[rtems commit] bsps/qoriq: Add MMU regions for PCIe based on fdt

[Christian Mauderer]

Module:    rtems
Branch:    master
Commit:    c9dda8cda8d6d13dcc07a5fea83da450c9bf809e
Changeset: http://git.rtems.org/rtems/commit/?id=c9dda8cda8d6d13dcc07a5fea83da450c9bf809e

Author:    Christian Mauderer <christian.mauderer at embedded-brains.de>
Date:      Wed Jul 26 13:38:45 2023 +0200

bsps/qoriq: Add MMU regions for PCIe based on fdt

Get the memory ranges for the PCIe from the FDT and add them to the MMU.
This is necessary so that the PCIe driver in libbsd can work.

---

 bsps/powerpc/qoriq/start/mmu-config.c | 88 +++++++++++++++++++++++++++++++++++
 1 file changed, 88 insertions(+)

diff --git a/bsps/powerpc/qoriq/start/mmu-config.c b/bsps/powerpc/qoriq/start/mmu-config.c
index 58b08c8349..15e4a83fc4 100644
--- a/bsps/powerpc/qoriq/start/mmu-config.c
+++ b/bsps/powerpc/qoriq/start/mmu-config.c
@@ -305,6 +305,92 @@ static void TEXT config_fdt_adjust(const void *fdt)
 	}
 }
 
+/*
+ * Each PCIe controller has a ranges attribute in the fdt like the following:
+ *
+ *   ranges = <0x2000000 0x00 0xc0000000 0x00 0xc0000000 0x00 0x20000000
+ *             0x1000000 0x00 0x00000000 0x00 0xffc20000 0x00 0x00010000>;
+ *             |------PCI address------| |-CPU address-| |-----size----|
+ *
+ * In theory, some fdt-attributes should be used to find out how long the PCI
+ * address (#address-cells of the PCIe node), the CPU address (#address-cells of
+ * the parent node) and the size (#size-cells of the PCIe node) are. In our case
+ * the structure is fixed because the pcie root controllers are a part of the
+ * chip. Therefore the sizes will never change and we can assume fixed lengths.
+ *
+ * The first cell of the PCI address holds a number of flags. A detailed
+ * explanation can be found for example here:
+ *
+ * https://web.archive.org/web/20240109080338/https://michael2012z.medium.com/understanding-pci-node-in-fdt-769a894a13cc
+ *
+ * We are only interested in the entry with the flags 0x02000000 which basically
+ * means that it is a non-relocatable, non-prefetchable, not-aliased 32 bit
+ * memory space on the first bus.
+ *
+ * The other two cells of the PCI address are a 64 Bit address viewed from PCI
+ * address space. The two CPU address cells are the same 64 Bit address viewed
+ * from CPU address space. For our controller these two should always be the
+ * same (no address translation). The last two cells give a size of the memory
+ * region (in theory in PCI address space but it has to be the same for CPU and
+ * PCI).
+ */
+static void TEXT add_pcie_regions(qoriq_mmu_context *context, const void *fdt)
+{
+	int node;
+
+	node = -1;
+
+	while (true) {
+		static const size_t range_length = 7 * 4;
+		const void *val;
+		int len;
+
+		node = fdt_node_offset_by_compatible(
+			fdt,
+			node,
+			"fsl,mpc8548-pcie"
+		);
+		if (node < 0) {
+			break;
+		}
+
+		val = fdt_getprop(fdt, node, "ranges", &len);
+		if (len % range_length != 0) {
+			continue;
+		}
+
+		while (len >= range_length) {
+			uint32_t pci_addr_flags;
+			uintptr_t pci_addr;
+			uintptr_t cpu_addr;
+			uintptr_t size;
+			const uint32_t *cells;
+
+			cells = val;
+			pci_addr_flags = fdt32_to_cpu(cells[0]);
+			pci_addr = fdt64_to_cpu(*(fdt64_t *)(&cells[1]));
+			cpu_addr = fdt64_to_cpu(*(fdt64_t *)(&cells[3]));
+			size = fdt64_to_cpu(*(fdt64_t *)(&cells[5]));
+
+			if (pci_addr_flags == 0x02000000 &&
+			    pci_addr == cpu_addr) {
+				/* Add as I/O memory */
+				qoriq_mmu_add(
+					context,
+					cpu_addr,
+					cpu_addr + size - 1,
+					0,
+					FSL_EIS_MAS2_I | FSL_EIS_MAS2_G,
+					FSL_EIS_MAS3_SR | FSL_EIS_MAS3_SW,
+					0
+				);
+			}
+			len -= range_length;
+			val += range_length;
+		}
+	}
+}
+
 void TEXT qoriq_mmu_config(bool boot_processor, int first_tlb, int scratch_tlb)
 {
 	qoriq_mmu_context context;
@@ -349,6 +435,8 @@ void TEXT qoriq_mmu_config(bool boot_processor, int first_tlb, int scratch_tlb)
 		}
 	}
 
+	add_pcie_regions(&context, fdt);
+
 	qoriq_mmu_partition(&context, max_count);
 	qoriq_mmu_write_to_tlb1(&context, first_tlb);
 }