powerpc/powernv: PCI support for p7IOC under OPAL v2
authorBenjamin Herrenschmidt <benh@kernel.crashing.org>
Tue, 15 Nov 2011 17:29:08 +0000 (17:29 +0000)
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>
Fri, 25 Nov 2011 03:53:15 +0000 (14:53 +1100)
This adds support for p7IOC (and possibly other IODA v1 IO Hubs)
using OPAL v2 interfaces.

We completely take over resource assignment and assign them using an
algorithm that hands out device BARs in a way that makes them fit in
individual segments of the M32 window of the bridge, which enables us
to assign individual PEs to devices and functions.

The current implementation gives out a PE per functions on PCIe, and a
PE for the entire bridge for PCIe to PCI-X bridges.

This can be adjusted / fine tuned later.

We also setup DMA resources (32-bit only for now) and MSIs (both 32-bit
and 64-bit MSI are supported).

The DMA allocation tries to divide the available 256M segments of the
32-bit DMA address space "fairly" among PEs. This is done using a
"weight" heuristic which assigns less value to things like OHCI USB
controllers than, for example SCSI RAID controllers. This algorithm
will probably want some fine tuning for specific devices or device
types.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
arch/powerpc/include/asm/pci-bridge.h
arch/powerpc/kernel/pci_dn.c
arch/powerpc/platforms/powernv/Makefile
arch/powerpc/platforms/powernv/pci-ioda.c [new file with mode: 0644]
arch/powerpc/platforms/powernv/pci.c
arch/powerpc/platforms/powernv/pci.h

index 56b879ab3a40211a26c8a7af9a4d606be2913e89..882b6aa6c85765f9bdd5df1f55c1cfad13d07ac1 100644 (file)
@@ -153,8 +153,8 @@ struct pci_dn {
 
        int     pci_ext_config_space;   /* for pci devices */
 
-#ifdef CONFIG_EEH
        struct  pci_dev *pcidev;        /* back-pointer to the pci device */
+#ifdef CONFIG_EEH
        int     class_code;             /* pci device class */
        int     eeh_mode;               /* See eeh.h for possible EEH_MODEs */
        int     eeh_config_addr;
@@ -164,6 +164,10 @@ struct pci_dn {
        int     eeh_false_positives;    /* # times this device reported #ff's */
        u32     config_space[16];       /* saved PCI config space */
 #endif
+#define IODA_INVALID_PE                (-1)
+#ifdef CONFIG_PPC_POWERNV
+       int     pe_number;
+#endif
 };
 
 /* Get the pointer to a device_node's pci_dn */
index 4e69deb89b37b752bfaa74aae679ecee82667aa7..dd9e4a04bf795b0529da827738d2fe8484f46be6 100644 (file)
@@ -50,6 +50,9 @@ void * __devinit update_dn_pci_info(struct device_node *dn, void *data)
        dn->data = pdn;
        pdn->node = dn;
        pdn->phb = phb;
+#ifdef CONFIG_PPC_POWERNV
+       pdn->pe_number = IODA_INVALID_PE;
+#endif
        regs = of_get_property(dn, "reg", NULL);
        if (regs) {
                /* First register entry is addr (00BBSS00)  */
index 31853008b4189f016e251910575cf706c0cee0d5..bcc3cb48a44ee341dc7fb871387ff3cf391fe90a 100644 (file)
@@ -2,4 +2,4 @@ obj-y                   += setup.o opal-takeover.o opal-wrappers.o opal.o
 obj-y                  += opal-rtc.o opal-nvram.o
 
 obj-$(CONFIG_SMP)      += smp.o
-obj-$(CONFIG_PCI)      += pci.o pci-p5ioc2.o
+obj-$(CONFIG_PCI)      += pci.o pci-p5ioc2.o pci-ioda.o
diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c
new file mode 100644 (file)
index 0000000..cf89f30
--- /dev/null
@@ -0,0 +1,1325 @@
+/*
+ * Support PCI/PCIe on PowerNV platforms
+ *
+ * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#define DEBUG
+
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include <linux/delay.h>
+#include <linux/string.h>
+#include <linux/init.h>
+#include <linux/bootmem.h>
+#include <linux/irq.h>
+#include <linux/io.h>
+#include <linux/msi.h>
+
+#include <asm/sections.h>
+#include <asm/io.h>
+#include <asm/prom.h>
+#include <asm/pci-bridge.h>
+#include <asm/machdep.h>
+#include <asm/ppc-pci.h>
+#include <asm/opal.h>
+#include <asm/iommu.h>
+#include <asm/tce.h>
+#include <asm/abs_addr.h>
+
+#include "powernv.h"
+#include "pci.h"
+
+struct resource_wrap {
+       struct list_head        link;
+       resource_size_t         size;
+       resource_size_t         align;
+       struct pci_dev          *dev;   /* Set if it's a device */
+       struct pci_bus          *bus;   /* Set if it's a bridge */
+};
+
+static int __pe_printk(const char *level, const struct pnv_ioda_pe *pe,
+                      struct va_format *vaf)
+{
+       char pfix[32];
+
+       if (pe->pdev)
+               strlcpy(pfix, dev_name(&pe->pdev->dev), sizeof(pfix));
+       else
+               sprintf(pfix, "%04x:%02x     ",
+                       pci_domain_nr(pe->pbus), pe->pbus->number);
+       return printk("pci %s%s: [PE# %.3d] %pV", level, pfix, pe->pe_number, vaf);
+}
+
+#define define_pe_printk_level(func, kern_level)               \
+static int func(const struct pnv_ioda_pe *pe, const char *fmt, ...)    \
+{                                                              \
+       struct va_format vaf;                                   \
+       va_list args;                                           \
+       int r;                                                  \
+                                                               \
+       va_start(args, fmt);                                    \
+                                                               \
+       vaf.fmt = fmt;                                          \
+       vaf.va = &args;                                         \
+                                                               \
+       r = __pe_printk(kern_level, pe, &vaf);                  \
+       va_end(args);                                           \
+                                                               \
+       return r;                                               \
+}                                                              \
+
+define_pe_printk_level(pe_err, KERN_ERR);
+define_pe_printk_level(pe_warn, KERN_WARNING);
+define_pe_printk_level(pe_info, KERN_INFO);
+
+
+/* Calculate resource usage & alignment requirement of a single
+ * device. This will also assign all resources within the device
+ * for a given type starting at 0 for the biggest one and then
+ * assigning in decreasing order of size.
+ */
+static void __devinit pnv_ioda_calc_dev(struct pci_dev *dev, unsigned int flags,
+                                       resource_size_t *size,
+                                       resource_size_t *align)
+{
+       resource_size_t start;
+       struct resource *r;
+       int i;
+
+       pr_devel("  -> CDR %s\n", pci_name(dev));
+
+       *size = *align = 0;
+
+       /* Clear the resources out and mark them all unset */
+       for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
+               r = &dev->resource[i];
+               if (!(r->flags & flags))
+                   continue;
+               if (r->start) {
+                       r->end -= r->start;
+                       r->start = 0;
+               }
+               r->flags |= IORESOURCE_UNSET;
+       }
+
+       /* We currently keep all memory resources together, we
+        * will handle prefetch & 64-bit separately in the future
+        * but for now we stick everybody in M32
+        */
+       start = 0;
+       for (;;) {
+               resource_size_t max_size = 0;
+               int max_no = -1;
+
+               /* Find next biggest resource */
+               for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
+                       r = &dev->resource[i];
+                       if (!(r->flags & IORESOURCE_UNSET) ||
+                           !(r->flags & flags))
+                               continue;
+                       if (resource_size(r) > max_size) {
+                               max_size = resource_size(r);
+                               max_no = i;
+                       }
+               }
+               if (max_no < 0)
+                       break;
+               r = &dev->resource[max_no];
+               if (max_size > *align)
+                       *align = max_size;
+               *size += max_size;
+               r->start = start;
+               start += max_size;
+               r->end = r->start + max_size - 1;
+               r->flags &= ~IORESOURCE_UNSET;
+               pr_devel("  ->     R%d %016llx..%016llx\n",
+                        max_no, r->start, r->end);
+       }
+       pr_devel("  <- CDR %s size=%llx align=%llx\n",
+                pci_name(dev), *size, *align);
+}
+
+/* Allocate a resource "wrap" for a given device or bridge and
+ * insert it at the right position in the sorted list
+ */
+static void __devinit pnv_ioda_add_wrap(struct list_head *list,
+                                       struct pci_bus *bus,
+                                       struct pci_dev *dev,
+                                       resource_size_t size,
+                                       resource_size_t align)
+{
+       struct resource_wrap *w1, *w = kzalloc(sizeof(*w), GFP_KERNEL);
+
+       w->size = size;
+       w->align = align;
+       w->dev = dev;
+       w->bus = bus;
+
+       list_for_each_entry(w1, list, link) {
+               if (w1->align < align) {
+                       list_add_tail(&w->link, &w1->link);
+                       return;
+               }
+       }
+       list_add_tail(&w->link, list);
+}
+
+/* Offset device resources of a given type */
+static void __devinit pnv_ioda_offset_dev(struct pci_dev *dev,
+                                         unsigned int flags,
+                                         resource_size_t offset)
+{
+       struct resource *r;
+       int i;
+
+       pr_devel("  -> ODR %s [%x] +%016llx\n", pci_name(dev), flags, offset);
+
+       for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
+               r = &dev->resource[i];
+               if (r->flags & flags) {
+                       dev->resource[i].start += offset;
+                       dev->resource[i].end += offset;
+               }
+       }
+
+       pr_devel("  <- ODR %s [%x] +%016llx\n", pci_name(dev), flags, offset);
+}
+
+/* Offset bus resources (& all children) of a given type */
+static void __devinit pnv_ioda_offset_bus(struct pci_bus *bus,
+                                         unsigned int flags,
+                                         resource_size_t offset)
+{
+       struct resource *r;
+       struct pci_dev *dev;
+       struct pci_bus *cbus;
+       int i;
+
+       pr_devel("  -> OBR %s [%x] +%016llx\n",
+                bus->self ? pci_name(bus->self) : "root", flags, offset);
+
+       for (i = 0; i < 2; i++) {
+               r = bus->resource[i];
+               if (r && (r->flags & flags)) {
+                       bus->resource[i]->start += offset;
+                       bus->resource[i]->end += offset;
+               }
+       }
+       list_for_each_entry(dev, &bus->devices, bus_list)
+               pnv_ioda_offset_dev(dev, flags, offset);
+       list_for_each_entry(cbus, &bus->children, node)
+               pnv_ioda_offset_bus(cbus, flags, offset);
+
+       pr_devel("  <- OBR %s [%x]\n",
+                bus->self ? pci_name(bus->self) : "root", flags);
+}
+
+/* This is the guts of our IODA resource allocation. This is called
+ * recursively for each bus in the system. It calculates all the
+ * necessary size and requirements for children and assign them
+ * resources such that:
+ *
+ *   - Each function fits in it's own contiguous set of IO/M32
+ *     segment
+ *
+ *   - All segments behind a P2P bridge are contiguous and obey
+ *     alignment constraints of those bridges
+ */
+static void __devinit pnv_ioda_calc_bus(struct pci_bus *bus, unsigned int flags,
+                                       resource_size_t *size,
+                                       resource_size_t *align)
+{
+       struct pci_controller *hose = pci_bus_to_host(bus);
+       struct pnv_phb *phb = hose->private_data;
+       resource_size_t dev_size, dev_align, start;
+       resource_size_t min_align, min_balign;
+       struct pci_dev *cdev;
+       struct pci_bus *cbus;
+       struct list_head head;
+       struct resource_wrap *w;
+       unsigned int bres;
+
+       *size = *align = 0;
+
+       pr_devel("-> CBR %s [%x]\n",
+                bus->self ? pci_name(bus->self) : "root", flags);
+
+       /* Calculate alignment requirements based on the type
+        * of resource we are working on
+        */
+       if (flags & IORESOURCE_IO) {
+               bres = 0;
+               min_align = phb->ioda.io_segsize;
+               min_balign = 0x1000;
+       } else {
+               bres = 1;
+               min_align = phb->ioda.m32_segsize;
+               min_balign = 0x100000;
+       }
+
+       /* Gather all our children resources ordered by alignment */
+       INIT_LIST_HEAD(&head);
+
+       /*   - Busses */
+       list_for_each_entry(cbus, &bus->children, node) {
+               pnv_ioda_calc_bus(cbus, flags, &dev_size, &dev_align);
+               pnv_ioda_add_wrap(&head, cbus, NULL, dev_size, dev_align);
+       }
+
+       /*   - Devices */
+       list_for_each_entry(cdev, &bus->devices, bus_list) {
+               pnv_ioda_calc_dev(cdev, flags, &dev_size, &dev_align);
+               /* Align them to segment size */
+               if (dev_align < min_align)
+                       dev_align = min_align;
+               pnv_ioda_add_wrap(&head, NULL, cdev, dev_size, dev_align);
+       }
+       if (list_empty(&head))
+               goto empty;
+
+       /* Now we can do two things: assign offsets to them within that
+        * level and get our total alignment & size requirements. The
+        * assignment algorithm is going to be uber-trivial for now, we
+        * can try to be smarter later at filling out holes.
+        */
+       start = bus->self ? 0 : bus->resource[bres]->start;
+
+       /* Don't hand out IO 0 */
+       if ((flags & IORESOURCE_IO) && !bus->self)
+               start += 0x1000;
+
+       while(!list_empty(&head)) {
+               w = list_first_entry(&head, struct resource_wrap, link);
+               list_del(&w->link);
+               if (w->size) {
+                       if (start) {
+                               start = ALIGN(start, w->align);
+                               if (w->dev)
+                                       pnv_ioda_offset_dev(w->dev,flags,start);
+                               else if (w->bus)
+                                       pnv_ioda_offset_bus(w->bus,flags,start);
+                       }
+                       if (w->align > *align)
+                               *align = w->align;
+               }
+               start += w->size;
+               kfree(w);
+       }
+       *size = start;
+
+       /* Align and setup bridge resources */
+       *align = max_t(resource_size_t, *align,
+                      max_t(resource_size_t, min_align, min_balign));
+       *size = ALIGN(*size,
+                     max_t(resource_size_t, min_align, min_balign));
+ empty:
+       /* Only setup P2P's, not the PHB itself */
+       if (bus->self) {
+               WARN_ON(bus->resource[bres] == NULL);
+               bus->resource[bres]->start = 0;
+               bus->resource[bres]->flags = (*size) ? flags : 0;
+               bus->resource[bres]->end = (*size) ? (*size - 1) : 0;
+
+               /* Clear prefetch bus resources for now */
+               bus->resource[2]->flags = 0;
+       }
+
+       pr_devel("<- CBR %s [%x] *size=%016llx *align=%016llx\n",
+                bus->self ? pci_name(bus->self) : "root", flags,*size,*align);
+}
+
+static struct pci_dn *pnv_ioda_get_pdn(struct pci_dev *dev)
+{
+       struct device_node *np;
+
+       np = pci_device_to_OF_node(dev);
+       if (!np)
+               return NULL;
+       return PCI_DN(np);
+}
+
+static void __devinit pnv_ioda_setup_pe_segments(struct pci_dev *dev)
+{
+       struct pci_controller *hose = pci_bus_to_host(dev->bus);
+       struct pnv_phb *phb = hose->private_data;
+       struct pci_dn *pdn = pnv_ioda_get_pdn(dev);
+       unsigned int pe, i;
+       resource_size_t pos;
+       struct resource io_res;
+       struct resource m32_res;
+       struct pci_bus_region region;
+       int rc;
+
+       /* Anything not referenced in the device-tree gets PE#0 */
+       pe = pdn ? pdn->pe_number : 0;
+
+       /* Calculate the device min/max */
+       io_res.start = m32_res.start = (resource_size_t)-1;
+       io_res.end = m32_res.end = 0;
+       io_res.flags = IORESOURCE_IO;
+       m32_res.flags = IORESOURCE_MEM;
+
+       for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
+               struct resource *r = NULL;
+               if (dev->resource[i].flags & IORESOURCE_IO)
+                       r = &io_res;
+               if (dev->resource[i].flags & IORESOURCE_MEM)
+                       r = &m32_res;
+               if (!r)
+                       continue;
+               if (dev->resource[i].start < r->start)
+                       r->start = dev->resource[i].start;
+               if (dev->resource[i].end > r->end)
+                       r->end = dev->resource[i].end;
+       }
+
+       /* Setup IO segments */
+       if (io_res.start < io_res.end) {
+               pcibios_resource_to_bus(dev, &region, &io_res);
+               pos = region.start;
+               i = pos / phb->ioda.io_segsize;
+               while(i < phb->ioda.total_pe && pos <= region.end) {
+                       if (phb->ioda.io_segmap[i]) {
+                               pr_err("%s: Trying to use IO seg #%d which is"
+                                      " already used by PE# %d\n",
+                                      pci_name(dev), i,
+                                      phb->ioda.io_segmap[i]);
+                               /* XXX DO SOMETHING TO DISABLE DEVICE ? */
+                               break;
+                       }
+                       phb->ioda.io_segmap[i] = pe;
+                       rc = opal_pci_map_pe_mmio_window(phb->opal_id, pe,
+                                                        OPAL_IO_WINDOW_TYPE,
+                                                        0, i);
+                       if (rc != OPAL_SUCCESS) {
+                               pr_err("%s: OPAL error %d setting up mapping"
+                                      " for IO seg# %d\n",
+                                      pci_name(dev), rc, i);
+                               /* XXX DO SOMETHING TO DISABLE DEVICE ? */
+                               break;
+                       }
+                       pos += phb->ioda.io_segsize;
+                       i++;
+               };
+       }
+
+       /* Setup M32 segments */
+       if (m32_res.start < m32_res.end) {
+               pcibios_resource_to_bus(dev, &region, &m32_res);
+               pos = region.start;
+               i = pos / phb->ioda.m32_segsize;
+               while(i < phb->ioda.total_pe && pos <= region.end) {
+                       if (phb->ioda.m32_segmap[i]) {
+                               pr_err("%s: Trying to use M32 seg #%d which is"
+                                      " already used by PE# %d\n",
+                                      pci_name(dev), i,
+                                      phb->ioda.m32_segmap[i]);
+                               /* XXX DO SOMETHING TO DISABLE DEVICE ? */
+                               break;
+                       }
+                       phb->ioda.m32_segmap[i] = pe;
+                       rc = opal_pci_map_pe_mmio_window(phb->opal_id, pe,
+                                                        OPAL_M32_WINDOW_TYPE,
+                                                        0, i);
+                       if (rc != OPAL_SUCCESS) {
+                               pr_err("%s: OPAL error %d setting up mapping"
+                                      " for M32 seg# %d\n",
+                                      pci_name(dev), rc, i);
+                               /* XXX DO SOMETHING TO DISABLE DEVICE ? */
+                               break;
+                       }
+                       pos += phb->ioda.m32_segsize;
+                       i++;
+               }
+       }
+}
+
+/* Check if a resource still fits in the total IO or M32 range
+ * for a given PHB
+ */
+static int __devinit pnv_ioda_resource_fit(struct pci_controller *hose,
+                                          struct resource *r)
+{
+       struct resource *bounds;
+
+       if (r->flags & IORESOURCE_IO)
+               bounds = &hose->io_resource;
+       else if (r->flags & IORESOURCE_MEM)
+               bounds = &hose->mem_resources[0];
+       else
+               return 1;
+
+       if (r->start >= bounds->start && r->end <= bounds->end)
+               return 1;
+       r->flags = 0;
+       return 0;
+}
+
+static void __devinit pnv_ioda_update_resources(struct pci_bus *bus)
+{
+       struct pci_controller *hose = pci_bus_to_host(bus);
+       struct pci_bus *cbus;
+       struct pci_dev *cdev;
+       unsigned int i;
+       u16 cmd;
+
+       /* Clear all device enables  */
+       list_for_each_entry(cdev, &bus->devices, bus_list) {
+               pci_read_config_word(cdev, PCI_COMMAND, &cmd);
+               cmd &= ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY|PCI_COMMAND_MASTER);
+               pci_write_config_word(cdev, PCI_COMMAND, cmd);
+       }
+
+       /* Check if bus resources fit in our IO or M32 range */
+       for (i = 0; bus->self && (i < 2); i++) {
+               struct resource *r = bus->resource[i];
+               if (r && !pnv_ioda_resource_fit(hose, r))
+                       pr_err("%s: Bus %d resource %d disabled, no room\n",
+                              pci_name(bus->self), bus->number, i);
+       }
+
+       /* Update self if it's not a PHB */
+       if (bus->self)
+               pci_setup_bridge(bus);
+
+       /* Update child devices */
+       list_for_each_entry(cdev, &bus->devices, bus_list) {
+               /* Check if resource fits, if not, disabled it */
+               for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
+                       struct resource *r = &cdev->resource[i];
+                       if (!pnv_ioda_resource_fit(hose, r))
+                               pr_err("%s: Resource %d disabled, no room\n",
+                                      pci_name(cdev), i);
+               }
+
+               /* Assign segments */
+               pnv_ioda_setup_pe_segments(cdev);
+
+               /* Update HW BARs */
+               for (i = 0; i <= PCI_ROM_RESOURCE; i++)
+                       pci_update_resource(cdev, i);
+       }
+
+       /* Update child busses */
+       list_for_each_entry(cbus, &bus->children, node)
+               pnv_ioda_update_resources(cbus);
+}
+
+static int __devinit pnv_ioda_alloc_pe(struct pnv_phb *phb)
+{
+       unsigned long pe;
+
+       do {
+               pe = find_next_zero_bit(phb->ioda.pe_alloc,
+                                       phb->ioda.total_pe, 0);
+               if (pe >= phb->ioda.total_pe)
+                       return IODA_INVALID_PE;
+       } while(test_and_set_bit(pe, phb->ioda.pe_alloc));
+
+       phb->ioda.pe_array[pe].pe_number = pe;
+       return pe;
+}
+
+static void __devinit pnv_ioda_free_pe(struct pnv_phb *phb, int pe)
+{
+       WARN_ON(phb->ioda.pe_array[pe].pdev);
+
+       memset(&phb->ioda.pe_array[pe], 0, sizeof(struct pnv_ioda_pe));
+       clear_bit(pe, phb->ioda.pe_alloc);
+}
+
+/* Currently those 2 are only used when MSIs are enabled, this will change
+ * but in the meantime, we need to protect them to avoid warnings
+ */
+#ifdef CONFIG_PCI_MSI
+static struct pnv_ioda_pe * __devinit __pnv_ioda_get_one_pe(struct pci_dev *dev)
+{
+       struct pci_controller *hose = pci_bus_to_host(dev->bus);
+       struct pnv_phb *phb = hose->private_data;
+       struct pci_dn *pdn = pnv_ioda_get_pdn(dev);
+
+       if (!pdn)
+               return NULL;
+       if (pdn->pe_number == IODA_INVALID_PE)
+               return NULL;
+       return &phb->ioda.pe_array[pdn->pe_number];
+}
+
+static struct pnv_ioda_pe * __devinit pnv_ioda_get_pe(struct pci_dev *dev)
+{
+       struct pnv_ioda_pe *pe = __pnv_ioda_get_one_pe(dev);
+
+       while (!pe && dev->bus->self) {
+               dev = dev->bus->self;
+               pe = __pnv_ioda_get_one_pe(dev);
+               if (pe)
+                       pe = pe->bus_pe;
+       }
+       return pe;
+}
+#endif /* CONFIG_PCI_MSI */
+
+static int __devinit pnv_ioda_configure_pe(struct pnv_phb *phb,
+                                          struct pnv_ioda_pe *pe)
+{
+       struct pci_dev *parent;
+       uint8_t bcomp, dcomp, fcomp;
+       long rc, rid_end, rid;
+
+       /* Bus validation ? */
+       if (pe->pbus) {
+               int count;
+
+               dcomp = OPAL_IGNORE_RID_DEVICE_NUMBER;
+               fcomp = OPAL_IGNORE_RID_FUNCTION_NUMBER;
+               parent = pe->pbus->self;
+               count = pe->pbus->subordinate - pe->pbus->secondary + 1;
+               switch(count) {
+               case  1: bcomp = OpalPciBusAll;         break;
+               case  2: bcomp = OpalPciBus7Bits;       break;
+               case  4: bcomp = OpalPciBus6Bits;       break;
+               case  8: bcomp = OpalPciBus5Bits;       break;
+               case 16: bcomp = OpalPciBus4Bits;       break;
+               case 32: bcomp = OpalPciBus3Bits;       break;
+               default:
+                       pr_err("%s: Number of subordinate busses %d"
+                              " unsupported\n",
+                              pci_name(pe->pbus->self), count);
+                       /* Do an exact match only */
+                       bcomp = OpalPciBusAll;
+               }
+               rid_end = pe->rid + (count << 8);
+       } else {
+               parent = pe->pdev->bus->self;
+               bcomp = OpalPciBusAll;
+               dcomp = OPAL_COMPARE_RID_DEVICE_NUMBER;
+               fcomp = OPAL_COMPARE_RID_FUNCTION_NUMBER;
+               rid_end = pe->rid + 1;
+       }
+
+       /* Associate PE in PELT */
+       rc = opal_pci_set_pe(phb->opal_id, pe->pe_number, pe->rid,
+                            bcomp, dcomp, fcomp, OPAL_MAP_PE);
+       if (rc) {
+               pe_err(pe, "OPAL error %ld trying to setup PELT table\n", rc);
+               return -ENXIO;
+       }
+       opal_pci_eeh_freeze_clear(phb->opal_id, pe->pe_number,
+                                 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
+
+       /* Add to all parents PELT-V */
+       while (parent) {
+               struct pci_dn *pdn = pnv_ioda_get_pdn(parent);
+               if (pdn && pdn->pe_number != IODA_INVALID_PE) {
+                       rc = opal_pci_set_peltv(phb->opal_id, pdn->pe_number,
+                                               pe->pe_number, 1);
+                       /* XXX What to do in case of error ? */
+               }
+               parent = parent->bus->self;
+       }
+       /* Setup reverse map */
+       for (rid = pe->rid; rid < rid_end; rid++)
+               phb->ioda.pe_rmap[rid] = pe->pe_number;
+
+       /* Setup one MVTs on IODA1 */
+       if (phb->type == PNV_PHB_IODA1) {
+               pe->mve_number = pe->pe_number;
+               rc = opal_pci_set_mve(phb->opal_id, pe->mve_number,
+                                     pe->pe_number);
+               if (rc) {
+                       pe_err(pe, "OPAL error %ld setting up MVE %d\n",
+                              rc, pe->mve_number);
+                       pe->mve_number = -1;
+               } else {
+                       rc = opal_pci_set_mve_enable(phb->opal_id,
+                                                    pe->mve_number, 1);
+                       if (rc) {
+                               pe_err(pe, "OPAL error %ld enabling MVE %d\n",
+                                      rc, pe->mve_number);
+                               pe->mve_number = -1;
+                       }
+               }
+       } else if (phb->type == PNV_PHB_IODA2)
+               pe->mve_number = 0;
+
+       return 0;
+}
+
+static void __devinit pnv_ioda_link_pe_by_weight(struct pnv_phb *phb,
+                                                struct pnv_ioda_pe *pe)
+{
+       struct pnv_ioda_pe *lpe;
+
+       list_for_each_entry(lpe, &phb->ioda.pe_list, link) {
+               if (lpe->dma_weight < pe->dma_weight) {
+                       list_add_tail(&pe->link, &lpe->link);
+                       return;
+               }
+       }
+       list_add_tail(&pe->link, &phb->ioda.pe_list);
+}
+
+static unsigned int pnv_ioda_dma_weight(struct pci_dev *dev)
+{
+       /* This is quite simplistic. The "base" weight of a device
+        * is 10. 0 means no DMA is to be accounted for it.
+        */
+
+       /* If it's a bridge, no DMA */
+       if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
+               return 0;
+
+       /* Reduce the weight of slow USB controllers */
+       if (dev->class == PCI_CLASS_SERIAL_USB_UHCI ||
+           dev->class == PCI_CLASS_SERIAL_USB_OHCI ||
+           dev->class == PCI_CLASS_SERIAL_USB_EHCI)
+               return 3;
+
+       /* Increase the weight of RAID (includes Obsidian) */
+       if ((dev->class >> 8) == PCI_CLASS_STORAGE_RAID)
+               return 15;
+
+       /* Default */
+       return 10;
+}
+
+static struct pnv_ioda_pe * __devinit pnv_ioda_setup_dev_PE(struct pci_dev *dev)
+{
+       struct pci_controller *hose = pci_bus_to_host(dev->bus);
+       struct pnv_phb *phb = hose->private_data;
+       struct pci_dn *pdn = pnv_ioda_get_pdn(dev);
+       struct pnv_ioda_pe *pe;
+       int pe_num;
+
+       if (!pdn) {
+               pr_err("%s: Device tree node not associated properly\n",
+                          pci_name(dev));
+               return NULL;
+       }
+       if (pdn->pe_number != IODA_INVALID_PE)
+               return NULL;
+
+       /* PE#0 has been pre-set */
+       if (dev->bus->number == 0)
+               pe_num = 0;
+       else
+               pe_num = pnv_ioda_alloc_pe(phb);
+       if (pe_num == IODA_INVALID_PE) {
+               pr_warning("%s: Not enough PE# available, disabling device\n",
+                          pci_name(dev));
+               return NULL;
+       }
+
+       /* NOTE: We get only one ref to the pci_dev for the pdn, not for the
+        * pointer in the PE data structure, both should be destroyed at the
+        * same time. However, this needs to be looked at more closely again
+        * once we actually start removing things (Hotplug, SR-IOV, ...)
+        *
+        * At some point we want to remove the PDN completely anyways
+        */
+       pe = &phb->ioda.pe_array[pe_num];
+       pci_dev_get(dev);
+       pdn->pcidev = dev;
+       pdn->pe_number = pe_num;
+       pe->pdev = dev;
+       pe->pbus = NULL;
+       pe->tce32_seg = -1;
+       pe->mve_number = -1;
+       pe->rid = dev->bus->number << 8 | pdn->devfn;
+
+       pe_info(pe, "Associated device to PE\n");
+
+       if (pnv_ioda_configure_pe(phb, pe)) {
+               /* XXX What do we do here ? */
+               if (pe_num)
+                       pnv_ioda_free_pe(phb, pe_num);
+               pdn->pe_number = IODA_INVALID_PE;
+               pe->pdev = NULL;
+               pci_dev_put(dev);
+               return NULL;
+       }
+
+       /* Assign a DMA weight to the device */
+       pe->dma_weight = pnv_ioda_dma_weight(dev);
+       if (pe->dma_weight != 0) {
+               phb->ioda.dma_weight += pe->dma_weight;
+               phb->ioda.dma_pe_count++;
+       }
+
+       /* Link the PE */
+       pnv_ioda_link_pe_by_weight(phb, pe);
+
+       return pe;
+}
+
+static void pnv_ioda_setup_same_PE(struct pci_bus *bus, struct pnv_ioda_pe *pe)
+{
+       struct pci_dev *dev;
+
+       list_for_each_entry(dev, &bus->devices, bus_list) {
+               struct pci_dn *pdn = pnv_ioda_get_pdn(dev);
+
+               if (pdn == NULL) {
+                       pr_warn("%s: No device node associated with device !\n",
+                               pci_name(dev));
+                       continue;
+               }
+               pci_dev_get(dev);
+               pdn->pcidev = dev;
+               pdn->pe_number = pe->pe_number;
+               pe->dma_weight += pnv_ioda_dma_weight(dev);
+               if (dev->subordinate)
+                       pnv_ioda_setup_same_PE(dev->subordinate, pe);
+       }
+}
+
+static void __devinit pnv_ioda_setup_bus_PE(struct pci_dev *dev,
+                                           struct pnv_ioda_pe *ppe)
+{
+       struct pci_controller *hose = pci_bus_to_host(dev->bus);
+       struct pnv_phb *phb = hose->private_data;
+       struct pci_bus *bus = dev->subordinate;
+       struct pnv_ioda_pe *pe;
+       int pe_num;
+
+       if (!bus) {
+               pr_warning("%s: Bridge without a subordinate bus !\n",
+                          pci_name(dev));
+               return;
+       }
+       pe_num = pnv_ioda_alloc_pe(phb);
+       if (pe_num == IODA_INVALID_PE) {
+               pr_warning("%s: Not enough PE# available, disabling bus\n",
+                          pci_name(dev));
+               return;
+       }
+
+       pe = &phb->ioda.pe_array[pe_num];
+       ppe->bus_pe = pe;
+       pe->pbus = bus;
+       pe->pdev = NULL;
+       pe->tce32_seg = -1;
+       pe->mve_number = -1;
+       pe->rid = bus->secondary << 8;
+       pe->dma_weight = 0;
+
+       pe_info(pe, "Secondary busses %d..%d associated with PE\n",
+               bus->secondary, bus->subordinate);
+
+       if (pnv_ioda_configure_pe(phb, pe)) {
+               /* XXX What do we do here ? */
+               if (pe_num)
+                       pnv_ioda_free_pe(phb, pe_num);
+               pe->pbus = NULL;
+               return;
+       }
+
+       /* Associate it with all child devices */
+       pnv_ioda_setup_same_PE(bus, pe);
+
+       /* Account for one DMA PE if at least one DMA capable device exist
+        * below the bridge
+        */
+       if (pe->dma_weight != 0) {
+               phb->ioda.dma_weight += pe->dma_weight;
+               phb->ioda.dma_pe_count++;
+       }
+
+       /* Link the PE */
+       pnv_ioda_link_pe_by_weight(phb, pe);
+}
+
+static void __devinit pnv_ioda_setup_PEs(struct pci_bus *bus)
+{
+       struct pci_dev *dev;
+       struct pnv_ioda_pe *pe;
+
+       list_for_each_entry(dev, &bus->devices, bus_list) {
+               pe = pnv_ioda_setup_dev_PE(dev);
+               if (pe == NULL)
+                       continue;
+               /* Leaving the PCIe domain ... single PE# */
+               if (dev->pcie_type == PCI_EXP_TYPE_PCI_BRIDGE)
+                       pnv_ioda_setup_bus_PE(dev, pe);
+               else if (dev->subordinate)
+                       pnv_ioda_setup_PEs(dev->subordinate);
+       }
+}
+
+static void __devinit pnv_pci_ioda_dma_dev_setup(struct pnv_phb *phb,
+                                                struct pci_dev *dev)
+{
+       /* We delay DMA setup after we have assigned all PE# */
+}
+
+static void __devinit pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe,
+                                            struct pci_bus *bus)
+{
+       struct pci_dev *dev;
+
+       list_for_each_entry(dev, &bus->devices, bus_list) {
+               set_iommu_table_base(&dev->dev, &pe->tce32_table);
+               if (dev->subordinate)
+                       pnv_ioda_setup_bus_dma(pe, dev->subordinate);
+       }
+}
+
+static void __devinit pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb,
+                                               struct pnv_ioda_pe *pe,
+                                               unsigned int base,
+                                               unsigned int segs)
+{
+
+       struct page *tce_mem = NULL;
+       const __be64 *swinvp;
+       struct iommu_table *tbl;
+       unsigned int i;
+       int64_t rc;
+       void *addr;
+
+       /* 256M DMA window, 4K TCE pages, 8 bytes TCE */
+#define TCE32_TABLE_SIZE       ((0x10000000 / 0x1000) * 8)
+
+       /* XXX FIXME: Handle 64-bit only DMA devices */
+       /* XXX FIXME: Provide 64-bit DMA facilities & non-4K TCE tables etc.. */
+       /* XXX FIXME: Allocate multi-level tables on PHB3 */
+
+       /* We shouldn't already have a 32-bit DMA associated */
+       if (WARN_ON(pe->tce32_seg >= 0))
+               return;
+
+       /* Grab a 32-bit TCE table */
+       pe->tce32_seg = base;
+       pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n",
+               (base << 28), ((base + segs) << 28) - 1);
+
+       /* XXX Currently, we allocate one big contiguous table for the
+        * TCEs. We only really need one chunk per 256M of TCE space
+        * (ie per segment) but that's an optimization for later, it
+        * requires some added smarts with our get/put_tce implementation
+        */
+       tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL,
+                                  get_order(TCE32_TABLE_SIZE * segs));
+       if (!tce_mem) {
+               pe_err(pe, " Failed to allocate a 32-bit TCE memory\n");
+               goto fail;
+       }
+       addr = page_address(tce_mem);
+       memset(addr, 0, TCE32_TABLE_SIZE * segs);
+
+       /* Configure HW */
+       for (i = 0; i < segs; i++) {
+               rc = opal_pci_map_pe_dma_window(phb->opal_id,
+                                             pe->pe_number,
+                                             base + i, 1,
+                                             __pa(addr) + TCE32_TABLE_SIZE * i,
+                                             TCE32_TABLE_SIZE, 0x1000);
+               if (rc) {
+                       pe_err(pe, " Failed to configure 32-bit TCE table,"
+                              " err %ld\n", rc);
+                       goto fail;
+               }
+       }
+
+       /* Setup linux iommu table */
+       tbl = &pe->tce32_table;
+       pnv_pci_setup_iommu_table(tbl, addr, TCE32_TABLE_SIZE * segs,
+                                 base << 28);
+
+       /* OPAL variant of P7IOC SW invalidated TCEs */
+       swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL);
+       if (swinvp) {
+               /* We need a couple more fields -- an address and a data
+                * to or.  Since the bus is only printed out on table free
+                * errors, and on the first pass the data will be a relative
+                * bus number, print that out instead.
+                */
+               tbl->it_busno = 0;
+               tbl->it_index = (unsigned long)ioremap(be64_to_cpup(swinvp), 8);
+               tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE
+                       | TCE_PCI_SWINV_PAIR;
+       }
+       iommu_init_table(tbl, phb->hose->node);
+
+       if (pe->pdev)
+               set_iommu_table_base(&pe->pdev->dev, tbl);
+       else
+               pnv_ioda_setup_bus_dma(pe, pe->pbus);
+
+       return;
+ fail:
+       /* XXX Failure: Try to fallback to 64-bit only ? */
+       if (pe->tce32_seg >= 0)
+               pe->tce32_seg = -1;
+       if (tce_mem)
+               __free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs));
+}
+
+static void __devinit pnv_ioda_setup_dma(struct pnv_phb *phb)
+{
+       struct pci_controller *hose = phb->hose;
+       unsigned int residual, remaining, segs, tw, base;
+       struct pnv_ioda_pe *pe;
+
+       /* If we have more PE# than segments available, hand out one
+        * per PE until we run out and let the rest fail. If not,
+        * then we assign at least one segment per PE, plus more based
+        * on the amount of devices under that PE
+        */
+       if (phb->ioda.dma_pe_count > phb->ioda.tce32_count)
+               residual = 0;
+       else
+               residual = phb->ioda.tce32_count -
+                       phb->ioda.dma_pe_count;
+
+       pr_info("PCI: Domain %04x has %ld available 32-bit DMA segments\n",
+               hose->global_number, phb->ioda.tce32_count);
+       pr_info("PCI: %d PE# for a total weight of %d\n",
+               phb->ioda.dma_pe_count, phb->ioda.dma_weight);
+
+       /* Walk our PE list and configure their DMA segments, hand them
+        * out one base segment plus any residual segments based on
+        * weight
+        */
+       remaining = phb->ioda.tce32_count;
+       tw = phb->ioda.dma_weight;
+       base = 0;
+       list_for_each_entry(pe, &phb->ioda.pe_list, link) {
+               if (!pe->dma_weight)
+                       continue;
+               if (!remaining) {
+                       pe_warn(pe, "No DMA32 resources available\n");
+                       continue;
+               }
+               segs = 1;
+               if (residual) {
+                       segs += ((pe->dma_weight * residual)  + (tw / 2)) / tw;
+                       if (segs > remaining)
+                               segs = remaining;
+               }
+               pe_info(pe, "DMA weight %d, assigned %d DMA32 segments\n",
+                       pe->dma_weight, segs);
+               pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs);
+               remaining -= segs;
+               base += segs;
+       }
+}
+
+#ifdef CONFIG_PCI_MSI
+static int pnv_pci_ioda_msi_setup(struct pnv_phb *phb, struct pci_dev *dev,
+                                 unsigned int hwirq, unsigned int is_64,
+                                 struct msi_msg *msg)
+{
+       struct pnv_ioda_pe *pe = pnv_ioda_get_pe(dev);
+       unsigned int xive_num = hwirq - phb->msi_base;
+       uint64_t addr64;
+       uint32_t addr32, data;
+       int rc;
+
+       /* No PE assigned ? bail out ... no MSI for you ! */
+       if (pe == NULL)
+               return -ENXIO;
+
+       /* Check if we have an MVE */
+       if (pe->mve_number < 0)
+               return -ENXIO;
+
+       /* Assign XIVE to PE */
+       rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num);
+       if (rc) {
+               pr_warn("%s: OPAL error %d setting XIVE %d PE\n",
+                       pci_name(dev), rc, xive_num);
+               return -EIO;
+       }
+
+       if (is_64) {
+               rc = opal_get_msi_64(phb->opal_id, pe->mve_number, xive_num, 1,
+                                    &addr64, &data);
+               if (rc) {
+                       pr_warn("%s: OPAL error %d getting 64-bit MSI data\n",
+                               pci_name(dev), rc);
+                       return -EIO;
+               }
+               msg->address_hi = addr64 >> 32;
+               msg->address_lo = addr64 & 0xfffffffful;
+       } else {
+               rc = opal_get_msi_32(phb->opal_id, pe->mve_number, xive_num, 1,
+                                    &addr32, &data);
+               if (rc) {
+                       pr_warn("%s: OPAL error %d getting 32-bit MSI data\n",
+                               pci_name(dev), rc);
+                       return -EIO;
+               }
+               msg->address_hi = 0;
+               msg->address_lo = addr32;
+       }
+       msg->data = data;
+
+       pr_devel("%s: %s-bit MSI on hwirq %x (xive #%d),"
+                " address=%x_%08x data=%x PE# %d\n",
+                pci_name(dev), is_64 ? "64" : "32", hwirq, xive_num,
+                msg->address_hi, msg->address_lo, data, pe->pe_number);
+
+       return 0;
+}
+
+static void pnv_pci_init_ioda_msis(struct pnv_phb *phb)
+{
+       unsigned int bmap_size;
+       const __be32 *prop = of_get_property(phb->hose->dn,
+                                            "ibm,opal-msi-ranges", NULL);
+       if (!prop) {
+               /* BML Fallback */
+               prop = of_get_property(phb->hose->dn, "msi-ranges", NULL);
+       }
+       if (!prop)
+               return;
+
+       phb->msi_base = be32_to_cpup(prop);
+       phb->msi_count = be32_to_cpup(prop + 1);
+       bmap_size = BITS_TO_LONGS(phb->msi_count) * sizeof(unsigned long);
+       phb->msi_map = zalloc_maybe_bootmem(bmap_size, GFP_KERNEL);
+       if (!phb->msi_map) {
+               pr_err("PCI %d: Failed to allocate MSI bitmap !\n",
+                      phb->hose->global_number);
+               return;
+       }
+       phb->msi_setup = pnv_pci_ioda_msi_setup;
+       phb->msi32_support = 1;
+       pr_info("  Allocated bitmap for %d MSIs (base IRQ 0x%x)\n",
+               phb->msi_count, phb->msi_base);
+}
+#else
+static void pnv_pci_init_ioda_msis(struct pnv_phb *phb) { }
+#endif /* CONFIG_PCI_MSI */
+
+/* This is the starting point of our IODA specific resource
+ * allocation process
+ */
+static void __devinit pnv_pci_ioda_fixup_phb(struct pci_controller *hose)
+{
+       resource_size_t size, align;
+       struct pci_bus *child;
+
+       /* Associate PEs per functions */
+       pnv_ioda_setup_PEs(hose->bus);
+
+       /* Calculate all resources */
+       pnv_ioda_calc_bus(hose->bus, IORESOURCE_IO, &size, &align);
+       pnv_ioda_calc_bus(hose->bus, IORESOURCE_MEM, &size, &align);
+
+       /* Apply then to HW */
+       pnv_ioda_update_resources(hose->bus);
+
+       /* Setup DMA */
+       pnv_ioda_setup_dma(hose->private_data);
+
+       /* Configure PCI Express settings */
+       list_for_each_entry(child, &hose->bus->children, node) {
+               struct pci_dev *self = child->self;
+               if (!self)
+                       continue;
+               pcie_bus_configure_settings(child, self->pcie_mpss);
+       }
+}
+
+/* Prevent enabling devices for which we couldn't properly
+ * assign a PE
+ */
+static int __devinit pnv_pci_enable_device_hook(struct pci_dev *dev)
+{
+       struct pci_dn *pdn = pnv_ioda_get_pdn(dev);
+
+       if (!pdn || pdn->pe_number == IODA_INVALID_PE)
+               return -EINVAL;
+       return 0;
+}
+
+static u32 pnv_ioda_bdfn_to_pe(struct pnv_phb *phb, struct pci_bus *bus,
+                              u32 devfn)
+{
+       return phb->ioda.pe_rmap[(bus->number << 8) | devfn];
+}
+
+void __init pnv_pci_init_ioda1_phb(struct device_node *np)
+{
+       struct pci_controller *hose;
+       static int primary = 1;
+       struct pnv_phb *phb;
+       unsigned long size, m32map_off, iomap_off, pemap_off;
+       const u64 *prop64;
+       u64 phb_id;
+       void *aux;
+       long rc;
+
+       pr_info(" Initializing IODA OPAL PHB %s\n", np->full_name);
+
+       prop64 = of_get_property(np, "ibm,opal-phbid", NULL);
+       if (!prop64) {
+               pr_err("  Missing \"ibm,opal-phbid\" property !\n");
+               return;
+       }
+       phb_id = be64_to_cpup(prop64);
+       pr_debug("  PHB-ID  : 0x%016llx\n", phb_id);
+
+       phb = alloc_bootmem(sizeof(struct pnv_phb));
+       if (phb) {
+               memset(phb, 0, sizeof(struct pnv_phb));
+               phb->hose = hose = pcibios_alloc_controller(np);
+       }
+       if (!phb || !phb->hose) {
+               pr_err("PCI: Failed to allocate PCI controller for %s\n",
+                      np->full_name);
+               return;
+       }
+
+       spin_lock_init(&phb->lock);
+       /* XXX Use device-tree */
+       hose->first_busno = 0;
+       hose->last_busno = 0xff;
+       hose->private_data = phb;
+       phb->opal_id = phb_id;
+       phb->type = PNV_PHB_IODA1;
+
+       /* We parse "ranges" now since we need to deduce the register base
+        * from the IO base
+        */
+       pci_process_bridge_OF_ranges(phb->hose, np, primary);
+       primary = 0;
+
+       /* Magic formula from Milton */
+       phb->regs = of_iomap(np, 0);
+       if (phb->regs == NULL)
+               pr_err("  Failed to map registers !\n");
+
+
+       /* XXX This is hack-a-thon. This needs to be changed so that:
+        *  - we obtain stuff like PE# etc... from device-tree
+        *  - we properly re-allocate M32 ourselves
+        *    (the OFW one isn't very good)
+        */
+
+       /* Initialize more IODA stuff */
+       phb->ioda.total_pe = 128;
+
+       phb->ioda.m32_size = resource_size(&hose->mem_resources[0]);
+       /* OFW Has already off top 64k of M32 space (MSI space) */
+       phb->ioda.m32_size += 0x10000;
+
+       phb->ioda.m32_segsize = phb->ioda.m32_size / phb->ioda.total_pe;
+       phb->ioda.m32_pci_base = hose->mem_resources[0].start -
+               hose->pci_mem_offset;
+       phb->ioda.io_size = hose->pci_io_size;
+       phb->ioda.io_segsize = phb->ioda.io_size / phb->ioda.total_pe;
+       phb->ioda.io_pci_base = 0; /* XXX calculate this ? */
+
+       /* Allocate aux data & arrays */
+       size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long));
+       m32map_off = size;
+       size += phb->ioda.total_pe;
+       iomap_off = size;
+       size += phb->ioda.total_pe;
+       pemap_off = size;
+       size += phb->ioda.total_pe * sizeof(struct pnv_ioda_pe);
+       aux = alloc_bootmem(size);
+       memset(aux, 0, size);
+       phb->ioda.pe_alloc = aux;
+       phb->ioda.m32_segmap = aux + m32map_off;
+       phb->ioda.io_segmap = aux + iomap_off;
+       phb->ioda.pe_array = aux + pemap_off;
+       set_bit(0, phb->ioda.pe_alloc);
+
+       INIT_LIST_HEAD(&phb->ioda.pe_list);
+
+       /* Calculate how many 32-bit TCE segments we have */
+       phb->ioda.tce32_count = phb->ioda.m32_pci_base >> 28;
+
+       /* Clear unusable m64 */
+       hose->mem_resources[1].flags = 0;
+       hose->mem_resources[1].start = 0;
+       hose->mem_resources[1].end = 0;
+       hose->mem_resources[2].flags = 0;
+       hose->mem_resources[2].start = 0;
+       hose->mem_resources[2].end = 0;
+
+#if 0
+       rc = opal_pci_set_phb_mem_window(opal->phb_id,
+                                        window_type,
+                                        window_num,
+                                        starting_real_address,
+                                        starting_pci_address,
+                                        segment_size);
+#endif
+
+       pr_info("  %d PE's M32: 0x%x [segment=0x%x] IO: 0x%x [segment=0x%x]\n",
+               phb->ioda.total_pe,
+               phb->ioda.m32_size, phb->ioda.m32_segsize,
+               phb->ioda.io_size, phb->ioda.io_segsize);
+
+       if (phb->regs)  {
+               pr_devel(" BUID     = 0x%016llx\n", in_be64(phb->regs + 0x100));
+               pr_devel(" PHB2_CR  = 0x%016llx\n", in_be64(phb->regs + 0x160));
+               pr_devel(" IO_BAR   = 0x%016llx\n", in_be64(phb->regs + 0x170));
+               pr_devel(" IO_BAMR  = 0x%016llx\n", in_be64(phb->regs + 0x178));
+               pr_devel(" IO_SAR   = 0x%016llx\n", in_be64(phb->regs + 0x180));
+               pr_devel(" M32_BAR  = 0x%016llx\n", in_be64(phb->regs + 0x190));
+               pr_devel(" M32_BAMR = 0x%016llx\n", in_be64(phb->regs + 0x198));
+               pr_devel(" M32_SAR  = 0x%016llx\n", in_be64(phb->regs + 0x1a0));
+       }
+       phb->hose->ops = &pnv_pci_ops;
+
+       /* Setup RID -> PE mapping function */
+       phb->bdfn_to_pe = pnv_ioda_bdfn_to_pe;
+
+       /* Setup TCEs */
+       phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup;
+
+       /* Setup MSI support */
+       pnv_pci_init_ioda_msis(phb);
+
+       /* We set both probe_only and PCI_REASSIGN_ALL_RSRC. This is an
+        * odd combination which essentially means that we skip all resource
+        * fixups and assignments in the generic code, and do it all
+        * ourselves here
+        */
+       pci_probe_only = 1;
+       ppc_md.pcibios_fixup_phb = pnv_pci_ioda_fixup_phb;
+       ppc_md.pcibios_enable_device_hook = pnv_pci_enable_device_hook;
+       pci_add_flags(PCI_REASSIGN_ALL_RSRC);
+
+       /* Reset IODA tables to a clean state */
+       rc = opal_pci_reset(phb_id, OPAL_PCI_IODA_RESET, OPAL_ASSERT_RESET);
+       if (rc)
+               pr_warning("  OPAL Error %ld performing IODA reset !\n", rc);
+       opal_pci_set_pe(phb_id, 0, 0, 7, 1, 1 , OPAL_MAP_PE);
+}
+
+void __init pnv_pci_init_ioda_hub(struct device_node *np)
+{
+       struct device_node *phbn;
+       const u64 *prop64;
+       u64 hub_id;
+
+       pr_info("Probing IODA IO-Hub %s\n", np->full_name);
+
+       prop64 = of_get_property(np, "ibm,opal-hubid", NULL);
+       if (!prop64) {
+               pr_err(" Missing \"ibm,opal-hubid\" property !\n");
+               return;
+       }
+       hub_id = be64_to_cpup(prop64);
+       pr_devel(" HUB-ID : 0x%016llx\n", hub_id);
+
+       /* Count child PHBs */
+       for_each_child_of_node(np, phbn) {
+               /* Look for IODA1 PHBs */
+               if (of_device_is_compatible(phbn, "ibm,ioda-phb"))
+                       pnv_pci_init_ioda1_phb(phbn);
+       }
+}
index baef772d41f4769ee2181f350fa280bfc30a89c6..c0ed379498a0e391bff0aaf5d64032de6b0fafe4 100644 (file)
@@ -467,12 +467,24 @@ void __init pnv_pci_init(void)
                init_pci_config_tokens();
                find_and_init_phbs();
 #endif /* CONFIG_PPC_POWERNV_RTAS */
-       } else {
-               /* OPAL is here, do our normal stuff */
+       }
+       /* OPAL is here, do our normal stuff */
+       else {
+               int found_ioda = 0;
+
+               /* Look for IODA IO-Hubs. We don't support mixing IODA
+                * and p5ioc2 due to the need to change some global
+                * probing flags
+                */
+               for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
+                       pnv_pci_init_ioda_hub(np);
+                       found_ioda = 1;
+               }
 
                /* Look for p5ioc2 IO-Hubs */
-               for_each_compatible_node(np, NULL, "ibm,p5ioc2")
-                       pnv_pci_init_p5ioc2_hub(np);
+               if (!found_ioda)
+                       for_each_compatible_node(np, NULL, "ibm,p5ioc2")
+                               pnv_pci_init_p5ioc2_hub(np);
        }
 
        /* Setup the linkage between OF nodes and PHBs */
index d4dbc495093659651646c65250e243a25dd2a8ee..28ae4ca512c4752db3ede399e37a581b98da0dc5 100644 (file)
@@ -9,6 +9,50 @@ enum pnv_phb_type {
        PNV_PHB_IODA2,
 };
 
+/* Data associated with a PE, including IOMMU tracking etc.. */
+struct pnv_ioda_pe {
+       /* A PE can be associated with a single device or an
+        * entire bus (& children). In the former case, pdev
+        * is populated, in the later case, pbus is.
+        */
+       struct pci_dev          *pdev;
+       struct pci_bus          *pbus;
+
+       /* Effective RID (device RID for a device PE and base bus
+        * RID with devfn 0 for a bus PE)
+        */
+       unsigned int            rid;
+
+       /* PE number */
+       unsigned int            pe_number;
+
+       /* "Weight" assigned to the PE for the sake of DMA resource
+        * allocations
+        */
+       unsigned int            dma_weight;
+
+       /* This is a PCI-E -> PCI-X bridge, this points to the
+        * corresponding bus PE
+        */
+       struct pnv_ioda_pe      *bus_pe;
+
+       /* "Base" iommu table, ie, 4K TCEs, 32-bit DMA */
+       int                     tce32_seg;
+       int                     tce32_segcount;
+       struct iommu_table      tce32_table;
+
+       /* XXX TODO: Add support for additional 64-bit iommus */
+
+       /* MSIs. MVE index is identical for for 32 and 64 bit MSI
+        * and -1 if not supported. (It's actually identical to the
+        * PE number)
+        */
+       int                     mve_number;
+
+       /* Link in list of PE#s */
+       struct list_head        link;
+};
+
 struct pnv_phb {
        struct pci_controller   *hose;
        enum pnv_phb_type       type;
@@ -34,6 +78,45 @@ struct pnv_phb {
                struct {
                        struct iommu_table iommu_table;
                } p5ioc2;
+
+               struct {
+                       /* Global bridge info */
+                       unsigned int            total_pe;
+                       unsigned int            m32_size;
+                       unsigned int            m32_segsize;
+                       unsigned int            m32_pci_base;
+                       unsigned int            io_size;
+                       unsigned int            io_segsize;
+                       unsigned int            io_pci_base;
+
+                       /* PE allocation bitmap */
+                       unsigned long           *pe_alloc;
+
+                       /* M32 & IO segment maps */
+                       unsigned int            *m32_segmap;
+                       unsigned int            *io_segmap;
+                       struct pnv_ioda_pe      *pe_array;
+
+                       /* Reverse map of PEs, will have to extend if
+                        * we are to support more than 256 PEs, indexed
+                        * bus { bus, devfn }
+                        */
+                       unsigned char           pe_rmap[0x10000];
+
+                       /* 32-bit TCE tables allocation */
+                       unsigned long           tce32_count;
+
+                       /* Total "weight" for the sake of DMA resources
+                        * allocation
+                        */
+                       unsigned int            dma_weight;
+                       unsigned int            dma_pe_count;
+
+                       /* Sorted list of used PE's, sorted at
+                        * boot for resource allocation purposes
+                        */
+                       struct list_head        pe_list;
+               } ioda;
        };
 };
 
@@ -43,6 +126,7 @@ extern void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
                                      void *tce_mem, u64 tce_size,
                                      u64 dma_offset);
 extern void pnv_pci_init_p5ioc2_hub(struct device_node *np);
+extern void pnv_pci_init_ioda_hub(struct device_node *np);
 
 
 #endif /* __POWERNV_PCI_H */