Skip to contentSkip to conversation

Introduce a vringh accessor for IO memory

RFC PATCH New Shunsuke Mie 3 participants 24 messages Dec 27, 2022 – Jan 11, 2023

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

cover Introduce a vringh accessor for IO memory

Vringh is a host-side implementation of virtio rings, and supports the vring located on three kinds of memories, userspace, kernel space and a space translated iotlb.

The goal of this patchset is to refactor vringh and introduce a new vringh accessor for the vring located on the io memory region. The io memory accessor (iomem) is used by a driver that is not published yet, but I'm planning to publish it. Drivers affected by these changes are not included in this patchset. e.g. caif_virtio and vdpa (sim_net, sim_blk and net/mlx5) drivers.

This patchset can separate into 3 parts:
1. Fix and prepare some code related vringh [1, 2, 3/6]
2. Unify the vringh APIs and change related [4, 5/6]
3. Support IOMEM to vringh [6/6]

This first part is preparation for the second part which has a little fix and changes. A test code for vringh named vringh_test is also updated along with the changes. In the second part, unify the vringh API for each accessors that are user, kern and iotlb. The main point is struct vringh_ops that fill the gap between all accessors. The final part introduces an iomem support to vringh according to the unified API in the second part.

Those changes are tested for the user accessor using vringh_test and kern and iomem using a non published driver, but I think I can add a link to a patchset for the driver in the next version of this patchset.

Shunsuke Mie (6):
vringh: fix a typo in comments for vringh_kiov
vringh: remove vringh_iov and unite to vringh_kiov tools/virtio: convert to new vringh user APIs
vringh: unify the APIs for all accessors tools/virtio: convert to use new unified vringh APIs
vringh: IOMEM support

 drivers/vhost/Kconfig      |   6 +
 drivers/vhost/vringh.c     | 721 ++++++++++++-------------------------
 include/linux/vringh.h     | 147 +++-----
 tools/virtio/vringh_test.c | 123 ++++---
 4 files changed, 356 insertions(+), 641 deletions(-)

--
2.25.1

#1 lore raw

RFC PATCH 1/9 vringh: fix a typo in comments for vringh_kiov

Probably it is a simple copy error from struct vring_iov.

Fixes f87d0fbb5798 ("vringh: host-side implementation of virtio rings.")
---
 include/linux/vringh.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/vringh.h +1 -1
index 212892cf9822..1991a02c6431 100644
@@ -92,7 +92,7 @@ struct vringh_iov {
 };
 
 /**
- * struct vringh_iov - kvec mangler.
+ * struct vringh_kiov - kvec mangler.
  *
  * Mangles kvec in place, and restores it.
  * Remaining data is iov + i, of used - i elements.
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
#2 lore raw

RFC PATCH 2/9 vringh: remove vringh_iov and unite to vringh_kiov

In reply to Shunsuke Mie

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

---
 drivers/vhost/vringh.c | 32 ++++++------------------------
 include/linux/vringh.h | 45 ++++--------------------------------------
 2 files changed, 10 insertions(+), 67 deletions(-)
drivers/vhost/vringh.c +6 -26
index 828c29306565..aa3cd27d2384 100644
@@ -691,8 +691,8 @@ EXPORT_SYMBOL(vringh_init_user);
  * calling vringh_iov_cleanup() to release the memory, even on error!
  */
 int vringh_getdesc_user(struct vringh *vrh,
-			struct vringh_iov *riov,
-			struct vringh_iov *wiov,
+			struct vringh_kiov *riov,
+			struct vringh_kiov *wiov,
 			bool (*getrange)(struct vringh *vrh,
 					 u64 addr, struct vringh_range *r),
 			u16 *head)
@@ -708,26 +708,6 @@ int vringh_getdesc_user(struct vringh *vrh,
 	if (err == vrh->vring.num)
 		return 0;
 
-	/* We need the layouts to be the identical for this to work */
-	BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
-	BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
-		     offsetof(struct vringh_iov, iov));
-	BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
-		     offsetof(struct vringh_iov, i));
-	BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
-		     offsetof(struct vringh_iov, used));
-	BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
-		     offsetof(struct vringh_iov, max_num));
-	BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
-	BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
-		     offsetof(struct kvec, iov_base));
-	BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
-		     offsetof(struct kvec, iov_len));
-	BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
-		     != sizeof(((struct kvec *)NULL)->iov_base));
-	BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
-		     != sizeof(((struct kvec *)NULL)->iov_len));
-
 	*head = err;
 	err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
 			   (struct vringh_kiov *)wiov,
@@ -740,14 +720,14 @@ int vringh_getdesc_user(struct vringh *vrh,
 EXPORT_SYMBOL(vringh_getdesc_user);
 
 /**
- * vringh_iov_pull_user - copy bytes from vring_iov.
+ * vringh_iov_pull_user - copy bytes from vring_kiov.
  * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
  * @dst: the place to copy.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len)
 {
 	return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
 			       dst, len, xfer_from_user);
@@ -755,14 +735,14 @@ ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
 EXPORT_SYMBOL(vringh_iov_pull_user);
 
 /**
- * vringh_iov_push_user - copy bytes into vring_iov.
+ * vringh_iov_push_user - copy bytes into vring_kiov.
  * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
  * @src: the place to copy from.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
 			     const void *src, size_t len)
 {
 	return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
include/linux/vringh.h +4 -41

66 lines not on this page · raw message

Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
#3 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

Quoted 10 lines · Shunsuke Mie

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

Thanks

Quoted 154 lines
---
 drivers/vhost/vringh.c | 32 ++++++------------------------
 include/linux/vringh.h | 45 ++++--------------------------------------
 2 files changed, 10 insertions(+), 67 deletions(-)
drivers/vhost/vringh.c +6 -26
index 828c29306565..aa3cd27d2384 100644
@@ -691,8 +691,8 @@ EXPORT_SYMBOL(vringh_init_user);
  * calling vringh_iov_cleanup() to release the memory, even on error!
  */
 int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                        bool (*getrange)(struct vringh *vrh,
                                         u64 addr, struct vringh_range *r),
                        u16 *head)
@@ -708,26 +708,6 @@ int vringh_getdesc_user(struct vringh *vrh,
        if (err == vrh->vring.num)
                return 0;
 
-       /* We need the layouts to be the identical for this to work */
-       BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
-                    offsetof(struct vringh_iov, iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
-                    offsetof(struct vringh_iov, i));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
-                    offsetof(struct vringh_iov, used));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
-                    offsetof(struct vringh_iov, max_num));
-       BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
-                    offsetof(struct kvec, iov_base));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
-                    offsetof(struct kvec, iov_len));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
-                    != sizeof(((struct kvec *)NULL)->iov_base));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
-                    != sizeof(((struct kvec *)NULL)->iov_len));
-
        *head = err;
        err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
                           (struct vringh_kiov *)wiov,
@@ -740,14 +720,14 @@ int vringh_getdesc_user(struct vringh *vrh,
 EXPORT_SYMBOL(vringh_getdesc_user);
 
 /**
- * vringh_iov_pull_user - copy bytes from vring_iov.
+ * vringh_iov_pull_user - copy bytes from vring_kiov.
  * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
  * @dst: the place to copy.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len)
 {
        return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
                               dst, len, xfer_from_user);
@@ -755,14 +735,14 @@ ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
 EXPORT_SYMBOL(vringh_iov_pull_user);
 
 /**
- * vringh_iov_push_user - copy bytes into vring_iov.
+ * vringh_iov_push_user - copy bytes into vring_kiov.
  * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
  * @src: the place to copy from.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                             const void *src, size_t len)
 {
        return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
include/linux/vringh.h +4 -42
index 1991a02c6431..733d948e8123 100644
@@ -79,18 +79,6 @@ struct vringh_range {
        u64 offset;
 };
 
-/**
- * struct vringh_iov - iovec mangler.
- *
- * Mangles iovec in place, and restores it.
- * Remaining data is iov + i, of used - i elements.
- */
-struct vringh_iov {
-       struct iovec *iov;
-       size_t consumed; /* Within iov[i] */
-       unsigned i, used, max_num;
-};
-
 /**
  * struct vringh_kiov - kvec mangler.
  *
@@ -113,44 +101,19 @@ int vringh_init_user(struct vringh *vrh, u64 features,
                     vring_avail_t __user *avail,
                     vring_used_t __user *used);
 
-static inline void vringh_iov_init(struct vringh_iov *iov,
-                                  struct iovec *iovec, unsigned num)
-{
-       iov->used = iov->i = 0;
-       iov->consumed = 0;
-       iov->max_num = num;
-       iov->iov = iovec;
-}
-
-static inline void vringh_iov_reset(struct vringh_iov *iov)
-{
-       iov->iov[iov->i].iov_len += iov->consumed;
-       iov->iov[iov->i].iov_base -= iov->consumed;
-       iov->consumed = 0;
-       iov->i = 0;
-}
-
-static inline void vringh_iov_cleanup(struct vringh_iov *iov)
-{
-       if (iov->max_num & VRINGH_IOV_ALLOCATED)
-               kfree(iov->iov);
-       iov->max_num = iov->used = iov->i = iov->consumed = 0;
-       iov->iov = NULL;
-}
-
 /* Convert a descriptor into iovecs. */
 int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                        bool (*getrange)(struct vringh *vrh,
                                         u64 addr, struct vringh_range *r),
                        u16 *head);
 
 /* Copy bytes from readable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len);
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len);
 
 /* Copy bytes into writable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                             const void *src, size_t len);
 
 /* Mark a descriptor as used. */
--
Quoted 2 lines

2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#4 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

Quoted 10 lines · Jason Wang
Quoted 9 lines · Shunsuke Mie

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

Quoted 6 lines
Quoted 1 line

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

Thanks

Possibly, but when doing changes like this one needs to be careful to avoid breaking all the inlining tricks vringh relies on for performance.

Signature · 6 lines
MST

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
#5 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

2022年12月27日(火) 16:05 Michael S. Tsirkin <[email protected]>:

Quoted 14 lines · Jason Wang
Quoted 12 lines · Shunsuke Mie
Quoted 10 lines

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

Quoted 8 lines
Quoted 4 lines

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

Thanks

Possibly, but when doing changes like this one needs to be careful to avoid breaking all the inlining tricks vringh relies on for performance.

Definitely, I'm evaluating the performance using vringh_test. I'll add a result of the evaluation. But, If there are other evaluation methods, could you please tell me?

Quoted 3 lines

--
MST

Best,
Shunsuke
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#6 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

Quoted 18 lines · Shunsuke Mie

2022年12月27日(火) 16:05 Michael S. Tsirkin <[email protected]>:

Quoted 17 lines · Jason Wang
Quoted 15 lines · Shunsuke Mie
Quoted 10 lines

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

Thanks

Possibly, but when doing changes like this one needs to be careful to avoid breaking all the inlining tricks vringh relies on for performance.

Definitely, I'm evaluating the performance using vringh_test. I'll add a result of the evaluation. But, If there are other evaluation methods, could you please tell me?

high level tests over virtio blk and net are possible, but let's start with vringh_test.

Quoted 6 lines
Quoted 3 lines

--
MST

Best,
Shunsuke

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#7 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

2022年12月27日(火) 16:56 Michael S. Tsirkin <[email protected]>:

Quoted 25 lines · Shunsuke Mie
Quoted 23 lines

2022年12月27日(火) 16:05 Michael S. Tsirkin <[email protected]>:

Quoted 22 lines · Jason Wang
Quoted 10 lines · Shunsuke Mie
Quoted 9 lines

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

Quoted 6 lines
Quoted 1 line

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

Thanks

Possibly, but when doing changes like this one needs to be careful to avoid breaking all the inlining tricks vringh relies on for performance.

Quoted 6 lines
Quoted 3 lines

Definitely, I'm evaluating the performance using vringh_test. I'll add a result of the evaluation. But, If there are other evaluation methods, could you please tell me?

high level tests over virtio blk and net are possible, but let's start with vringh_test.

Ok, I'll do it.

Quoted 7 lines
Quoted 6 lines
Quoted 3 lines

--
MST

Best,
Shunsuke

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#8 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

In reply to Jason Wang

2022年12月27日(火) 15:04 Jason Wang <[email protected]>:

Quoted 8 lines · Shunsuke Mie
Quoted 6 lines

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

Quoted 7 lines
Quoted 4 lines

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

I didn't see the iov iterator yet, thank you for informing me. Is that iov_iter? https://lwn.net/Articles/625077/

Quoted 154 lines

Thanks

Quoted 152 lines
---
 drivers/vhost/vringh.c | 32 ++++++------------------------
 include/linux/vringh.h | 45 ++++--------------------------------------
 2 files changed, 10 insertions(+), 67 deletions(-)
drivers/vhost/vringh.c +6 -26
index 828c29306565..aa3cd27d2384 100644
@@ -691,8 +691,8 @@ EXPORT_SYMBOL(vringh_init_user);
  * calling vringh_iov_cleanup() to release the memory, even on error!
  */
 int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                        bool (*getrange)(struct vringh *vrh,
                                         u64 addr, struct vringh_range *r),
                        u16 *head)
@@ -708,26 +708,6 @@ int vringh_getdesc_user(struct vringh *vrh,
        if (err == vrh->vring.num)
                return 0;
 
-       /* We need the layouts to be the identical for this to work */
-       BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
-                    offsetof(struct vringh_iov, iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
-                    offsetof(struct vringh_iov, i));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
-                    offsetof(struct vringh_iov, used));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
-                    offsetof(struct vringh_iov, max_num));
-       BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
-                    offsetof(struct kvec, iov_base));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
-                    offsetof(struct kvec, iov_len));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
-                    != sizeof(((struct kvec *)NULL)->iov_base));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
-                    != sizeof(((struct kvec *)NULL)->iov_len));
-
        *head = err;
        err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
                           (struct vringh_kiov *)wiov,
@@ -740,14 +720,14 @@ int vringh_getdesc_user(struct vringh *vrh,
 EXPORT_SYMBOL(vringh_getdesc_user);
 
 /**
- * vringh_iov_pull_user - copy bytes from vring_iov.
+ * vringh_iov_pull_user - copy bytes from vring_kiov.
  * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
  * @dst: the place to copy.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len)
 {
        return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
                               dst, len, xfer_from_user);
@@ -755,14 +735,14 @@ ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
 EXPORT_SYMBOL(vringh_iov_pull_user);
 
 /**
- * vringh_iov_push_user - copy bytes into vring_iov.
+ * vringh_iov_push_user - copy bytes into vring_kiov.
  * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
  * @src: the place to copy from.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                             const void *src, size_t len)
 {
        return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
include/linux/vringh.h +4 -41
index 1991a02c6431..733d948e8123 100644
@@ -79,18 +79,6 @@ struct vringh_range {
        u64 offset;
 };
 
-/**
- * struct vringh_iov - iovec mangler.
- *
- * Mangles iovec in place, and restores it.
- * Remaining data is iov + i, of used - i elements.
- */
-struct vringh_iov {
-       struct iovec *iov;
-       size_t consumed; /* Within iov[i] */
-       unsigned i, used, max_num;
-};
-
 /**
  * struct vringh_kiov - kvec mangler.
  *
@@ -113,44 +101,19 @@ int vringh_init_user(struct vringh *vrh, u64 features,
                     vring_avail_t __user *avail,
                     vring_used_t __user *used);
 
-static inline void vringh_iov_init(struct vringh_iov *iov,
-                                  struct iovec *iovec, unsigned num)
-{
-       iov->used = iov->i = 0;
-       iov->consumed = 0;
-       iov->max_num = num;
-       iov->iov = iovec;
-}
-
-static inline void vringh_iov_reset(struct vringh_iov *iov)
-{
-       iov->iov[iov->i].iov_len += iov->consumed;
-       iov->iov[iov->i].iov_base -= iov->consumed;
-       iov->consumed = 0;
-       iov->i = 0;
-}
-
-static inline void vringh_iov_cleanup(struct vringh_iov *iov)
-{
-       if (iov->max_num & VRINGH_IOV_ALLOCATED)
-               kfree(iov->iov);
-       iov->max_num = iov->used = iov->i = iov->consumed = 0;
-       iov->iov = NULL;
-}
-
 /* Convert a descriptor into iovecs. */
 int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                        bool (*getrange)(struct vringh *vrh,
                                         u64 addr, struct vringh_range *r),
                        u16 *head);
 
 /* Copy bytes from readable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len);
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len);
 
 /* Copy bytes into writable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                             const void *src, size_t len);
 
Quoted 5 lines
Quoted 4 lines
include/linux/vringh.h +0 -1
 /* Mark a descriptor as used. */
--

2.25.1

Best,
Shunsuke
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#9 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

Quoted 13 lines · Shunsuke Mie

2022年12月27日(火) 15:04 Jason Wang <[email protected]>:

Quoted 11 lines · Shunsuke Mie
Quoted 9 lines

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

I didn't see the iov iterator yet, thank you for informing me. Is that iov_iter? https://lwn.net/Articles/625077/

Exactly.

Thanks

Quoted 154 lines
Quoted 154 lines

Thanks

Quoted 152 lines
---
 drivers/vhost/vringh.c | 32 ++++++------------------------
 include/linux/vringh.h | 45 ++++--------------------------------------
 2 files changed, 10 insertions(+), 67 deletions(-)
drivers/vhost/vringh.c +6 -26
index 828c29306565..aa3cd27d2384 100644
@@ -691,8 +691,8 @@ EXPORT_SYMBOL(vringh_init_user);
  * calling vringh_iov_cleanup() to release the memory, even on error!
  */
 int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                        bool (*getrange)(struct vringh *vrh,
                                         u64 addr, struct vringh_range *r),
                        u16 *head)
@@ -708,26 +708,6 @@ int vringh_getdesc_user(struct vringh *vrh,
        if (err == vrh->vring.num)
                return 0;
 
-       /* We need the layouts to be the identical for this to work */
-       BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
-                    offsetof(struct vringh_iov, iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
-                    offsetof(struct vringh_iov, i));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
-                    offsetof(struct vringh_iov, used));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
-                    offsetof(struct vringh_iov, max_num));
-       BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
-                    offsetof(struct kvec, iov_base));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
-                    offsetof(struct kvec, iov_len));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
-                    != sizeof(((struct kvec *)NULL)->iov_base));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
-                    != sizeof(((struct kvec *)NULL)->iov_len));
-
        *head = err;
        err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
                           (struct vringh_kiov *)wiov,
@@ -740,14 +720,14 @@ int vringh_getdesc_user(struct vringh *vrh,
 EXPORT_SYMBOL(vringh_getdesc_user);
 
 /**
- * vringh_iov_pull_user - copy bytes from vring_iov.
+ * vringh_iov_pull_user - copy bytes from vring_kiov.
  * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
  * @dst: the place to copy.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len)
 {
        return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
                               dst, len, xfer_from_user);
@@ -755,14 +735,14 @@ ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
 EXPORT_SYMBOL(vringh_iov_pull_user);
 
 /**
- * vringh_iov_push_user - copy bytes into vring_iov.
+ * vringh_iov_push_user - copy bytes into vring_kiov.
  * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
  * @src: the place to copy from.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                             const void *src, size_t len)
 {
        return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
include/linux/vringh.h +4 -41
index 1991a02c6431..733d948e8123 100644
@@ -79,18 +79,6 @@ struct vringh_range {
        u64 offset;
 };
 
-/**
- * struct vringh_iov - iovec mangler.
- *
- * Mangles iovec in place, and restores it.
- * Remaining data is iov + i, of used - i elements.
- */
-struct vringh_iov {
-       struct iovec *iov;
-       size_t consumed; /* Within iov[i] */
-       unsigned i, used, max_num;
-};
-
 /**
  * struct vringh_kiov - kvec mangler.
  *
@@ -113,44 +101,19 @@ int vringh_init_user(struct vringh *vrh, u64 features,
                     vring_avail_t __user *avail,
                     vring_used_t __user *used);
 
-static inline void vringh_iov_init(struct vringh_iov *iov,
-                                  struct iovec *iovec, unsigned num)
-{
-       iov->used = iov->i = 0;
-       iov->consumed = 0;
-       iov->max_num = num;
-       iov->iov = iovec;
-}
-
-static inline void vringh_iov_reset(struct vringh_iov *iov)
-{
-       iov->iov[iov->i].iov_len += iov->consumed;
-       iov->iov[iov->i].iov_base -= iov->consumed;
-       iov->consumed = 0;
-       iov->i = 0;
-}
-
-static inline void vringh_iov_cleanup(struct vringh_iov *iov)
-{
-       if (iov->max_num & VRINGH_IOV_ALLOCATED)
-               kfree(iov->iov);
-       iov->max_num = iov->used = iov->i = iov->consumed = 0;
-       iov->iov = NULL;
-}
-
 /* Convert a descriptor into iovecs. */
 int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                        bool (*getrange)(struct vringh *vrh,
                                         u64 addr, struct vringh_range *r),
                        u16 *head);
 
 /* Copy bytes from readable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len);
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len);
 
 /* Copy bytes into writable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                             const void *src, size_t len);
 
Quoted 8 lines
Quoted 5 lines
Quoted 4 lines
include/linux/vringh.h +0 -1
 /* Mark a descriptor as used. */
--

2.25.1

Best,
Shunsuke

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#10 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

Quoted 11 lines · Jason Wang
Quoted 10 lines · Shunsuke Mie

2022年12月27日(火) 15:04 Jason Wang <[email protected]>:

Quoted 9 lines · Shunsuke Mie
Quoted 8 lines

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
Quoted 1 line

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

I didn't see the iov iterator yet, thank you for informing me. Is that iov_iter? https://lwn.net/Articles/625077/

Exactly.

I've investigated the iov_iter, vhost and related APIs. As a result, I think that it is not easy to switch to use the iov_iter. Because, the design of vhost and vringh is different.

The iov_iter has vring desc info and meta data of transfer method. The vhost provides generic transfer function for the iov_iter. In constrast, vringh_iov just has vring desc info. The vringh provides transfer functions for each methods.

In the future, it is better to use common data structure and APIs between vhost and vringh (or merge completely), but it requires a lot of changes, so I'd like to just organize data structure in vringh as a first step in this patch.

Best

Quoted 163 lines

Thanks

Quoted 161 lines
Quoted 154 lines

Thanks

Quoted 152 lines
---
  drivers/vhost/vringh.c | 32 ++++++------------------------
  include/linux/vringh.h | 45 ++++--------------------------------------
  2 files changed, 10 insertions(+), 67 deletions(-)
drivers/vhost/vringh.c +6 -26
index 828c29306565..aa3cd27d2384 100644
@@ -691,8 +691,8 @@ EXPORT_SYMBOL(vringh_init_user);
   * calling vringh_iov_cleanup() to release the memory, even on error!
   */
  int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                         bool (*getrange)(struct vringh *vrh,
                                          u64 addr, struct vringh_range *r),
                         u16 *head)
@@ -708,26 +708,6 @@ int vringh_getdesc_user(struct vringh *vrh,
         if (err == vrh->vring.num)
                 return 0;
 
-       /* We need the layouts to be the identical for this to work */
-       BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
-                    offsetof(struct vringh_iov, iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
-                    offsetof(struct vringh_iov, i));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
-                    offsetof(struct vringh_iov, used));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
-                    offsetof(struct vringh_iov, max_num));
-       BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
-                    offsetof(struct kvec, iov_base));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
-                    offsetof(struct kvec, iov_len));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
-                    != sizeof(((struct kvec *)NULL)->iov_base));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
-                    != sizeof(((struct kvec *)NULL)->iov_len));
-
         *head = err;
         err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
                            (struct vringh_kiov *)wiov,
@@ -740,14 +720,14 @@ int vringh_getdesc_user(struct vringh *vrh,
  EXPORT_SYMBOL(vringh_getdesc_user);
 
  /**
- * vringh_iov_pull_user - copy bytes from vring_iov.
+ * vringh_iov_pull_user - copy bytes from vring_kiov.
   * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
   * @dst: the place to copy.
   * @len: the maximum length to copy.
   *
   * Returns the bytes copied <= len or a negative errno.
   */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len)
  {
         return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
                                dst, len, xfer_from_user);
@@ -755,14 +735,14 @@ ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
  EXPORT_SYMBOL(vringh_iov_pull_user);
 
  /**
- * vringh_iov_push_user - copy bytes into vring_iov.
+ * vringh_iov_push_user - copy bytes into vring_kiov.
   * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
   * @src: the place to copy from.
   * @len: the maximum length to copy.
   *
   * Returns the bytes copied <= len or a negative errno.
   */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                              const void *src, size_t len)
  {
         return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
include/linux/vringh.h +4 -41
index 1991a02c6431..733d948e8123 100644
@@ -79,18 +79,6 @@ struct vringh_range {
         u64 offset;
  };
 
-/**
- * struct vringh_iov - iovec mangler.
- *
- * Mangles iovec in place, and restores it.
- * Remaining data is iov + i, of used - i elements.
- */
-struct vringh_iov {
-       struct iovec *iov;
-       size_t consumed; /* Within iov[i] */
-       unsigned i, used, max_num;
-};
-
  /**
   * struct vringh_kiov - kvec mangler.
   *
@@ -113,44 +101,19 @@ int vringh_init_user(struct vringh *vrh, u64 features,
                      vring_avail_t __user *avail,
                      vring_used_t __user *used);
 
-static inline void vringh_iov_init(struct vringh_iov *iov,
-                                  struct iovec *iovec, unsigned num)
-{
-       iov->used = iov->i = 0;
-       iov->consumed = 0;
-       iov->max_num = num;
-       iov->iov = iovec;
-}
-
-static inline void vringh_iov_reset(struct vringh_iov *iov)
-{
-       iov->iov[iov->i].iov_len += iov->consumed;
-       iov->iov[iov->i].iov_base -= iov->consumed;
-       iov->consumed = 0;
-       iov->i = 0;
-}
-
-static inline void vringh_iov_cleanup(struct vringh_iov *iov)
-{
-       if (iov->max_num & VRINGH_IOV_ALLOCATED)
-               kfree(iov->iov);
-       iov->max_num = iov->used = iov->i = iov->consumed = 0;
-       iov->iov = NULL;
-}
-
  /* Convert a descriptor into iovecs. */
  int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                         bool (*getrange)(struct vringh *vrh,
                                          u64 addr, struct vringh_range *r),
                         u16 *head);
 
  /* Copy bytes from readable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len);
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len);
 
  /* Copy bytes into writable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                              const void *src, size_t len);
 
Quoted 4 lines
Quoted 4 lines
include/linux/vringh.h +0 -1
  /* Mark a descriptor as used. */
--

2.25.1

Best,
Shunsuke

#11 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

Quoted 17 lines · Shunsuke Mie
Quoted 14 lines · Jason Wang
Quoted 13 lines · Shunsuke Mie

2022年12月27日(火) 15:04 Jason Wang <[email protected]>:

Quoted 12 lines · Shunsuke Mie
Quoted 9 lines

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

Quoted 7 lines
Quoted 3 lines
Quoted 2 lines

I didn't see the iov iterator yet, thank you for informing me. Is that iov_iter? https://lwn.net/Articles/625077/

Exactly.

I've investigated the iov_iter, vhost and related APIs. As a result, I think that it is not easy to switch to use the iov_iter. Because, the design of vhost and vringh is different.

Yes, but just to make sure we are on the same page, the reason I suggest iov_iter for vringh is that the vringh itself has customized iter equivalent, e.g it has iter for kernel,user, or even iotlb. At least the kernel and userspace part could be switched to iov_iter. Note that it has nothing to do with vhost.

Quoted 10 lines

The iov_iter has vring desc info and meta data of transfer method. The vhost provides generic transfer function for the iov_iter. In constrast, vringh_iov just has vring desc info. The vringh provides transfer functions for each methods.

In the future, it is better to use common data structure and APIs between vhost and vringh (or merge completely), but it requires a lot of changes, so I'd like to just
organize data structure in vringh as a first step in this patch.

That's fine.

Thansk

Quoted 168 lines

Best

Quoted 163 lines

Thanks

Quoted 161 lines
Quoted 154 lines

Thanks

Quoted 152 lines
---
  drivers/vhost/vringh.c | 32 ++++++------------------------
  include/linux/vringh.h | 45 ++++--------------------------------------
  2 files changed, 10 insertions(+), 67 deletions(-)
drivers/vhost/vringh.c +6 -26
index 828c29306565..aa3cd27d2384 100644
@@ -691,8 +691,8 @@ EXPORT_SYMBOL(vringh_init_user);
   * calling vringh_iov_cleanup() to release the memory, even on error!
   */
  int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                         bool (*getrange)(struct vringh *vrh,
                                          u64 addr, struct vringh_range *r),
                         u16 *head)
@@ -708,26 +708,6 @@ int vringh_getdesc_user(struct vringh *vrh,
         if (err == vrh->vring.num)
                 return 0;
 
-       /* We need the layouts to be the identical for this to work */
-       BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
-                    offsetof(struct vringh_iov, iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
-                    offsetof(struct vringh_iov, i));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
-                    offsetof(struct vringh_iov, used));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
-                    offsetof(struct vringh_iov, max_num));
-       BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
-                    offsetof(struct kvec, iov_base));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
-                    offsetof(struct kvec, iov_len));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
-                    != sizeof(((struct kvec *)NULL)->iov_base));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
-                    != sizeof(((struct kvec *)NULL)->iov_len));
-
         *head = err;
         err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
                            (struct vringh_kiov *)wiov,
@@ -740,14 +720,14 @@ int vringh_getdesc_user(struct vringh *vrh,
  EXPORT_SYMBOL(vringh_getdesc_user);
 
  /**
- * vringh_iov_pull_user - copy bytes from vring_iov.
+ * vringh_iov_pull_user - copy bytes from vring_kiov.
   * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
   * @dst: the place to copy.
   * @len: the maximum length to copy.
   *
   * Returns the bytes copied <= len or a negative errno.
   */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len)
  {
         return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
                                dst, len, xfer_from_user);
@@ -755,14 +735,14 @@ ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
  EXPORT_SYMBOL(vringh_iov_pull_user);
 
  /**
- * vringh_iov_push_user - copy bytes into vring_iov.
+ * vringh_iov_push_user - copy bytes into vring_kiov.
   * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
   * @src: the place to copy from.
   * @len: the maximum length to copy.
   *
   * Returns the bytes copied <= len or a negative errno.
   */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                              const void *src, size_t len)
  {
         return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
include/linux/vringh.h +4 -41
index 1991a02c6431..733d948e8123 100644
@@ -79,18 +79,6 @@ struct vringh_range {
         u64 offset;
  };
 
-/**
- * struct vringh_iov - iovec mangler.
- *
- * Mangles iovec in place, and restores it.
- * Remaining data is iov + i, of used - i elements.
- */
-struct vringh_iov {
-       struct iovec *iov;
-       size_t consumed; /* Within iov[i] */
-       unsigned i, used, max_num;
-};
-
  /**
   * struct vringh_kiov - kvec mangler.
   *
@@ -113,44 +101,19 @@ int vringh_init_user(struct vringh *vrh, u64 features,
                      vring_avail_t __user *avail,
                      vring_used_t __user *used);
 
-static inline void vringh_iov_init(struct vringh_iov *iov,
-                                  struct iovec *iovec, unsigned num)
-{
-       iov->used = iov->i = 0;
-       iov->consumed = 0;
-       iov->max_num = num;
-       iov->iov = iovec;
-}
-
-static inline void vringh_iov_reset(struct vringh_iov *iov)
-{
-       iov->iov[iov->i].iov_len += iov->consumed;
-       iov->iov[iov->i].iov_base -= iov->consumed;
-       iov->consumed = 0;
-       iov->i = 0;
-}
-
-static inline void vringh_iov_cleanup(struct vringh_iov *iov)
-{
-       if (iov->max_num & VRINGH_IOV_ALLOCATED)
-               kfree(iov->iov);
-       iov->max_num = iov->used = iov->i = iov->consumed = 0;
-       iov->iov = NULL;
-}
-
  /* Convert a descriptor into iovecs. */
  int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                         bool (*getrange)(struct vringh *vrh,
                                          u64 addr, struct vringh_range *r),
                         u16 *head);
 
  /* Copy bytes from readable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len);
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len);
 
  /* Copy bytes into writable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                              const void *src, size_t len);
 
Quoted 4 lines
Quoted 4 lines
include/linux/vringh.h +0 -1
  /* Mark a descriptor as used. */
--

2.25.1

Best,
Shunsuke

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#12 lore raw

Re: [RFC PATCH 2/9] vringh: remove vringh_iov and unite to vringh_kiov

Quoted 20 lines · Jason Wang
Quoted 19 lines · Shunsuke Mie
Quoted 17 lines · Jason Wang
Quoted 7 lines · Shunsuke Mie

2022年12月27日(火) 15:04 Jason Wang <[email protected]>:

Quoted 6 lines · Shunsuke Mie
Quoted 5 lines

struct vringh_iov is defined to hold userland addresses. However, to use common function, __vring_iov, finally the vringh_iov converts to the vringh_kiov with simple cast. It includes compile time check code to make sure it can be cast correctly.

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines

To simplify the code, this patch removes the struct vringh_iov and unifies APIs to struct vringh_kiov.

While at this, I wonder if we need to go further, that is, switch to using an iov iterator instead of a vringh customized one.

I didn't see the iov iterator yet, thank you for informing me. Is that iov_iter? https://lwn.net/Articles/625077/

Exactly.

Quoted 8 lines
Quoted 3 lines

I've investigated the iov_iter, vhost and related APIs. As a result, I think that it is not easy to switch to use the iov_iter. Because, the design of vhost and vringh is different.

Yes, but just to make sure we are on the same page, the reason I suggest iov_iter for vringh is that the vringh itself has customized iter equivalent, e.g it has iter for kernel,user, or even iotlb. At least the kernel and userspace part could be switched to iov_iter. Note that it has nothing to do with vhost.

I agree. It can be switch to use iov_iter, but I think we need to change fundamentally. There are duplicated code on vhost and vringh to access vring, and some helper functions...

Anyway, I'd like to focus vringh in this patchset. Thank you for your

comments and suggestions!

Best

Quoted 179 lines
Quoted 9 lines

The iov_iter has vring desc info and meta data of transfer method. The vhost provides generic transfer function for the iov_iter. In constrast, vringh_iov just has vring desc info. The vringh provides transfer functions for each methods.

In the future, it is better to use common data structure and APIs between vhost and vringh (or merge completely), but it requires a lot of changes, so I'd like to just
organize data structure in vringh as a first step in this patch.

That's fine.

Thansk

Quoted 166 lines

Best

Quoted 163 lines

Thanks

Quoted 161 lines
Quoted 154 lines

Thanks

Quoted 152 lines
---
   drivers/vhost/vringh.c | 32 ++++++------------------------
   include/linux/vringh.h | 45 ++++--------------------------------------
   2 files changed, 10 insertions(+), 67 deletions(-)
drivers/vhost/vringh.c +6 -26
index 828c29306565..aa3cd27d2384 100644
@@ -691,8 +691,8 @@ EXPORT_SYMBOL(vringh_init_user);
    * calling vringh_iov_cleanup() to release the memory, even on error!
    */
   int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                          bool (*getrange)(struct vringh *vrh,
                                           u64 addr, struct vringh_range *r),
                          u16 *head)
@@ -708,26 +708,6 @@ int vringh_getdesc_user(struct vringh *vrh,
          if (err == vrh->vring.num)
                  return 0;
 
-       /* We need the layouts to be the identical for this to work */
-       BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
-                    offsetof(struct vringh_iov, iov));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
-                    offsetof(struct vringh_iov, i));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
-                    offsetof(struct vringh_iov, used));
-       BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
-                    offsetof(struct vringh_iov, max_num));
-       BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
-                    offsetof(struct kvec, iov_base));
-       BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
-                    offsetof(struct kvec, iov_len));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
-                    != sizeof(((struct kvec *)NULL)->iov_base));
-       BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
-                    != sizeof(((struct kvec *)NULL)->iov_len));
-
          *head = err;
          err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
                             (struct vringh_kiov *)wiov,
@@ -740,14 +720,14 @@ int vringh_getdesc_user(struct vringh *vrh,
   EXPORT_SYMBOL(vringh_getdesc_user);
 
   /**
- * vringh_iov_pull_user - copy bytes from vring_iov.
+ * vringh_iov_pull_user - copy bytes from vring_kiov.
    * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
    * @dst: the place to copy.
    * @len: the maximum length to copy.
    *
    * Returns the bytes copied <= len or a negative errno.
    */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len)
   {
          return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
                                 dst, len, xfer_from_user);
@@ -755,14 +735,14 @@ ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
   EXPORT_SYMBOL(vringh_iov_pull_user);
 
   /**
- * vringh_iov_push_user - copy bytes into vring_iov.
+ * vringh_iov_push_user - copy bytes into vring_kiov.
    * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
    * @src: the place to copy from.
    * @len: the maximum length to copy.
    *
    * Returns the bytes copied <= len or a negative errno.
    */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                               const void *src, size_t len)
   {
          return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
include/linux/vringh.h +4 -41
index 1991a02c6431..733d948e8123 100644
@@ -79,18 +79,6 @@ struct vringh_range {
          u64 offset;
   };
 
-/**
- * struct vringh_iov - iovec mangler.
- *
- * Mangles iovec in place, and restores it.
- * Remaining data is iov + i, of used - i elements.
- */
-struct vringh_iov {
-       struct iovec *iov;
-       size_t consumed; /* Within iov[i] */
-       unsigned i, used, max_num;
-};
-
   /**
    * struct vringh_kiov - kvec mangler.
    *
@@ -113,44 +101,19 @@ int vringh_init_user(struct vringh *vrh, u64 features,
                       vring_avail_t __user *avail,
                       vring_used_t __user *used);
 
-static inline void vringh_iov_init(struct vringh_iov *iov,
-                                  struct iovec *iovec, unsigned num)
-{
-       iov->used = iov->i = 0;
-       iov->consumed = 0;
-       iov->max_num = num;
-       iov->iov = iovec;
-}
-
-static inline void vringh_iov_reset(struct vringh_iov *iov)
-{
-       iov->iov[iov->i].iov_len += iov->consumed;
-       iov->iov[iov->i].iov_base -= iov->consumed;
-       iov->consumed = 0;
-       iov->i = 0;
-}
-
-static inline void vringh_iov_cleanup(struct vringh_iov *iov)
-{
-       if (iov->max_num & VRINGH_IOV_ALLOCATED)
-               kfree(iov->iov);
-       iov->max_num = iov->used = iov->i = iov->consumed = 0;
-       iov->iov = NULL;
-}
-
   /* Convert a descriptor into iovecs. */
   int vringh_getdesc_user(struct vringh *vrh,
-                       struct vringh_iov *riov,
-                       struct vringh_iov *wiov,
+                       struct vringh_kiov *riov,
+                       struct vringh_kiov *wiov,
                          bool (*getrange)(struct vringh *vrh,
                                           u64 addr, struct vringh_range *r),
                          u16 *head);
 
   /* Copy bytes from readable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len);
+ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len);
 
   /* Copy bytes into writable vsg, consuming it (and incrementing wiov->i). */
-ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
                               const void *src, size_t len);
 
Quoted 4 lines
Quoted 4 lines
include/linux/vringh.h +0 -1
   /* Mark a descriptor as used. */
--

2.25.1

Best,
Shunsuke

#13 lore raw

RFC PATCH 3/9 tools/virtio: convert to new vringh user APIs

In reply to Shunsuke Mie

struct vringh_iov is being remove, so convert vringh_test to use the vringh user APIs. This has it change to use struct vringh_kiov instead of the struct vringh_iov.

---
 tools/virtio/vringh_test.c | 34 +++++++++++++++++-----------------
 1 file changed, 17 insertions(+), 17 deletions(-)
tools/virtio/vringh_test.c +17 -17
index 98ff808d6f0c..6c9533b8a2ca 100644
@@ -193,8 +193,8 @@ static int parallel_test(u64 features,
 			errx(1, "Could not set affinity to cpu %u", first_cpu);
 
 		while (xfers < NUM_XFERS) {
-			struct iovec host_riov[2], host_wiov[2];
-			struct vringh_iov riov, wiov;
+			struct kvec host_riov[2], host_wiov[2];
+			struct vringh_kiov riov, wiov;
 			u16 head, written;
 
 			if (fast_vringh) {
@@ -216,10 +216,10 @@ static int parallel_test(u64 features,
 				written = 0;
 				goto complete;
 			} else {
-				vringh_iov_init(&riov,
+				vringh_kiov_init(&riov,
 						host_riov,
 						ARRAY_SIZE(host_riov));
-				vringh_iov_init(&wiov,
+				vringh_kiov_init(&wiov,
 						host_wiov,
 						ARRAY_SIZE(host_wiov));
 
@@ -442,8 +442,8 @@ int main(int argc, char *argv[])
 	struct virtqueue *vq;
 	struct vringh vrh;
 	struct scatterlist guest_sg[RINGSIZE], *sgs[2];
-	struct iovec host_riov[2], host_wiov[2];
-	struct vringh_iov riov, wiov;
+	struct kvec host_riov[2], host_wiov[2];
+	struct vringh_kiov riov, wiov;
 	struct vring_used_elem used[RINGSIZE];
 	char buf[28];
 	u16 head;
@@ -517,8 +517,8 @@ int main(int argc, char *argv[])
 	__kmalloc_fake = NULL;
 
 	/* Host retreives it. */
-	vringh_iov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
-	vringh_iov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
+	vringh_kiov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
+	vringh_kiov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
 
 	err = vringh_getdesc_user(&vrh, &riov, &wiov, getrange, &head);
 	if (err != 1)
@@ -586,8 +586,8 @@ int main(int argc, char *argv[])
 	__kmalloc_fake = NULL;
 
 	/* Host picks it up (allocates new iov). */
-	vringh_iov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
-	vringh_iov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
+	vringh_kiov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
+	vringh_kiov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
 
 	err = vringh_getdesc_user(&vrh, &riov, &wiov, getrange, &head);
 	if (err != 1)
@@ -613,8 +613,8 @@ int main(int argc, char *argv[])
 		assert(err < 3 || buf[2] == (char)(i + 2));
 	}
 	assert(riov.i == riov.used);
-	vringh_iov_cleanup(&riov);
-	vringh_iov_cleanup(&wiov);
+	vringh_kiov_cleanup(&riov);
+	vringh_kiov_cleanup(&wiov);
 
 	/* Complete using multi interface, just because we can. */
 	used[0].id = head;
@@ -638,8 +638,8 @@ int main(int argc, char *argv[])
 	}
 
 	/* Now get many, and consume them all at once. */
-	vringh_iov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
-	vringh_iov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
+	vringh_kiov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
+	vringh_kiov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
 
 	for (i = 0; i < RINGSIZE; i++) {
 		err = vringh_getdesc_user(&vrh, &riov, &wiov, getrange, &head);
@@ -723,8 +723,8 @@ int main(int argc, char *argv[])
 		d[5].flags = 0;
 
 		/* Host picks it up (allocates new iov). */
-		vringh_iov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
-		vringh_iov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
+		vringh_kiov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
+		vringh_kiov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
 
 		err = vringh_getdesc_user(&vrh, &riov, &wiov, getrange, &head);
 		if (err != 1)
@@ -744,7 +744,7 @@ int main(int argc, char *argv[])
 		/* Data should be linear. */
 		for (i = 0; i < err; i++)
 			assert(buf[i] == i);
-		vringh_iov_cleanup(&riov);
+		vringh_kiov_cleanup(&riov);
 	}
 
 	/* Don't leak memory... */
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
#14 lore raw

RFC PATCH 4/9 vringh: unify the APIs for all accessors

In reply to Shunsuke Mie

Each vringh memory accessors that are for user, kern and iotlb has own interfaces that calls common code. But some codes are duplicated and that becomes loss extendability.

Introduce a struct vringh_ops and provide a common APIs for all accessors. It can bee easily extended vringh code for new memory accessor and simplified a caller code.

---
 drivers/vhost/vringh.c | 667 +++++++++++------------------------------
 include/linux/vringh.h | 100 +++---
 2 files changed, 225 insertions(+), 542 deletions(-)
drivers/vhost/vringh.c +178 -489
index aa3cd27d2384..ebfd3644a1a3 100644
@@ -35,15 +35,12 @@ static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
 }
 
 /* Returns vring->num if empty, -ve on error. */
-static inline int __vringh_get_head(const struct vringh *vrh,
-				    int (*getu16)(const struct vringh *vrh,
-						  u16 *val, const __virtio16 *p),
-				    u16 *last_avail_idx)
+static inline int __vringh_get_head(const struct vringh *vrh, u16 *last_avail_idx)
 {
 	u16 avail_idx, i, head;
 	int err;
 
-	err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
+	err = vrh->ops.getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
 	if (err) {
 		vringh_bad("Failed to access avail idx at %p",
 			   &vrh->vring.avail->idx);
@@ -58,7 +55,7 @@ static inline int __vringh_get_head(const struct vringh *vrh,
 
 	i = *last_avail_idx & (vrh->vring.num - 1);
 
-	err = getu16(vrh, &head, &vrh->vring.avail->ring[i]);
+	err = vrh->ops.getu16(vrh, &head, &vrh->vring.avail->ring[i]);
 	if (err) {
 		vringh_bad("Failed to read head: idx %d address %p",
 			   *last_avail_idx, &vrh->vring.avail->ring[i]);
@@ -131,12 +128,10 @@ static inline ssize_t vringh_iov_xfer(struct vringh *vrh,
 
 /* May reduce *len if range is shorter. */
 static inline bool range_check(struct vringh *vrh, u64 addr, size_t *len,
-			       struct vringh_range *range,
-			       bool (*getrange)(struct vringh *,
-						u64, struct vringh_range *))
+			       struct vringh_range *range)
 {
 	if (addr < range->start || addr > range->end_incl) {
-		if (!getrange(vrh, addr, range))
+		if (!vrh->ops.getrange(vrh, addr, range))
 			return false;
 	}
 	BUG_ON(addr < range->start || addr > range->end_incl);
@@ -165,9 +160,7 @@ static inline bool range_check(struct vringh *vrh, u64 addr, size_t *len,
 }
 
 static inline bool no_range_check(struct vringh *vrh, u64 addr, size_t *len,
-				  struct vringh_range *range,
-				  bool (*getrange)(struct vringh *,
-						   u64, struct vringh_range *))
+				  struct vringh_range *range)
 {
 	return true;
 }
@@ -244,17 +237,7 @@ static u16 __cold return_from_indirect(const struct vringh *vrh, int *up_next,
 }
 
 static int slow_copy(struct vringh *vrh, void *dst, const void *src,
-		     bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
-				    struct vringh_range *range,
-				    bool (*getrange)(struct vringh *vrh,
-						     u64,
-						     struct vringh_range *)),
-		     bool (*getrange)(struct vringh *vrh,
-				      u64 addr,
-				      struct vringh_range *r),
-		     struct vringh_range *range,
-		     int (*copy)(const struct vringh *vrh,
-				 void *dst, const void *src, size_t len))
+		     struct vringh_range *range)
 {
 	size_t part, len = sizeof(struct vring_desc);
 
@@ -265,10 +248,10 @@ static int slow_copy(struct vringh *vrh, void *dst, const void *src,
 		part = len;
 		addr = (u64)(unsigned long)src - range->offset;
 
-		if (!rcheck(vrh, addr, &part, range, getrange))
+		if (!vrh->ops.range_check(vrh, addr, &part, range))
 			return -EINVAL;
 
-		err = copy(vrh, dst, src, part);
+		err = vrh->ops.copydesc(vrh, dst, src, part);
 		if (err)
 			return err;
 
@@ -279,18 +262,35 @@ static int slow_copy(struct vringh *vrh, void *dst, const void *src,
 	return 0;
 }
 
+static int __vringh_init(struct vringh *vrh, u64 features, unsigned int num,
+			 bool weak_barriers, gfp_t gfp, struct vring_desc *desc,
+			 struct vring_avail *avail, struct vring_used *used)
+{
+	/* Sane power of 2 please! */
+	if (!num || num > 0xffff || (num & (num - 1))) {
+		vringh_bad("Bad ring size %u", num);
+		return -EINVAL;
+	}
+
+	vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
+	vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
+	vrh->weak_barriers = weak_barriers;
+	vrh->completed = 0;
+	vrh->last_avail_idx = 0;
+	vrh->last_used_idx = 0;
+	vrh->vring.num = num;
+	vrh->vring.desc = desc;
+	vrh->vring.avail = avail;
+	vrh->vring.used = used;
+	vrh->desc_gfp = gfp;
+
+	return 0;
+}
+
 static inline int
 __vringh_iov(struct vringh *vrh, u16 i,
 	     struct vringh_kiov *riov,
-	     struct vringh_kiov *wiov,
-	     bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
-			    struct vringh_range *range,
-			    bool (*getrange)(struct vringh *, u64,
-					     struct vringh_range *)),
-	     bool (*getrange)(struct vringh *, u64, struct vringh_range *),
-	     gfp_t gfp,
-	     int (*copy)(const struct vringh *vrh,
-			 void *dst, const void *src, size_t len))
+	     struct vringh_kiov *wiov, gfp_t gfp)
 {
 	int err, count = 0, indirect_count = 0, up_next, desc_max;
 	struct vring_desc desc, *descs;
@@ -317,10 +317,9 @@ __vringh_iov(struct vringh *vrh, u16 i,
 		size_t len;
 
 		if (unlikely(slow))
-			err = slow_copy(vrh, &desc, &descs[i], rcheck, getrange,
-					&slowrange, copy);
+			err = slow_copy(vrh, &desc, &descs[i], &slowrange);
 		else
-			err = copy(vrh, &desc, &descs[i], sizeof(desc));
+			err = vrh->ops.copydesc(vrh, &desc, &descs[i], sizeof(desc));
 		if (unlikely(err))
 			goto fail;
 
@@ -330,7 +329,7 @@ __vringh_iov(struct vringh *vrh, u16 i,
 
 			/* Make sure it's OK, and get offset. */
 			len = vringh32_to_cpu(vrh, desc.len);
-			if (!rcheck(vrh, a, &len, &range, getrange)) {
+			if (!vrh->ops.range_check(vrh, a, &len, &range)) {
 				err = -EINVAL;
 				goto fail;
 			}
@@ -382,8 +381,7 @@ __vringh_iov(struct vringh *vrh, u16 i,
 	again:
 		/* Make sure it's OK, and get offset. */
 		len = vringh32_to_cpu(vrh, desc.len);
-		if (!rcheck(vrh, vringh64_to_cpu(vrh, desc.addr), &len, &range,
-			    getrange)) {
+		if (!vrh->ops.range_check(vrh, vringh64_to_cpu(vrh, desc.addr), &len, &range)) {
 			err = -EINVAL;
 			goto fail;
 		}
@@ -436,13 +434,7 @@ __vringh_iov(struct vringh *vrh, u16 i,
 
 static inline int __vringh_complete(struct vringh *vrh,
 				    const struct vring_used_elem *used,
-				    unsigned int num_used,
-				    int (*putu16)(const struct vringh *vrh,
-						  __virtio16 *p, u16 val),
-				    int (*putused)(const struct vringh *vrh,
-						   struct vring_used_elem *dst,
-						   const struct vring_used_elem
-						   *src, unsigned num))
+				    unsigned int num_used)
 {
 	struct vring_used *used_ring;
 	int err;
@@ -456,12 +448,12 @@ static inline int __vringh_complete(struct vringh *vrh,
 	/* Compiler knows num_used == 1 sometimes, hence extra check */
 	if (num_used > 1 && unlikely(off + num_used >= vrh->vring.num)) {
 		u16 part = vrh->vring.num - off;
-		err = putused(vrh, &used_ring->ring[off], used, part);
+		err = vrh->ops.putused(vrh, &used_ring->ring[off], used, part);
 		if (!err)
-			err = putused(vrh, &used_ring->ring[0], used + part,
+			err = vrh->ops.putused(vrh, &used_ring->ring[0], used + part,
 				      num_used - part);
 	} else
-		err = putused(vrh, &used_ring->ring[off], used, num_used);
+		err = vrh->ops.putused(vrh, &used_ring->ring[off], used, num_used);
 
 	if (err) {
 		vringh_bad("Failed to write %u used entries %u at %p",
@@ -472,7 +464,7 @@ static inline int __vringh_complete(struct vringh *vrh,
 	/* Make sure buffer is written before we update index. */
 	virtio_wmb(vrh->weak_barriers);
 
-	err = putu16(vrh, &vrh->vring.used->idx, used_idx + num_used);
+	err = vrh->ops.putu16(vrh, &vrh->vring.used->idx, used_idx + num_used);
 	if (err) {
 		vringh_bad("Failed to update used index at %p",
 			   &vrh->vring.used->idx);
@@ -483,11 +475,13 @@ static inline int __vringh_complete(struct vringh *vrh,
 	return 0;
 }
 
-
-static inline int __vringh_need_notify(struct vringh *vrh,
-				       int (*getu16)(const struct vringh *vrh,
-						     u16 *val,
-						     const __virtio16 *p))
+/**
+ * vringh_need_notify - must we tell the other side about used buffers?
+ * @vrh: the vring we've called vringh_complete() on.
+ *
+ * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
+ */
+int vringh_need_notify(struct vringh *vrh)
 {
 	bool notify;
 	u16 used_event;
@@ -501,7 +495,7 @@ static inline int __vringh_need_notify(struct vringh *vrh,
 	/* Old-style, without event indices. */
 	if (!vrh->event_indices) {
 		u16 flags;
-		err = getu16(vrh, &flags, &vrh->vring.avail->flags);
+		err = vrh->ops.getu16(vrh, &flags, &vrh->vring.avail->flags);
 		if (err) {
 			vringh_bad("Failed to get flags at %p",
 				   &vrh->vring.avail->flags);
@@ -511,7 +505,7 @@ static inline int __vringh_need_notify(struct vringh *vrh,
 	}
 
 	/* Modern: we know when other side wants to know. */
-	err = getu16(vrh, &used_event, &vring_used_event(&vrh->vring));
+	err = vrh->ops.getu16(vrh, &used_event, &vring_used_event(&vrh->vring));
 	if (err) {
 		vringh_bad("Failed to get used event idx at %p",
 			   &vring_used_event(&vrh->vring));
@@ -530,24 +524,28 @@ static inline int __vringh_need_notify(struct vringh *vrh,
 	vrh->completed = 0;
 	return notify;
 }
+EXPORT_SYMBOL(vringh_need_notify);
 
-static inline bool __vringh_notify_enable(struct vringh *vrh,
-					  int (*getu16)(const struct vringh *vrh,
-							u16 *val, const __virtio16 *p),
-					  int (*putu16)(const struct vringh *vrh,
-							__virtio16 *p, u16 val))
+/**
+ * vringh_notify_enable - we want to know if something changes.
+ * @vrh: the vring.
+ *
+ * This always enables notifications, but returns false if there are
+ * now more buffers available in the vring.
+ */
+bool vringh_notify_enable(struct vringh *vrh)
 {
 	u16 avail;
 
 	if (!vrh->event_indices) {
 		/* Old-school; update flags. */
-		if (putu16(vrh, &vrh->vring.used->flags, 0) != 0) {
+		if (vrh->ops.putu16(vrh, &vrh->vring.used->flags, 0) != 0) {
 			vringh_bad("Clearing used flags %p",
 				   &vrh->vring.used->flags);
 			return true;
 		}
 	} else {
-		if (putu16(vrh, &vring_avail_event(&vrh->vring),
+		if (vrh->ops.putu16(vrh, &vring_avail_event(&vrh->vring),
 			   vrh->last_avail_idx) != 0) {
 			vringh_bad("Updating avail event index %p",
 				   &vring_avail_event(&vrh->vring));
@@ -559,7 +557,7 @@ static inline bool __vringh_notify_enable(struct vringh *vrh,
 	 * sure it's written, then check again. */
 	virtio_mb(vrh->weak_barriers);
 
-	if (getu16(vrh, &avail, &vrh->vring.avail->idx) != 0) {
+	if (vrh->ops.getu16(vrh, &avail, &vrh->vring.avail->idx) != 0) {
 		vringh_bad("Failed to check avail idx at %p",
 			   &vrh->vring.avail->idx);
 		return true;
@@ -570,20 +568,27 @@ static inline bool __vringh_notify_enable(struct vringh *vrh,
 	 * notification anyway). */
 	return avail == vrh->last_avail_idx;
 }
+EXPORT_SYMBOL(vringh_notify_enable);
 
-static inline void __vringh_notify_disable(struct vringh *vrh,
-					   int (*putu16)(const struct vringh *vrh,
-							 __virtio16 *p, u16 val))
+/**
+ * vringh_notify_disable - don't tell us if something changes.
+ * @vrh: the vring.
+ *
+ * This is our normal running state: we disable and then only enable when
+ * we're going to sleep.
+ */
+void vringh_notify_disable(struct vringh *vrh)
 {
 	if (!vrh->event_indices) {
 		/* Old-school; update flags. */
-		if (putu16(vrh, &vrh->vring.used->flags,
+		if (vrh->ops.putu16(vrh, &vrh->vring.used->flags,
 			   VRING_USED_F_NO_NOTIFY)) {
 			vringh_bad("Setting used flags %p",
 				   &vrh->vring.used->flags);
 		}
 	}
 }
+EXPORT_SYMBOL(vringh_notify_disable);
 
 /* Userspace access helpers: in this case, addresses are really userspace. */
 static inline int getu16_user(const struct vringh *vrh, u16 *val, const __virtio16 *p)
@@ -630,6 +635,16 @@ static inline int xfer_to_user(const struct vringh *vrh,
 		-EFAULT : 0;
 }
 
+static struct vringh_ops user_vringh_ops = {
+	.getu16 = getu16_user,
+	.putu16 = putu16_user,
+	.xfer_from = xfer_from_user,
+	.xfer_to = xfer_to_user,
+	.putused = putused_user,
+	.copydesc = copydesc_user,
+	.range_check = range_check,
+};
+
 /**
  * vringh_init_user - initialize a vringh for a userspace vring.
  * @vrh: the vringh to initialize.
@@ -639,6 +654,7 @@ static inline int xfer_to_user(const struct vringh *vrh,
  * @desc: the userpace descriptor pointer.
  * @avail: the userpace avail pointer.
  * @used: the userpace used pointer.
+ * @getrange: a function that return a range that vring can access.
  *
  * Returns an error if num is invalid: you should check pointers
  * yourself!
@@ -647,36 +663,32 @@ int vringh_init_user(struct vringh *vrh, u64 features,
 		     unsigned int num, bool weak_barriers,
 		     vring_desc_t __user *desc,
 		     vring_avail_t __user *avail,
-		     vring_used_t __user *used)
+		     vring_used_t __user *used,
+			 bool (*getrange)(struct vringh *vrh, u64 addr, struct vringh_range *r))
 {
-	/* Sane power of 2 please! */
-	if (!num || num > 0xffff || (num & (num - 1))) {
-		vringh_bad("Bad ring size %u", num);
-		return -EINVAL;
-	}
+	int err;
+
+	err = __vringh_init(vrh, features, num, weak_barriers, GFP_KERNEL,
+			(__force struct vring_desc *)desc,
+			(__force struct vring_avail *)avail,
+			(__force struct vring_used *)used);
+	if (err)
+		return err;
+
+	memcpy(&vrh->ops, &user_vringh_ops, sizeof(user_vringh_ops));
+	vrh->ops.getrange = getrange;
 
-	vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
-	vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
-	vrh->weak_barriers = weak_barriers;
-	vrh->completed = 0;
-	vrh->last_avail_idx = 0;
-	vrh->last_used_idx = 0;
-	vrh->vring.num = num;
-	/* vring expects kernel addresses, but only used via accessors. */
-	vrh->vring.desc = (__force struct vring_desc *)desc;
-	vrh->vring.avail = (__force struct vring_avail *)avail;
-	vrh->vring.used = (__force struct vring_used *)used;
 	return 0;
 }
 EXPORT_SYMBOL(vringh_init_user);
 
 /**
- * vringh_getdesc_user - get next available descriptor from userspace ring.
- * @vrh: the userspace vring.
+ * vringh_getdesc - get next available descriptor from ring.
+ * @vrh: the vringh to get desc.
  * @riov: where to put the readable descriptors (or NULL)
  * @wiov: where to put the writable descriptors (or NULL)
  * @getrange: function to call to check ranges.
- * @head: head index we received, for passing to vringh_complete_user().
+ * @head: head index we received, for passing to vringh_complete().
  *
  * Returns 0 if there was no descriptor, 1 if there was, or -errno.
  *
@@ -690,17 +702,15 @@ EXPORT_SYMBOL(vringh_init_user);
  * When you don't have to use riov and wiov anymore, you should clean up them
  * calling vringh_iov_cleanup() to release the memory, even on error!
  */
-int vringh_getdesc_user(struct vringh *vrh,
+int vringh_getdesc(struct vringh *vrh,
 			struct vringh_kiov *riov,
 			struct vringh_kiov *wiov,
-			bool (*getrange)(struct vringh *vrh,
-					 u64 addr, struct vringh_range *r),
 			u16 *head)
 {
 	int err;
 
 	*head = vrh->vring.num;
-	err = __vringh_get_head(vrh, getu16_user, &vrh->last_avail_idx);
+	err = __vringh_get_head(vrh, &vrh->last_avail_idx);
 	if (err < 0)
 		return err;
 
@@ -709,137 +719,100 @@ int vringh_getdesc_user(struct vringh *vrh,
 		return 0;
 
 	*head = err;
-	err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
-			   (struct vringh_kiov *)wiov,
-			   range_check, getrange, GFP_KERNEL, copydesc_user);
+	err = __vringh_iov(vrh, *head, riov, wiov, GFP_KERNEL);
 	if (err)
 		return err;
 
 	return 1;
 }
-EXPORT_SYMBOL(vringh_getdesc_user);
+EXPORT_SYMBOL(vringh_getdesc);
 
 /**
- * vringh_iov_pull_user - copy bytes from vring_kiov.
- * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
+ * vringh_iov_pull - copy bytes from vring_kiov.
+ * @vrh: the vringh to load data.
+ * @riov: the riov as passed to vringh_getdesc() (updated as we consume)
  * @dst: the place to copy.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_pull_user(struct vringh_kiov *riov, void *dst, size_t len)
+ssize_t vringh_iov_pull(struct vringh *vrh, struct vringh_kiov *riov, void *dst, size_t len)
 {
 	return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
-			       dst, len, xfer_from_user);
+			       dst, len, vrh->ops.xfer_from);
 }
-EXPORT_SYMBOL(vringh_iov_pull_user);
+EXPORT_SYMBOL(vringh_iov_pull);
 
 /**
- * vringh_iov_push_user - copy bytes into vring_kiov.
- * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
+ * vringh_iov_push - copy bytes into vring_kiov.
+ * @vrh: the vringh to store data.
+ * @wiov: the wiov as passed to vringh_getdesc() (updated as we consume)
  * @src: the place to copy from.
  * @len: the maximum length to copy.
  *
  * Returns the bytes copied <= len or a negative errno.
  */
-ssize_t vringh_iov_push_user(struct vringh_kiov *wiov,
+ssize_t vringh_iov_push(struct vringh *vrh, struct vringh_kiov *wiov,
 			     const void *src, size_t len)
 {
 	return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
-			       (void *)src, len, xfer_to_user);
+			       (void *)src, len, vrh->ops.xfer_to);
 }
-EXPORT_SYMBOL(vringh_iov_push_user);
+EXPORT_SYMBOL(vringh_iov_push);
 
 /**
- * vringh_abandon_user - we've decided not to handle the descriptor(s).
+ * vringh_abandon - we've decided not to handle the descriptor(s).
  * @vrh: the vring.
  * @num: the number of descriptors to put back (ie. num
  *	 vringh_get_user() to undo).
  *
  * The next vringh_get_user() will return the old descriptor(s) again.
  */
-void vringh_abandon_user(struct vringh *vrh, unsigned int num)
+void vringh_abandon(struct vringh *vrh, unsigned int num)
 {
 	/* We only update vring_avail_event(vr) when we want to be notified,
 	 * so we haven't changed that yet. */
 	vrh->last_avail_idx -= num;
 }
-EXPORT_SYMBOL(vringh_abandon_user);
+EXPORT_SYMBOL(vringh_abandon);
 
 /**
- * vringh_complete_user - we've finished with descriptor, publish it.
+ * vringh_complete - we've finished with descriptor, publish it.
  * @vrh: the vring.
- * @head: the head as filled in by vringh_getdesc_user.
+ * @head: the head as filled in by vringh_getdesc.
  * @len: the length of data we have written.
  *
- * You should check vringh_need_notify_user() after one or more calls
+ * You should check vringh_need_notify() after one or more calls
  * to this function.
  */
-int vringh_complete_user(struct vringh *vrh, u16 head, u32 len)
+int vringh_complete(struct vringh *vrh, u16 head, u32 len)
 {
 	struct vring_used_elem used;
 
 	used.id = cpu_to_vringh32(vrh, head);
 	used.len = cpu_to_vringh32(vrh, len);
-	return __vringh_complete(vrh, &used, 1, putu16_user, putused_user);
+	return __vringh_complete(vrh, &used, 1);
 }
-EXPORT_SYMBOL(vringh_complete_user);
+EXPORT_SYMBOL(vringh_complete);
 
 /**
- * vringh_complete_multi_user - we've finished with many descriptors.
+ * vringh_complete_multi - we've finished with many descriptors.
  * @vrh: the vring.
  * @used: the head, length pairs.
  * @num_used: the number of used elements.
  *
- * You should check vringh_need_notify_user() after one or more calls
+ * You should check vringh_need_notify() after one or more calls
  * to this function.
  */
-int vringh_complete_multi_user(struct vringh *vrh,
+int vringh_complete_multi(struct vringh *vrh,
 			       const struct vring_used_elem used[],
 			       unsigned num_used)
 {
-	return __vringh_complete(vrh, used, num_used,
-				 putu16_user, putused_user);
-}
-EXPORT_SYMBOL(vringh_complete_multi_user);
-
-/**
- * vringh_notify_enable_user - we want to know if something changes.
- * @vrh: the vring.
- *
- * This always enables notifications, but returns false if there are
- * now more buffers available in the vring.
- */
-bool vringh_notify_enable_user(struct vringh *vrh)
-{
-	return __vringh_notify_enable(vrh, getu16_user, putu16_user);
+	return __vringh_complete(vrh, used, num_used);
 }
-EXPORT_SYMBOL(vringh_notify_enable_user);
+EXPORT_SYMBOL(vringh_complete_multi);
 
-/**
- * vringh_notify_disable_user - don't tell us if something changes.
- * @vrh: the vring.
- *
- * This is our normal running state: we disable and then only enable when
- * we're going to sleep.
- */
-void vringh_notify_disable_user(struct vringh *vrh)
-{
-	__vringh_notify_disable(vrh, putu16_user);
-}
-EXPORT_SYMBOL(vringh_notify_disable_user);
 
-/**
- * vringh_need_notify_user - must we tell the other side about used buffers?
- * @vrh: the vring we've called vringh_complete_user() on.
- *
- * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
- */
-int vringh_need_notify_user(struct vringh *vrh)
-{
-	return __vringh_need_notify(vrh, getu16_user);
-}
-EXPORT_SYMBOL(vringh_need_notify_user);
 
 /* Kernelspace access helpers. */
 static inline int getu16_kern(const struct vringh *vrh,
@@ -885,6 +858,17 @@ static inline int kern_xfer(const struct vringh *vrh, void *dst,
 	return 0;
 }
 
+static const struct vringh_ops kern_vringh_ops = {
+	.getu16 = getu16_kern,
+	.putu16 = putu16_kern,
+	.xfer_from = xfer_kern,
+	.xfer_to = xfer_kern,
+	.putused = putused_kern,
+	.copydesc = copydesc_kern,
+	.range_check = no_range_check,
+	.getrange = NULL,
+};
+
 /**
  * vringh_init_kern - initialize a vringh for a kernelspace vring.
  * @vrh: the vringh to initialize.
@@ -898,179 +882,22 @@ static inline int kern_xfer(const struct vringh *vrh, void *dst,
  * Returns an error if num is invalid.
  */
 int vringh_init_kern(struct vringh *vrh, u64 features,
-		     unsigned int num, bool weak_barriers,
+		     unsigned int num, bool weak_barriers, gfp_t gfp,
 		     struct vring_desc *desc,
 		     struct vring_avail *avail,
 		     struct vring_used *used)
-{
-	/* Sane power of 2 please! */
-	if (!num || num > 0xffff || (num & (num - 1))) {
-		vringh_bad("Bad ring size %u", num);
-		return -EINVAL;
-	}
-
-	vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
-	vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
-	vrh->weak_barriers = weak_barriers;
-	vrh->completed = 0;
-	vrh->last_avail_idx = 0;
-	vrh->last_used_idx = 0;
-	vrh->vring.num = num;
-	vrh->vring.desc = desc;
-	vrh->vring.avail = avail;
-	vrh->vring.used = used;
-	return 0;
-}
-EXPORT_SYMBOL(vringh_init_kern);
-
-/**
- * vringh_getdesc_kern - get next available descriptor from kernelspace ring.
- * @vrh: the kernelspace vring.
- * @riov: where to put the readable descriptors (or NULL)
- * @wiov: where to put the writable descriptors (or NULL)
- * @head: head index we received, for passing to vringh_complete_kern().
- * @gfp: flags for allocating larger riov/wiov.
- *
- * Returns 0 if there was no descriptor, 1 if there was, or -errno.
- *
- * Note that on error return, you can tell the difference between an
- * invalid ring and a single invalid descriptor: in the former case,
- * *head will be vrh->vring.num.  You may be able to ignore an invalid
- * descriptor, but there's not much you can do with an invalid ring.
- *
- * Note that you can reuse riov and wiov with subsequent calls. Content is
- * overwritten and memory reallocated if more space is needed.
- * When you don't have to use riov and wiov anymore, you should clean up them
- * calling vringh_kiov_cleanup() to release the memory, even on error!
- */
-int vringh_getdesc_kern(struct vringh *vrh,
-			struct vringh_kiov *riov,
-			struct vringh_kiov *wiov,
-			u16 *head,
-			gfp_t gfp)
 {
 	int err;
 
-	err = __vringh_get_head(vrh, getu16_kern, &vrh->last_avail_idx);
-	if (err < 0)
-		return err;
-
-	/* Empty... */
-	if (err == vrh->vring.num)
-		return 0;
-
-	*head = err;
-	err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
-			   gfp, copydesc_kern);
+	err = __vringh_init(vrh, features, num, weak_barriers, gfp, desc, avail, used);
 	if (err)
 		return err;
 
-	return 1;
-}
-EXPORT_SYMBOL(vringh_getdesc_kern);
-
-/**
- * vringh_iov_pull_kern - copy bytes from vring_iov.
- * @riov: the riov as passed to vringh_getdesc_kern() (updated as we consume)
- * @dst: the place to copy.
- * @len: the maximum length to copy.
- *
- * Returns the bytes copied <= len or a negative errno.
- */
-ssize_t vringh_iov_pull_kern(struct vringh_kiov *riov, void *dst, size_t len)
-{
-	return vringh_iov_xfer(NULL, riov, dst, len, xfer_kern);
-}
-EXPORT_SYMBOL(vringh_iov_pull_kern);
-
-/**
- * vringh_iov_push_kern - copy bytes into vring_iov.
- * @wiov: the wiov as passed to vringh_getdesc_kern() (updated as we consume)
- * @src: the place to copy from.
- * @len: the maximum length to copy.
- *
- * Returns the bytes copied <= len or a negative errno.
- */
-ssize_t vringh_iov_push_kern(struct vringh_kiov *wiov,
-			     const void *src, size_t len)
-{
-	return vringh_iov_xfer(NULL, wiov, (void *)src, len, kern_xfer);
-}
-EXPORT_SYMBOL(vringh_iov_push_kern);
+	memcpy(&vrh->ops, &kern_vringh_ops, sizeof(kern_vringh_ops));
 
-/**
- * vringh_abandon_kern - we've decided not to handle the descriptor(s).
- * @vrh: the vring.
- * @num: the number of descriptors to put back (ie. num
- *	 vringh_get_kern() to undo).
- *
- * The next vringh_get_kern() will return the old descriptor(s) again.
- */
-void vringh_abandon_kern(struct vringh *vrh, unsigned int num)
-{
-	/* We only update vring_avail_event(vr) when we want to be notified,
-	 * so we haven't changed that yet. */
-	vrh->last_avail_idx -= num;
-}
-EXPORT_SYMBOL(vringh_abandon_kern);
-
-/**
- * vringh_complete_kern - we've finished with descriptor, publish it.
- * @vrh: the vring.
- * @head: the head as filled in by vringh_getdesc_kern.
- * @len: the length of data we have written.
- *
- * You should check vringh_need_notify_kern() after one or more calls
- * to this function.
- */
-int vringh_complete_kern(struct vringh *vrh, u16 head, u32 len)
-{
-	struct vring_used_elem used;
-
-	used.id = cpu_to_vringh32(vrh, head);
-	used.len = cpu_to_vringh32(vrh, len);
-
-	return __vringh_complete(vrh, &used, 1, putu16_kern, putused_kern);
-}
-EXPORT_SYMBOL(vringh_complete_kern);
-
-/**
- * vringh_notify_enable_kern - we want to know if something changes.
- * @vrh: the vring.
- *
- * This always enables notifications, but returns false if there are
- * now more buffers available in the vring.
- */
-bool vringh_notify_enable_kern(struct vringh *vrh)
-{
-	return __vringh_notify_enable(vrh, getu16_kern, putu16_kern);
-}
-EXPORT_SYMBOL(vringh_notify_enable_kern);
-
-/**
- * vringh_notify_disable_kern - don't tell us if something changes.
- * @vrh: the vring.
- *
- * This is our normal running state: we disable and then only enable when
- * we're going to sleep.
- */
-void vringh_notify_disable_kern(struct vringh *vrh)
-{
-	__vringh_notify_disable(vrh, putu16_kern);
-}
-EXPORT_SYMBOL(vringh_notify_disable_kern);
-
-/**
- * vringh_need_notify_kern - must we tell the other side about used buffers?
- * @vrh: the vring we've called vringh_complete_kern() on.
- *
- * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
- */
-int vringh_need_notify_kern(struct vringh *vrh)
-{
-	return __vringh_need_notify(vrh, getu16_kern);
+	return 0;
 }
-EXPORT_SYMBOL(vringh_need_notify_kern);
+EXPORT_SYMBOL(vringh_init_kern);
 
 #if IS_REACHABLE(CONFIG_VHOST_IOTLB)
 
@@ -1122,7 +949,7 @@ static int iotlb_translate(const struct vringh *vrh,
 	return ret;
 }
 
-static inline int copy_from_iotlb(const struct vringh *vrh, void *dst,
+static int copy_from_iotlb(const struct vringh *vrh, void *dst,
 				  void *src, size_t len)
 {
 	u64 total_translated = 0;
@@ -1155,7 +982,7 @@ static inline int copy_from_iotlb(const struct vringh *vrh, void *dst,
 	return total_translated;
 }
 
-static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
+static int copy_to_iotlb(const struct vringh *vrh, void *dst,
 				void *src, size_t len)
 {
 	u64 total_translated = 0;
@@ -1188,7 +1015,7 @@ static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
 	return total_translated;
 }
 
-static inline int getu16_iotlb(const struct vringh *vrh,
+static int getu16_iotlb(const struct vringh *vrh,
 			       u16 *val, const __virtio16 *p)
 {
 	struct bio_vec iov;
@@ -1209,7 +1036,7 @@ static inline int getu16_iotlb(const struct vringh *vrh,
 	return 0;
 }
 
-static inline int putu16_iotlb(const struct vringh *vrh,
+static int putu16_iotlb(const struct vringh *vrh,
 			       __virtio16 *p, u16 val)
 {
 	struct bio_vec iov;
@@ -1230,7 +1057,7 @@ static inline int putu16_iotlb(const struct vringh *vrh,
 	return 0;
 }
 
-static inline int copydesc_iotlb(const struct vringh *vrh,
+static int copydesc_iotlb(const struct vringh *vrh,
 				 void *dst, const void *src, size_t len)
 {
 	int ret;
@@ -1242,7 +1069,7 @@ static inline int copydesc_iotlb(const struct vringh *vrh,
 	return 0;
 }
 
-static inline int xfer_from_iotlb(const struct vringh *vrh, void *src,
+static int xfer_from_iotlb(const struct vringh *vrh, void *src,
 				  void *dst, size_t len)
 {
 	int ret;
@@ -1254,7 +1081,7 @@ static inline int xfer_from_iotlb(const struct vringh *vrh, void *src,
 	return 0;
 }
 
-static inline int xfer_to_iotlb(const struct vringh *vrh,
+static int xfer_to_iotlb(const struct vringh *vrh,
 			       void *dst, void *src, size_t len)
 {
 	int ret;
@@ -1266,7 +1093,7 @@ static inline int xfer_to_iotlb(const struct vringh *vrh,
 	return 0;
 }
 
-static inline int putused_iotlb(const struct vringh *vrh,
+static int putused_iotlb(const struct vringh *vrh,
 				struct vring_used_elem *dst,
 				const struct vring_used_elem *src,
 				unsigned int num)
@@ -1281,6 +1108,17 @@ static inline int putused_iotlb(const struct vringh *vrh,
 	return 0;
 }
 
+static const struct vringh_ops iotlb_vringh_ops = {
+	.getu16 = getu16_iotlb,
+	.putu16 = putu16_iotlb,
+	.xfer_from = xfer_from_iotlb,
+	.xfer_to = xfer_to_iotlb,
+	.putused = putused_iotlb,
+	.copydesc = copydesc_iotlb,
+	.range_check = no_range_check,
+	.getrange = NULL,
+};
+
 /**
  * vringh_init_iotlb - initialize a vringh for a ring with IOTLB.
  * @vrh: the vringh to initialize.
@@ -1294,13 +1132,20 @@ static inline int putused_iotlb(const struct vringh *vrh,
  * Returns an error if num is invalid.
  */
 int vringh_init_iotlb(struct vringh *vrh, u64 features,
-		      unsigned int num, bool weak_barriers,
+		      unsigned int num, bool weak_barriers, gfp_t gfp,
 		      struct vring_desc *desc,
 		      struct vring_avail *avail,
 		      struct vring_used *used)
 {
-	return vringh_init_kern(vrh, features, num, weak_barriers,
-				desc, avail, used);
+	int err;
+
+	err = __vringh_init(vrh, features, num, weak_barriers, gfp, desc, avail, used);
+	if (err)
+		return err;
+
+	memcpy(&vrh->ops, &iotlb_vringh_ops, sizeof(iotlb_vringh_ops));
+
+	return 0;
 }
 EXPORT_SYMBOL(vringh_init_iotlb);
 
@@ -1318,162 +1163,6 @@ void vringh_set_iotlb(struct vringh *vrh, struct vhost_iotlb *iotlb,
 }
 EXPORT_SYMBOL(vringh_set_iotlb);
 
-/**
- * vringh_getdesc_iotlb - get next available descriptor from ring with
- * IOTLB.
- * @vrh: the kernelspace vring.
- * @riov: where to put the readable descriptors (or NULL)
- * @wiov: where to put the writable descriptors (or NULL)
- * @head: head index we received, for passing to vringh_complete_iotlb().
- * @gfp: flags for allocating larger riov/wiov.
- *
- * Returns 0 if there was no descriptor, 1 if there was, or -errno.
- *
- * Note that on error return, you can tell the difference between an
- * invalid ring and a single invalid descriptor: in the former case,
- * *head will be vrh->vring.num.  You may be able to ignore an invalid
- * descriptor, but there's not much you can do with an invalid ring.
- *
- * Note that you can reuse riov and wiov with subsequent calls. Content is
- * overwritten and memory reallocated if more space is needed.
- * When you don't have to use riov and wiov anymore, you should clean up them
- * calling vringh_kiov_cleanup() to release the memory, even on error!
- */
-int vringh_getdesc_iotlb(struct vringh *vrh,
-			 struct vringh_kiov *riov,
-			 struct vringh_kiov *wiov,
-			 u16 *head,
-			 gfp_t gfp)
-{
-	int err;
-
-	err = __vringh_get_head(vrh, getu16_iotlb, &vrh->last_avail_idx);
-	if (err < 0)
-		return err;
-
-	/* Empty... */
-	if (err == vrh->vring.num)
-		return 0;
-
-	*head = err;
-	err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
-			   gfp, copydesc_iotlb);
-	if (err)
-		return err;
-
-	return 1;
-}
-EXPORT_SYMBOL(vringh_getdesc_iotlb);
-
-/**
- * vringh_iov_pull_iotlb - copy bytes from vring_iov.
- * @vrh: the vring.
- * @riov: the riov as passed to vringh_getdesc_iotlb() (updated as we consume)
- * @dst: the place to copy.
- * @len: the maximum length to copy.
- *
- * Returns the bytes copied <= len or a negative errno.
- */
-ssize_t vringh_iov_pull_iotlb(struct vringh *vrh,
-			      struct vringh_kiov *riov,
-			      void *dst, size_t len)
-{
-	return vringh_iov_xfer(vrh, riov, dst, len, xfer_from_iotlb);
-}
-EXPORT_SYMBOL(vringh_iov_pull_iotlb);
-
-/**
- * vringh_iov_push_iotlb - copy bytes into vring_iov.
- * @vrh: the vring.
- * @wiov: the wiov as passed to vringh_getdesc_iotlb() (updated as we consume)
- * @src: the place to copy from.
- * @len: the maximum length to copy.
- *
- * Returns the bytes copied <= len or a negative errno.
- */
-ssize_t vringh_iov_push_iotlb(struct vringh *vrh,
-			      struct vringh_kiov *wiov,
-			      const void *src, size_t len)
-{
-	return vringh_iov_xfer(vrh, wiov, (void *)src, len, xfer_to_iotlb);
-}
-EXPORT_SYMBOL(vringh_iov_push_iotlb);
-
-/**
- * vringh_abandon_iotlb - we've decided not to handle the descriptor(s).
- * @vrh: the vring.
- * @num: the number of descriptors to put back (ie. num
- *	 vringh_get_iotlb() to undo).
- *
- * The next vringh_get_iotlb() will return the old descriptor(s) again.
- */
-void vringh_abandon_iotlb(struct vringh *vrh, unsigned int num)
-{
-	/* We only update vring_avail_event(vr) when we want to be notified,
-	 * so we haven't changed that yet.
-	 */
-	vrh->last_avail_idx -= num;
-}
-EXPORT_SYMBOL(vringh_abandon_iotlb);
-
-/**
- * vringh_complete_iotlb - we've finished with descriptor, publish it.
- * @vrh: the vring.
- * @head: the head as filled in by vringh_getdesc_iotlb.
- * @len: the length of data we have written.
- *
- * You should check vringh_need_notify_iotlb() after one or more calls
- * to this function.
- */
-int vringh_complete_iotlb(struct vringh *vrh, u16 head, u32 len)
-{
-	struct vring_used_elem used;
-
-	used.id = cpu_to_vringh32(vrh, head);
-	used.len = cpu_to_vringh32(vrh, len);
-
-	return __vringh_complete(vrh, &used, 1, putu16_iotlb, putused_iotlb);
-}
-EXPORT_SYMBOL(vringh_complete_iotlb);
-
-/**
- * vringh_notify_enable_iotlb - we want to know if something changes.
- * @vrh: the vring.
- *
- * This always enables notifications, but returns false if there are
- * now more buffers available in the vring.
- */
-bool vringh_notify_enable_iotlb(struct vringh *vrh)
-{
-	return __vringh_notify_enable(vrh, getu16_iotlb, putu16_iotlb);
-}
-EXPORT_SYMBOL(vringh_notify_enable_iotlb);
-
-/**
- * vringh_notify_disable_iotlb - don't tell us if something changes.
- * @vrh: the vring.
- *
- * This is our normal running state: we disable and then only enable when
- * we're going to sleep.
- */
-void vringh_notify_disable_iotlb(struct vringh *vrh)
-{
-	__vringh_notify_disable(vrh, putu16_iotlb);
-}
-EXPORT_SYMBOL(vringh_notify_disable_iotlb);
-
-/**
- * vringh_need_notify_iotlb - must we tell the other side about used buffers?
- * @vrh: the vring we've called vringh_complete_iotlb() on.
- *
- * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
- */
-int vringh_need_notify_iotlb(struct vringh *vrh)
-{
-	return __vringh_need_notify(vrh, getu16_iotlb);
-}
-EXPORT_SYMBOL(vringh_need_notify_iotlb);
-
 #endif
 
 MODULE_LICENSE("GPL");
include/linux/vringh.h +47 -53

156 lines not on this page · raw message

Signature · 7 lines
2.25.1


_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
#15 lore raw

Re: [RFC PATCH 4/9] vringh: unify the APIs for all accessors

Quoted 26 lines · Shunsuke Mie

Each vringh memory accessors that are for user, kern and iotlb has own interfaces that calls common code. But some codes are duplicated and that becomes loss extendability.

Introduce a struct vringh_ops and provide a common APIs for all accessors. It can bee easily extended vringh code for new memory accessor and simplified a caller code.

---
 drivers/vhost/vringh.c | 667 +++++++++++------------------------------
 include/linux/vringh.h | 100 +++---
 2 files changed, 225 insertions(+), 542 deletions(-)
drivers/vhost/vringh.c +0 -4
index aa3cd27d2384..ebfd3644a1a3 100644
@@ -35,15 +35,12 @@ static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
 }
 
 /* Returns vring->num if empty, -ve on error. */
-static inline int __vringh_get_head(const struct vringh *vrh,
-				    int (*getu16)(const struct vringh *vrh,
-						  u16 *val, const __virtio16 *p),
-				    u16 *last_avail_idx)

drivers/vhost/vringh.c static __printf(1,2) __cold void vringh_bad(const char *fmt, ...) · line 46

Quoted 10 lines
drivers/vhost/vringh.c +2 -1
+static inline int __vringh_get_head(const struct vringh *vrh, u16 *last_avail_idx)
 {
 	u16 avail_idx, i, head;
 	int err;
 
-	err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
+	err = vrh->ops.getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
 	if (err) {
 		vringh_bad("Failed to access avail idx at %p",
 			   &vrh->vring.avail->idx);

I like that this patch removes more lines of code than it adds.

However one of the design points of vringh abstractions is that they were carefully written to be very low overhead.
This is why we are passing function pointers to inline functions - compiler can optimize that out.

I think that introducing ops indirect functions calls here is going to break these assumptions and hurt performance.
Unless compiler can somehow figure it out and optimize?
I don't see how it's possible with ops pointer in memory
but maybe I'm wrong.

Was any effort taken to test effect of these patches on performance?

Thanks!

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#16 lore raw

Re: [RFC PATCH 4/9] vringh: unify the APIs for all accessors

2022年12月27日(火) 16:04 Michael S. Tsirkin <[email protected]>:

Quoted 46 lines · Shunsuke Mie
Quoted 26 lines

Each vringh memory accessors that are for user, kern and iotlb has own interfaces that calls common code. But some codes are duplicated and that becomes loss extendability.

Introduce a struct vringh_ops and provide a common APIs for all accessors. It can bee easily extended vringh code for new memory accessor and simplified a caller code.

---
 drivers/vhost/vringh.c | 667 +++++++++++------------------------------
 include/linux/vringh.h | 100 +++---
 2 files changed, 225 insertions(+), 542 deletions(-)
drivers/vhost/vringh.c +0 -4
index aa3cd27d2384..ebfd3644a1a3 100644
@@ -35,15 +35,12 @@ static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
 }
 
 /* Returns vring->num if empty, -ve on error. */
-static inline int __vringh_get_head(const struct vringh *vrh,
-                                 int (*getu16)(const struct vringh *vrh,
-                                               u16 *val, const __virtio16 *p),
-                                 u16 *last_avail_idx)

drivers/vhost/vringh.c static __printf(1,2) __cold void vringh_bad(const char *fmt, ...) · line 46

Quoted 10 lines
drivers/vhost/vringh.c +2 -1
+static inline int __vringh_get_head(const struct vringh *vrh, u16 *last_avail_idx)
 {
      u16 avail_idx, i, head;
      int err;
 
-     err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
+     err = vrh->ops.getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
      if (err) {
              vringh_bad("Failed to access avail idx at %p",
                         &vrh->vring.avail->idx);

I like that this patch removes more lines of code than it adds.

However one of the design points of vringh abstractions is that they were carefully written to be very low overhead.
This is why we are passing function pointers to inline functions - compiler can optimize that out.

Quoted 5 lines

I think that introducing ops indirect functions calls here is going to break these assumptions and hurt performance.
Unless compiler can somehow figure it out and optimize?
I don't see how it's possible with ops pointer in memory
but maybe I'm wrong.

I think your concern is correct. I have to understand the compiler optimization and redesign this approach If it is needed.

Quoted 1 line

Was any effort taken to test effect of these patches on performance?

I just tested vringh_test and already faced little performance reduction. I have to investigate that, as you said.

Thank you for your comments.

Quoted 3 lines

Thanks!

Best,
Shunsuke.

#17 lore raw

Re: [RFC PATCH 4/9] vringh: unify the APIs for all accessors

2022年12月27日(火) 16:49 Shunsuke Mie <[email protected]>:

Quoted 53 lines

2022年12月27日(火) 16:04 Michael S. Tsirkin <[email protected]>:

Quoted 46 lines · Shunsuke Mie
Quoted 26 lines

Each vringh memory accessors that are for user, kern and iotlb has own interfaces that calls common code. But some codes are duplicated and that becomes loss extendability.

Introduce a struct vringh_ops and provide a common APIs for all accessors. It can bee easily extended vringh code for new memory accessor and simplified a caller code.

---
 drivers/vhost/vringh.c | 667 +++++++++++------------------------------
 include/linux/vringh.h | 100 +++---
 2 files changed, 225 insertions(+), 542 deletions(-)
drivers/vhost/vringh.c +0 -4
index aa3cd27d2384..ebfd3644a1a3 100644
@@ -35,15 +35,12 @@ static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
 }
 
 /* Returns vring->num if empty, -ve on error. */
-static inline int __vringh_get_head(const struct vringh *vrh,
-                                 int (*getu16)(const struct vringh *vrh,
-                                               u16 *val, const __virtio16 *p),
-                                 u16 *last_avail_idx)

drivers/vhost/vringh.c static __printf(1,2) __cold void vringh_bad(const char *fmt, ...) · line 46

Quoted 10 lines
drivers/vhost/vringh.c +2 -1
+static inline int __vringh_get_head(const struct vringh *vrh, u16 *last_avail_idx)
 {
      u16 avail_idx, i, head;
      int err;
 
-     err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
+     err = vrh->ops.getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
      if (err) {
              vringh_bad("Failed to access avail idx at %p",
                         &vrh->vring.avail->idx);

I like that this patch removes more lines of code than it adds.

However one of the design points of vringh abstractions is that they were carefully written to be very low overhead.
This is why we are passing function pointers to inline functions - compiler can optimize that out.

Quoted 5 lines

I think that introducing ops indirect functions calls here is going to break these assumptions and hurt performance.
Unless compiler can somehow figure it out and optimize?
I don't see how it's possible with ops pointer in memory
but maybe I'm wrong.

Quoted 5 lines

I think your concern is correct. I have to understand the compiler optimization and redesign this approach If it is needed.

Quoted 1 line

Was any effort taken to test effect of these patches on performance?

I just tested vringh_test and already faced little performance reduction. I have to investigate that, as you said.

I attempted to test with perf. I found that the performance of patched code is almost the same as the upstream one. However, I have to investigate way this patch leads to this result, also the profiling should be run on more powerful
machines too.

environment:

$ grep 'model name' /proc/cpuinfo
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz

results:
* for patched code
Performance counter stats for 'nice -n -20 ./vringh_test_patched
--parallel --eventidx --fast-vringh --indirect --virtio-1' (20 runs):

          3,028.05 msec task-clock                #    0.995 CPUs
utilized            ( +-  0.12% )
            78,150      context-switches          #   25.691 K/sec
               ( +-  0.00% )
                 5      cpu-migrations            #    1.644 /sec
               ( +-  3.33% )
               190      page-faults               #   62.461 /sec
               ( +-  0.41% )
     6,919,025,222      cycles                    #    2.275 GHz
               ( +-  0.13% )
     8,990,220,160      instructions              #    1.29  insn per
cycle           ( +-  0.04% )
     1,788,326,786      branches                  #  587.899 M/sec
               ( +-  0.05% )
         4,557,398      branch-misses             #    0.25% of all

branches ( +- 0.43% )

           3.04359 +- 0.00378 seconds time elapsed  ( +-  0.12% )

* for upstream code
Performance counter stats for 'nice -n -20 ./vringh_test_base
--parallel --eventidx --fast-vringh --indirect --virtio-1' (10 runs):

          3,058.41 msec task-clock                #    0.999 CPUs
utilized            ( +-  0.14% )
            78,149      context-switches          #   25.545 K/sec
               ( +-  0.00% )
                 5      cpu-migrations            #    1.634 /sec
               ( +-  2.67% )
               194      page-faults               #   63.414 /sec
               ( +-  0.43% )
     6,988,713,963      cycles                    #    2.284 GHz
               ( +-  0.14% )
     8,512,533,269      instructions              #    1.22  insn per
cycle           ( +-  0.04% )
     1,638,375,371      branches                  #  535.549 M/sec
               ( +-  0.05% )
         4,428,866      branch-misses             #    0.27% of all

branches ( +- 22.57% )

           3.06085 +- 0.00420 seconds time elapsed  ( +-  0.14% )
Quoted 6 lines

Thank you for your comments.

Quoted 3 lines

Thanks!

Best,
Shunsuke.

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#18 lore raw

Re: [RFC PATCH 4/9] vringh: unify the APIs for all accessors

Quoted 115 lines · Shunsuke Mie

2022年12月27日(火) 16:49 Shunsuke Mie <[email protected]>:

Quoted 53 lines

2022年12月27日(火) 16:04 Michael S. Tsirkin <[email protected]>:

Quoted 46 lines · Shunsuke Mie
Quoted 26 lines

Each vringh memory accessors that are for user, kern and iotlb has own interfaces that calls common code. But some codes are duplicated and that becomes loss extendability.

Introduce a struct vringh_ops and provide a common APIs for all accessors. It can bee easily extended vringh code for new memory accessor and simplified a caller code.

---
 drivers/vhost/vringh.c | 667 +++++++++++------------------------------
 include/linux/vringh.h | 100 +++---
 2 files changed, 225 insertions(+), 542 deletions(-)
drivers/vhost/vringh.c +0 -4
index aa3cd27d2384..ebfd3644a1a3 100644
@@ -35,15 +35,12 @@ static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
 }
 
 /* Returns vring->num if empty, -ve on error. */
-static inline int __vringh_get_head(const struct vringh *vrh,
-                                 int (*getu16)(const struct vringh *vrh,
-                                               u16 *val, const __virtio16 *p),
-                                 u16 *last_avail_idx)

drivers/vhost/vringh.c static __printf(1,2) __cold void vringh_bad(const char *fmt, ...) · line 46

Quoted 10 lines
drivers/vhost/vringh.c +2 -1
+static inline int __vringh_get_head(const struct vringh *vrh, u16 *last_avail_idx)
 {
      u16 avail_idx, i, head;
      int err;
 
-     err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
+     err = vrh->ops.getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
      if (err) {
              vringh_bad("Failed to access avail idx at %p",
                         &vrh->vring.avail->idx);

I like that this patch removes more lines of code than it adds.

However one of the design points of vringh abstractions is that they were carefully written to be very low overhead.
This is why we are passing function pointers to inline functions - compiler can optimize that out.

Quoted 5 lines

I think that introducing ops indirect functions calls here is going to break these assumptions and hurt performance.
Unless compiler can somehow figure it out and optimize?
I don't see how it's possible with ops pointer in memory
but maybe I'm wrong.

Quoted 5 lines

I think your concern is correct. I have to understand the compiler optimization and redesign this approach If it is needed.

Quoted 1 line

Was any effort taken to test effect of these patches on performance?

I just tested vringh_test and already faced little performance reduction. I have to investigate that, as you said.

I attempted to test with perf. I found that the performance of patched code is almost the same as the upstream one. However, I have to investigate way this patch leads to this result, also the profiling should be run on more powerful
machines too.

environment:

$ grep 'model name' /proc/cpuinfo
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz

results:
* for patched code
Performance counter stats for 'nice -n -20 ./vringh_test_patched
--parallel --eventidx --fast-vringh --indirect --virtio-1' (20 runs):

          3,028.05 msec task-clock                #    0.995 CPUs
utilized            ( +-  0.12% )
            78,150      context-switches          #   25.691 K/sec
               ( +-  0.00% )
                 5      cpu-migrations            #    1.644 /sec
               ( +-  3.33% )
               190      page-faults               #   62.461 /sec
               ( +-  0.41% )
     6,919,025,222      cycles                    #    2.275 GHz
               ( +-  0.13% )
     8,990,220,160      instructions              #    1.29  insn per
cycle           ( +-  0.04% )
     1,788,326,786      branches                  #  587.899 M/sec
               ( +-  0.05% )
         4,557,398      branch-misses             #    0.25% of all

branches ( +- 0.43% )

           3.04359 +- 0.00378 seconds time elapsed  ( +-  0.12% )

* for upstream code
Performance counter stats for 'nice -n -20 ./vringh_test_base
--parallel --eventidx --fast-vringh --indirect --virtio-1' (10 runs):

          3,058.41 msec task-clock                #    0.999 CPUs
utilized            ( +-  0.14% )
            78,149      context-switches          #   25.545 K/sec
               ( +-  0.00% )
                 5      cpu-migrations            #    1.634 /sec
               ( +-  2.67% )
               194      page-faults               #   63.414 /sec
               ( +-  0.43% )
     6,988,713,963      cycles                    #    2.284 GHz
               ( +-  0.14% )
     8,512,533,269      instructions              #    1.22  insn per
cycle           ( +-  0.04% )
     1,638,375,371      branches                  #  535.549 M/sec
               ( +-  0.05% )
         4,428,866      branch-misses             #    0.27% of all
Quoted 3 lines

branches ( +- 22.57% )

           3.06085 +- 0.00420 seconds time elapsed  ( +-  0.14% )

How you compiled it also matters. ATM we don't enable retpolines and it did not matter since we didn't have indirect calls, but we should. Didn't yet investigate how to do that for virtio tools.

Quoted 6 lines
Quoted 6 lines

Thank you for your comments.

Quoted 3 lines

Thanks!

Best,
Shunsuke.

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#19 lore raw

Re: [RFC PATCH 4/9] vringh: unify the APIs for all accessors

2022年12月27日(火) 23:37 Michael S. Tsirkin <[email protected]>:

Quoted 119 lines · Shunsuke Mie
Quoted 117 lines

2022年12月27日(火) 16:49 Shunsuke Mie <[email protected]>:

Quoted 53 lines

2022年12月27日(火) 16:04 Michael S. Tsirkin <[email protected]>:

Quoted 46 lines · Shunsuke Mie
Quoted 26 lines

Each vringh memory accessors that are for user, kern and iotlb has own interfaces that calls common code. But some codes are duplicated and that becomes loss extendability.

Introduce a struct vringh_ops and provide a common APIs for all accessors. It can bee easily extended vringh code for new memory accessor and simplified a caller code.

---
 drivers/vhost/vringh.c | 667 +++++++++++------------------------------
 include/linux/vringh.h | 100 +++---
 2 files changed, 225 insertions(+), 542 deletions(-)
drivers/vhost/vringh.c +0 -4
index aa3cd27d2384..ebfd3644a1a3 100644
@@ -35,15 +35,12 @@ static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
 }
 
 /* Returns vring->num if empty, -ve on error. */
-static inline int __vringh_get_head(const struct vringh *vrh,
-                                 int (*getu16)(const struct vringh *vrh,
-                                               u16 *val, const __virtio16 *p),
-                                 u16 *last_avail_idx)

drivers/vhost/vringh.c static __printf(1,2) __cold void vringh_bad(const char *fmt, ...) · line 46

Quoted 10 lines
drivers/vhost/vringh.c +2 -1
+static inline int __vringh_get_head(const struct vringh *vrh, u16 *last_avail_idx)
 {
      u16 avail_idx, i, head;
      int err;
 
-     err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
+     err = vrh->ops.getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
      if (err) {
              vringh_bad("Failed to access avail idx at %p",
                         &vrh->vring.avail->idx);

I like that this patch removes more lines of code than it adds.

However one of the design points of vringh abstractions is that they were carefully written to be very low overhead.
This is why we are passing function pointers to inline functions - compiler can optimize that out.

Quoted 5 lines

I think that introducing ops indirect functions calls here is going to break these assumptions and hurt performance.
Unless compiler can somehow figure it out and optimize?
I don't see how it's possible with ops pointer in memory
but maybe I'm wrong.

Quoted 5 lines

I think your concern is correct. I have to understand the compiler optimization and redesign this approach If it is needed.

Quoted 1 line

Was any effort taken to test effect of these patches on performance?

I just tested vringh_test and already faced little performance reduction. I have to investigate that, as you said.

I attempted to test with perf. I found that the performance of patched code is almost the same as the upstream one. However, I have to investigate way this patch leads to this result, also the profiling should be run on more powerful
machines too.

environment:

$ grep 'model name' /proc/cpuinfo
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz

results:
* for patched code
Performance counter stats for 'nice -n -20 ./vringh_test_patched
--parallel --eventidx --fast-vringh --indirect --virtio-1' (20 runs):

          3,028.05 msec task-clock                #    0.995 CPUs
utilized            ( +-  0.12% )
            78,150      context-switches          #   25.691 K/sec
               ( +-  0.00% )
                 5      cpu-migrations            #    1.644 /sec
               ( +-  3.33% )
               190      page-faults               #   62.461 /sec
               ( +-  0.41% )
     6,919,025,222      cycles                    #    2.275 GHz
               ( +-  0.13% )
     8,990,220,160      instructions              #    1.29  insn per
cycle           ( +-  0.04% )
     1,788,326,786      branches                  #  587.899 M/sec
               ( +-  0.05% )
         4,557,398      branch-misses             #    0.25% of all

branches ( +- 0.43% )

           3.04359 +- 0.00378 seconds time elapsed  ( +-  0.12% )

* for upstream code
Performance counter stats for 'nice -n -20 ./vringh_test_base
--parallel --eventidx --fast-vringh --indirect --virtio-1' (10 runs):

          3,058.41 msec task-clock                #    0.999 CPUs
utilized            ( +-  0.14% )
            78,149      context-switches          #   25.545 K/sec
               ( +-  0.00% )
                 5      cpu-migrations            #    1.634 /sec
               ( +-  2.67% )
               194      page-faults               #   63.414 /sec
               ( +-  0.43% )
     6,988,713,963      cycles                    #    2.284 GHz
               ( +-  0.14% )
     8,512,533,269      instructions              #    1.22  insn per
cycle           ( +-  0.04% )
     1,638,375,371      branches                  #  535.549 M/sec
               ( +-  0.05% )
         4,428,866      branch-misses             #    0.27% of all

branches ( +- 22.57% )

Quoted 6 lines
Quoted 1 line
           3.06085 +- 0.00420 seconds time elapsed  ( +-  0.14% )

How you compiled it also matters. ATM we don't enable retpolines and it did not matter since we didn't have indirect calls, but we should. Didn't yet investigate how to do that for virtio tools.

I think the retpolines certainly affect performance. Thank you for pointing it out. I'd like to start the investigation that how to apply the retpolines to the
virtio tools.

Quoted 7 lines
Quoted 6 lines
Quoted 6 lines

Thank you for your comments.

Quoted 3 lines

Thanks!

Best,
Shunsuke.

#20 lore raw

Re: [RFC PATCH 4/9] vringh: unify the APIs for all accessors

Quoted 123 lines · Shunsuke Mie

2022年12月27日(火) 23:37 Michael S. Tsirkin <[email protected]>:

Quoted 122 lines · Shunsuke Mie
Quoted 118 lines

2022年12月27日(火) 16:49 Shunsuke Mie <[email protected]>:

Quoted 53 lines

2022年12月27日(火) 16:04 Michael S. Tsirkin <[email protected]>:

Quoted 46 lines · Shunsuke Mie
Quoted 26 lines

Each vringh memory accessors that are for user, kern and iotlb has own interfaces that calls common code. But some codes are duplicated and that becomes loss extendability.

Introduce a struct vringh_ops and provide a common APIs for all accessors. It can bee easily extended vringh code for new memory accessor and simplified a caller code.

---
 drivers/vhost/vringh.c | 667 +++++++++++------------------------------
 include/linux/vringh.h | 100 +++---
 2 files changed, 225 insertions(+), 542 deletions(-)
drivers/vhost/vringh.c +0 -4
index aa3cd27d2384..ebfd3644a1a3 100644
@@ -35,15 +35,12 @@ static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
 }
 
 /* Returns vring->num if empty, -ve on error. */
-static inline int __vringh_get_head(const struct vringh *vrh,
-                                 int (*getu16)(const struct vringh *vrh,
-                                               u16 *val, const __virtio16 *p),
-                                 u16 *last_avail_idx)

drivers/vhost/vringh.c static __printf(1,2) __cold void vringh_bad(const char *fmt, ...) · line 46

Quoted 10 lines
drivers/vhost/vringh.c +2 -1
+static inline int __vringh_get_head(const struct vringh *vrh, u16 *last_avail_idx)
 {
      u16 avail_idx, i, head;
      int err;
 
-     err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
+     err = vrh->ops.getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
      if (err) {
              vringh_bad("Failed to access avail idx at %p",
                         &vrh->vring.avail->idx);

I like that this patch removes more lines of code than it adds.

However one of the design points of vringh abstractions is that they were carefully written to be very low overhead.
This is why we are passing function pointers to inline functions - compiler can optimize that out.

Quoted 5 lines

I think that introducing ops indirect functions calls here is going to break these assumptions and hurt performance.
Unless compiler can somehow figure it out and optimize?
I don't see how it's possible with ops pointer in memory
but maybe I'm wrong.

Quoted 5 lines

I think your concern is correct. I have to understand the compiler optimization and redesign this approach If it is needed.

Quoted 1 line

Was any effort taken to test effect of these patches on performance?

I just tested vringh_test and already faced little performance reduction. I have to investigate that, as you said.

I attempted to test with perf. I found that the performance of patched code is almost the same as the upstream one. However, I have to investigate way this patch leads to this result, also the profiling should be run on more powerful
machines too.

environment:

$ grep 'model name' /proc/cpuinfo
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz

results:
* for patched code
Performance counter stats for 'nice -n -20 ./vringh_test_patched
--parallel --eventidx --fast-vringh --indirect --virtio-1' (20 runs):

          3,028.05 msec task-clock                #    0.995 CPUs
utilized            ( +-  0.12% )
            78,150      context-switches          #   25.691 K/sec
               ( +-  0.00% )
                 5      cpu-migrations            #    1.644 /sec
               ( +-  3.33% )
               190      page-faults               #   62.461 /sec
               ( +-  0.41% )
     6,919,025,222      cycles                    #    2.275 GHz
               ( +-  0.13% )
     8,990,220,160      instructions              #    1.29  insn per
cycle           ( +-  0.04% )
     1,788,326,786      branches                  #  587.899 M/sec
               ( +-  0.05% )
         4,557,398      branch-misses             #    0.25% of all

branches ( +- 0.43% )

           3.04359 +- 0.00378 seconds time elapsed  ( +-  0.12% )

* for upstream code
Performance counter stats for 'nice -n -20 ./vringh_test_base
--parallel --eventidx --fast-vringh --indirect --virtio-1' (10 runs):

          3,058.41 msec task-clock                #    0.999 CPUs
utilized            ( +-  0.14% )
            78,149      context-switches          #   25.545 K/sec
               ( +-  0.00% )
                 5      cpu-migrations            #    1.634 /sec
               ( +-  2.67% )
               194      page-faults               #   63.414 /sec
               ( +-  0.43% )
     6,988,713,963      cycles                    #    2.284 GHz
               ( +-  0.14% )
     8,512,533,269      instructions              #    1.22  insn per
cycle           ( +-  0.04% )
     1,638,375,371      branches                  #  535.549 M/sec
               ( +-  0.05% )
         4,428,866      branch-misses             #    0.27% of all

branches ( +- 22.57% )

           3.06085 +- 0.00420 seconds time elapsed  ( +-  0.14% )
Quoted 13 lines
Quoted 3 lines

How you compiled it also matters. ATM we don't enable retpolines and it did not matter since we didn't have indirect calls, but we should. Didn't yet investigate how to do that for virtio tools.

I think the retpolines certainly affect performance. Thank you for pointing it out. I'd like to start the investigation that how to apply the retpolines to the
virtio tools.

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

Thank you for your comments.

Quoted 3 lines

Thanks!

Best,
Shunsuke.

This isn't all that trivial if we want this at runtime. But compile time is kind of easy.
See Documentation/admin-guide/hw-vuln/spectre.rst

Signature · 6 lines
MST

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
#21 lore raw

Re: [RFC PATCH 4/9] vringh: unify the APIs for all accessors

Quoted 127 lines · Michael S. Tsirkin
Quoted 126 lines · Shunsuke Mie

2022年12月27日(火) 23:37 Michael S. Tsirkin <[email protected]>:

Quoted 115 lines · Shunsuke Mie
Quoted 114 lines

2022年12月27日(火) 16:49 Shunsuke Mie <[email protected]>:

Quoted 50 lines

2022年12月27日(火) 16:04 Michael S. Tsirkin <[email protected]>:

Quoted 44 lines · Shunsuke Mie
Quoted 26 lines

Each vringh memory accessors that are for user, kern and iotlb has own interfaces that calls common code. But some codes are duplicated and that becomes loss extendability.

Introduce a struct vringh_ops and provide a common APIs for all accessors. It can bee easily extended vringh code for new memory accessor and simplified a caller code.

---
  drivers/vhost/vringh.c | 667 +++++++++++------------------------------
  include/linux/vringh.h | 100 +++---
  2 files changed, 225 insertions(+), 542 deletions(-)
drivers/vhost/vringh.c +0 -4
index aa3cd27d2384..ebfd3644a1a3 100644
@@ -35,15 +35,12 @@ static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
  }
 
  /* Returns vring->num if empty, -ve on error. */
-static inline int __vringh_get_head(const struct vringh *vrh,
-                                 int (*getu16)(const struct vringh *vrh,
-                                               u16 *val, const __virtio16 *p),
-                                 u16 *last_avail_idx)

drivers/vhost/vringh.c static __printf(1,2) __cold void vringh_bad(const char *fmt, ...) · line 46

Quoted 10 lines
drivers/vhost/vringh.c +2 -1
+static inline int __vringh_get_head(const struct vringh *vrh, u16 *last_avail_idx)
  {
       u16 avail_idx, i, head;
       int err;
 
-     err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
+     err = vrh->ops.getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
       if (err) {
               vringh_bad("Failed to access avail idx at %p",
                          &vrh->vring.avail->idx);

I like that this patch removes more lines of code than it adds.

However one of the design points of vringh abstractions is that they were carefully written to be very low overhead.
This is why we are passing function pointers to inline functions - compiler can optimize that out.

Quoted 5 lines

I think that introducing ops indirect functions calls here is going to break these assumptions and hurt performance.
Unless compiler can somehow figure it out and optimize?
I don't see how it's possible with ops pointer in memory
but maybe I'm wrong.

Quoted 5 lines

I think your concern is correct. I have to understand the compiler optimization and redesign this approach If it is needed.

Quoted 1 line

Was any effort taken to test effect of these patches on performance?

I just tested vringh_test and already faced little performance reduction. I have to investigate that, as you said.

I attempted to test with perf. I found that the performance of patched code is almost the same as the upstream one. However, I have to investigate way this patch leads to this result, also the profiling should be run on more powerful
machines too.

environment:

$ grep 'model name' /proc/cpuinfo
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz
model name      : Intel(R) Core(TM) i3-7020U CPU @ 2.30GHz

results:
* for patched code
Performance counter stats for 'nice -n -20 ./vringh_test_patched
--parallel --eventidx --fast-vringh --indirect --virtio-1' (20 runs):

           3,028.05 msec task-clock                #    0.995 CPUs
utilized            ( +-  0.12% )
             78,150      context-switches          #   25.691 K/sec
                ( +-  0.00% )
                  5      cpu-migrations            #    1.644 /sec
                ( +-  3.33% )
                190      page-faults               #   62.461 /sec
                ( +-  0.41% )
      6,919,025,222      cycles                    #    2.275 GHz
                ( +-  0.13% )
      8,990,220,160      instructions              #    1.29  insn per
cycle           ( +-  0.04% )
      1,788,326,786      branches                  #  587.899 M/sec
                ( +-  0.05% )
          4,557,398      branch-misses             #    0.25% of all

branches ( +- 0.43% )

            3.04359 +- 0.00378 seconds time elapsed  ( +-  0.12% )

* for upstream code
Performance counter stats for 'nice -n -20 ./vringh_test_base
--parallel --eventidx --fast-vringh --indirect --virtio-1' (10 runs):

           3,058.41 msec task-clock                #    0.999 CPUs
utilized            ( +-  0.14% )
             78,149      context-switches          #   25.545 K/sec
                ( +-  0.00% )
                  5      cpu-migrations            #    1.634 /sec
                ( +-  2.67% )
                194      page-faults               #   63.414 /sec
                ( +-  0.43% )
      6,988,713,963      cycles                    #    2.284 GHz
                ( +-  0.14% )
      8,512,533,269      instructions              #    1.22  insn per
cycle           ( +-  0.04% )
      1,638,375,371      branches                  #  535.549 M/sec
                ( +-  0.05% )
          4,428,866      branch-misses             #    0.27% of all

branches ( +- 22.57% )

Quoted 5 lines
Quoted 1 line
            3.06085 +- 0.00420 seconds time elapsed  ( +-  0.14% )

How you compiled it also matters. ATM we don't enable retpolines and it did not matter since we didn't have indirect calls, but we should. Didn't yet investigate how to do that for virtio tools.

I think the retpolines certainly affect performance. Thank you for pointing it out. I'd like to start the investigation that how to apply the retpolines to the
virtio tools.

Quoted 1 line
Quoted 1 line
Quoted 1 line

Thank you for your comments.

Quoted 8 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 3 lines

Thanks!

Best,
Shunsuke.

This isn't all that trivial if we want this at runtime. But compile time is kind of easy.
See Documentation/admin-guide/hw-vuln/spectre.rst

Thank you for showing it.

I followed the document and added options to CFLAGS to the tools Makefile.

That is

---

tools/virtio/Makefile +0 -0
index 1b25cc7c64bb..7b7139d97d74 100644
@@ -4,7 +4,7 @@ test: virtio_test vringh_test

 virtio_test: virtio_ring.o virtio_test.o
 vringh_test: vringh_test.o vringh.o virtio_ring.o

-CFLAGS += -g -O2 -Werror -Wno-maybe-uninitialized -Wall -I. -I../include/ -I ../../usr/include/ -Wno-pointer-sign -fno-strict-overflow -fno-strict-aliasing -fno-common -MMD -U_FORTIFY_SOURCE -include ../../include/linux/kconfig.h
+CFLAGS += -g -O2 -Werror -Wno-maybe-uninitialized -Wall -I. -I../include/ -I ../../usr/include/ -Wno-pointer-sign -fno-strict-overflow -fno-strict-aliasing -fno-common -MMD -U_FORTIFY_SOURCE -include ../../include/linux/kconfig.h -mfunction-return=thunk -fcf-protection=none -mindirect-branch-register CFLAGS += -pthread
 LDFLAGS += -pthread
 vpath %.c ../../drivers/virtio ../../drivers/vhost
---

And results of evaluation are following:

- base with retpoline

$ sudo perf stat --repeat 20 -- nice -n -20 ./vringh_test_retp_origin --parallel --eventidx --fast-vringh
Using CPUS 0 and 3
Guest: notified 0, pinged 98040
Host: notified 98040, pinged 0
...

 Performance counter stats for 'nice -n -20 ./vringh_test_retp_origin --parallel --eventidx --fast-vringh' (20 runs):

          6,228.33 msec task-clock                #    1.004 CPUs utilized            ( +-  0.05% ) 196,110      context-switches          #   31.616 K/sec                    ( +-  0.00% ) 6      cpu-migrations            #    0.967 /sec                     ( +-  2.39% ) 205      page-faults               #   33.049 /sec                     ( +-  0.46% ) 14,218,527,987      cycles                    #    2.292 GHz                      ( +-  0.05% ) 10,342,897,254      instructions              #    0.73  insn per cycle           ( +-  0.02% ) 2,310,572,989      branches                  #  372.500 M/sec                    ( +-  0.03% ) 178,273,068      branch-misses             #    7.72% of all branches          ( +-  0.04% )

           6.20406 +- 0.00308 seconds time elapsed  ( +-  0.05% )

- patched (unified APIs) with retpoline

$ sudo perf stat --repeat 20 -- nice -n -20 ./vringh_test_retp_patched --parallel --eventidx --fast-vringh
Using CPUS 0 and 3
Guest: notified 0, pinged 98040
Host: notified 98040, pinged 0
...

 Performance counter stats for 'nice -n -20 ./vringh_test_retp_patched --parallel --eventidx --fast-vringh' (20 runs):

          6,103.94 msec task-clock                #    1.001 CPUs utilized            ( +-  0.03% ) 196,125      context-switches          #   32.165 K/sec                    ( +-  0.00% ) 7      cpu-migrations            #    1.148 /sec                     ( +-  1.56% ) 196      page-faults               #   32.144 /sec                     ( +-  0.41% ) 13,933,055,778      cycles                    #    2.285 GHz                      ( +-  0.03% ) 10,309,004,718      instructions              #    0.74  insn per cycle           ( +-  0.03% ) 2,368,447,519      branches                  #  388.425 M/sec                    ( +-  0.04% ) 211,364,886      branch-misses             #    8.94% of all branches          ( +-  0.05% )

           6.09888 +- 0.00155 seconds time elapsed  ( +-  0.03% )

As a result, at the patched code, the branch-misses was increased but elapsed time became faster than the based code. The number of page-faults was a little different. I'm suspicious of that the page-fault penalty leads the performance result.

I think that a pattern of memory access for data is same with those, but for instruction is different. Actually a code size (.text segment) was a little smaller. 0x6a65 and 0x63f5.

$ readelf -a ./vringh_test_retp_origin |grep .text -1
       0000000000000008  0000000000000008  AX       0     0     8
  [14] .text             PROGBITS         0000000000001230 00001230
       0000000000006a65  0000000000000000  AX       0     0     16
--
   02     .interp .note.gnu.build-id .note.ABI-tag .gnu.hash .dynsym .dynstr .gnu.version .gnu.version_r .rela.dyn .rela.plt
   03     .init .plt .plt.got .text .fini
   04     .rodata .eh_frame_hdr .eh_frame
$ readelf -a ./vringh_test_retp_patched |grep .text -1
       0000000000000008  0000000000000008  AX       0     0     8
  [14] .text             PROGBITS         0000000000001230 00001230
       00000000000063f5  0000000000000000  AX       0     0     16
--
   02     .interp .note.gnu.build-id .note.ABI-tag .gnu.hash .dynsym .dynstr .gnu.version .gnu.version_r .rela.dyn .rela.plt
   03     .init .plt .plt.got .text .fini
   04     .rodata .eh_frame_hdr .eh_frame

I'll keep this investigation. I was wondering if you could comment me.

Best

Quoted 1 line

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#22 lore raw

RFC PATCH 5/9 tools/virtio: convert to use new unified vringh APIs

In reply to Shunsuke Mie

vringh_*_user APIs is being removed without vringh_init_user(). so change to use new APIs.

---
 tools/virtio/vringh_test.c | 89 +++++++++++++++++++-------------------
 1 file changed, 44 insertions(+), 45 deletions(-)
tools/virtio/vringh_test.c +44 -45
index 6c9533b8a2ca..068c6d5aa4fd 100644
@@ -187,7 +187,7 @@ static int parallel_test(u64 features,
 
 		vring_init(&vrh.vring, RINGSIZE, host_map, ALIGN);
 		vringh_init_user(&vrh, features, RINGSIZE, true,
-				 vrh.vring.desc, vrh.vring.avail, vrh.vring.used);
+				 vrh.vring.desc, vrh.vring.avail, vrh.vring.used, getrange);
 		CPU_SET(first_cpu, &cpu_set);
 		if (sched_setaffinity(getpid(), sizeof(cpu_set), &cpu_set))
 			errx(1, "Could not set affinity to cpu %u", first_cpu);
@@ -202,9 +202,9 @@ static int parallel_test(u64 features,
 					err = vringh_get_head(&vrh, &head);
 					if (err != 0)
 						break;
-					err = vringh_need_notify_user(&vrh);
+					err = vringh_need_notify(&vrh);
 					if (err < 0)
-						errx(1, "vringh_need_notify_user: %i",
+						errx(1, "vringh_need_notify: %i",
 						     err);
 					if (err) {
 						write(to_guest[1], "", 1);
@@ -223,46 +223,45 @@ static int parallel_test(u64 features,
 						host_wiov,
 						ARRAY_SIZE(host_wiov));
 
-				err = vringh_getdesc_user(&vrh, &riov, &wiov,
-							  getrange, &head);
+				err = vringh_getdesc(&vrh, &riov, &wiov, &head);
 			}
 			if (err == 0) {
-				err = vringh_need_notify_user(&vrh);
+				err = vringh_need_notify(&vrh);
 				if (err < 0)
-					errx(1, "vringh_need_notify_user: %i",
+					errx(1, "vringh_need_notify: %i",
 					     err);
 				if (err) {
 					write(to_guest[1], "", 1);
 					notifies++;
 				}
 
-				if (!vringh_notify_enable_user(&vrh))
+				if (!vringh_notify_enable(&vrh))
 					continue;
 
 				/* Swallow all notifies at once. */
 				if (read(to_host[0], buf, sizeof(buf)) < 1)
 					break;
 
-				vringh_notify_disable_user(&vrh);
+				vringh_notify_disable(&vrh);
 				receives++;
 				continue;
 			}
 			if (err != 1)
-				errx(1, "vringh_getdesc_user: %i", err);
+				errx(1, "vringh_getdesc: %i", err);
 
 			/* We simply copy bytes. */
 			if (riov.used) {
-				rlen = vringh_iov_pull_user(&riov, rbuf,
+				rlen = vringh_iov_pull(&vrh, &riov, rbuf,
 							    sizeof(rbuf));
 				if (rlen != 4)
-					errx(1, "vringh_iov_pull_user: %i",
+					errx(1, "vringh_iov_pull: %i",
 					     rlen);
 				assert(riov.i == riov.used);
 				written = 0;
 			} else {
-				err = vringh_iov_push_user(&wiov, rbuf, rlen);
+				err = vringh_iov_push(&vrh, &wiov, rbuf, rlen);
 				if (err != rlen)
-					errx(1, "vringh_iov_push_user: %i",
+					errx(1, "vringh_iov_push: %i",
 					     err);
 				assert(wiov.i == wiov.used);
 				written = err;
@@ -270,14 +269,14 @@ static int parallel_test(u64 features,
 		complete:
 			xfers++;
 
-			err = vringh_complete_user(&vrh, head, written);
+			err = vringh_complete(&vrh, head, written);
 			if (err != 0)
-				errx(1, "vringh_complete_user: %i", err);
+				errx(1, "vringh_complete: %i", err);
 		}
 
-		err = vringh_need_notify_user(&vrh);
+		err = vringh_need_notify(&vrh);
 		if (err < 0)
-			errx(1, "vringh_need_notify_user: %i", err);
+			errx(1, "vringh_need_notify: %i", err);
 		if (err) {
 			write(to_guest[1], "", 1);
 			notifies++;
@@ -493,12 +492,12 @@ int main(int argc, char *argv[])
 	/* Set up host side. */
 	vring_init(&vrh.vring, RINGSIZE, __user_addr_min, ALIGN);
 	vringh_init_user(&vrh, vdev.features, RINGSIZE, true,
-			 vrh.vring.desc, vrh.vring.avail, vrh.vring.used);
+			 vrh.vring.desc, vrh.vring.avail, vrh.vring.used, getrange);
 
 	/* No descriptor to get yet... */
-	err = vringh_getdesc_user(&vrh, &riov, &wiov, getrange, &head);
+	err = vringh_getdesc(&vrh, &riov, &wiov, &head);
 	if (err != 0)
-		errx(1, "vringh_getdesc_user: %i", err);
+		errx(1, "vringh_getdesc: %i", err);
 
 	/* Guest puts in a descriptor. */
 	memcpy(__user_addr_max - 1, "a", 1);
@@ -520,9 +519,9 @@ int main(int argc, char *argv[])
 	vringh_kiov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
 	vringh_kiov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
 
-	err = vringh_getdesc_user(&vrh, &riov, &wiov, getrange, &head);
+	err = vringh_getdesc(&vrh, &riov, &wiov, &head);
 	if (err != 1)
-		errx(1, "vringh_getdesc_user: %i", err);
+		errx(1, "vringh_getdesc: %i", err);
 
 	assert(riov.used == 1);
 	assert(riov.iov[0].iov_base == __user_addr_max - 1);
@@ -539,25 +538,25 @@ int main(int argc, char *argv[])
 		assert(wiov.iov[1].iov_len == 1);
 	}
 
-	err = vringh_iov_pull_user(&riov, buf, 5);
+	err = vringh_iov_pull(&vrh, &riov, buf, 5);
 	if (err != 1)
-		errx(1, "vringh_iov_pull_user: %i", err);
+		errx(1, "vringh_iov_pull: %i", err);
 	assert(buf[0] == 'a');
 	assert(riov.i == 1);
-	assert(vringh_iov_pull_user(&riov, buf, 5) == 0);
+	assert(vringh_iov_pull(&vrh, &riov, buf, 5) == 0);
 
 	memcpy(buf, "bcdef", 5);
-	err = vringh_iov_push_user(&wiov, buf, 5);
+	err = vringh_iov_push(&vrh, &wiov, buf, 5);
 	if (err != 2)
-		errx(1, "vringh_iov_push_user: %i", err);
+		errx(1, "vringh_iov_push: %i", err);
 	assert(memcmp(__user_addr_max - 3, "bc", 2) == 0);
 	assert(wiov.i == wiov.used);
-	assert(vringh_iov_push_user(&wiov, buf, 5) == 0);
+	assert(vringh_iov_push(&vrh, &wiov, buf, 5) == 0);
 
 	/* Host is done. */
-	err = vringh_complete_user(&vrh, head, err);
+	err = vringh_complete(&vrh, head, err);
 	if (err != 0)
-		errx(1, "vringh_complete_user: %i", err);
+		errx(1, "vringh_complete: %i", err);
 
 	/* Guest should see used token now. */
 	__kfree_ignore_start = __user_addr_min + vring_size(RINGSIZE, ALIGN);
@@ -589,9 +588,9 @@ int main(int argc, char *argv[])
 	vringh_kiov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
 	vringh_kiov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
 
-	err = vringh_getdesc_user(&vrh, &riov, &wiov, getrange, &head);
+	err = vringh_getdesc(&vrh, &riov, &wiov, &head);
 	if (err != 1)
-		errx(1, "vringh_getdesc_user: %i", err);
+		errx(1, "vringh_getdesc: %i", err);
 
 	assert(riov.max_num & VRINGH_IOV_ALLOCATED);
 	assert(riov.iov != host_riov);
@@ -605,9 +604,9 @@ int main(int argc, char *argv[])
 
 	/* Pull data back out (in odd chunks), should be as expected. */
 	for (i = 0; i < RINGSIZE * USER_MEM/4; i += 3) {
-		err = vringh_iov_pull_user(&riov, buf, 3);
+		err = vringh_iov_pull(&vrh, &riov, buf, 3);
 		if (err != 3 && i + err != RINGSIZE * USER_MEM/4)
-			errx(1, "vringh_iov_pull_user large: %i", err);
+			errx(1, "vringh_iov_pulllarge: %i", err);
 		assert(buf[0] == (char)i);
 		assert(err < 2 || buf[1] == (char)(i + 1));
 		assert(err < 3 || buf[2] == (char)(i + 2));
@@ -619,9 +618,9 @@ int main(int argc, char *argv[])
 	/* Complete using multi interface, just because we can. */
 	used[0].id = head;
 	used[0].len = 0;
-	err = vringh_complete_multi_user(&vrh, used, 1);
+	err = vringh_complete_multi(&vrh, used, 1);
 	if (err)
-		errx(1, "vringh_complete_multi_user(1): %i", err);
+		errx(1, "vringh_complete_multi(1): %i", err);
 
 	/* Free up those descriptors. */
 	ret = virtqueue_get_buf(vq, &i);
@@ -642,17 +641,17 @@ int main(int argc, char *argv[])
 	vringh_kiov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
 
 	for (i = 0; i < RINGSIZE; i++) {
-		err = vringh_getdesc_user(&vrh, &riov, &wiov, getrange, &head);
+		err = vringh_getdesc(&vrh, &riov, &wiov, &head);
 		if (err != 1)
-			errx(1, "vringh_getdesc_user: %i", err);
+			errx(1, "vringh_getdesc: %i", err);
 		used[i].id = head;
 		used[i].len = 0;
 	}
 	/* Make sure it wraps around ring, to test! */
 	assert(vrh.vring.used->idx % RINGSIZE != 0);
-	err = vringh_complete_multi_user(&vrh, used, RINGSIZE);
+	err = vringh_complete_multi(&vrh, used, RINGSIZE);
 	if (err)
-		errx(1, "vringh_complete_multi_user: %i", err);
+		errx(1, "vringh_complete_multi: %i", err);
 
 	/* Free those buffers. */
 	for (i = 0; i < RINGSIZE; i++) {
@@ -726,19 +725,19 @@ int main(int argc, char *argv[])
 		vringh_kiov_init(&riov, host_riov, ARRAY_SIZE(host_riov));
 		vringh_kiov_init(&wiov, host_wiov, ARRAY_SIZE(host_wiov));
 
-		err = vringh_getdesc_user(&vrh, &riov, &wiov, getrange, &head);
+		err = vringh_getdesc(&vrh, &riov, &wiov, &head);
 		if (err != 1)
-			errx(1, "vringh_getdesc_user: %i", err);
+			errx(1, "vringh_getdesc: %i", err);
 
 		if (head != 0)
-			errx(1, "vringh_getdesc_user: head %i not 0", head);
+			errx(1, "vringh_getdesc: head %i not 0", head);
 
 		assert(riov.max_num & VRINGH_IOV_ALLOCATED);
 		if (getrange != getrange_slow)
 			assert(riov.used == 7);
 		else
 			assert(riov.used == 28);
-		err = vringh_iov_pull_user(&riov, buf, 29);
+		err = vringh_iov_pull(&vrh, &riov, buf, 29);
 		assert(err == 28);
 
 		/* Data should be linear. */
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
#23 lore raw

RFC PATCH 6/9 caif_virtio: convert to new unified vringh APIs

In reply to Shunsuke Mie

vringh_*_kern APIs are being removed without vringh_init_kern(), so change to use new APIs.

---
 drivers/net/caif/caif_virtio.c | 26 ++++++++++----------------
 1 file changed, 10 insertions(+), 16 deletions(-)
drivers/net/caif/caif_virtio.c +10 -16
index 0b0f234b0b50..f9dd79807afa 100644
@@ -265,18 +265,12 @@ static int cfv_rx_poll(struct napi_struct *napi, int quota)
 		 */
 		if (riov->i == riov->used) {
 			if (cfv->ctx.head != USHRT_MAX) {
-				vringh_complete_kern(cfv->vr_rx,
-						     cfv->ctx.head,
-						     0);
+				vringh_complete(cfv->vr_rx, cfv->ctx.head, 0);
 				cfv->ctx.head = USHRT_MAX;
 			}
 
-			err = vringh_getdesc_kern(
-				cfv->vr_rx,
-				riov,
-				NULL,
-				&cfv->ctx.head,
-				GFP_ATOMIC);
+			err = vringh_getdesc(cfv->vr_rx, riov, NULL,
+					     &cfv->ctx.head);
 
 			if (err <= 0)
 				goto exit;
@@ -317,9 +311,9 @@ static int cfv_rx_poll(struct napi_struct *napi, int quota)
 
 		/* Really out of packets? (stolen from virtio_net)*/
 		napi_complete(napi);
-		if (unlikely(!vringh_notify_enable_kern(cfv->vr_rx)) &&
+		if (unlikely(!vringh_notify_enable(cfv->vr_rx)) &&
 		    napi_schedule_prep(napi)) {
-			vringh_notify_disable_kern(cfv->vr_rx);
+			vringh_notify_disable(cfv->vr_rx);
 			__napi_schedule(napi);
 		}
 		break;
@@ -329,7 +323,7 @@ static int cfv_rx_poll(struct napi_struct *napi, int quota)
 		dev_kfree_skb(skb);
 		/* Stop NAPI poll on OOM, we hope to be polled later */
 		napi_complete(napi);
-		vringh_notify_enable_kern(cfv->vr_rx);
+		vringh_notify_enable(cfv->vr_rx);
 		break;
 
 	default:
@@ -337,12 +331,12 @@ static int cfv_rx_poll(struct napi_struct *napi, int quota)
 		netdev_warn(cfv->ndev, "Bad ring, disable device\n");
 		cfv->ndev->stats.rx_dropped = riov->used - riov->i;
 		napi_complete(napi);
-		vringh_notify_disable_kern(cfv->vr_rx);
+		vringh_notify_disable(cfv->vr_rx);
 		netif_carrier_off(cfv->ndev);
 		break;
 	}
 out:
-	if (rxcnt && vringh_need_notify_kern(cfv->vr_rx) > 0)
+	if (rxcnt && vringh_need_notify(cfv->vr_rx) > 0)
 		vringh_notify(cfv->vr_rx);
 	return rxcnt;
 }
@@ -352,7 +346,7 @@ static void cfv_recv(struct virtio_device *vdev, struct vringh *vr_rx)
 	struct cfv_info *cfv = vdev->priv;
 
 	++cfv->stats.rx_kicks;
-	vringh_notify_disable_kern(cfv->vr_rx);
+	vringh_notify_disable(cfv->vr_rx);
 	napi_schedule(&cfv->napi);
 }
 
@@ -460,7 +454,7 @@ static int cfv_netdev_close(struct net_device *netdev)
 	/* Disable interrupts, queues and NAPI polling */
 	netif_carrier_off(netdev);
 	virtqueue_disable_cb(cfv->vq_tx);
-	vringh_notify_disable_kern(cfv->vr_rx);
+	vringh_notify_disable(cfv->vr_rx);
 	napi_disable(&cfv->napi);
 
 	/* Release any TX buffers on both used and available rings */
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization