[PATCH v2] qemu-kvm: add irqfd support

[PATCH v2] qemu-kvm: add irqfd support

[Gregory Haskins]

(applies to kvm-userspace.git 84474e4f39)

This adds userspace support for the irqfd mechanism in the kernel, published
here:

http://lkml.org/lkml/2009/4/24/6

This is v2.  Changes since v1:

*) bug-fix to prevent returning "0" for success on allocation

---------------------

There are no current users, though the future virtual-bus v4 has plans to
adopt it.

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
---

 libkvm/libkvm.c |   33 +++++++++++++++++++++++++++++++++
 libkvm/libkvm.h |   13 +++++++++++++
 2 files changed, 46 insertions(+), 0 deletions(-)

diff --git a/libkvm/libkvm.c b/libkvm/libkvm.c
index 0610e3f..8c994a3 100644
--- a/libkvm/libkvm.c
+++ b/libkvm/libkvm.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 #include <sys/ioctl.h>
 #include <inttypes.h>
+#include <sys/eventfd.h>
 #include "libkvm.h"
 
 #if defined(__x86_64__) || defined(__i386__)
@@ -1440,3 +1441,35 @@ int kvm_assign_set_msix_entry(kvm_context_t kvm,
         return ret;
 }
 #endif
+
+static int assign_irqfd(kvm_context_t kvm, int fd, int gsi)
+{
+	int r;
+	struct kvm_irqfd data = {
+		.fd  = fd,
+		.gsi = gsi,
+	};
+
+	r = ioctl(kvm->vm_fd, KVM_ASSIGN_IRQFD, &data);
+	if (r == -1)
+		r = -errno;
+	return r;
+}	
+
+int kvm_irqfd(kvm_context_t kvm, int gsi)
+{
+	int fd, r;
+
+	if (!kvm_check_extension(kvm, KVM_CAP_IRQFD))
+		return -ENOENT;
+
+	fd = eventfd(0, 0);
+	if (fd < 0)
+		return fd;
+
+	r = assign_irqfd(kvm, fd, gsi);
+	if (r < 0)
+		return r;
+
+	return fd;
+}
diff --git a/libkvm/libkvm.h b/libkvm/libkvm.h
index ce6f054..404c028 100644
--- a/libkvm/libkvm.h
+++ b/libkvm/libkvm.h
@@ -856,6 +856,19 @@ int kvm_commit_irq_routes(kvm_context_t kvm);
  */
 int kvm_get_irq_route_gsi(kvm_context_t kvm);
 
+/*!
+ * \brief Create a file descriptor for injecting interrupts
+ *
+ * Creates an eventfd based file-descriptor that maps to a specific GSI
+ * in the guest.  eventfd compliant signaling (write() from userspace, or
+ * eventfd_signal() from kernelspace) will cause the GSI to inject
+ * itself into the guest at the next available window.
+ *
+ * \param kvm Pointer to the current kvm_context
+ * \param gsi GSI to assign to this fd
+ */
+int kvm_irqfd(kvm_context_t kvm, int gsi);
+
 #ifdef KVM_CAP_DEVICE_MSIX
 int kvm_assign_set_msix_nr(kvm_context_t kvm,
 			   struct kvm_assigned_msix_nr *msix_nr);

Re: [PATCH v2] qemu-kvm: add irqfd support

[Michael S. Tsirkin]

On Fri, Apr 24, 2009 at 12:30:17AM -0400, Gregory Haskins wrote:
> +int kvm_irqfd(kvm_context_t kvm, int gsi)
> +{
> +	int fd, r;
> +
> +	if (!kvm_check_extension(kvm, KVM_CAP_IRQFD))
> +		return -ENOENT;
> +
> +	fd = eventfd(0, 0);
> +	if (fd < 0)
> +		return fd;
> +
> +	r = assign_irqfd(kvm, fd, gsi);
> +	if (r < 0)
> +		return r;

Do we need to close fd on error?

> +
> +	return fd;
> +}

Re: [PATCH v2] qemu-kvm: add irqfd support

[Gregory Haskins]

[-- Attachment #1: Type: text/plain, Size: 694 bytes --]

Michael S. Tsirkin wrote:
> On Fri, Apr 24, 2009 at 12:30:17AM -0400, Gregory Haskins wrote:
>   
>> +int kvm_irqfd(kvm_context_t kvm, int gsi)
>> +{
>> +	int fd, r;
>> +
>> +	if (!kvm_check_extension(kvm, KVM_CAP_IRQFD))
>> +		return -ENOENT;
>> +
>> +	fd = eventfd(0, 0);
>> +	if (fd < 0)
>> +		return fd;
>> +
>> +	r = assign_irqfd(kvm, fd, gsi);
>> +	if (r < 0)
>> +		return r;
>>     
>
> Do we need to close fd on error?
>   
Good catch.  This will be changing in v3 to do the allocation in kernel,
but I will be sure to properly cleanup this type of leak, however that
ends up looking.

Thanks Michael.
-Greg

>   
>> +
>> +	return fd;
>> +}
>>     



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 266 bytes --]