[Qemu-devel] [PATCH qemu] spapr_pci: Alias MMIO windows to PHB address space

[Alexey Kardashevskiy]

At the moment the sPAPR PHB MMIO space does not have an address space
object as it does not really need one - guest accesses it via virtual
addresses (and we provide mappings to the CPU space), the device drivers
in QEMU access MMIO directly (as they own MRs). The only case when
a driver in QEMU might need an address space (AS) is DMA so we provide one
with only IOMMUs on it.

So PHB AS looks like:
address-space: address@hidden
  0000000000000000-ffffffffffffffff (prio 0, i/o): address@hidden
    0000000000000000-ffffffffffffffff (prio 0, i/o): tce-root-80000001
      0800000000000000-080000007fffffff (prio 0, i/o): tce-iommu-80000001
    0000000000000000-ffffffffffffffff (prio 0, i/o): tce-root-80000000
      0000000000000000-000000003fffffff (prio 0, i/o): tce-iommu-80000000
    0000040000000000-000004000000ffff (prio 0, i/o): msi

with no mention of any PCI device, and so does the flatview:

 FlatView #2
  AS "address@hidden", root: address@hidden
  AS "nec-usb-xhci", root: bus master container
  Root memory region: address@hidden
   0000000000000000-000000003fffffff (prio 0, i/o): tce-iommu-80000000

This adds 2 more aliases (for MMIO32 and MMIO64 windows) on PHB AS so
whatever appears on the bus can be seen in "info mtree [-f]".

Here are the chunks from "diff" for a QEMU with an emulater XHCI adapter:

 address-space: address@hidden
   0000000000000000-ffffffffffffffff (prio 0, i/o): address@hidden
     0000000000000000-ffffffffffffffff (prio 0, i/o): tce-root-80000001
       0800000000000000-080000007fffffff (prio 0, i/o): tce-iommu-80000001
     0000000000000000-ffffffffffffffff (prio 0, i/o): tce-root-80000000
       0000000000000000-000000003fffffff (prio 0, i/o): tce-iommu-80000000
+    0000000080000000-00000000ffffffff (prio 0, i/o): alias address@hidden 
@address@hidden
+2-alias 0000000000000000-000000007fffffff
     0000040000000000-000004000000ffff (prio 0, i/o): msi
+    0000210000000000-000021ffffffffff (prio 0, i/o): alias address@hidden 
@address@hidden
+4-alias 0000000000000000-000000ffffffffff


 FlatView #2
  AS "address@hidden", root: address@hidden
  AS "nec-usb-xhci", root: bus master container
  Root memory region: address@hidden
   0000000000000000-000000003fffffff (prio 0, i/o): tce-iommu-80000000
+  0000000080000000-000000008000003f (prio 0, i/o): capabilities
+  0000000080000040-000000008000043f (prio 0, i/o): operational
+  0000000080000440-000000008000044f (prio 0, i/o): usb3 port #1
+  0000000080000450-000000008000045f (prio 0, i/o): usb3 port #2
+  0000000080000460-000000008000046f (prio 0, i/o): usb3 port #3
+  0000000080000470-000000008000047f (prio 0, i/o): usb3 port #4
+  0000000080000480-000000008000048f (prio 0, i/o): usb2 port #1
+  0000000080000490-000000008000049f (prio 0, i/o): usb2 port #2
+  00000000800004a0-00000000800004af (prio 0, i/o): usb2 port #3
+  00000000800004b0-00000000800004bf (prio 0, i/o): usb2 port #4
+  0000000080001000-000000008000121f (prio 0, i/o): runtime
+  0000000080002000-000000008000281f (prio 0, i/o): doorbell
+  0000000080003000-00000000800030ff (prio 0, i/o): msix-table
+  0000000080003800-0000000080003807 (prio 0, i/o): msix-pba
   0000040000000000-000004000000ffff (prio 0, i/o): msi
   0800000000000000-080000007fffffff (prio 0, i/o): tce-iommu-80000001

The major point for this is to help debugging overlapped regions.

Signed-off-by: Alexey Kardashevskiy <address@hidden>
---

Should not we do this in the common code really?
Does this look useful at all? I was quite confused when I realized that
I do not see an AS with DMA windows and MMIO while they co-exist
on a bus.


---
 include/hw/pci-host/spapr.h |  1 +
 hw/ppc/spapr_pci.c          | 12 ++++++++++++
 2 files changed, 13 insertions(+)

diff --git a/include/hw/pci-host/spapr.h b/include/hw/pci-host/spapr.h
index 0fae4fc..4550192 100644
--- a/include/hw/pci-host/spapr.h
+++ b/include/hw/pci-host/spapr.h
@@ -57,6 +57,7 @@ struct sPAPRPHBState {
     uint64_t mem64_win_pciaddr;
     hwaddr io_win_addr, io_win_size;
     MemoryRegion mem32window, mem64window, iowindow, msiwindow;
+    MemoryRegion mem32windowpci, mem64windowpci;
 
     uint32_t dma_liobn[SPAPR_PCI_DMA_MAX_WINDOWS];
     hwaddr dma_win_addr, dma_win_size;
diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c
index 37f18b3..d142f74 100644
--- a/hw/ppc/spapr_pci.c
+++ b/hw/ppc/spapr_pci.c
@@ -1642,6 +1642,18 @@ static void spapr_phb_realize(DeviceState *dev, Error 
**errp)
     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
                        sphb->dtbusname);
 
+    /* Map memory space onto the PHB address space */
+    namebuf = g_strdup_printf("%s.mmio32-alias-pci", sphb->dtbusname);
+    memory_region_init_alias(&sphb->mem32windowpci, OBJECT(sphb), namebuf,
+                             &sphb->mem32window, 0, sphb->mem_win_size);
+    memory_region_add_subregion(&sphb->iommu_root, 
SPAPR_PCI_MEM_WIN_BUS_OFFSET,
+                                &sphb->mem32windowpci);
+    namebuf = g_strdup_printf("%s.mmio64-alias-pci", sphb->dtbusname);
+    memory_region_init_alias(&sphb->mem64windowpci, OBJECT(sphb), namebuf,
+                             &sphb->mem64window, 0, sphb->mem64_win_size);
+    memory_region_add_subregion(&sphb->iommu_root, sphb->mem64_win_pciaddr,
+                                &sphb->mem64windowpci);
+
     /*
      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
      * we need to allocate some memory to catch those writes coming
-- 
2.11.0