[PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

From Yishai,

This series exposes API to enable a dynamic allocation and management of a
UAR which now becomes to be a regular uobject.

Moving to that mode enables allocating a UAR only upon demand and drop the
redundant static allocation of UARs upon context creation.

In addition, it allows master and secondary processes that own the same command
FD to allocate and manage UARs according to their needs, this can’t be achieved
today.

As part of this option, QP & CQ creation flows were adapted to support this
dynamic UAR mode once asked by user space.

Once this mode is asked by mlx5 user space driver on a given context, it will
be mutual exclusive, means both the static and legacy dynamic modes for using
UARs will be blocked.

The legacy modes are supported for backward compatible reasons, looking
forward we expect this new mode to be the default.

Thanks

Yishai Hadas (4):
  IB/mlx5: Expose UAR object and its alloc/destroy commands
  IB/mlx5: Extend CQ creation to get uar page index from user space
  IB/mlx5: Extend QP creation to get uar page index from user space
  IB/mlx5: Move to fully dynamic UAR mode once user space supports it

 drivers/infiniband/hw/mlx5/cq.c           |  21 ++-
 drivers/infiniband/hw/mlx5/main.c         | 185 ++++++++++++++++++++--
 drivers/infiniband/hw/mlx5/mlx5_ib.h      |   2 +
 drivers/infiniband/hw/mlx5/qp.c           |  33 ++--
 include/linux/mlx5/driver.h               |   1 +
 include/rdma/uverbs_ioctl.h               |   2 +-
 include/uapi/rdma/mlx5-abi.h              |   6 +
 include/uapi/rdma/mlx5_user_ioctl_cmds.h  |  18 +++
 include/uapi/rdma/mlx5_user_ioctl_verbs.h |   5 +
 9 files changed, 246 insertions(+), 27 deletions(-)

--
2.24.1

[PATCH mlx5-next 4/4] IB/mlx5: Move to fully dynamic UAR mode once user space supports it

[Leon Romanovsky]

From: Yishai Hadas <yishaih@mellanox.com>

Move to fully dynamic UAR mode once user space supports it.
In this case we prevent any legacy mode of UARs on the allocated context
and prevent redundant allocation of the static ones.

Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Reviewed-by: Michael Guralnik <michaelgur@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/cq.c   |  8 ++++++--
 drivers/infiniband/hw/mlx5/main.c | 13 ++++++++++++-
 drivers/infiniband/hw/mlx5/qp.c   |  6 ++++++
 include/linux/mlx5/driver.h       |  1 +
 include/uapi/rdma/mlx5-abi.h      |  1 +
 5 files changed, 26 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
index eafedc2f697b..146ba2966744 100644
--- a/drivers/infiniband/hw/mlx5/cq.c
+++ b/drivers/infiniband/hw/mlx5/cq.c
@@ -764,10 +764,14 @@ static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
 	MLX5_SET(cqc, cqc, log_page_size,
 		 page_shift - MLX5_ADAPTER_PAGE_SHIFT);
 
-	if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX)
+	if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX) {
 		*index = ucmd.uar_page_index;
-	else
+	} else if (context->bfregi.lib_uar_dyn) {
+		err = -EINVAL;
+		goto err_cqb;
+	} else {
 		*index = context->bfregi.sys_pages[0];
+	}
 
 	if (ucmd.cqe_comp_en == 1) {
 		int mini_cqe_format;
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index e8787af2d74d..e355e06bf3ac 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -1787,6 +1787,7 @@ static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx,
 				     max_cqe_version);
 	u32 dump_fill_mkey;
 	bool lib_uar_4k;
+	bool lib_uar_dyn;
 
 	if (!dev->ib_active)
 		return -EAGAIN;
@@ -1845,8 +1846,14 @@ static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx,
 	}
 
 	lib_uar_4k = req.lib_caps & MLX5_LIB_CAP_4K_UAR;
+	lib_uar_dyn = req.lib_caps & MLX5_LIB_CAP_DYN_UAR;
 	bfregi = &context->bfregi;
 
+	if (lib_uar_dyn) {
+		bfregi->lib_uar_dyn = lib_uar_dyn;
+		goto uar_done;
+	}
+
 	/* updates req->total_num_bfregs */
 	err = calc_total_bfregs(dev, lib_uar_4k, &req, bfregi);
 	if (err)
@@ -1873,6 +1880,7 @@ static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx,
 	if (err)
 		goto out_sys_pages;
 
+uar_done:
 	if (req.flags & MLX5_IB_ALLOC_UCTX_DEVX) {
 		err = mlx5_ib_devx_create(dev, true);
 		if (err < 0)
@@ -1894,7 +1902,7 @@ static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx,
 	INIT_LIST_HEAD(&context->db_page_list);
 	mutex_init(&context->db_page_mutex);
 
-	resp.tot_bfregs = req.total_num_bfregs;
+	resp.tot_bfregs = lib_uar_dyn ? 0 : req.total_num_bfregs;
 	resp.num_ports = dev->num_ports;
 
 	if (offsetofend(typeof(resp), cqe_version) <= udata->outlen)
@@ -2142,6 +2150,9 @@ static int uar_mmap(struct mlx5_ib_dev *dev, enum mlx5_ib_mmap_cmd cmd,
 	int max_valid_idx = dyn_uar ? bfregi->num_sys_pages :
 				bfregi->num_static_sys_pages;
 
+	if (bfregi->lib_uar_dyn)
+		return -EINVAL;
+
 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 		return -EINVAL;
 
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 380ba3321851..319d514a2223 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -697,6 +697,9 @@ static int alloc_bfreg(struct mlx5_ib_dev *dev,
 {
 	int bfregn = -ENOMEM;
 
+	if (bfregi->lib_uar_dyn)
+		return -EINVAL;
+
 	mutex_lock(&bfregi->lock);
 	if (bfregi->ver >= 2) {
 		bfregn = alloc_high_class_bfreg(dev, bfregi);
@@ -768,6 +771,9 @@ int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
 	u32 index_of_sys_page;
 	u32 offset;
 
+	if (bfregi->lib_uar_dyn)
+		return -EINVAL;
+
 	bfregs_per_sys_page = get_uars_per_sys_page(dev, bfregi->lib_uar_4k) *
 				MLX5_NON_FP_BFREGS_PER_UAR;
 	index_of_sys_page = bfregn / bfregs_per_sys_page;
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index 3f10a9633012..e4ab0eb9d202 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -224,6 +224,7 @@ struct mlx5_bfreg_info {
 	struct mutex		lock;
 	u32			ver;
 	bool			lib_uar_4k;
+	u8			lib_uar_dyn : 1;
 	u32			num_sys_pages;
 	u32			num_static_sys_pages;
 	u32			total_num_bfregs;
diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-abi.h
index a65d60b44829..df1cc3641bda 100644
--- a/include/uapi/rdma/mlx5-abi.h
+++ b/include/uapi/rdma/mlx5-abi.h
@@ -79,6 +79,7 @@ struct mlx5_ib_alloc_ucontext_req {
 
 enum mlx5_lib_caps {
 	MLX5_LIB_CAP_4K_UAR	= (__u64)1 << 0,
+	MLX5_LIB_CAP_DYN_UAR	= (__u64)1 << 1,
 };
 
 enum mlx5_ib_alloc_uctx_v2_flags {
-- 
2.24.1

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Jason Gunthorpe]

On Wed, Mar 18, 2020 at 02:43:25PM +0200, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@mellanox.com>
> 
> From Yishai,
> 
> This series exposes API to enable a dynamic allocation and management of a
> UAR which now becomes to be a regular uobject.
> 
> Moving to that mode enables allocating a UAR only upon demand and drop the
> redundant static allocation of UARs upon context creation.
> 
> In addition, it allows master and secondary processes that own the same command
> FD to allocate and manage UARs according to their needs, this can’t be achieved
> today.
> 
> As part of this option, QP & CQ creation flows were adapted to support this
> dynamic UAR mode once asked by user space.
> 
> Once this mode is asked by mlx5 user space driver on a given context, it will
> be mutual exclusive, means both the static and legacy dynamic modes for using
> UARs will be blocked.
> 
> The legacy modes are supported for backward compatible reasons, looking
> forward we expect this new mode to be the default.

We are starting to accumulate a lot of code that is now old-rdma-core
only.

I have been wondering if we should add something like

#if CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION < 21
#endif

So we can keep track of what is actually a used code flow and what is
now hard to test legacy code.

eg this config would also disable the write interface(), turn off
compat write interfaces as they are switched to use ioctl, etc, etc.

Jason

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Leon Romanovsky]

On Wed, Mar 18, 2020 at 09:54:59AM -0300, Jason Gunthorpe wrote:
> On Wed, Mar 18, 2020 at 02:43:25PM +0200, Leon Romanovsky wrote:
> > From: Leon Romanovsky <leonro@mellanox.com>
> >
> > From Yishai,
> >
> > This series exposes API to enable a dynamic allocation and management of a
> > UAR which now becomes to be a regular uobject.
> >
> > Moving to that mode enables allocating a UAR only upon demand and drop the
> > redundant static allocation of UARs upon context creation.
> >
> > In addition, it allows master and secondary processes that own the same command
> > FD to allocate and manage UARs according to their needs, this can’t be achieved
> > today.
> >
> > As part of this option, QP & CQ creation flows were adapted to support this
> > dynamic UAR mode once asked by user space.
> >
> > Once this mode is asked by mlx5 user space driver on a given context, it will
> > be mutual exclusive, means both the static and legacy dynamic modes for using
> > UARs will be blocked.
> >
> > The legacy modes are supported for backward compatible reasons, looking
> > forward we expect this new mode to be the default.
>
> We are starting to accumulate a lot of code that is now old-rdma-core
> only.

Agree

>
> I have been wondering if we should add something like
>
> #if CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION < 21
> #endif

From one side it will definitely help to see old code, but from another
it will create many ifdef inside of the code with a very little chance
of testing. Also we will continue to have the same problem to decide when
we can delete this code.

>
> So we can keep track of what is actually a used code flow and what is
> now hard to test legacy code.
>
> eg this config would also disable the write interface(), turn off
> compat write interfaces as they are switched to use ioctl, etc, etc.

What about if we introduce one ifdef, let's say CONFIG_INFINIBAND_LEGACY
and put everything that will be declared as legacy to that bucket? And
once every 5 (???) years delete everything from that bucket.

>
> Jason

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Jason Gunthorpe]

On Wed, Mar 18, 2020 at 03:14:50PM +0200, Leon Romanovsky wrote:
> On Wed, Mar 18, 2020 at 09:54:59AM -0300, Jason Gunthorpe wrote:
> > On Wed, Mar 18, 2020 at 02:43:25PM +0200, Leon Romanovsky wrote:
> > > From: Leon Romanovsky <leonro@mellanox.com>
> > >
> > > From Yishai,
> > >
> > > This series exposes API to enable a dynamic allocation and management of a
> > > UAR which now becomes to be a regular uobject.
> > >
> > > Moving to that mode enables allocating a UAR only upon demand and drop the
> > > redundant static allocation of UARs upon context creation.
> > >
> > > In addition, it allows master and secondary processes that own the same command
> > > FD to allocate and manage UARs according to their needs, this can’t be achieved
> > > today.
> > >
> > > As part of this option, QP & CQ creation flows were adapted to support this
> > > dynamic UAR mode once asked by user space.
> > >
> > > Once this mode is asked by mlx5 user space driver on a given context, it will
> > > be mutual exclusive, means both the static and legacy dynamic modes for using
> > > UARs will be blocked.
> > >
> > > The legacy modes are supported for backward compatible reasons, looking
> > > forward we expect this new mode to be the default.
> >
> > We are starting to accumulate a lot of code that is now old-rdma-core
> > only.
> 
> Agree
> 
> >
> > I have been wondering if we should add something like
> >
> > #if CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION < 21
> > #endif
> 
> From one side it will definitely help to see old code, but from another
> it will create many ifdef inside of the code with a very little chance
> of testing. Also we will continue to have the same problem to decide when
> we can delete this code.

Well, it doesn't have to be an #ifdef, eg just sticking

if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
     return -ENOPROTOOPT;

at the top of obsolete functions would go a long way

> > So we can keep track of what is actually a used code flow and what is
> > now hard to test legacy code.
> >
> > eg this config would also disable the write interface(), turn off
> > compat write interfaces as they are switched to use ioctl, etc, etc.
> 
> What about if we introduce one ifdef, let's say CONFIG_INFINIBAND_LEGACY
> and put everything that will be declared as legacy to that bucket? And
> once every 5 (???) years delete everything from that bucket.

It is much harder to see what is really old vs only a little old

I'm not sure we can ever completely delete any of this, but at least
the distros can make an informed choice to either do more detailed
test of old libraries or disable those code paths.

Jason

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Leon Romanovsky]

On Wed, Mar 18, 2020 at 10:21:00AM -0300, Jason Gunthorpe wrote:
> On Wed, Mar 18, 2020 at 03:14:50PM +0200, Leon Romanovsky wrote:
> > On Wed, Mar 18, 2020 at 09:54:59AM -0300, Jason Gunthorpe wrote:
> > > On Wed, Mar 18, 2020 at 02:43:25PM +0200, Leon Romanovsky wrote:
> > > > From: Leon Romanovsky <leonro@mellanox.com>
> > > >
> > > > From Yishai,
> > > >
> > > > This series exposes API to enable a dynamic allocation and management of a
> > > > UAR which now becomes to be a regular uobject.
> > > >
> > > > Moving to that mode enables allocating a UAR only upon demand and drop the
> > > > redundant static allocation of UARs upon context creation.
> > > >
> > > > In addition, it allows master and secondary processes that own the same command
> > > > FD to allocate and manage UARs according to their needs, this can’t be achieved
> > > > today.
> > > >
> > > > As part of this option, QP & CQ creation flows were adapted to support this
> > > > dynamic UAR mode once asked by user space.
> > > >
> > > > Once this mode is asked by mlx5 user space driver on a given context, it will
> > > > be mutual exclusive, means both the static and legacy dynamic modes for using
> > > > UARs will be blocked.
> > > >
> > > > The legacy modes are supported for backward compatible reasons, looking
> > > > forward we expect this new mode to be the default.
> > >
> > > We are starting to accumulate a lot of code that is now old-rdma-core
> > > only.
> >
> > Agree
> >
> > >
> > > I have been wondering if we should add something like
> > >
> > > #if CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION < 21
> > > #endif
> >
> > From one side it will definitely help to see old code, but from another
> > it will create many ifdef inside of the code with a very little chance
> > of testing. Also we will continue to have the same problem to decide when
> > we can delete this code.
>
> Well, it doesn't have to be an #ifdef, eg just sticking
>
> if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
>      return -ENOPROTOOPT;
>
> at the top of obsolete functions would go a long way

First, how will you set this min_version? hordcoded in the kernel code?
Second, it will work for simple flows, but can be extremely complex
if your code looks like:
if (old_version)
 do something
if (new version)
 do something else

You will need to add logic to handle this -ENOPROTOOPT error value.

>
> > > So we can keep track of what is actually a used code flow and what is
> > > now hard to test legacy code.
> > >
> > > eg this config would also disable the write interface(), turn off
> > > compat write interfaces as they are switched to use ioctl, etc, etc.
> >
> > What about if we introduce one ifdef, let's say CONFIG_INFINIBAND_LEGACY
> > and put everything that will be declared as legacy to that bucket? And
> > once every 5 (???) years delete everything from that bucket.
>
> It is much harder to see what is really old vs only a little old
>
> I'm not sure we can ever completely delete any of this, but at least
> the distros can make an informed choice to either do more detailed
> test of old libraries or disable those code paths.

It will be nice to hear how distros decide to disable/drop the code.

Thanks

>
> Jason

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Jason Gunthorpe]

On Wed, Mar 18, 2020 at 03:56:31PM +0200, Leon Romanovsky wrote:
> On Wed, Mar 18, 2020 at 10:21:00AM -0300, Jason Gunthorpe wrote:
> > On Wed, Mar 18, 2020 at 03:14:50PM +0200, Leon Romanovsky wrote:
> > > On Wed, Mar 18, 2020 at 09:54:59AM -0300, Jason Gunthorpe wrote:
> > > > On Wed, Mar 18, 2020 at 02:43:25PM +0200, Leon Romanovsky wrote:
> > > > > From: Leon Romanovsky <leonro@mellanox.com>
> > > > >
> > > > > From Yishai,
> > > > >
> > > > > This series exposes API to enable a dynamic allocation and management of a
> > > > > UAR which now becomes to be a regular uobject.
> > > > >
> > > > > Moving to that mode enables allocating a UAR only upon demand and drop the
> > > > > redundant static allocation of UARs upon context creation.
> > > > >
> > > > > In addition, it allows master and secondary processes that own the same command
> > > > > FD to allocate and manage UARs according to their needs, this can’t be achieved
> > > > > today.
> > > > >
> > > > > As part of this option, QP & CQ creation flows were adapted to support this
> > > > > dynamic UAR mode once asked by user space.
> > > > >
> > > > > Once this mode is asked by mlx5 user space driver on a given context, it will
> > > > > be mutual exclusive, means both the static and legacy dynamic modes for using
> > > > > UARs will be blocked.
> > > > >
> > > > > The legacy modes are supported for backward compatible reasons, looking
> > > > > forward we expect this new mode to be the default.
> > > >
> > > > We are starting to accumulate a lot of code that is now old-rdma-core
> > > > only.
> > >
> > > Agree
> > >
> > > >
> > > > I have been wondering if we should add something like
> > > >
> > > > #if CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION < 21
> > > > #endif
> > >
> > > From one side it will definitely help to see old code, but from another
> > > it will create many ifdef inside of the code with a very little chance
> > > of testing. Also we will continue to have the same problem to decide when
> > > we can delete this code.
> >
> > Well, it doesn't have to be an #ifdef, eg just sticking
> >
> > if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
> >      return -ENOPROTOOPT;
> >
> > at the top of obsolete functions would go a long way
> 
> First, how will you set this min_version? hordcoded in the kernel
> code?

Yes, when a rdma-core release obsoletes the code path then it can
become annotated.

> Second, it will work for simple flows, but can be extremely complex
> if your code looks like:
> if (old_version)
>  do something
> if (new version)
>  do something else

Well, we'd avoid making such complications, it would be something like

if (flag & foo) {
   if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
      return -ENOPROTOOPT;
  [keep going as before]
}

At least we now know this conditional path isn't used / isn't covered
by testing

Doug? What does a distro think?

Jason

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Leon Romanovsky]

On Wed, Mar 18, 2020 at 11:00:01AM -0300, Jason Gunthorpe wrote:
> On Wed, Mar 18, 2020 at 03:56:31PM +0200, Leon Romanovsky wrote:
> > On Wed, Mar 18, 2020 at 10:21:00AM -0300, Jason Gunthorpe wrote:
> > > On Wed, Mar 18, 2020 at 03:14:50PM +0200, Leon Romanovsky wrote:
> > > > On Wed, Mar 18, 2020 at 09:54:59AM -0300, Jason Gunthorpe wrote:
> > > > > On Wed, Mar 18, 2020 at 02:43:25PM +0200, Leon Romanovsky wrote:
> > > > > > From: Leon Romanovsky <leonro@mellanox.com>
> > > > > >
> > > > > > From Yishai,
> > > > > >
> > > > > > This series exposes API to enable a dynamic allocation and management of a
> > > > > > UAR which now becomes to be a regular uobject.
> > > > > >
> > > > > > Moving to that mode enables allocating a UAR only upon demand and drop the
> > > > > > redundant static allocation of UARs upon context creation.
> > > > > >
> > > > > > In addition, it allows master and secondary processes that own the same command
> > > > > > FD to allocate and manage UARs according to their needs, this can’t be achieved
> > > > > > today.
> > > > > >
> > > > > > As part of this option, QP & CQ creation flows were adapted to support this
> > > > > > dynamic UAR mode once asked by user space.
> > > > > >
> > > > > > Once this mode is asked by mlx5 user space driver on a given context, it will
> > > > > > be mutual exclusive, means both the static and legacy dynamic modes for using
> > > > > > UARs will be blocked.
> > > > > >
> > > > > > The legacy modes are supported for backward compatible reasons, looking
> > > > > > forward we expect this new mode to be the default.
> > > > >
> > > > > We are starting to accumulate a lot of code that is now old-rdma-core
> > > > > only.
> > > >
> > > > Agree
> > > >
> > > > >
> > > > > I have been wondering if we should add something like
> > > > >
> > > > > #if CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION < 21
> > > > > #endif
> > > >
> > > > From one side it will definitely help to see old code, but from another
> > > > it will create many ifdef inside of the code with a very little chance
> > > > of testing. Also we will continue to have the same problem to decide when
> > > > we can delete this code.
> > >
> > > Well, it doesn't have to be an #ifdef, eg just sticking
> > >
> > > if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
> > >      return -ENOPROTOOPT;
> > >
> > > at the top of obsolete functions would go a long way
> >
> > First, how will you set this min_version? hordcoded in the kernel
> > code?
>
> Yes, when a rdma-core release obsoletes the code path then it can
> become annotated.
>
> > Second, it will work for simple flows, but can be extremely complex
> > if your code looks like:
> > if (old_version)
> >  do something
> > if (new version)
> >  do something else
>
> Well, we'd avoid making such complications, it would be something like
>
> if (flag & foo) {
>    if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
>       return -ENOPROTOOPT;
>   [keep going as before]
> }
>
> At least we now know this conditional path isn't used / isn't covered
> by testing

I'm ok with this approach because it helps us to find those dead
paths, but have last question, shouldn't this be achieved with
proper documentation of every flag instead of adding CONFIG_..?

>
> Doug? What does a distro think?
>
> Jason

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Jason Gunthorpe]

On Wed, Mar 18, 2020 at 04:09:32PM +0200, Leon Romanovsky wrote:
> On Wed, Mar 18, 2020 at 11:00:01AM -0300, Jason Gunthorpe wrote:
> > On Wed, Mar 18, 2020 at 03:56:31PM +0200, Leon Romanovsky wrote:
> > > On Wed, Mar 18, 2020 at 10:21:00AM -0300, Jason Gunthorpe wrote:
> > > > On Wed, Mar 18, 2020 at 03:14:50PM +0200, Leon Romanovsky wrote:
> > > > > On Wed, Mar 18, 2020 at 09:54:59AM -0300, Jason Gunthorpe wrote:
> > > > > > On Wed, Mar 18, 2020 at 02:43:25PM +0200, Leon Romanovsky wrote:
> > > > > > > From: Leon Romanovsky <leonro@mellanox.com>
> > > > > > >
> > > > > > > From Yishai,
> > > > > > >
> > > > > > > This series exposes API to enable a dynamic allocation and management of a
> > > > > > > UAR which now becomes to be a regular uobject.
> > > > > > >
> > > > > > > Moving to that mode enables allocating a UAR only upon demand and drop the
> > > > > > > redundant static allocation of UARs upon context creation.
> > > > > > >
> > > > > > > In addition, it allows master and secondary processes that own the same command
> > > > > > > FD to allocate and manage UARs according to their needs, this can’t be achieved
> > > > > > > today.
> > > > > > >
> > > > > > > As part of this option, QP & CQ creation flows were adapted to support this
> > > > > > > dynamic UAR mode once asked by user space.
> > > > > > >
> > > > > > > Once this mode is asked by mlx5 user space driver on a given context, it will
> > > > > > > be mutual exclusive, means both the static and legacy dynamic modes for using
> > > > > > > UARs will be blocked.
> > > > > > >
> > > > > > > The legacy modes are supported for backward compatible reasons, looking
> > > > > > > forward we expect this new mode to be the default.
> > > > > >
> > > > > > We are starting to accumulate a lot of code that is now old-rdma-core
> > > > > > only.
> > > > >
> > > > > Agree
> > > > >
> > > > > >
> > > > > > I have been wondering if we should add something like
> > > > > >
> > > > > > #if CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION < 21
> > > > > > #endif
> > > > >
> > > > > From one side it will definitely help to see old code, but from another
> > > > > it will create many ifdef inside of the code with a very little chance
> > > > > of testing. Also we will continue to have the same problem to decide when
> > > > > we can delete this code.
> > > >
> > > > Well, it doesn't have to be an #ifdef, eg just sticking
> > > >
> > > > if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
> > > >      return -ENOPROTOOPT;
> > > >
> > > > at the top of obsolete functions would go a long way
> > >
> > > First, how will you set this min_version? hordcoded in the kernel
> > > code?
> >
> > Yes, when a rdma-core release obsoletes the code path then it can
> > become annotated.
> >
> > > Second, it will work for simple flows, but can be extremely complex
> > > if your code looks like:
> > > if (old_version)
> > >  do something
> > > if (new version)
> > >  do something else
> >
> > Well, we'd avoid making such complications, it would be something like
> >
> > if (flag & foo) {
> >    if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
> >       return -ENOPROTOOPT;
> >   [keep going as before]
> > }
> >
> > At least we now know this conditional path isn't used / isn't covered
> > by testing
> 
> I'm ok with this approach because it helps us to find those dead
> paths, but have last question, shouldn't this be achieved with
> proper documentation of every flag instead of adding CONFIG_..?

How do you mean?

The other half of this idea is to disable obsolete un tested code to
avoid potential bugs. Which requires CONFIG_?

Jason

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Leon Romanovsky]

On Wed, Mar 18, 2020 at 11:12:08AM -0300, Jason Gunthorpe wrote:
> On Wed, Mar 18, 2020 at 04:09:32PM +0200, Leon Romanovsky wrote:
> > On Wed, Mar 18, 2020 at 11:00:01AM -0300, Jason Gunthorpe wrote:
> > > On Wed, Mar 18, 2020 at 03:56:31PM +0200, Leon Romanovsky wrote:
> > > > On Wed, Mar 18, 2020 at 10:21:00AM -0300, Jason Gunthorpe wrote:
> > > > > On Wed, Mar 18, 2020 at 03:14:50PM +0200, Leon Romanovsky wrote:
> > > > > > On Wed, Mar 18, 2020 at 09:54:59AM -0300, Jason Gunthorpe wrote:
> > > > > > > On Wed, Mar 18, 2020 at 02:43:25PM +0200, Leon Romanovsky wrote:
> > > > > > > > From: Leon Romanovsky <leonro@mellanox.com>
> > > > > > > >
> > > > > > > > From Yishai,
> > > > > > > >
> > > > > > > > This series exposes API to enable a dynamic allocation and management of a
> > > > > > > > UAR which now becomes to be a regular uobject.
> > > > > > > >
> > > > > > > > Moving to that mode enables allocating a UAR only upon demand and drop the
> > > > > > > > redundant static allocation of UARs upon context creation.
> > > > > > > >
> > > > > > > > In addition, it allows master and secondary processes that own the same command
> > > > > > > > FD to allocate and manage UARs according to their needs, this can’t be achieved
> > > > > > > > today.
> > > > > > > >
> > > > > > > > As part of this option, QP & CQ creation flows were adapted to support this
> > > > > > > > dynamic UAR mode once asked by user space.
> > > > > > > >
> > > > > > > > Once this mode is asked by mlx5 user space driver on a given context, it will
> > > > > > > > be mutual exclusive, means both the static and legacy dynamic modes for using
> > > > > > > > UARs will be blocked.
> > > > > > > >
> > > > > > > > The legacy modes are supported for backward compatible reasons, looking
> > > > > > > > forward we expect this new mode to be the default.
> > > > > > >
> > > > > > > We are starting to accumulate a lot of code that is now old-rdma-core
> > > > > > > only.
> > > > > >
> > > > > > Agree
> > > > > >
> > > > > > >
> > > > > > > I have been wondering if we should add something like
> > > > > > >
> > > > > > > #if CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION < 21
> > > > > > > #endif
> > > > > >
> > > > > > From one side it will definitely help to see old code, but from another
> > > > > > it will create many ifdef inside of the code with a very little chance
> > > > > > of testing. Also we will continue to have the same problem to decide when
> > > > > > we can delete this code.
> > > > >
> > > > > Well, it doesn't have to be an #ifdef, eg just sticking
> > > > >
> > > > > if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
> > > > >      return -ENOPROTOOPT;
> > > > >
> > > > > at the top of obsolete functions would go a long way
> > > >
> > > > First, how will you set this min_version? hordcoded in the kernel
> > > > code?
> > >
> > > Yes, when a rdma-core release obsoletes the code path then it can
> > > become annotated.
> > >
> > > > Second, it will work for simple flows, but can be extremely complex
> > > > if your code looks like:
> > > > if (old_version)
> > > >  do something
> > > > if (new version)
> > > >  do something else
> > >
> > > Well, we'd avoid making such complications, it would be something like
> > >
> > > if (flag & foo) {
> > >    if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 21)
> > >       return -ENOPROTOOPT;
> > >   [keep going as before]
> > > }
> > >
> > > At least we now know this conditional path isn't used / isn't covered
> > > by testing
> >
> > I'm ok with this approach because it helps us to find those dead
> > paths, but have last question, shouldn't this be achieved with
> > proper documentation of every flag instead of adding CONFIG_..?
>
> How do you mean?
>
> The other half of this idea is to disable obsolete un tested code to
> avoid potential bugs. Which requires CONFIG_?

The second part is achievable by distros when they will decide to
support starting from version X. The same decision is not so easy
to do in the upstream.

Let's take as an example this feature. It will be set as default from
rdma-core v29 and the legacy code will be guarded by
"if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 29)". When will change
CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION to be above 29? So we will
delete such legacy code.

In the world where we are not breaking user space, it will never happen.

It means that upstream doesn't get anything from those CONFIG_*s.

Thanks

>
> Jason

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Jason Gunthorpe]

On Wed, Mar 18, 2020 at 04:24:55PM +0200, Leon Romanovsky wrote:
> > > I'm ok with this approach because it helps us to find those dead
> > > paths, but have last question, shouldn't this be achieved with
> > > proper documentation of every flag instead of adding CONFIG_..?
> >
> > How do you mean?
> >
> > The other half of this idea is to disable obsolete un tested code to
> > avoid potential bugs. Which requires CONFIG_?
> 
> The second part is achievable by distros when they will decide to
> support starting from version X. The same decision is not so easy
> to do in the upstream.

Upstream will probably carry the code for a long, long time, that
doesn't mean the distros don't get value by using a shorter time
window

> Let's take as an example this feature. It will be set as default from
> rdma-core v29 and the legacy code will be guarded by
> "if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 29)". When will change
> CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION to be above 29? So we will
> delete such legacy code.

First the distros will decide in their own kconfigs where they want to
set the value.

Then the upstream kernel will decide some default value

Then maybe we could talk about lowest values when enough of the user
community uses a higher value

Jason

Re: [PATCH rdma-next 0/4] Introduce dynamic UAR allocation mode

[Leon Romanovsky]

On Wed, Mar 18, 2020 at 11:39:03AM -0300, Jason Gunthorpe wrote:
> On Wed, Mar 18, 2020 at 04:24:55PM +0200, Leon Romanovsky wrote:
> > > > I'm ok with this approach because it helps us to find those dead
> > > > paths, but have last question, shouldn't this be achieved with
> > > > proper documentation of every flag instead of adding CONFIG_..?
> > >
> > > How do you mean?
> > >
> > > The other half of this idea is to disable obsolete un tested code to
> > > avoid potential bugs. Which requires CONFIG_?
> >
> > The second part is achievable by distros when they will decide to
> > support starting from version X. The same decision is not so easy
> > to do in the upstream.
>
> Upstream will probably carry the code for a long, long time, that
> doesn't mean the distros don't get value by using a shorter time
> window

Sure

>
> > Let's take as an example this feature. It will be set as default from
> > rdma-core v29 and the legacy code will be guarded by
> > "if (CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION >= 29)". When will change
> > CONFIG_INFINIBAND_MIN_RDMA_CORE_VERSION to be above 29? So we will
> > delete such legacy code.
>
> First the distros will decide in their own kconfigs where they want to
> set the value.
>
> Then the upstream kernel will decide some default value
>
> Then maybe we could talk about lowest values when enough of the user
> community uses a higher value

I think that you over-optimistic here, but let's hear other voices here.

Thanks

>
> Jason

Re: [PATCH mlx5-next 4/4] IB/mlx5: Move to fully dynamic UAR mode once user space supports it

[Saeed Mahameed]

On Wed, 2020-03-18 at 14:43 +0200, Leon Romanovsky wrote:
> From: Yishai Hadas <yishaih@mellanox.com>
> 
> Move to fully dynamic UAR mode once user space supports it.
> In this case we prevent any legacy mode of UARs on the allocated
> context
> and prevent redundant allocation of the static ones.
> 
> Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
> Reviewed-by: Michael Guralnik <michaelgur@mellanox.com>
> Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
> ---
>  drivers/infiniband/hw/mlx5/cq.c   |  8 ++++++--
>  drivers/infiniband/hw/mlx5/main.c | 13 ++++++++++++-
>  drivers/infiniband/hw/mlx5/qp.c   |  6 ++++++
>  include/linux/mlx5/driver.h       |  1 +
>  include/uapi/rdma/mlx5-abi.h      |  1 +
>  5 files changed, 26 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/infiniband/hw/mlx5/cq.c
> b/drivers/infiniband/hw/mlx5/cq.c
> index eafedc2f697b..146ba2966744 100644
> --- a/drivers/infiniband/hw/mlx5/cq.c
> +++ b/drivers/infiniband/hw/mlx5/cq.c
> @@ -764,10 +764,14 @@ static int create_cq_user(struct mlx5_ib_dev
> *dev, struct ib_udata *udata,
>  	MLX5_SET(cqc, cqc, log_page_size,
>  		 page_shift - MLX5_ADAPTER_PAGE_SHIFT);
>  
> -	if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX)
> +	if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX) {
>  		*index = ucmd.uar_page_index;
> -	else
> +	} else if (context->bfregi.lib_uar_dyn) {
> +		err = -EINVAL;
> +		goto err_cqb;
> +	} else {
>  		*index = context->bfregi.sys_pages[0];
> +	}
>  
>  	if (ucmd.cqe_comp_en == 1) {
>  		int mini_cqe_format;
> diff --git a/drivers/infiniband/hw/mlx5/main.c
> b/drivers/infiniband/hw/mlx5/main.c
> index e8787af2d74d..e355e06bf3ac 100644
> --- a/drivers/infiniband/hw/mlx5/main.c
> +++ b/drivers/infiniband/hw/mlx5/main.c
> @@ -1787,6 +1787,7 @@ static int mlx5_ib_alloc_ucontext(struct
> ib_ucontext *uctx,
>  				     max_cqe_version);
>  	u32 dump_fill_mkey;
>  	bool lib_uar_4k;
> +	bool lib_uar_dyn;
>  
>  	if (!dev->ib_active)
>  		return -EAGAIN;
> @@ -1845,8 +1846,14 @@ static int mlx5_ib_alloc_ucontext(struct
> ib_ucontext *uctx,
>  	}
>  
>  	lib_uar_4k = req.lib_caps & MLX5_LIB_CAP_4K_UAR;
> +	lib_uar_dyn = req.lib_caps & MLX5_LIB_CAP_DYN_UAR;
>  	bfregi = &context->bfregi;
>  
> +	if (lib_uar_dyn) {
> +		bfregi->lib_uar_dyn = lib_uar_dyn;
> +		goto uar_done;
> +	}
> +
>  	/* updates req->total_num_bfregs */
>  	err = calc_total_bfregs(dev, lib_uar_4k, &req, bfregi);
>  	if (err)
> @@ -1873,6 +1880,7 @@ static int mlx5_ib_alloc_ucontext(struct
> ib_ucontext *uctx,
>  	if (err)
>  		goto out_sys_pages;
>  
> +uar_done:
>  	if (req.flags & MLX5_IB_ALLOC_UCTX_DEVX) {
>  		err = mlx5_ib_devx_create(dev, true);
>  		if (err < 0)
> @@ -1894,7 +1902,7 @@ static int mlx5_ib_alloc_ucontext(struct
> ib_ucontext *uctx,
>  	INIT_LIST_HEAD(&context->db_page_list);
>  	mutex_init(&context->db_page_mutex);
>  
> -	resp.tot_bfregs = req.total_num_bfregs;
> +	resp.tot_bfregs = lib_uar_dyn ? 0 : req.total_num_bfregs;
>  	resp.num_ports = dev->num_ports;
>  
>  	if (offsetofend(typeof(resp), cqe_version) <= udata->outlen)
> @@ -2142,6 +2150,9 @@ static int uar_mmap(struct mlx5_ib_dev *dev,
> enum mlx5_ib_mmap_cmd cmd,
>  	int max_valid_idx = dyn_uar ? bfregi->num_sys_pages :
>  				bfregi->num_static_sys_pages;
>  
> +	if (bfregi->lib_uar_dyn)
> +		return -EINVAL;
> +
>  	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
>  		return -EINVAL;
>  
> diff --git a/drivers/infiniband/hw/mlx5/qp.c
> b/drivers/infiniband/hw/mlx5/qp.c
> index 380ba3321851..319d514a2223 100644
> --- a/drivers/infiniband/hw/mlx5/qp.c
> +++ b/drivers/infiniband/hw/mlx5/qp.c
> @@ -697,6 +697,9 @@ static int alloc_bfreg(struct mlx5_ib_dev *dev,
>  {
>  	int bfregn = -ENOMEM;
>  
> +	if (bfregi->lib_uar_dyn)
> +		return -EINVAL;
> +
>  	mutex_lock(&bfregi->lock);
>  	if (bfregi->ver >= 2) {
>  		bfregn = alloc_high_class_bfreg(dev, bfregi);
> @@ -768,6 +771,9 @@ int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
>  	u32 index_of_sys_page;
>  	u32 offset;
>  
> +	if (bfregi->lib_uar_dyn)
> +		return -EINVAL;
> +
>  	bfregs_per_sys_page = get_uars_per_sys_page(dev, bfregi-
> >lib_uar_4k) *
>  				MLX5_NON_FP_BFREGS_PER_UAR;
>  	index_of_sys_page = bfregn / bfregs_per_sys_page;
> diff --git a/include/linux/mlx5/driver.h
> b/include/linux/mlx5/driver.h
> index 3f10a9633012..e4ab0eb9d202 100644
> --- a/include/linux/mlx5/driver.h
> +++ b/include/linux/mlx5/driver.h
> @@ -224,6 +224,7 @@ struct mlx5_bfreg_info {
>  	struct mutex		lock;
>  	u32			ver;
>  	bool			lib_uar_4k;
> +	u8			lib_uar_dyn : 1;
>  	u32			num_sys_pages;
>  	u32			num_static_sys_pages
>  	u32			total_num_bfregs;

this struct is not used in mlx5_core, shall we move it to mlx5_ib as
part of this patch? 
so next time you need to update it you don't need to bother netdev busy
people about it ;-)

> diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-
> abi.h
> index a65d60b44829..df1cc3641bda 100644
> --- a/include/uapi/rdma/mlx5-abi.h
> +++ b/include/uapi/rdma/mlx5-abi.h
> @@ -79,6 +79,7 @@ struct mlx5_ib_alloc_ucontext_req {
>  
>  enum mlx5_lib_caps {
>  	MLX5_LIB_CAP_4K_UAR	= (__u64)1 << 0,
> +	MLX5_LIB_CAP_DYN_UAR	= (__u64)1 << 1,
>  };
>  
>  enum mlx5_ib_alloc_uctx_v2_flags {

Re: [PATCH mlx5-next 4/4] IB/mlx5: Move to fully dynamic UAR mode once user space supports it

[Leon Romanovsky]

On Wed, Mar 18, 2020 at 11:38:32PM +0000, Saeed Mahameed wrote:
> On Wed, 2020-03-18 at 14:43 +0200, Leon Romanovsky wrote:
> > From: Yishai Hadas <yishaih@mellanox.com>
> >
> > Move to fully dynamic UAR mode once user space supports it.
> > In this case we prevent any legacy mode of UARs on the allocated
> > context
> > and prevent redundant allocation of the static ones.
> >
> > Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
> > Reviewed-by: Michael Guralnik <michaelgur@mellanox.com>
> > Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
> > ---
> >  drivers/infiniband/hw/mlx5/cq.c   |  8 ++++++--
> >  drivers/infiniband/hw/mlx5/main.c | 13 ++++++++++++-
> >  drivers/infiniband/hw/mlx5/qp.c   |  6 ++++++
> >  include/linux/mlx5/driver.h       |  1 +
> >  include/uapi/rdma/mlx5-abi.h      |  1 +
> >  5 files changed, 26 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/infiniband/hw/mlx5/cq.c
> > b/drivers/infiniband/hw/mlx5/cq.c
> > index eafedc2f697b..146ba2966744 100644
> > --- a/drivers/infiniband/hw/mlx5/cq.c
> > +++ b/drivers/infiniband/hw/mlx5/cq.c
> > @@ -764,10 +764,14 @@ static int create_cq_user(struct mlx5_ib_dev
> > *dev, struct ib_udata *udata,
> >  	MLX5_SET(cqc, cqc, log_page_size,
> >  		 page_shift - MLX5_ADAPTER_PAGE_SHIFT);
> >
> > -	if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX)
> > +	if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX) {
> >  		*index = ucmd.uar_page_index;
> > -	else
> > +	} else if (context->bfregi.lib_uar_dyn) {
> > +		err = -EINVAL;
> > +		goto err_cqb;
> > +	} else {
> >  		*index = context->bfregi.sys_pages[0];
> > +	}
> >
> >  	if (ucmd.cqe_comp_en == 1) {
> >  		int mini_cqe_format;
> > diff --git a/drivers/infiniband/hw/mlx5/main.c
> > b/drivers/infiniband/hw/mlx5/main.c
> > index e8787af2d74d..e355e06bf3ac 100644
> > --- a/drivers/infiniband/hw/mlx5/main.c
> > +++ b/drivers/infiniband/hw/mlx5/main.c
> > @@ -1787,6 +1787,7 @@ static int mlx5_ib_alloc_ucontext(struct
> > ib_ucontext *uctx,
> >  				     max_cqe_version);
> >  	u32 dump_fill_mkey;
> >  	bool lib_uar_4k;
> > +	bool lib_uar_dyn;
> >
> >  	if (!dev->ib_active)
> >  		return -EAGAIN;
> > @@ -1845,8 +1846,14 @@ static int mlx5_ib_alloc_ucontext(struct
> > ib_ucontext *uctx,
> >  	}
> >
> >  	lib_uar_4k = req.lib_caps & MLX5_LIB_CAP_4K_UAR;
> > +	lib_uar_dyn = req.lib_caps & MLX5_LIB_CAP_DYN_UAR;
> >  	bfregi = &context->bfregi;
> >
> > +	if (lib_uar_dyn) {
> > +		bfregi->lib_uar_dyn = lib_uar_dyn;
> > +		goto uar_done;
> > +	}
> > +
> >  	/* updates req->total_num_bfregs */
> >  	err = calc_total_bfregs(dev, lib_uar_4k, &req, bfregi);
> >  	if (err)
> > @@ -1873,6 +1880,7 @@ static int mlx5_ib_alloc_ucontext(struct
> > ib_ucontext *uctx,
> >  	if (err)
> >  		goto out_sys_pages;
> >
> > +uar_done:
> >  	if (req.flags & MLX5_IB_ALLOC_UCTX_DEVX) {
> >  		err = mlx5_ib_devx_create(dev, true);
> >  		if (err < 0)
> > @@ -1894,7 +1902,7 @@ static int mlx5_ib_alloc_ucontext(struct
> > ib_ucontext *uctx,
> >  	INIT_LIST_HEAD(&context->db_page_list);
> >  	mutex_init(&context->db_page_mutex);
> >
> > -	resp.tot_bfregs = req.total_num_bfregs;
> > +	resp.tot_bfregs = lib_uar_dyn ? 0 : req.total_num_bfregs;
> >  	resp.num_ports = dev->num_ports;
> >
> >  	if (offsetofend(typeof(resp), cqe_version) <= udata->outlen)
> > @@ -2142,6 +2150,9 @@ static int uar_mmap(struct mlx5_ib_dev *dev,
> > enum mlx5_ib_mmap_cmd cmd,
> >  	int max_valid_idx = dyn_uar ? bfregi->num_sys_pages :
> >  				bfregi->num_static_sys_pages;
> >
> > +	if (bfregi->lib_uar_dyn)
> > +		return -EINVAL;
> > +
> >  	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
> >  		return -EINVAL;
> >
> > diff --git a/drivers/infiniband/hw/mlx5/qp.c
> > b/drivers/infiniband/hw/mlx5/qp.c
> > index 380ba3321851..319d514a2223 100644
> > --- a/drivers/infiniband/hw/mlx5/qp.c
> > +++ b/drivers/infiniband/hw/mlx5/qp.c
> > @@ -697,6 +697,9 @@ static int alloc_bfreg(struct mlx5_ib_dev *dev,
> >  {
> >  	int bfregn = -ENOMEM;
> >
> > +	if (bfregi->lib_uar_dyn)
> > +		return -EINVAL;
> > +
> >  	mutex_lock(&bfregi->lock);
> >  	if (bfregi->ver >= 2) {
> >  		bfregn = alloc_high_class_bfreg(dev, bfregi);
> > @@ -768,6 +771,9 @@ int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
> >  	u32 index_of_sys_page;
> >  	u32 offset;
> >
> > +	if (bfregi->lib_uar_dyn)
> > +		return -EINVAL;
> > +
> >  	bfregs_per_sys_page = get_uars_per_sys_page(dev, bfregi-
> > >lib_uar_4k) *
> >  				MLX5_NON_FP_BFREGS_PER_UAR;
> >  	index_of_sys_page = bfregn / bfregs_per_sys_page;
> > diff --git a/include/linux/mlx5/driver.h
> > b/include/linux/mlx5/driver.h
> > index 3f10a9633012..e4ab0eb9d202 100644
> > --- a/include/linux/mlx5/driver.h
> > +++ b/include/linux/mlx5/driver.h
> > @@ -224,6 +224,7 @@ struct mlx5_bfreg_info {
> >  	struct mutex		lock;
> >  	u32			ver;
> >  	bool			lib_uar_4k;
> > +	u8			lib_uar_dyn : 1;
> >  	u32			num_sys_pages;
> >  	u32			num_static_sys_pages
> >  	u32			total_num_bfregs;
>
> this struct is not used in mlx5_core, shall we move it to mlx5_ib as
> part of this patch?
> so next time you need to update it you don't need to bother netdev busy
> people about it ;-)

Thanks, completely agree. I'll do it now.