On 9/24/2019 5:41 PM, Pankaj Dubey wrote:
-----Original Message-----
From: Vidya Sagar <vidyas@xxxxxxxxxx>
Sent: Tuesday, September 24, 2019 4:57 PM
To: Pankaj Dubey <pankaj.dubey@xxxxxxxxxxx>; 'Gustavo Pimentel'
<Gustavo.Pimentel@xxxxxxxxxxxx>; 'Andrew Murray'
<andrew.murray@xxxxxxx>
Cc: linux-pci@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
jingoohan1@xxxxxxxxx; lorenzo.pieralisi@xxxxxxx; bhelgaas@xxxxxxxxxx;
'Anvesh Salveru' <anvesh.s@xxxxxxxxxxx>
Subject: Re: [PATCH v2] PCI: dwc: Add support to add GEN3 related equalization
quirks
On 9/24/2019 2:58 PM, Pankaj Dubey wrote:
-----Original Message-----
From: Vidya Sagar <vidyas@xxxxxxxxxx>
Sent: Thursday, September 19, 2019 4:54 PM
Subject: Re: [PATCH v2] PCI: dwc: Add support to add GEN3 related
equalization quirks
On 9/16/2019 6:22 PM, Gustavo Pimentel wrote:
On Mon, Sep 16, 2019 at 13:24:1, Andrew Murray
<andrew.murray@xxxxxxx>
wrote:
On Mon, Sep 16, 2019 at 04:36:33PM +0530, Pankaj Dubey wrote:
-----Original Message-----
From: Andrew Murray <andrew.murray@xxxxxxx>
Sent: Monday, September 16, 2019 3:46 PM
To: Pankaj Dubey <pankaj.dubey@xxxxxxxxxxx>
Cc: linux-pci@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
jingoohan1@xxxxxxxxx; gustavo.pimentel@xxxxxxxxxxxx;
lorenzo.pieralisi@xxxxxxx; bhelgaas@xxxxxxxxxx; Anvesh Salveru
<anvesh.s@xxxxxxxxxxx>
Subject: Re: [PATCH v2] PCI: dwc: Add support to add GEN3 related
equalization
quirks
On Fri, Sep 13, 2019 at 04:09:50PM +0530, Pankaj Dubey wrote:
From: Anvesh Salveru <anvesh.s@xxxxxxxxxxx>
In some platforms, PCIe PHY may have issues which will prevent
linkup to happen in GEN3 or higher speed. In case equalization
fails, link will fallback to GEN1.
DesignWare controller gives flexibility to disable GEN3
equalization completely or only phase 2 and 3 of equalization.
This patch enables the DesignWare driver to disable the PCIe
GEN3 equalization by enabling one of the following quirks:
- DWC_EQUALIZATION_DISABLE: To disable GEN3 equalization all
phases
I don't think Gen-3 equalization can be skipped altogether.
PCIe Spec Rev 4.0 Ver 1.0 in Section-4.2.3 has the following statement.
"All the Lanes that are associated with the LTSSM (i.e., those Lanes
that are currently operational or may be operational in the future
due to Link
Upconfigure) must participate in the Equalization procedure"
and in Section-4.2.6.4.2.1.1 it says
"Note: A transition to Recovery.RcvrLock might be used in the case
where the Downstream Port determines that Phase 2 and Phase 3 are not
needed based on the platform and channel characteristics."
Based on the above statements, I think it is Ok to skip only Phases
2&3 of equalization but not 0&1.
I even checked with our hardware engineers and it seems
DWC_EQUALIZATION_DISABLE is present only for debugging purpose in
hardware simulations and shouldn't be used on real silicon otherwise it seems.
In DesignWare manual we don't see any comment that this feature is for
debugging purpose only.
Agree and as I mentioned even I got to know about it offline.
Even if it is meant for debugging purpose, if for some reason in an SoC, Gen3/4
linkup is failing due to equalization, and if disabling equalization is helping then
IMO it is OK to do it.
Well, I don't have specific reservations to not have it. We can use this as a fall
back option.
Just to re-confirm we tested one of the NVMe device on Jatson AGX Xavier RC
with equalization disabled. We do see linkup works well in GEN3. As we have
added this feature as a platform-quirk so only platforms that required this
feature can enable it.
Curious to know...You did it because link didn't come up with equalization
enabled? or just as an experiment?
We did this, just as an experiment.
Ok. Thanks for the clarification.
Reviewed-by: Vidya Sagar <vidyas@xxxxxxxxxx>
Snippet of lspci (from Jatson AGX Xavier RC) is given below, showing
EQ is completely disabled and GEN3 linkup
-----
0005:01:00.0 Non-Volatile memory controller: Lite-On Technology
Corporation Device 21f1 (rev 01) (prog-if 02 [NVM Express])
Subsystem: Marvell Technology Group Ltd. Device 1093
<snip>
LnkCap: Port #0, Speed 8GT/s, Width x4, ASPM L1, Exit Latency L0s
<512ns, L1 <64us
ClockPM+ Surprise- LLActRep- BwNot- ASPMOptComp+
LnkCtl: ASPM Disabled; RCB 64 bytes Disabled- CommClk+
ExtSynch- ClockPM- AutWidDis- BWInt- AutBWInt-
LnkSta: Speed 8GT/s, Width x4, TrErr- Train- SlotClk+ DLActive-
BWMgmt- ABWMgmt-
DevCap2: Completion Timeout: Not Supported, TimeoutDis+, LTR+,
OBFF Via message
DevCtl2: Completion Timeout: 50us to 50ms, TimeoutDis-, LTR+,
OBFF Disabled
LnkCtl2: Target Link Speed: 8GT/s, EnterCompliance- SpeedDis-
Transmit Margin: Normal Operating Range,
EnterModifiedCompliance- ComplianceSOS-
Compliance De-emphasis: -6dB
LnkSta2: Current De-emphasis Level: -6dB, EqualizationComplete-,
EqualizationPhase1-
EqualizationPhase2-, EqualizationPhase3-,
LinkEqualizationRequest-
-----
- Vidya Sagar
- DWC_EQ_PHASE_2_3_DISABLE: To disable GEN3 equalization
phase 2 & 3
Platform drivers can set these quirks via "quirk" variable of "dw_pcie"
struct.
Signed-off-by: Anvesh Salveru <anvesh.s@xxxxxxxxxxx>
Signed-off-by: Pankaj Dubey <pankaj.dubey@xxxxxxxxxxx>
---
Patchset v1 can be found at:
- 1/2: https://urldefense.proofpoint.com/v2/url?u=https-
3A__lkml.org_lkml_2019_9_10_443&d=DwIBAg&c=DPL6_X_6JkXFx7AXWqB0tg
&r=bkWxpLoW-f-
E3EdiDCCa0_h0PicsViasSlvIpzZvPxs&m=MtEKKeJsQvi2UM1eSZUv2vPLLxrYU0aI1
Ry4ICIDaiQ&s=s_nPmMNbQFswYRxQgBkeg4H9J_0FEtzRE-0AruC5WI4&e=
- 2/2:
https://urldefense.proofpoint.com/v2/url?u=https-3A__lkml.org_lk
ml
_2019_9_10_444&d=DwIBAg&c=DPL6_X_6JkXFx7AXWqB0tg&r=bkWxpLoW-
f-
E3Ed
iDCCa0_h0PicsViasSlvIpzZvPxs&m=MtEKKeJsQvi2UM1eSZUv2vPLLxrYU0aI1Ry
4ICIDaiQ&s=kkfdwcX6bYcLrnJSgw_GcMMGAjnDTMtN2v6svWuANpk&e=
Changes w.r.t v1:
- Squashed two patches from v1 into one as suggested by Gustavo
- Addressed review comments from Andrew
drivers/pci/controller/dwc/pcie-designware.c | 12
++++++++++++ drivers/pci/controller/dwc/pcie-designware.h | 9
+++++++++
2 files changed, 21 insertions(+)
diff --git a/drivers/pci/controller/dwc/pcie-designware.c
b/drivers/pci/controller/dwc/pcie-designware.c
index 7d25102..97fb18d 100644
--- a/drivers/pci/controller/dwc/pcie-designware.c
+++ b/drivers/pci/controller/dwc/pcie-designware.c
@@ -466,4 +466,16 @@ void dw_pcie_setup(struct dw_pcie *pci)
break;
}
dw_pcie_writel_dbi(pci, PCIE_LINK_WIDTH_SPEED_CONTROL,
val);
+
+ if (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);
+ }
+
+ if (pci->quirk & DWC_EQ_PHASE_2_3_DISABLE) {
+ val = dw_pcie_readl_dbi(pci,
PCIE_PORT_GEN3_RELATED);
+ val |= PORT_LOGIC_GEN3_EQ_PHASE_2_3_DISABLE;
+ dw_pcie_writel_dbi(pci, PCIE_PORT_GEN3_RELATED,
val);
+ }
}
diff --git a/drivers/pci/controller/dwc/pcie-designware.h
b/drivers/pci/controller/dwc/pcie-designware.h
index ffed084..e428b62 100644
--- a/drivers/pci/controller/dwc/pcie-designware.h
+++ b/drivers/pci/controller/dwc/pcie-designware.h
@@ -29,6 +29,10 @@
#define LINK_WAIT_MAX_IATU_RETRIES 5
#define LINK_WAIT_IATU 9
+/* Parameters for GEN3 related quirks */
+#define DWC_EQUALIZATION_DISABLE BIT(1)
+#define DWC_EQ_PHASE_2_3_DISABLE BIT(2)
+
/* Synopsys-specific PCIe configuration registers */
#define PCIE_PORT_LINK_CONTROL 0x710
#define PORT_LINK_MODE_MASK GENMASK(21, 16)
@@ -60,6 +64,10 @@
#define PCIE_MSI_INTR0_MASK 0x82C
#define PCIE_MSI_INTR0_STATUS 0x830
+#define PCIE_PORT_GEN3_RELATED 0x890
I hadn't noticed this in the previous version - what is the
proper name
for this
register? Does it end in _RELATED?
As per SNPS databook the name of the register is "GEN3_RELATED_OFF".
It is port logic register so, to keep similarity with other port
logic registers in this file we named it as "PCIE_PORT_GEN3_RELATED".
OK.
Reviewed-by: Andrew Murray <andrew.murray@xxxxxxx>
Also is the SNPS databook publicly available? I'd be interested in
reading it.
The databook isn't openly available, sorry.
Gustavo
Thanks,
Andrew Murray
Thanks,
Andrew Murray
+#define PORT_LOGIC_GEN3_EQ_PHASE_2_3_DISABLE BIT(9)
+#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 +252,7 @@ struct dw_pcie {
struct dw_pcie_ep ep;
const struct dw_pcie_ops *ops;
unsigned int version;
+ unsigned int quirk;
};
#define to_dw_pcie_from_pp(port) container_of((port), struct
dw_pcie,
pp)
--
2.7.4
