects timers to have a nominal value (and I expect this is the
> > case for most people).
> > 
> > Making sure the FAST_LINK_MODE bit is cleared when configuring the RC
> > solves this problem.
> > 
> > Fixes: 9c0ef6d34fdb ("PCI: amlogic: Add the Amlogic Meson PCIe controller driver")
> > Signed-off-by: Marc Zyngier <maz@kernel.org>
> > ---
> >  drivers/pci/controller/dwc/pci-meson.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/pci/controller/dwc/pci-meson.c b/drivers/pci/controller/dwc/pci-meson.c
> > index 3715dceca1bf..ca59ba9e0ecd 100644
> > --- a/drivers/pci/controller/dwc/pci-meson.c
> > +++ b/drivers/pci/controller/dwc/pci-meson.c
> > @@ -289,11 +289,11 @@ static void meson_pcie_init_dw(struct meson_pcie *mp)
> >  meson_cfg_writel(mp, val, PCIE_CFG0);
> >  
> >  val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
> > - val &= ~LINK_CAPABLE_MASK;
> > + val &= ~(LINK_CAPABLE_MASK | FAST_LINK_MODE);
> >  meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
> >  
> >  val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
> > - val |= LINK_CAPABLE_X1 | FAST_LINK_MODE;
> > + val |= LINK_CAPABLE_X1;
> >  meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
> >  
> >  val = meson_elb_readl(mp, PCIE_GEN2_CTRL_OFF);
>  
> Yue, Kevin: any comment on this?
>  
> I found that the issue is reproducible even without a PCIe switch,
> depending on the single device I plug in this machine (an Intel SSD
> works fine, while a Marvell Ethernet adapter never shows up) as the
> LTSSM times out much earlier than it really should (HW timers running
> too quickly). Applying this patch makes every single device I have
> lying around work fine.
>  
> Thanks,
>  
> M.
> -- 
> Jazz is not dead. It just smells funny...
>  
> [2  <text/html; utf-8 (quoted-printable)>]
Neil Armstrong May 11, 2020, 8:58 a.m. UTC | #4 
On 29/04/2020 18:42, Marc Zyngier wrote:
> My vim3l board stubbornly refuses to play ball with a bog
> standard PCIe switch (ASM1184e), spitting all kind of errors
> ranging from link never coming up to crazy things like downstream
> ports falling off the face of the planet.
> 
> Upon investigating how the PCIe RC is configured, I found the
> following nugget: the Sysnopsys DWC PCIe Reference Manual, in the
> section dedicated to the PLCR register, describes bit 7 (FAST_LINK_MODE)
> as:
> 
> "Sets all internal timers to fast mode for simulation purposes."
> 
> I completely understand the need for setting this bit from a simulation
> perspective, but what I have on my desk is actual silicon, which
> expects timers to have a nominal value (and I expect this is the
> case for most people).
> 
> Making sure the FAST_LINK_MODE bit is cleared when configuring the RC
> solves this problem.
> 
> Fixes: 9c0ef6d34fdb ("PCI: amlogic: Add the Amlogic Meson PCIe controller driver")
> Signed-off-by: Marc Zyngier <maz@kernel.org>
> ---
>  drivers/pci/controller/dwc/pci-meson.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/pci/controller/dwc/pci-meson.c b/drivers/pci/controller/dwc/pci-meson.c
> index 3715dceca1bf..ca59ba9e0ecd 100644
> --- a/drivers/pci/controller/dwc/pci-meson.c
> +++ b/drivers/pci/controller/dwc/pci-meson.c
> @@ -289,11 +289,11 @@ static void meson_pcie_init_dw(struct meson_pcie *mp)
>  	meson_cfg_writel(mp, val, PCIE_CFG0);
>  
>  	val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
> -	val &= ~LINK_CAPABLE_MASK;
> +	val &= ~(LINK_CAPABLE_MASK | FAST_LINK_MODE);
>  	meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
>  
>  	val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
> -	val |= LINK_CAPABLE_X1 | FAST_LINK_MODE;
> +	val |= LINK_CAPABLE_X1;
>  	meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
>  
>  	val = meson_elb_readl(mp, PCIE_GEN2_CTRL_OFF);
> 

I don't have HW to test on non NVMe, but I'm reading the same as you in the
DWC PCIe Reference Manual, and it seems coherent.

Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>

Neil
Lorenzo Pieralisi May 11, 2020, 10:22 a.m. UTC | #5 
On Wed, Apr 29, 2020 at 05:42:30PM +0100, Marc Zyngier wrote:
> My vim3l board stubbornly refuses to play ball with a bog
> standard PCIe switch (ASM1184e), spitting all kind of errors
> ranging from link never coming up to crazy things like downstream
> ports falling off the face of the planet.
> 
> Upon investigating how the PCIe RC is configured, I found the
> following nugget: the Sysnopsys DWC PCIe Reference Manual, in the
> section dedicated to the PLCR register, describes bit 7 (FAST_LINK_MODE)
> as:
> 
> "Sets all internal timers to fast mode for simulation purposes."
> 
> I completely understand the need for setting this bit from a simulation
> perspective, but what I have on my desk is actual silicon, which
> expects timers to have a nominal value (and I expect this is the
> case for most people).
> 
> Making sure the FAST_LINK_MODE bit is cleared when configuring the RC
> solves this problem.
> 
> Fixes: 9c0ef6d34fdb ("PCI: amlogic: Add the Amlogic Meson PCIe controller driver")
> Signed-off-by: Marc Zyngier <maz@kernel.org>
> ---
>  drivers/pci/controller/dwc/pci-meson.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)

Reworded the commit log (even if yours was more fun :)) and applied
to pci/dwc, thanks !

Lorenzo

> diff --git a/drivers/pci/controller/dwc/pci-meson.c b/drivers/pci/controller/dwc/pci-meson.c
> index 3715dceca1bf..ca59ba9e0ecd 100644
> --- a/drivers/pci/controller/dwc/pci-meson.c
> +++ b/drivers/pci/controller/dwc/pci-meson.c
> @@ -289,11 +289,11 @@ static void meson_pcie_init_dw(struct meson_pcie *mp)
>  	meson_cfg_writel(mp, val, PCIE_CFG0);
>  
>  	val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
> -	val &= ~LINK_CAPABLE_MASK;
> +	val &= ~(LINK_CAPABLE_MASK | FAST_LINK_MODE);
>  	meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
>  
>  	val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
> -	val |= LINK_CAPABLE_X1 | FAST_LINK_MODE;
> +	val |= LINK_CAPABLE_X1;
>  	meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
>  
>  	val = meson_elb_readl(mp, PCIE_GEN2_CTRL_OFF);
> -- 
> 2.26.2
>
diff mbox series

Patch

diff --git a/drivers/pci/controller/dwc/pci-meson.c b/drivers/pci/controller/dwc/pci-meson.c
index 3715dceca1bf..ca59ba9e0ecd 100644
--- a/drivers/pci/controller/dwc/pci-meson.c
+++ b/drivers/pci/controller/dwc/pci-meson.c
@@ -289,11 +289,11 @@  static void meson_pcie_init_dw(struct meson_pcie *mp)
 	meson_cfg_writel(mp, val, PCIE_CFG0);
 
 	val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
-	val &= ~LINK_CAPABLE_MASK;
+	val &= ~(LINK_CAPABLE_MASK | FAST_LINK_MODE);
 	meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
 
 	val = meson_elb_readl(mp, PCIE_PORT_LINK_CTRL_OFF);
-	val |= LINK_CAPABLE_X1 | FAST_LINK_MODE;
+	val |= LINK_CAPABLE_X1;
 	meson_elb_writel(mp, val, PCIE_PORT_LINK_CTRL_OFF);
 
 	val = meson_elb_readl(mp, PCIE_GEN2_CTRL_OFF);