Re: [Qemu-devel] [Qemu-ppc] [PATCH qemu v13 07/16] vfio, memory: Notify IOMMU about starting/stopping being used by VFIO

[David Gibson <david@gibson.dropbear.id.au>]

[-- Attachment #1: Type: text/plain, Size: 8992 bytes --]

On Thu, Mar 03, 2016 at 05:01:31PM +1100, Alexey Kardashevskiy wrote:
> On 03/03/2016 04:28 PM, David Gibson wrote:
> >On Tue, Mar 01, 2016 at 08:10:32PM +1100, Alexey Kardashevskiy wrote:
> >>This adds a vfio_votify() callback to inform an IOMMU (and then its owner)
> >>that VFIO started using the IOMMU. This is used by the pseries machine to
> >>enable/disable in-kernel acceleration of TCE hypercalls.
> >>
> >>Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> >
> >Hmm.. the current approach of having a hook when vfio-pci devices are
> >attached is pretty ugly, but what exactly the case that it doesn't
> >handle and this approach does?
> 
> Sorry, I am not following you here. What hook do you mean here?
> 
> My hook fixes the case when I want to enable/disable KVM acceleration,
> without these patches, I need to re-count how many vfio-pci devices are
> there and it is more ugly with PCI hotplug/unplug...
> 
> 
> >This two tiered notify system for a single bit is also kinda ugly.
> >
> >>---
> >>  hw/ppc/spapr_iommu.c   |  9 +++++++++
> >>  hw/ppc/spapr_pci.c     | 14 ++++++++------
> >>  hw/vfio/common.c       |  7 +++++++
> >>  include/exec/memory.h  |  2 ++
> >>  include/hw/ppc/spapr.h |  4 ++++
> >>  5 files changed, 30 insertions(+), 6 deletions(-)
> >>
> >>diff --git a/hw/ppc/spapr_iommu.c b/hw/ppc/spapr_iommu.c
> >>index 8a88a74..67a8356 100644
> >>--- a/hw/ppc/spapr_iommu.c
> >>+++ b/hw/ppc/spapr_iommu.c
> >>@@ -136,6 +136,13 @@ static IOMMUTLBEntry spapr_tce_translate_iommu(MemoryRegion *iommu, hwaddr addr,
> >>      return ret;
> >>  }
> >>
> >>+static int spapr_tce_vfio_notify(MemoryRegion *iommu, bool attached)
> >>+{
> >>+    sPAPRTCETable *tcet = container_of(iommu, sPAPRTCETable, iommu);
> >>+
> >>+    return spapr_tce_vfio_notify_owner(tcet->owner, tcet, attached);
> >
> >I'm guessing the "owner" is the PHB, but I'm not entirely clear.
> >
> >Could you use the QOM parent to get the the PHB instead of storing it
> >explicitly?
> 
> 
> I am pretty sure I am not allowed to use the QOM parent, this is why there
> is no object_get_parent() helper.

Hmm.. I thought I had this discussion before and accessing the qom
parent from qmp was bad, but it was ok for internal code use.  But I
may be getting muddled with older qdev stuff.

> 
> >
> >>+}
> >>+
> >>  static int spapr_tce_table_post_load(void *opaque, int version_id)
> >>  {
> >>      sPAPRTCETable *tcet = SPAPR_TCE_TABLE(opaque);
> >>@@ -167,6 +174,7 @@ static const VMStateDescription vmstate_spapr_tce_table = {
> >>
> >>  static MemoryRegionIOMMUOps spapr_iommu_ops = {
> >>      .translate = spapr_tce_translate_iommu,
> >>+    .vfio_notify = spapr_tce_vfio_notify,
> >>  };
> >>
> >>  static int spapr_tce_table_realize(DeviceState *dev)
> >>@@ -235,6 +243,7 @@ sPAPRTCETable *spapr_tce_new_table(DeviceState *owner, uint32_t liobn)
> >>
> >>      tcet = SPAPR_TCE_TABLE(object_new(TYPE_SPAPR_TCE_TABLE));
> >>      tcet->liobn = liobn;
> >>+    tcet->owner = owner;
> >>
> >>      snprintf(tmp, sizeof(tmp), "tce-table-%x", liobn);
> >>      object_property_add_child(OBJECT(owner), tmp, OBJECT(tcet), NULL);
> >>diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c
> >>index ee0fecf..b0cd148 100644
> >>--- a/hw/ppc/spapr_pci.c
> >>+++ b/hw/ppc/spapr_pci.c
> >>@@ -1084,6 +1084,14 @@ static int spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
> >>      return 0;
> >>  }
> >>
> >>+int spapr_tce_vfio_notify_owner(DeviceState *dev, sPAPRTCETable *tcet,
> >>+                                bool attached)
> >>+{
> >>+    spapr_tce_set_need_vfio(tcet, attached);
> >
> >Hmm.. you go to the trouble of storing owner in dev, then don't
> >actually use it.
> 
> 
> Yeah, I need to clean this, I removed spapr_tce_vfio_notify_owner() from my
> working branch already and call spapr_tce_set_need_vfio() directly from
> spapr_tce_vfio_notify().

Ok.

> 
> 
> >
> >>+    return 0;
> >>+}
> >>+
> >>  /* create OF node for pci device and required OF DT properties */
> >>  static int spapr_create_pci_child_dt(sPAPRPHBState *phb, PCIDevice *dev,
> >>                                       void *fdt, int node_offset)
> >>@@ -1118,12 +1126,6 @@ static void spapr_phb_add_pci_device(sPAPRDRConnector *drc,
> >>      void *fdt = NULL;
> >>      int fdt_start_offset = 0, fdt_size;
> >>
> >>-    if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
> >>-        sPAPRTCETable *tcet = spapr_tce_find_by_liobn(phb->dma_liobn);
> >>-
> >>-        spapr_tce_set_need_vfio(tcet, true);
> >>-    }
> >>-
> >>      if (dev->hotplugged) {
> >>          fdt = create_device_tree(&fdt_size);
> >>          fdt_start_offset = spapr_create_pci_child_dt(phb, pdev, fdt, 0);
> >>diff --git a/hw/vfio/common.c b/hw/vfio/common.c
> >>index 9bf4c3b..ca3fd47 100644
> >>--- a/hw/vfio/common.c
> >>+++ b/hw/vfio/common.c
> >>@@ -384,6 +384,7 @@ static void vfio_listener_region_add(MemoryListener *listener,
> >>          QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
> >>
> >>          memory_region_register_iommu_notifier(giommu->iommu, &giommu->n);
> >>+        giommu->iommu->iommu_ops->vfio_notify(section->mr, true);
> >>          memory_region_iommu_replay(giommu->iommu, &giommu->n,
> >>                                     vfio_container_granularity(container),
> >>                                     false);
> >>@@ -430,6 +431,7 @@ static void vfio_listener_region_del(MemoryListener *listener,
> >>      VFIOContainer *container = container_of(listener, VFIOContainer, listener);
> >>      hwaddr iova, end;
> >>      int ret;
> >>+    MemoryRegion *iommu = NULL;
> >>
> >>      if (vfio_listener_skipped_section(section)) {
> >>          trace_vfio_listener_region_del_skip(
> >>@@ -451,6 +453,7 @@ static void vfio_listener_region_del(MemoryListener *listener,
> >>          QLIST_FOREACH(giommu, &container->giommu_list, giommu_next) {
> >>              if (giommu->iommu == section->mr) {
> >>                  memory_region_unregister_iommu_notifier(&giommu->n);
> >>+                iommu = giommu->iommu;
> >>                  QLIST_REMOVE(giommu, giommu_next);
> >>                  g_free(giommu);
> >>                  break;
> >>@@ -483,6 +486,10 @@ static void vfio_listener_region_del(MemoryListener *listener,
> >>                       "0x%"HWADDR_PRIx") = %d (%m)",
> >>                       container, iova, end - iova, ret);
> >>      }
> >>+
> >>+    if (iommu && iommu->iommu_ops && iommu->iommu_ops->vfio_notify) {
> >>+        iommu->iommu_ops->vfio_notify(section->mr, false);
> >>+    }
> >
> >So, if an IOMMU is removed from the guest, this will turn off VFIO
> >enablement.  However, IIUC this won't get caled in the more likely
> >case that the address space stays the same, but the VFIO device is
> >removed.
> 
> 
> When VFIO device is removed, its listener gets removed and this is supposed
> to end up calling vfio_listener_region_del().

Ah, ok region_del() gets called on all existing regions when the
listener is removed.  I guess that makes sense since region_add() is
called on the existing regions when the listener is registered.



> 
> 
> >
> >>  }
> >>
> >>  static const MemoryListener vfio_memory_listener = {
> >>diff --git a/include/exec/memory.h b/include/exec/memory.h
> >>index d5284c2..9f82629 100644
> >>--- a/include/exec/memory.h
> >>+++ b/include/exec/memory.h
> >>@@ -150,6 +150,8 @@ typedef struct MemoryRegionIOMMUOps MemoryRegionIOMMUOps;
> >>  struct MemoryRegionIOMMUOps {
> >>      /* Return a TLB entry that contains a given address. */
> >>      IOMMUTLBEntry (*translate)(MemoryRegion *iommu, hwaddr addr, bool is_write);
> >>+    /* Called when VFIO starts/stops using this */
> >>+    int (*vfio_notify)(MemoryRegion *iommu, bool attached);
> >>  };
> >>
> >>  typedef struct CoalescedMemoryRange CoalescedMemoryRange;
> >>diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
> >>index 8aa0c45..5d2f8f4 100644
> >>--- a/include/hw/ppc/spapr.h
> >>+++ b/include/hw/ppc/spapr.h
> >>@@ -550,6 +550,7 @@ struct sPAPRTCETable {
> >>      int fd;
> >>      MemoryRegion root, iommu;
> >>      struct VIOsPAPRDevice *vdev; /* for @bypass migration compatibility only */
> >>+    DeviceState *owner;
> >>      QLIST_ENTRY(sPAPRTCETable) list;
> >>  };
> >>
> >>@@ -629,4 +630,7 @@ int spapr_rng_populate_dt(void *fdt);
> >>   */
> >>  #define SPAPR_LMB_FLAGS_ASSIGNED 0x00000008
> >>
> >>+int spapr_tce_vfio_notify_owner(DeviceState *dev, sPAPRTCETable *tcet,
> >>+                                bool attached);
> >>+
> >>  #endif /* !defined (__HW_SPAPR_H__) */
> >
> 
> 

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]