Skip to contentSkip to conversation

PCI: rcar: Fix incorrect programming of OB windows

RFC PATCH Applied Andrew Murray 4 participants 6 messages Oct 4, 2019 – Apr 24, 2020

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 PCI: rcar: Fix incorrect programming of OB windows

The outbound windows (PCIEPAUR(x), PCIEPALR(x)) describe a mapping between a CPU address (which is determined by the window number 'x') and a programmed PCI address - Thus allowing the controller to translate CPU accesses into PCI accesses.

However the existing code incorrectly writes the CPU address - lets fix this by writing the PCI address instead.

For memory transactions, existing DT users describe a 1:1 identity mapping and thus this change should have no effect. However the same isn't true for I/O.

Fixes c25da4778803 ("PCI: rcar: Add Renesas R-Car PCIe driver")

---
This hasn't been tested, so keen for someone to give it a try.

Also keen for someone to confirm my understanding that the RCar windows expect PCI addresses and that res->start refers to CPU addresses. If this is correct then it's possible the I/O doesn't work correctly.

---
 drivers/pci/controller/pcie-rcar.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
drivers/pci/controller/pcie-rcar.c +5 -4
index f6a669a9af41..b28d726b4aba 100644
@@ -332,11 +332,12 @@ static struct pci_ops rcar_pcie_ops = {
 };
 
 static void rcar_pcie_setup_window(int win, struct rcar_pcie *pcie,
-				   struct resource *res)
+				   struct resource_entry *window)
 {
 	/* Setup PCIe address space mappings for each resource */
 	resource_size_t size;
 	resource_size_t res_start;
+	struct resource *res = window->res;
 	u32 mask;
 
 	rcar_pci_write_reg(pcie, 0x00000000, PCIEPTCTLR(win));
@@ -350,9 +351,9 @@ static void rcar_pcie_setup_window(int win, struct rcar_pcie *pcie,
 	rcar_pci_write_reg(pcie, mask << 7, PCIEPAMR(win));
 
 	if (res->flags & IORESOURCE_IO)
-		res_start = pci_pio_to_address(res->start);
+		res_start = pci_pio_to_address(res->start) - window->offset;
 	else
-		res_start = res->start;
+		res_start = res->start - window->offset;
 
 	rcar_pci_write_reg(pcie, upper_32_bits(res_start), PCIEPAUR(win));
 	rcar_pci_write_reg(pcie, lower_32_bits(res_start) & ~0x7F,
@@ -381,7 +382,7 @@ static int rcar_pcie_setup(struct list_head *resource, struct rcar_pcie *pci)
 		switch (resource_type(res)) {
 		case IORESOURCE_IO:
 		case IORESOURCE_MEM:
-			rcar_pcie_setup_window(i, pci, res);
+			rcar_pcie_setup_window(i, pci, win);
 			i++;
 			break;
 		case IORESOURCE_BUS:
Signature · 1 line
2.21.0
#1 lore raw
Quoted 15 lines · Andrew Murray

The outbound windows (PCIEPAUR(x), PCIEPALR(x)) describe a mapping between a CPU address (which is determined by the window number 'x') and a programmed PCI address - Thus allowing the controller to translate CPU accesses into PCI accesses.

However the existing code incorrectly writes the CPU address - lets fix this by writing the PCI address instead.

For memory transactions, existing DT users describe a 1:1 identity mapping and thus this change should have no effect. However the same isn't true for I/O.

Fixes c25da4778803 ("PCI: rcar: Add Renesas R-Car PCIe driver")
Quoted 6 lines

---
This hasn't been tested, so keen for someone to give it a try.

Also keen for someone to confirm my understanding that the RCar windows expect PCI addresses and that res->start refers to CPU addresses. If this is correct then it's possible the I/O doesn't work correctly.

Marek/Yoshihiro - any feedback on this?

Thanks,

Andrew Murray

Quoted 46 lines
---
 drivers/pci/controller/pcie-rcar.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
drivers/pci/controller/pcie-rcar.c +5 -4
index f6a669a9af41..b28d726b4aba 100644
@@ -332,11 +332,12 @@ static struct pci_ops rcar_pcie_ops = {
 };
 
 static void rcar_pcie_setup_window(int win, struct rcar_pcie *pcie,
-				   struct resource *res)
+				   struct resource_entry *window)
 {
 	/* Setup PCIe address space mappings for each resource */
 	resource_size_t size;
 	resource_size_t res_start;
+	struct resource *res = window->res;
 	u32 mask;
 
 	rcar_pci_write_reg(pcie, 0x00000000, PCIEPTCTLR(win));
@@ -350,9 +351,9 @@ static void rcar_pcie_setup_window(int win, struct rcar_pcie *pcie,
 	rcar_pci_write_reg(pcie, mask << 7, PCIEPAMR(win));
 
 	if (res->flags & IORESOURCE_IO)
-		res_start = pci_pio_to_address(res->start);
+		res_start = pci_pio_to_address(res->start) - window->offset;
 	else
-		res_start = res->start;
+		res_start = res->start - window->offset;
 
 	rcar_pci_write_reg(pcie, upper_32_bits(res_start), PCIEPAUR(win));
 	rcar_pci_write_reg(pcie, lower_32_bits(res_start) & ~0x7F,
@@ -381,7 +382,7 @@ static int rcar_pcie_setup(struct list_head *resource, struct rcar_pcie *pci)
 		switch (resource_type(res)) {
 		case IORESOURCE_IO:
 		case IORESOURCE_MEM:
-			rcar_pcie_setup_window(i, pci, res);
+			rcar_pcie_setup_window(i, pci, win);
 			i++;
 			break;
 		case IORESOURCE_BUS:
Signature · 1 line
2.21.0
Reviewed-by Tested-by
#2 lore raw
Quoted 16 lines · Andrew Murray
Quoted 15 lines · Andrew Murray

The outbound windows (PCIEPAUR(x), PCIEPALR(x)) describe a mapping between a CPU address (which is determined by the window number 'x') and a programmed PCI address - Thus allowing the controller to translate CPU accesses into PCI accesses.

However the existing code incorrectly writes the CPU address - lets fix this by writing the PCI address instead.

For memory transactions, existing DT users describe a 1:1 identity mapping and thus this change should have no effect. However the same isn't true for I/O.

Fixes c25da4778803 ("PCI: rcar: Add Renesas R-Car PCIe driver")
Quoted 8 lines
Quoted 6 lines

---
This hasn't been tested, so keen for someone to give it a try.

Also keen for someone to confirm my understanding that the RCar windows expect PCI addresses and that res->start refers to CPU addresses. If this is correct then it's possible the I/O doesn't work correctly.

Marek/Yoshihiro - any feedback on this?

It does indeed look correct,

# On R8A77951 Salvator-XS with Intel 8086:f1a5 600P SSD # On R8A77965 Salvator-XS with Intel 8086:10d3 82574L NIC

#3 lore raw
Quoted 24 lines · Marek Vasut
Quoted 23 lines · Andrew Murray
Quoted 21 lines · Andrew Murray

The outbound windows (PCIEPAUR(x), PCIEPALR(x)) describe a mapping between a CPU address (which is determined by the window number 'x') and a programmed PCI address - Thus allowing the controller to translate CPU accesses into PCI accesses.

However the existing code incorrectly writes the CPU address - lets fix this by writing the PCI address instead.

For memory transactions, existing DT users describe a 1:1 identity mapping and thus this change should have no effect. However the same isn't true for I/O.

Fixes c25da4778803 ("PCI: rcar: Add Renesas R-Car PCIe driver")

---
This hasn't been tested, so keen for someone to give it a try.

Also keen for someone to confirm my understanding that the RCar windows expect PCI addresses and that res->start refers to CPU addresses. If this is correct then it's possible the I/O doesn't work correctly.

Quoted 8 lines
Quoted 1 line

Marek/Yoshihiro - any feedback on this?

It does indeed look correct,

# On R8A77951 Salvator-XS with Intel 8086:f1a5 600P SSD # On R8A77965 Salvator-XS with Intel 8086:10d3 82574L NIC

Thanks for testing - much appreciated!

Andrew Murray

#4 lore raw
Quoted 30 lines · Andrew Murray
Quoted 29 lines · Marek Vasut
Quoted 16 lines · Andrew Murray
Quoted 15 lines · Andrew Murray

The outbound windows (PCIEPAUR(x), PCIEPALR(x)) describe a mapping between a CPU address (which is determined by the window number 'x') and a programmed PCI address - Thus allowing the controller to translate CPU accesses into PCI accesses.

However the existing code incorrectly writes the CPU address - lets fix this by writing the PCI address instead.

For memory transactions, existing DT users describe a 1:1 identity mapping and thus this change should have no effect. However the same isn't true for I/O.

Fixes c25da4778803 ("PCI: rcar: Add Renesas R-Car PCIe driver")
Quoted 8 lines
Quoted 6 lines

---
This hasn't been tested, so keen for someone to give it a try.

Also keen for someone to confirm my understanding that the RCar windows expect PCI addresses and that res->start refers to CPU addresses. If this is correct then it's possible the I/O doesn't work correctly.

Marek/Yoshihiro - any feedback on this?

It does indeed look correct,

Quoted 7 lines
Quoted 3 lines

# On R8A77951 Salvator-XS with Intel 8086:f1a5 600P SSD # On R8A77965 Salvator-XS with Intel 8086:10d3 82574L NIC

Thanks for testing - much appreciated!

Andrew Murray

Can this be applied then ?

#5 lore raw
In reply to Andrew Murray
Quoted 18 lines · Andrew Murray

The outbound windows (PCIEPAUR(x), PCIEPALR(x)) describe a mapping between a CPU address (which is determined by the window number 'x') and a programmed PCI address - Thus allowing the controller to translate CPU accesses into PCI accesses.

However the existing code incorrectly writes the CPU address - lets fix this by writing the PCI address instead.

For memory transactions, existing DT users describe a 1:1 identity mapping and thus this change should have no effect. However the same isn't true for I/O.

Fixes c25da4778803 ("PCI: rcar: Add Renesas R-Car PCIe driver")

---
This hasn't been tested, so keen for someone to give it a try.

Quoted 6 lines

Also keen for someone to confirm my understanding that the RCar windows expect PCI addresses and that res->start refers to CPU addresses. If this is correct then it's possible the I/O doesn't work correctly.

---
 drivers/pci/controller/pcie-rcar.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

Applied to pci/rcar, thanks !

Lorenzo

Quoted 42 lines
drivers/pci/controller/pcie-rcar.c +5 -4
index f6a669a9af41..b28d726b4aba 100644
@@ -332,11 +332,12 @@ static struct pci_ops rcar_pcie_ops = {
 };
 
 static void rcar_pcie_setup_window(int win, struct rcar_pcie *pcie,
-				   struct resource *res)
+				   struct resource_entry *window)
 {
 	/* Setup PCIe address space mappings for each resource */
 	resource_size_t size;
 	resource_size_t res_start;
+	struct resource *res = window->res;
 	u32 mask;
 
 	rcar_pci_write_reg(pcie, 0x00000000, PCIEPTCTLR(win));
@@ -350,9 +351,9 @@ static void rcar_pcie_setup_window(int win, struct rcar_pcie *pcie,
 	rcar_pci_write_reg(pcie, mask << 7, PCIEPAMR(win));
 
 	if (res->flags & IORESOURCE_IO)
-		res_start = pci_pio_to_address(res->start);
+		res_start = pci_pio_to_address(res->start) - window->offset;
 	else
-		res_start = res->start;
+		res_start = res->start - window->offset;
 
 	rcar_pci_write_reg(pcie, upper_32_bits(res_start), PCIEPAUR(win));
 	rcar_pci_write_reg(pcie, lower_32_bits(res_start) & ~0x7F,
@@ -381,7 +382,7 @@ static int rcar_pcie_setup(struct list_head *resource, struct rcar_pcie *pci)
 		switch (resource_type(res)) {
 		case IORESOURCE_IO:
 		case IORESOURCE_MEM:
-			rcar_pcie_setup_window(i, pci, res);
+			rcar_pcie_setup_window(i, pci, win);
 			i++;
 			break;
 		case IORESOURCE_BUS:
Signature · 1 line
2.21.0