[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] [PATCH 11/88] spapr_pci: map the MSI window in each PHB
From: |
Michael Roth |
Subject: |
[Qemu-devel] [PATCH 11/88] spapr_pci: map the MSI window in each PHB |
Date: |
Thu, 8 Jan 2015 11:33:15 -0600 |
From: Greg Kurz <address@hidden>
On sPAPR, virtio devices are connected to the PCI bus and use MSI-X.
Commit cc943c36faa192cd4b32af8fe5edb31894017d35 has modified MSI-X
so that writes are made using the bus master address space and follow
the IOMMU path.
Unfortunately, the IOMMU address space address space does not have an
MSI window: the notification is silently dropped in unassigned_mem_write
instead of reaching the guest... The most visible effect is that all
virtio devices are non-functional on sPAPR since then. :(
This patch does the following:
1) map the MSI window into the IOMMU address space for each PHB
- since each PHB instantiates its own IOMMU address space, we
can safely map the window at a fixed address (SPAPR_PCI_MSI_WINDOW)
- no real need to keep the MSI window setup in a separate function,
the spapr_pci_msi_init() code moves to spapr_phb_realize().
2) kill the global MSI window as it is not needed in the end
Signed-off-by: Greg Kurz <address@hidden>
Signed-off-by: Alexander Graf <address@hidden>
(cherry picked from commit 8c46f7ec85a4dd9663489b2fa2b425cd7b3653e1)
Signed-off-by: Michael Roth <address@hidden>
---
hw/ppc/spapr.c | 1 -
hw/ppc/spapr_pci.c | 53 ++++++++++++++++++++-------------------------
include/hw/pci-host/spapr.h | 2 +-
include/hw/ppc/spapr.h | 2 --
4 files changed, 25 insertions(+), 33 deletions(-)
diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index d01978f..4196a70 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -1377,7 +1377,6 @@ static void ppc_spapr_init(MachineState *machine)
spapr_create_nvram(spapr);
/* Set up PCI */
- spapr_pci_msi_init(spapr, SPAPR_PCI_MSI_WINDOW);
spapr_pci_rtas_init();
phb = spapr_create_phb(spapr, 0);
diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c
index 9ed39a9..dadba5f 100644
--- a/hw/ppc/spapr_pci.c
+++ b/hw/ppc/spapr_pci.c
@@ -341,7 +341,7 @@ static void rtas_ibm_change_msi(PowerPCCPU *cpu,
sPAPREnvironment *spapr,
}
/* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
- spapr_msi_setmsg(pdev, spapr->msi_win_addr, ret_intr_type ==
RTAS_TYPE_MSIX,
+ spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type ==
RTAS_TYPE_MSIX,
irq, req_num);
/* Add MSI device to cache */
@@ -465,34 +465,6 @@ static const MemoryRegionOps spapr_msi_ops = {
.endianness = DEVICE_LITTLE_ENDIAN
};
-void spapr_pci_msi_init(sPAPREnvironment *spapr, hwaddr addr)
-{
- uint64_t window_size = 4096;
-
- /*
- * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
- * we need to allocate some memory to catch those writes coming
- * from msi_notify()/msix_notify().
- * As MSIMessage:addr is going to be the same and MSIMessage:data
- * is going to be a VIRQ number, 4 bytes of the MSI MR will only
- * be used.
- *
- * For KVM we want to ensure that this memory is a full page so that
- * our memory slot is of page size granularity.
- */
-#ifdef CONFIG_KVM
- if (kvm_enabled()) {
- window_size = getpagesize();
- }
-#endif
-
- spapr->msi_win_addr = addr;
- memory_region_init_io(&spapr->msiwindow, NULL, &spapr_msi_ops, spapr,
- "msi", window_size);
- memory_region_add_subregion(get_system_memory(), spapr->msi_win_addr,
- &spapr->msiwindow);
-}
-
/*
* PHB PCI device
*/
@@ -512,6 +484,7 @@ static void spapr_phb_realize(DeviceState *dev, Error
**errp)
char *namebuf;
int i;
PCIBus *bus;
+ uint64_t msi_window_size = 4096;
if (sphb->index != -1) {
hwaddr windows_base;
@@ -604,6 +577,28 @@ static void spapr_phb_realize(DeviceState *dev, Error
**errp)
address_space_init(&sphb->iommu_as, &sphb->iommu_root,
sphb->dtbusname);
+ /*
+ * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
+ * we need to allocate some memory to catch those writes coming
+ * from msi_notify()/msix_notify().
+ * As MSIMessage:addr is going to be the same and MSIMessage:data
+ * is going to be a VIRQ number, 4 bytes of the MSI MR will only
+ * be used.
+ *
+ * For KVM we want to ensure that this memory is a full page so that
+ * our memory slot is of page size granularity.
+ */
+#ifdef CONFIG_KVM
+ if (kvm_enabled()) {
+ msi_window_size = getpagesize();
+ }
+#endif
+
+ memory_region_init_io(&sphb->msiwindow, NULL, &spapr_msi_ops, spapr,
+ "msi", msi_window_size);
+ memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
+ &sphb->msiwindow);
+
pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
diff --git a/include/hw/pci-host/spapr.h b/include/hw/pci-host/spapr.h
index 32f0aa7..4ea2a0d 100644
--- a/include/hw/pci-host/spapr.h
+++ b/include/hw/pci-host/spapr.h
@@ -70,7 +70,7 @@ struct sPAPRPHBState {
MemoryRegion memspace, iospace;
hwaddr mem_win_addr, mem_win_size, io_win_addr, io_win_size;
- MemoryRegion memwindow, iowindow;
+ MemoryRegion memwindow, iowindow, msiwindow;
uint32_t dma_liobn;
AddressSpace iommu_as;
diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
index bbba51a..832ad6b 100644
--- a/include/hw/ppc/spapr.h
+++ b/include/hw/ppc/spapr.h
@@ -13,8 +13,6 @@ struct sPAPRNVRAM;
typedef struct sPAPREnvironment {
struct VIOsPAPRBus *vio_bus;
QLIST_HEAD(, sPAPRPHBState) phbs;
- hwaddr msi_win_addr;
- MemoryRegion msiwindow;
struct sPAPRNVRAM *nvram;
XICSState *icp;
--
1.9.1
- [Qemu-devel] [PATCH 03/88] exec: file_ram_alloc(): print error when prealloc fails, (continued)
- [Qemu-devel] [PATCH 03/88] exec: file_ram_alloc(): print error when prealloc fails, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 10/88] virtio-pci: enable bus master for old guests, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 05/88] Introduce cpu_clean_all_dirty, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 08/88] kvmclock: Add comment explaining why we need cpu_clean_all_dirty(), Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 09/88] pci: Use bus master address space for delivering MSI/MSI-X messages, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 07/88] kvmclock: Ensure time in migration never goes backward, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 06/88] kvmclock: Ensure proper env->tsc value for kvmclock_current_nsec calculation, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 12/88] hw/arm/virt: fix pl011 and pl031 irq flags, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 13/88] gdbstub: init mon_chr through qemu_chr_alloc, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 14/88] qapi: add visit_start_union and visit_end_union, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 11/88] spapr_pci: map the MSI window in each PHB,
Michael Roth <=
- [Qemu-devel] [PATCH 15/88] qapi: dealloc visitor, implement visit_start_union, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 16/88] tests: add QMP input visitor test for unions with no discriminator, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 17/88] qemu-iotests: Test missing "driver" key for blockdev-add, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 18/88] monitor: Reset HMP mon->rs in CHR_EVENT_OPEN, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 20/88] vhost-user: fix VIRTIO_NET_F_MRG_RXBUF negotiation, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 22/88] ivshmem: validate incoming_posn value from server, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 24/88] ivshmem: Fix fd leak on error, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 25/88] pc-dimm: Don't check dimm->node when there is non-NUMA config, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 26/88] tests: avoid running duplicate qom-tests, Michael Roth, 2015/01/08
- [Qemu-devel] [PATCH 33/88] virtio/vhost-scsi: fix virtio-scsi/vhost-scsi child refcount in transports, Michael Roth, 2015/01/08