Skip to contentSkip to conversation

powerpc/5200: Bugfix for PCI mapping of memory and IMMR

PATCH In review Grant Likely 4 participants 5 messages Jan 27, 2009 – Jan 30, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 powerpc/5200: Bugfix for PCI mapping of memory and IMMR

From: Grant Likely <[email protected]>

This patch ensures that memory gets properly mapped into the PCI address space. Without this patch, the memory window BAR is left at whatever value happened to be loaded into the BAR when Linux was booted. Without this patch, memory could end up getting mapped at any of the 1G address boundaries instead of at '0' where Linux expects it.

Similarly, this patch also ensures that the internally memory mapped registers (IMMR) are mapped to the correct PCI address range.

Without this patch, PCI appears to work correctly until a PCI device is inserted which DMAs into memory.

---

This is a bugfix that I intend to merge into 2.6.29 and once it is mainlined get it added to the stable queue. If you have a 5200 system, please test and make sure it works for you.

Thanks
g.

 arch/powerpc/platforms/52xx/mpc52xx_pci.c |   24 ++++++++++--------------
 1 files changed, 10 insertions(+), 14 deletions(-)
arch/powerpc/platforms/52xx/mpc52xx_pci.c +10 -14
index c3f2c21..87ff522 100644
@@ -20,14 +20,6 @@
 
 
 /* ======================================================================== */
-/* PCI windows config                                                       */
-/* ======================================================================== */
-
-#define MPC52xx_PCI_TARGET_IO	0xf0000000
-#define MPC52xx_PCI_TARGET_MEM	0x00000000
-
-
-/* ======================================================================== */
 /* Structures mapping & Defines for PCI Unit                                */
 /* ======================================================================== */
 
@@ -244,7 +236,7 @@ static struct pci_ops mpc52xx_pci_ops = {
 
 static void __init
 mpc52xx_pci_setup(struct pci_controller *hose,
-                  struct mpc52xx_pci __iomem *pci_regs)
+                  struct mpc52xx_pci __iomem *pci_regs, phys_addr_t pci_phys)
 {
 	struct resource *res;
 	u32 tmp;
@@ -314,10 +306,14 @@ mpc52xx_pci_setup(struct pci_controller *hose,
 	/* Set all the IWCR fields at once; they're in the same reg */
 	out_be32(&pci_regs->iwcr, MPC52xx_PCI_IWCR_PACK(iwcr0, iwcr1, iwcr2));
 
-	out_be32(&pci_regs->tbatr0,
-		MPC52xx_PCI_TBATR_ENABLE | MPC52xx_PCI_TARGET_IO );
-	out_be32(&pci_regs->tbatr1,
-		MPC52xx_PCI_TBATR_ENABLE | MPC52xx_PCI_TARGET_MEM );
+	/* Map IMMR onto PCI bus */
+	pci_phys &= 0xfffc0000; /* bar0 has only 14 significant bits */
+	out_be32(&pci_regs->tbatr0, MPC52xx_PCI_TBATR_ENABLE | pci_phys);
+	out_be32(&pci_regs->bar0, PCI_BASE_ADDRESS_MEM_PREFETCH | pci_phys);
+
+	/* Map memory onto PCI bus */
+	out_be32(&pci_regs->tbatr1, MPC52xx_PCI_TBATR_ENABLE);
+	out_be32(&pci_regs->bar1, PCI_BASE_ADDRESS_MEM_PREFETCH);
 
 	out_be32(&pci_regs->tcr, MPC52xx_PCI_TCR_LD | MPC52xx_PCI_TCR_WCT8);
 
@@ -414,7 +410,7 @@ mpc52xx_add_bridge(struct device_node *node)
 
 	/* Finish setting up PCI using values obtained by
 	 * pci_proces_bridge_OF_ranges */
-	mpc52xx_pci_setup(hose, pci_regs);
+	mpc52xx_pci_setup(hose, pci_regs, rsrc.start);
 
 	return 0;
 }
#1 lore raw
Quoted 22 lines · Grant Likely

From: Grant Likely <[email protected]>

This patch ensures that memory gets properly mapped into the PCI address space. Without this patch, the memory window BAR is left at whatever value happened to be loaded into the BAR when Linux was booted. Without this patch, memory could end up getting mapped at any of the 1G address boundaries instead of at '0' where Linux expects it.

Similarly, this patch also ensures that the internally memory mapped registers (IMMR) are mapped to the correct PCI address range.

Without this patch, PCI appears to work correctly until a PCI device is inserted which DMAs into memory.

---

This is a bugfix that I intend to merge into 2.6.29 and once it is mainlined get it added to the stable queue. If you have a 5200 system, please test and make sure it works for you.

Quoted 6 lines

Thanks
g.

 arch/powerpc/platforms/52xx/mpc52xx_pci.c |   24 ++++++++++--------------
 1 files changed, 10 insertions(+), 14 deletions(-)

This doesn't affect Efika, right?

What's the expected impact and how come nobody tried a DMA-capable PCI card in an MPC5200B board before? :)

Signature · 2 lines
Matt Sealey <[email protected]>
Genesi, Manager, Developer Relations
#2 lore raw
Quoted 24 lines · Matt Sealey
Quoted 22 lines · Grant Likely

From: Grant Likely <[email protected]>

This patch ensures that memory gets properly mapped into the PCI address space. Without this patch, the memory window BAR is left at whatever value happened to be loaded into the BAR when Linux was booted. Without this patch, memory could end up getting mapped at any of the 1G address boundaries instead of at '0' where Linux expects it.

Similarly, this patch also ensures that the internally memory mapped registers (IMMR) are mapped to the correct PCI address range.

Without this patch, PCI appears to work correctly until a PCI device is inserted which DMAs into memory.

---

This is a bugfix that I intend to merge into 2.6.29 and once it is mainlined get it added to the stable queue. If you have a 5200 system, please test and make sure it works for you.

Quoted 8 lines
Quoted 6 lines

Thanks
g.

 arch/powerpc/platforms/52xx/mpc52xx_pci.c |   24 ++++++++++--------------
 1 files changed, 10 insertions(+), 14 deletions(-)

This doesn't affect Efika, right?

Correct

Quoted 2 lines

What's the expected impact and how come nobody tried a DMA-capable PCI card in an MPC5200B board before? :)

Most of the board I have don't hit the bug, but only because on most boards U-Boot is setting the BAR to a sane value. I discovered the bug on a custom board where u-boot didn't do the right thing with PCI.

g.

Signature · 2 lines
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.
#3 lore raw
In reply to Matt Sealey
Quoted 33 lines · Matt Sealey
Quoted 22 lines · Grant Likely

From: Grant Likely <[email protected]>

This patch ensures that memory gets properly mapped into the PCI address space. Without this patch, the memory window BAR is left at whatever value happened to be loaded into the BAR when Linux was booted. Without this patch, memory could end up getting mapped at any of the 1G address boundaries instead of at '0' where Linux expects it.

Similarly, this patch also ensures that the internally memory mapped registers (IMMR) are mapped to the correct PCI address range.

Without this patch, PCI appears to work correctly until a PCI device is inserted which DMAs into memory.

---

This is a bugfix that I intend to merge into 2.6.29 and once it is mainlined get it added to the stable queue. If you have a 5200 system, please test and make sure it works for you.

Quoted 7 lines

Thanks
g.

 arch/powerpc/platforms/52xx/mpc52xx_pci.c |   24 ++++++++++--------------
 1 files changed, 10 insertions(+), 14 deletions(-)
Quoted 6 lines

This doesn't affect Efika, right?

What's the expected impact and how come nobody tried a DMA-capable PCI card in an MPC5200B board before? :)

We're running a MPC5200B based board with an external FPGA pumping as much data into the CPU's memory as possible using scatter-gather DMA. No problems so far ... besides the lousy PCI target performance of the 5200, of course. Of course we've been always running latest u-boot.

regards,
Andre

MATRIX VISION GmbH, Talstraße 16, DE-71570 Oppenweiler
Registergericht: Amtsgericht Stuttgart, HRB 271090
Geschäftsführer: Gerhard Thullner, Werner Armingeon, Uwe Furtner, Hans-Joachim Reich

Tested-by on v1 1/1
#4 lore raw
In reply to Grant Likely
Quoted 9 lines · Grant Likely

From: Grant Likely <[email protected]>

This patch ensures that memory gets properly mapped into the PCI address space. Without this patch, the memory window BAR is left at whatever value happened to be loaded into the BAR when Linux was booted. Without this patch, memory could end up getting mapped at any of the 1G address boundaries instead of at '0' where Linux expects it.

Quoted 7 lines

Similarly, this patch also ensures that the internally memory mapped registers (IMMR) are mapped to the correct PCI address range.

Without this patch, PCI appears to work correctly until a PCI device is inserted which DMAs into memory.

No regression with a phyCORE-MPC5200B-tiny. U-Boot did probably the right thing here...

Quoted 70 lines

---

This is a bugfix that I intend to merge into 2.6.29 and once it is mainlined get it added to the stable queue. If you have a 5200 system, please test and make sure it works for you.

Thanks
g.

 arch/powerpc/platforms/52xx/mpc52xx_pci.c |   24 ++++++++++--------------
 1 files changed, 10 insertions(+), 14 deletions(-)
arch/powerpc/platforms/52xx/mpc52xx_pci.c +10 -14
index c3f2c21..87ff522 100644
@@ -20,14 +20,6 @@
 
 
 /* ======================================================================== */
-/* PCI windows config                                                       */
-/* ======================================================================== */
-
-#define MPC52xx_PCI_TARGET_IO	0xf0000000
-#define MPC52xx_PCI_TARGET_MEM	0x00000000
-
-
-/* ======================================================================== */
 /* Structures mapping & Defines for PCI Unit                                */
 /* ======================================================================== */
 
@@ -244,7 +236,7 @@ static struct pci_ops mpc52xx_pci_ops = {
 
 static void __init
 mpc52xx_pci_setup(struct pci_controller *hose,
-                  struct mpc52xx_pci __iomem *pci_regs)
+                  struct mpc52xx_pci __iomem *pci_regs, phys_addr_t pci_phys)
 {
 	struct resource *res;
 	u32 tmp;
@@ -314,10 +306,14 @@ mpc52xx_pci_setup(struct pci_controller *hose,
 	/* Set all the IWCR fields at once; they're in the same reg */
 	out_be32(&pci_regs->iwcr, MPC52xx_PCI_IWCR_PACK(iwcr0, iwcr1, iwcr2));
 
-	out_be32(&pci_regs->tbatr0,
-		MPC52xx_PCI_TBATR_ENABLE | MPC52xx_PCI_TARGET_IO );
-	out_be32(&pci_regs->tbatr1,
-		MPC52xx_PCI_TBATR_ENABLE | MPC52xx_PCI_TARGET_MEM );
+	/* Map IMMR onto PCI bus */
+	pci_phys &= 0xfffc0000; /* bar0 has only 14 significant bits */
+	out_be32(&pci_regs->tbatr0, MPC52xx_PCI_TBATR_ENABLE | pci_phys);
+	out_be32(&pci_regs->bar0, PCI_BASE_ADDRESS_MEM_PREFETCH | pci_phys);
+
+	/* Map memory onto PCI bus */
+	out_be32(&pci_regs->tbatr1, MPC52xx_PCI_TBATR_ENABLE);
+	out_be32(&pci_regs->bar1, PCI_BASE_ADDRESS_MEM_PREFETCH);
 
 	out_be32(&pci_regs->tcr, MPC52xx_PCI_TCR_LD | MPC52xx_PCI_TCR_WCT8);
 
@@ -414,7 +410,7 @@ mpc52xx_add_bridge(struct device_node *node)
 
 	/* Finish setting up PCI using values obtained by
 	 * pci_proces_bridge_OF_ranges */
-	mpc52xx_pci_setup(hose, pci_regs);
+	mpc52xx_pci_setup(hose, pci_regs, rsrc.start);
 
 	return 0;
 }
Signature · 2 lines
  Dipl.-Ing. Wolfram Sang | http://www.pengutronix.de
 Pengutronix - Linux Solutions for Science and Industry