memory: emulate ioeventfd

The ioeventfd mechanism is used by vhost, dataplane, and virtio-pci to
turn guest MMIO/PIO writes into eventfd file descriptor events.  This
allows arbitrary threads to be notified when the guest writes to a
specific MMIO/PIO address.

qtest and TCG do not support ioeventfd because memory writes are not
checked against registered ioeventfds in QEMU.  This patch implements
this in memory_region_dispatch_write() so qtest can use ioeventfd.

Also this patch fixes vhost aborting on some misconfigured old kernels
like 3.18.0 on ARM. It is possible to explicitly enable CONFIG_EVENTFD
in expert settings, while MMIO binding support in KVM will still be
missing.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Pavel Fedin <p.fedin@samsung.com>
---
RFC => PATCH:
- Add !kvm_eventfds_enabled() conditions to bypass eventfd injection when not needed
- Renamed "ioeventfd" to "eventfd", just to make words shorter
- Add a one-shot warning about missing MMIO bindings in KVM
---
 kvm-all.c |  6 ++++--
 memory.c  | 42 ++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+), 2 deletions(-)

On Fri, Nov 20, 2015 at 12:37:16PM +0300, Pavel Fedin wrote:
> The ioeventfd mechanism is used by vhost, dataplane, and virtio-pci to
> turn guest MMIO/PIO writes into eventfd file descriptor events.  This
> allows arbitrary threads to be notified when the guest writes to a
> specific MMIO/PIO address.
> 
> qtest and TCG do not support ioeventfd because memory writes are not
> checked against registered ioeventfds in QEMU.  This patch implements
> this in memory_region_dispatch_write() so qtest can use ioeventfd.
> 
> Also this patch fixes vhost aborting on some misconfigured old kernels
> like 3.18.0 on ARM. It is possible to explicitly enable CONFIG_EVENTFD
> in expert settings, while MMIO binding support in KVM will still be
> missing.
> 
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> Signed-off-by: Pavel Fedin <p.fedin@samsung.com>

Reviewed-by: Michael S. Tsirkin <mst@redhat.com>

> ---
> RFC => PATCH:
> - Add !kvm_eventfds_enabled() conditions to bypass eventfd injection when not needed
> - Renamed "ioeventfd" to "eventfd", just to make words shorter
> - Add a one-shot warning about missing MMIO bindings in KVM
> ---
>  kvm-all.c |  6 ++++--
>  memory.c  | 42 ++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 46 insertions(+), 2 deletions(-)
> 
> diff --git a/kvm-all.c b/kvm-all.c
> index ddb007a..70f5cec 100644
> --- a/kvm-all.c
> +++ b/kvm-all.c
> @@ -1633,8 +1633,10 @@ static int kvm_init(MachineState *ms)
>  
>      kvm_state = s;
>  
> -    s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add;
> -    s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del;
> +    if (kvm_eventfds_allowed) {
> +        s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add;
> +        s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del;
> +    }
>      s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region;
>      s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region;
>  
> diff --git a/memory.c b/memory.c
> index e193658..4d138fb 100644
> --- a/memory.c
> +++ b/memory.c
> @@ -18,12 +18,14 @@
>  #include "exec/ioport.h"
>  #include "qapi/visitor.h"
>  #include "qemu/bitops.h"
> +#include "qemu/error-report.h"
>  #include "qom/object.h"
>  #include "trace.h"
>  #include <assert.h>
>  
>  #include "exec/memory-internal.h"
>  #include "exec/ram_addr.h"
> +#include "sysemu/kvm.h"
>  #include "sysemu/sysemu.h"
>  
>  //#define DEBUG_UNASSIGNED
> @@ -1141,6 +1143,32 @@ MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
>      return r;
>  }
>  
> +/* Return true if an eventfd was signalled */
> +static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
> +                                                    hwaddr addr,
> +                                                    uint64_t data,
> +                                                    unsigned size,
> +                                                    MemTxAttrs attrs)
> +{
> +    MemoryRegionIoeventfd ioeventfd = {
> +        .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
> +        .data = data,
> +    };
> +    unsigned i;
> +
> +    for (i = 0; i < mr->ioeventfd_nb; i++) {
> +        ioeventfd.match_data = mr->ioeventfds[i].match_data;
> +        ioeventfd.e = mr->ioeventfds[i].e;
> +
> +        if (memory_region_ioeventfd_equal(ioeventfd, mr->ioeventfds[i])) {
> +            event_notifier_set(ioeventfd.e);
> +            return true;
> +        }
> +    }
> +
> +    return false;
> +}
> +
>  MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
>                                           hwaddr addr,
>                                           uint64_t data,
> @@ -1154,6 +1182,11 @@ MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
>  
>      adjust_endianness(mr, &data, size);
>  
> +    if ((!kvm_eventfds_enabled()) &&
> +        memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
> +        return MEMTX_OK;
> +    }
> +
>      if (mr->ops->write) {
>          return access_with_adjusted_size(addr, &data, size,
>                                           mr->ops->impl.min_access_size,
> @@ -1672,6 +1705,8 @@ void memory_region_clear_global_locking(MemoryRegion *mr)
>      mr->global_locking = false;
>  }
>  
> +static bool userspace_eventfd_warning;
> +
>  void memory_region_add_eventfd(MemoryRegion *mr,
>                                 hwaddr addr,
>                                 unsigned size,
> @@ -1688,6 +1723,13 @@ void memory_region_add_eventfd(MemoryRegion *mr,
>      };
>      unsigned i;
>  
> +    if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
> +                            userspace_eventfd_warning))) {
> +        userspace_eventfd_warning = true;
> +        error_report("Using eventfd without MMIO binding in KVM. "
> +                     "Suboptimal performance expected");
> +    }
> +
>      if (size) {
>          adjust_endianness(mr, &mrfd.data, size);
>      }
> -- 
> 1.9.5.msysgit.0
>

On 20/11/2015 10:37, Pavel Fedin wrote:
> The ioeventfd mechanism is used by vhost, dataplane, and virtio-pci to
> turn guest MMIO/PIO writes into eventfd file descriptor events.  This
> allows arbitrary threads to be notified when the guest writes to a
> specific MMIO/PIO address.
> 
> qtest and TCG do not support ioeventfd because memory writes are not
> checked against registered ioeventfds in QEMU.  This patch implements
> this in memory_region_dispatch_write() so qtest can use ioeventfd.
> 
> Also this patch fixes vhost aborting on some misconfigured old kernels
> like 3.18.0 on ARM. It is possible to explicitly enable CONFIG_EVENTFD
> in expert settings, while MMIO binding support in KVM will still be
> missing.
> 
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> Signed-off-by: Pavel Fedin <p.fedin@samsung.com>
> ---
> RFC => PATCH:
> - Add !kvm_eventfds_enabled() conditions to bypass eventfd injection when not needed
> - Renamed "ioeventfd" to "eventfd", just to make words shorter
> - Add a one-shot warning about missing MMIO bindings in KVM
> ---
>  kvm-all.c |  6 ++++--
>  memory.c  | 42 ++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 46 insertions(+), 2 deletions(-)
> 
> diff --git a/kvm-all.c b/kvm-all.c
> index ddb007a..70f5cec 100644
> --- a/kvm-all.c
> +++ b/kvm-all.c
> @@ -1633,8 +1633,10 @@ static int kvm_init(MachineState *ms)
>  
>      kvm_state = s;
>  
> -    s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add;
> -    s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del;
> +    if (kvm_eventfds_allowed) {
> +        s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add;
> +        s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del;
> +    }
>      s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region;
>      s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region;
>  
> diff --git a/memory.c b/memory.c
> index e193658..4d138fb 100644
> --- a/memory.c
> +++ b/memory.c
> @@ -18,12 +18,14 @@
>  #include "exec/ioport.h"
>  #include "qapi/visitor.h"
>  #include "qemu/bitops.h"
> +#include "qemu/error-report.h"
>  #include "qom/object.h"
>  #include "trace.h"
>  #include <assert.h>
>  
>  #include "exec/memory-internal.h"
>  #include "exec/ram_addr.h"
> +#include "sysemu/kvm.h"
>  #include "sysemu/sysemu.h"
>  
>  //#define DEBUG_UNASSIGNED
> @@ -1141,6 +1143,32 @@ MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
>      return r;
>  }
>  
> +/* Return true if an eventfd was signalled */
> +static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
> +                                                    hwaddr addr,
> +                                                    uint64_t data,
> +                                                    unsigned size,
> +                                                    MemTxAttrs attrs)
> +{
> +    MemoryRegionIoeventfd ioeventfd = {
> +        .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
> +        .data = data,
> +    };
> +    unsigned i;
> +
> +    for (i = 0; i < mr->ioeventfd_nb; i++) {
> +        ioeventfd.match_data = mr->ioeventfds[i].match_data;
> +        ioeventfd.e = mr->ioeventfds[i].e;
> +
> +        if (memory_region_ioeventfd_equal(ioeventfd, mr->ioeventfds[i])) {
> +            event_notifier_set(ioeventfd.e);
> +            return true;
> +        }
> +    }
> +
> +    return false;
> +}
> +
>  MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
>                                           hwaddr addr,
>                                           uint64_t data,
> @@ -1154,6 +1182,11 @@ MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
>  
>      adjust_endianness(mr, &data, size);
>  
> +    if ((!kvm_eventfds_enabled()) &&
> +        memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
> +        return MEMTX_OK;
> +    }
> +
>      if (mr->ops->write) {
>          return access_with_adjusted_size(addr, &data, size,
>                                           mr->ops->impl.min_access_size,
> @@ -1672,6 +1705,8 @@ void memory_region_clear_global_locking(MemoryRegion *mr)
>      mr->global_locking = false;
>  }
>  
> +static bool userspace_eventfd_warning;
> +
>  void memory_region_add_eventfd(MemoryRegion *mr,
>                                 hwaddr addr,
>                                 unsigned size,
> @@ -1688,6 +1723,13 @@ void memory_region_add_eventfd(MemoryRegion *mr,
>      };
>      unsigned i;
>  
> +    if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
> +                            userspace_eventfd_warning))) {
> +        userspace_eventfd_warning = true;
> +        error_report("Using eventfd without MMIO binding in KVM. "
> +                     "Suboptimal performance expected");
> +    }
> +
>      if (size) {
>          adjust_endianness(mr, &mrfd.data, size);
>      }
> 

Queued for 2.6.

Paolo