MIPS: Octeon: Allow more than 3.75GB of memory with PCIe

We reserve the 3.75GB - 4GB region of PCIe address space for device to
device transfers, making the corresponding physical memory under
direct mapping unavailable for DMA.

To allow for PCIe DMA to all physical memory we map this chunk of
physical memory with BAR1.  Because of the resulting discontinuity in
the mapping function, we remove a page of memory at each end of the
range so multi-page DMA buffers can never be allocated that span the
range.

Signed-off-by: David Daney <ddaney@caviumnetworks.com>
To: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/1535/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

1 files changed, 31 insertions, 6 deletions

diff --git a/arch/mips/pci/pcie-octeon.c b/arch/mips/pci/pcie-octeon.c
index 6aa5c542d52d..861361e0c9af 100644
--- a/arch/mips/pci/pcie-octeon.c
+++ b/arch/mips/pci/pcie-octeon.c
@@ -402,6 +402,10 @@ static void __cvmx_pcie_rc_initialize_config_space(int pcie_port)
         npei_ctl_status2.s.mps = 0;
         /* Max read request size = 128 bytes for best Octeon DMA performance */
         npei_ctl_status2.s.mrrs = 0;
+        if (pcie_port)
+                npei_ctl_status2.s.c1_b1_s = 3; /* Port1 BAR1 Size 256MB */
+        else
+                npei_ctl_status2.s.c0_b1_s = 3; /* Port0 BAR1 Size 256MB */
         cvmx_write_csr(CVMX_PEXP_NPEI_CTL_STATUS2, npei_ctl_status2.u64);
 
         /* ECRC Generation (PCIE*_CFG070[GE,CE]) */
@@ -666,6 +670,8 @@ static int __cvmx_pcie_rc_initialize_link(int pcie_port)
 static int cvmx_pcie_rc_initialize(int pcie_port)
 {
         int i;
+        int base;
+        u64 addr_swizzle;
         union cvmx_ciu_soft_prst ciu_soft_prst;
         union cvmx_pescx_bist_status pescx_bist_status;
         union cvmx_pescx_bist_status2 pescx_bist_status2;
@@ -674,6 +680,7 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
         union cvmx_npei_mem_access_subidx mem_access_subid;
         union cvmx_npei_dbg_data npei_dbg_data;
         union cvmx_pescx_ctl_status2 pescx_ctl_status2;
+        union cvmx_npei_bar1_indexx bar1_index;
 
         /*
          * Make sure we aren't trying to setup a target mode interface
@@ -918,12 +925,30 @@ static int cvmx_pcie_rc_initialize(int pcie_port)
         /* Set Octeon's BAR0 to decode 0-16KB. It overlaps with Bar2 */
         cvmx_write_csr(CVMX_PESCX_P2N_BAR0_START(pcie_port), 0);
 
-        /*
-         * Disable Octeon's BAR1. It isn't needed in RC mode since
-         * BAR2 maps all of memory. BAR2 also maps 256MB-512MB into
-         * the 2nd 256MB of memory.
-         */
-        cvmx_write_csr(CVMX_PESCX_P2N_BAR1_START(pcie_port), -1);
+        /* BAR1 follows BAR2 with a gap. */
+        cvmx_write_csr(CVMX_PESCX_P2N_BAR1_START(pcie_port), CVMX_PCIE_BAR1_RC_BASE);
+
+        bar1_index.u32 = 0;
+        bar1_index.s.addr_idx = (CVMX_PCIE_BAR1_PHYS_BASE >> 22);
+        bar1_index.s.ca = 1;       /* Not Cached */
+        bar1_index.s.end_swp = 1;  /* Endian Swap mode */
+        bar1_index.s.addr_v = 1;   /* Valid entry */
+
+        base = pcie_port ? 16 : 0;
+
+        /* Big endian swizzle for 32-bit PEXP_NCB register. */
+#ifdef __MIPSEB__
+        addr_swizzle = 4;
+#else
+        addr_swizzle = 0;
+#endif
+        for (i = 0; i < 16; i++) {
+                cvmx_write64_uint32((CVMX_PEXP_NPEI_BAR1_INDEXX(base) ^ addr_swizzle),
+                                    bar1_index.u32);
+                base++;
+                /* 256MB / 16 >> 22 == 4 */
+                bar1_index.s.addr_idx += (((1ull << 28) / 16ull) >> 22);
+        }
 
         /*
          * Set Octeon's BAR2 to decode 0-2^39. Bar0 and Bar1 take