Skip to contentSkip to conversation

PCI: dwc: Relatively simple fixes and cleanups

PATCH RESEND Superseded Serge Semin 3 participants 14 messages Mar 13, 2023

Subsystems (5)

v2 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
#0 lore raw

v2 cover PCI: dwc: Relatively simple fixes and cleanups

It turns out the recent DW PCIe-related patchset was merged in with several relatively trivial issues left unsettled (noted by Bjorn and Manivannan). All of these lefovers have been fixed in this patchset. Namely the series starts with two bug-fixes. The first one concerns the improper link-mode initialization in case if the CDM-check is enabled. The second unfortunate mistake I made in the IP-core version type helper. In particular instead of testing the IP-core version type the macro function referred to the just IP-core version which obviously wasn't what I intended.

Afterwards two @Mani-noted fixes follow. Firstly the dma-ranges related warning message is fixed to start with "DMA-ranges" word instead of "Dma-ranges". Secondly the Baikal-T1 PCIe Host driver is converted to perform the asynchronous probe type which saved us of about 15% of bootup time if no any PCIe peripheral device attached to the port.

Then the patchset contains the Baikal-T1 PCIe driver fix. The corresponding patch removes the false error message printed during the controller probe procedure. I accidentally added the unconditional dev_err_probe() method invocation. It was obviously wrong.

Then two trivial cleanups are introduced. The first one concerns the duplicated fast-link-mode flag unsetting. The second one implies dropping a redundant empty line from the dw_pcie_link_set_max_speed() function.

The series continues with a patch inspired by the last @Bjorn note regarding the generic resources request interface. As @Bjorn correctly said it would be nice to have the new interface used wider in the DW PCIe subsystem. Aside with the Baikal-T1 PCIe Host driver the Toshiba Visconti PCIe driver can be easily converted to using the generic clock names. That's what is done in the noted patch.

The patchset is closed with a series of MAINTAINERS-list related patches. Firstly after getting the DW PCIe RP/EP DT-schemas refactored I forgot to update the MAINTAINER-list with the new files added in the framework of that procedure. All the snps,dw-pcie* schemas shall be maintained by the DW PCIe core driver maintainers. Secondly seeing how long it took for my patchsets to review and not having any comments from the original driver maintainers I'd suggest to add myself as the reviewer to the DW PCIe and eDMA drivers. Thus hopefully the new updates review process will be performed with much less latencies. For the same reason I would also like to suggest to add @Manivannan as the DW PCIe/eDMA drivers maintainer if he isn't against that idea. What do you think about the last suggestion?

Changelog v2:
- Rebase onto the kernel 6.3-rc2.

Serge Semin (11):
PCI: dwc: Fix port link CSR improper init if CDM check enabled
PCI: dwc: Fix erroneous version type test helper
PCI: dwc: Fix inbound iATU entries out-of-bounds warning message
PCI: bt1: Enable async probe type
PCI: bt1: Fix printing false error message
PCI: dwc: Drop duplicated fast-link-mode flag unsetting
PCI: dwc: Drop empty line from dw_pcie_link_set_max_speed()
PCI: visconti: Convert to using generic resources getter
MAINTAINERS: Add all generic DW PCIe RP/EP DT-schemas
MAINTAINERS: Add myself as the DW PCIe core reviewer
MAINTAINERS: Add myself as the DW eDMA driver reviewer

 MAINTAINERS                                   |  5 ++-
 drivers/pci/controller/dwc/pcie-bt1.c         |  5 ++-
 .../pci/controller/dwc/pcie-designware-host.c |  2 +-
 drivers/pci/controller/dwc/pcie-designware.c  |  3 +-
 drivers/pci/controller/dwc/pcie-designware.h  |  7 +++-
 drivers/pci/controller/dwc/pcie-visconti.c    | 37 +++++++++----------
 6 files changed, 31 insertions(+), 28 deletions(-)
Signature · 1 line
2.39.2
#1 lore raw

PATCH RESEND v2 1/11 PCI: dwc: Fix port link CSR improper init if CDM check enabled

If CDM_CHECK capability is set then the local variable 'val' will be overwritten in the dw_pcie_setup() method in the PL_CHK register initialization procedure. Thus further variable usage in the framework of the PCIE_PORT_LINK_CONTROL register initialization must imply the variable re-initialization. Alas it hasn't been taken into account in the commit ec7b952f453c ("PCI: dwc: Always enable CDM check if "snps,enable-cdm-check" exists"). Due to that the PCIE_PORT_LINK_CONTROL register will be written with improper value in case if the CDM-check is enabled. Let's fix this by re-initializing the 'val' variable with the PCIE_PORT_LINK_CONTROL CSR content before link-mode initialization.

Fixes ec7b952f453c ("PCI: dwc: Always enable CDM check if "snps,enable-cdm-check" exists")
---
 drivers/pci/controller/dwc/pcie-designware.c | 1 +
 1 file changed, 1 insertion(+)
drivers/pci/controller/dwc/pcie-designware.c +1 -0
index 53a16b8b6ac2..04fedd257ef1 100644
@@ -1019,6 +1019,7 @@ void dw_pcie_setup(struct dw_pcie *pci)
 	}
 
 	/* Set the number of lanes */
+	val = dw_pcie_readl_dbi(pci, PCIE_PORT_LINK_CONTROL);
 	val &= ~PORT_LINK_FAST_LINK_MODE;
 	val &= ~PORT_LINK_MODE_MASK;
 	switch (pci->num_lanes) {
Signature · 1 line
2.39.2
#2 lore raw

PATCH RESEND v2 2/11 PCI: dwc: Fix erroneous version type test helper

In reply to Serge Semin

Due to an unfortunate mistake the macro function actually checks the IP-core version instead of the IP-core version type which isn't what originally implied. Fix it by introducing a new helper __dw_pcie_ver_type_cmp() with the same semantic as the __dw_pcie_ver_cmp() counterpart except it refers to the dw_pcie.type field in order to perform the passed comparison operation.

Fixes 0b0a780d52ad ("PCI: dwc: Add macros to compare Synopsys IP core versions")
---
 drivers/pci/controller/dwc/pcie-designware.h | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/pci/controller/dwc/pcie-designware.h +5 -2
index 79713ce075cc..adad0ea61799 100644
@@ -37,17 +37,20 @@
 #define __dw_pcie_ver_cmp(_pci, _ver, _op) \
 	((_pci)->version _op DW_PCIE_VER_ ## _ver)
 
+#define __dw_pcie_ver_type_cmp(_pci, _type, _op) \
+	((_pci)->type _op DW_PCIE_VER_TYPE_ ## _type)
+
 #define dw_pcie_ver_is(_pci, _ver) __dw_pcie_ver_cmp(_pci, _ver, ==)
 
 #define dw_pcie_ver_is_ge(_pci, _ver) __dw_pcie_ver_cmp(_pci, _ver, >=)
 
 #define dw_pcie_ver_type_is(_pci, _ver, _type) \
 	(__dw_pcie_ver_cmp(_pci, _ver, ==) && \
-	 __dw_pcie_ver_cmp(_pci, TYPE_ ## _type, ==))
+	 __dw_pcie_ver_type_cmp(_pci, _type, ==))
 
 #define dw_pcie_ver_type_is_ge(_pci, _ver, _type) \
 	(__dw_pcie_ver_cmp(_pci, _ver, ==) && \
-	 __dw_pcie_ver_cmp(_pci, TYPE_ ## _type, >=))
+	 __dw_pcie_ver_type_cmp(_pci, _type, >=))
 
 /* DWC PCIe controller capabilities */
 #define DW_PCIE_CAP_REQ_RES		0
Signature · 1 line
2.39.2
#3 lore raw

PATCH RESEND v2 3/11 PCI: dwc: Fix inbound iATU entries out-of-bounds warning message

In reply to Serge Semin

The message is printed if the number of requested inbound iATU windows exceed the device capability. In that case the message should either refer to the "dma-ranges" DT property or to the DMA-ranges mapping. We suggest to use the later version as a counterpart to the just CPU-ranges mapping. In any case the current "Dma-ranges" phrase seems incorrect.

---
 drivers/pci/controller/dwc/pcie-designware-host.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/pci/controller/dwc/pcie-designware-host.c +1 -1
index 9952057c8819..5718b4bb67f0 100644
@@ -723,7 +723,7 @@ static int dw_pcie_iatu_setup(struct dw_pcie_rp *pp)
 	}
 
 	if (pci->num_ib_windows <= i)
-		dev_warn(pci->dev, "Dma-ranges exceed inbound iATU size (%u)\n",
+		dev_warn(pci->dev, "DMA-ranges exceed inbound iATU size (%u)\n",
 			 pci->num_ib_windows);
 
 	return 0;
Signature · 1 line
2.39.2
#4 lore raw

PATCH RESEND v2 4/11 PCI: bt1: Enable async probe type

In reply to Serge Semin

It's safe to enable the asyncronous probe type since the PCIe peripheral devices probing order isn't essential for booting the system. Moreover enabling that feature saves 0.5 seconds of bootup time if no any device attached to the PCIe root port. It's a significant performance gain seeing the total bootup time takes about 3 seconds.

---
 drivers/pci/controller/dwc/pcie-bt1.c | 1 +
 1 file changed, 1 insertion(+)
drivers/pci/controller/dwc/pcie-bt1.c +1 -0
index 95a723a6fd46..e36a20bf82cf 100644
@@ -638,6 +638,7 @@ static struct platform_driver bt1_pcie_driver = {
 	.driver = {
 		.name	= "bt1-pcie",
 		.of_match_table = bt1_pcie_of_match,
+		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
 	},
 };
 module_platform_driver(bt1_pcie_driver);
Signature · 1 line
2.39.2
#5 lore raw

PATCH RESEND v2 5/11 PCI: bt1: Fix printing false error message

In reply to Serge Semin

The dev_err_probe() method is supposed to be invoked only if any error is happened. It was definitely wrong to call it unconditionally. Due to that the DWC PCIe host initialization error-message is printed all the time the Baikal-T1 PCIe controller is probed even if no error actually happened.

Fixes ba6ed462dcf4 ("PCI: dwc: Add Baikal-T1 PCIe controller support")
---
 drivers/pci/controller/dwc/pcie-bt1.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/pci/controller/dwc/pcie-bt1.c +3 -1
index e36a20bf82cf..6557141873ad 100644
@@ -597,8 +597,10 @@ static int bt1_pcie_add_port(struct bt1_pcie *btpci)
 	dw_pcie_cap_set(&btpci->dw, REQ_RES);
 
 	ret = dw_pcie_host_init(&btpci->dw.pp);
+	if (ret)
+		dev_err_probe(dev, ret, "Failed to initialize DWC PCIe host\n");
 
-	return dev_err_probe(dev, ret, "Failed to initialize DWC PCIe host\n");
+	return ret;
 }
 
 static void bt1_pcie_del_port(struct bt1_pcie *btpci)
Signature · 1 line
2.39.2
#6 lore raw

PATCH RESEND v2 6/11 PCI: dwc: Drop duplicated fast-link-mode flag unsetting

In reply to Serge Semin

Most likely by mistake the PORT_LINK_FAST_LINK_MODE flag unsetting was added twice in the commit cff9244432e8 ("PCI: dwc: Ensure FAST_LINK_MODE is cleared"): first it is cleared right after the content of the PCIE_PORT_LINK_CONTROL register is read, second it's cleared in the framework of the link-mode initialization procedure. The later action is redundant. Let's drop it.

---
 drivers/pci/controller/dwc/pcie-designware.c | 1 -
 1 file changed, 1 deletion(-)
drivers/pci/controller/dwc/pcie-designware.c +0 -1
index 04fedd257ef1..6061824f7b7e 100644
@@ -1020,7 +1020,6 @@ void dw_pcie_setup(struct dw_pcie *pci)
 
 	/* Set the number of lanes */
 	val = dw_pcie_readl_dbi(pci, PCIE_PORT_LINK_CONTROL);
-	val &= ~PORT_LINK_FAST_LINK_MODE;
 	val &= ~PORT_LINK_MODE_MASK;
 	switch (pci->num_lanes) {
 	case 1:
Signature · 1 line
2.39.2
#7 lore raw

PATCH RESEND v2 7/11 PCI: dwc: Drop empty line from dw_pcie_link_set_max_speed()

In reply to Serge Semin

Well, just drop a redundant empty line from the tail of the denoted function which by mistake was added in commit 39bc5006501c ("PCI: dwc:
Centralize link gen setting").

---
 drivers/pci/controller/dwc/pcie-designware.c | 1 -
 1 file changed, 1 deletion(-)
drivers/pci/controller/dwc/pcie-designware.c +0 -1
index 6061824f7b7e..83ecc764330c 100644
@@ -729,7 +729,6 @@ static void dw_pcie_link_set_max_speed(struct dw_pcie *pci, u32 link_gen)
 
 	cap &= ~((u32)PCI_EXP_LNKCAP_SLS);
 	dw_pcie_writel_dbi(pci, offset + PCI_EXP_LNKCAP, cap | link_speed);
-
 }
 
 void dw_pcie_iatu_detect(struct dw_pcie *pci)
Signature · 1 line
2.39.2
#8 lore raw

PATCH RESEND v2 8/11 PCI: visconti: Convert to using generic resources getter

In reply to Serge Semin

The generic resources request infrastructure has been recently added to the DW PCIe core driver. Since the DT-bindings of the Toshibo Visconti PCIe Host controller is fully compatible with the generic names set let's convert the driver to using that infrastructure. It won't take much effort since the low-level device driver implies the resources request only with no additional manipulations involving them. So just drop the locally defined clocks request procedures, activate the generic resources request capability and make sure the mandatory resources have been requested by the DW PCIe core driver.

---
 drivers/pci/controller/dwc/pcie-visconti.c | 37 ++++++++++------------
 1 file changed, 17 insertions(+), 20 deletions(-)
drivers/pci/controller/dwc/pcie-visconti.c +17 -20
index 71026fefa366..ae1517b52c58 100644
@@ -29,9 +29,6 @@ struct visconti_pcie {
 	void __iomem *ulreg_base;
 	void __iomem *smu_base;
 	void __iomem *mpu_base;
-	struct clk *refclk;
-	struct clk *coreclk;
-	struct clk *auxclk;
 };
 
 #define PCIE_UL_REG_S_PCIE_MODE		0x00F4
@@ -198,6 +195,21 @@ static int visconti_pcie_host_init(struct dw_pcie_rp *pp)
 	int err;
 	u32 val;
 
+	if (!pcie->pci.core_clks[DW_PCIE_REF_CLK].clk) {
+		dev_err(pci->dev, "Missing ref clock source\n");
+		return -ENOENT;
+	}
+
+	if (!pcie->pci.core_clks[DW_PCIE_CORE_CLK].clk) {
+		dev_err(pci->dev, "Missing core clock source\n");
+		return -ENOENT;
+	}
+
+	if (!pcie->pci.core_clks[DW_PCIE_AUX_CLK].clk) {
+		dev_err(pci->dev, "Missing aux clock source\n");
+		return -ENOENT;
+	}
+
 	visconti_smu_writel(pcie,
 			    PISMU_CKON_PCIE_AUX_CLK | PISMU_CKON_PCIE_MSTR_ACLK,
 			    PISMU_CKON_PCIE);
@@ -242,8 +254,6 @@ static const struct dw_pcie_host_ops visconti_pcie_host_ops = {
 static int visconti_get_resources(struct platform_device *pdev,
 				  struct visconti_pcie *pcie)
 {
-	struct device *dev = &pdev->dev;
-
 	pcie->ulreg_base = devm_platform_ioremap_resource_byname(pdev, "ulreg");
 	if (IS_ERR(pcie->ulreg_base))
 		return PTR_ERR(pcie->ulreg_base);
@@ -256,21 +266,6 @@ static int visconti_get_resources(struct platform_device *pdev,
 	if (IS_ERR(pcie->mpu_base))
 		return PTR_ERR(pcie->mpu_base);
 
-	pcie->refclk = devm_clk_get(dev, "ref");
-	if (IS_ERR(pcie->refclk))
-		return dev_err_probe(dev, PTR_ERR(pcie->refclk),
-				     "Failed to get ref clock\n");
-
-	pcie->coreclk = devm_clk_get(dev, "core");
-	if (IS_ERR(pcie->coreclk))
-		return dev_err_probe(dev, PTR_ERR(pcie->coreclk),
-				     "Failed to get core clock\n");
-
-	pcie->auxclk = devm_clk_get(dev, "aux");
-	if (IS_ERR(pcie->auxclk))
-		return dev_err_probe(dev, PTR_ERR(pcie->auxclk),
-				     "Failed to get aux clock\n");
-
 	return 0;
 }
 
@@ -304,6 +299,8 @@ static int visconti_pcie_probe(struct platform_device *pdev)
 	pci->dev = dev;
 	pci->ops = &dw_pcie_ops;
 
+	dw_pcie_cap_set(pci, REQ_RES);
+
 	ret = visconti_get_resources(pdev, pcie);
 	if (ret)
 		return ret;
Signature · 1 line
2.39.2
#9 lore raw

PATCH RESEND v2 9/11 MAINTAINERS: Add all generic DW PCIe RP/EP DT-schemas

In reply to Serge Semin

Recently the DT-schema of the DW PCIe Root Port and End-point controllers has been refactored by detaching the common bindings into a separate schema. The provided modification must be reflected in the MAINTAINERS list so the patch submitters would be aware of the new files maintainers. Let's do that by adding the maintained files wildcard pattern like snps,dw-pcie*.yaml, which is applicable for all the old DW PCIe DT-schema files and the new one.

---
 MAINTAINERS | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
MAINTAINERS +1 -2
index ec57c42ed544..489fd4b4c7ae 100644
@@ -16063,8 +16063,7 @@ M:	Jingoo Han <[email protected]>
 M:	Gustavo Pimentel <[email protected]>
 L:	[email protected]
 S:	Maintained
-F:	Documentation/devicetree/bindings/pci/snps,dw-pcie.yaml
-F:	Documentation/devicetree/bindings/pci/snps,dw-pcie-ep.yaml
+F:	Documentation/devicetree/bindings/pci/snps,dw-pcie*.yaml
 F:	drivers/pci/controller/dwc/*designware*
 
 PCI DRIVER FOR TI DRA7XX/J721E
Signature · 1 line
2.39.2
#10 lore raw

PATCH RESEND v2 10/11 MAINTAINERS: Add myself as the DW PCIe core reviewer

In reply to Serge Semin

No actions have been spotted from the driver maintainers for almost two years now. It significantly delays the review process of the relatively often incoming updates. Since that IP-core has been used in several our SoCs adding myself to the list of reviewers will help in the evolving the driver faster and in catching any potential problem as early as possible.

---
 MAINTAINERS | 1 +
 1 file changed, 1 insertion(+)
MAINTAINERS +1 -0
index 489fd4b4c7ae..51adcafa0f0c 100644
@@ -16061,6 +16061,7 @@ F:	drivers/pci/controller/dwc/pci-exynos.c
 PCI DRIVER FOR SYNOPSYS DESIGNWARE
 M:	Jingoo Han <[email protected]>
 M:	Gustavo Pimentel <[email protected]>
+R:	Serge Semin <[email protected]>
 L:	[email protected]
 S:	Maintained
 F:	Documentation/devicetree/bindings/pci/snps,dw-pcie*.yaml
Signature · 1 line
2.39.2
#11 lore raw

PATCH RESEND v2 11/11 MAINTAINERS: Add myself as the DW eDMA driver reviewer

In reply to Serge Semin

The driver maintainer has been inactive for almost two years now. It doesn't positively affect the new patches tests and reviews process. Since the DW eDMA engine has been embedded into the PCIe controllers in several our SoCs we will be interested in helping with the updates review.

---
 MAINTAINERS | 1 +
 1 file changed, 1 insertion(+)
MAINTAINERS +1 -0
index 51adcafa0f0c..fb7bada6cdd0 100644
@@ -5862,6 +5862,7 @@ F:	drivers/mtd/nand/raw/denali*
 
 DESIGNWARE EDMA CORE IP DRIVER
 M:	Gustavo Pimentel <[email protected]>
+R:	Serge Semin <[email protected]>
 L:	[email protected]
 S:	Maintained
 F:	drivers/dma/dw-edma/
Signature · 1 line
2.39.2
#13 lore raw

Hi Bjorn

Quoted 10 lines · Bjorn Helgaas
Quoted 4 lines · Serge Semin

...

Changelog v2:
- Rebase onto the kernel 6.3-rc2.

This is fine, but just FYI that there's no need to rebase past -rc1 because PCI patches are applied on topic branches based on the PCI "main" branch, typically -rc1.

Thanks for reminding about that. I am not that keen of early rc's because there is a higher risk to catch instability in unexpected places. Normally I wait for rc2 or newer kernel is released in order to rebase onto that mainline kernel. I am using the Linus' master because most of time I get to develop and submit patchsets for several subsystems concurrently (currently it's PCIe, EDAC, net and MIPS-arch) and having them rebased on top of each subsystem' next-branch would be even more risky. Of course I understand that the subsystem main/topic branches may have some conflicting changes. So normally I make sure that the submitted changes are applicable against the subsystem trees if/when it's necessary. (Can't deny though I forget to do that from time to time.)

-Serge(y)

Quoted 2 lines

Bjorn