diff mbox series

[rdma-next,02/10] RDMA/core: Enable Relaxed Ordering in __ib_alloc_pd()

Message ID 20210405052404.213889-3-leon@kernel.org (mailing list archive)
State New
Headers show
Series Enable relaxed ordering for ULPs | expand

Commit Message

Leon Romanovsky April 5, 2021, 5:23 a.m. UTC
From: Avihai Horon <avihaih@nvidia.com>

Enable Relaxed Ordering in __ib_alloc_pd() allocation of the
local_dma_lkey.

This will take effect only for devices that don't pre-allocate the lkey
but allocate it per PD allocation.

Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/core/verbs.c              | 3 ++-
 drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c | 1 +
 2 files changed, 3 insertions(+), 1 deletion(-)

Comments

Tom Talpey April 5, 2021, 6:01 p.m. UTC | #1
On 4/5/2021 1:23 AM, Leon Romanovsky wrote:
> From: Avihai Horon <avihaih@nvidia.com>
> 
> Enable Relaxed Ordering in __ib_alloc_pd() allocation of the
> local_dma_lkey.
> 
> This will take effect only for devices that don't pre-allocate the lkey
> but allocate it per PD allocation.
> 
> Signed-off-by: Avihai Horon <avihaih@nvidia.com>
> Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
> Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> ---
>   drivers/infiniband/core/verbs.c              | 3 ++-
>   drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c | 1 +
>   2 files changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
> index a1782f8a6ca0..9b719f7d6fd5 100644
> --- a/drivers/infiniband/core/verbs.c
> +++ b/drivers/infiniband/core/verbs.c
> @@ -287,7 +287,8 @@ struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
>   	if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
>   		pd->local_dma_lkey = device->local_dma_lkey;
>   	else
> -		mr_access_flags |= IB_ACCESS_LOCAL_WRITE;
> +		mr_access_flags |=
> +			IB_ACCESS_LOCAL_WRITE | IB_ACCESS_RELAXED_ORDERING;

So, do local_dma_lkey's get relaxed ordering unconditionally?

>   	if (flags & IB_PD_UNSAFE_GLOBAL_RKEY) {
>   		pr_warn("%s: enabling unsafe global rkey\n", caller);
> diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c
> index b3fa783698a0..d74827694f92 100644
> --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c
> +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c
> @@ -66,6 +66,7 @@ struct ib_mr *pvrdma_get_dma_mr(struct ib_pd *pd, int acc)
>   	int ret;
>   
>   	/* Support only LOCAL_WRITE flag for DMA MRs */
> +	acc &= ~IB_ACCESS_RELAXED_ORDERING;
>   	if (acc & ~IB_ACCESS_LOCAL_WRITE) {
>   		dev_warn(&dev->pdev->dev,
>   			 "unsupported dma mr access flags %#x\n", acc);

Why does the pvrdma driver require relaxed ordering to be off?

Tom.
Adit Ranadive April 5, 2021, 8:40 p.m. UTC | #2
On 4/5/21 11:01 AM, Tom Talpey wrote:
> On 4/5/2021 1:23 AM, Leon Romanovsky wrote:
>> From: Avihai Horon <avihaih@nvidia.com>
>>
>> Enable Relaxed Ordering in __ib_alloc_pd() allocation of the
>> local_dma_lkey.
>>
>> This will take effect only for devices that don't pre-allocate the lkey
>> but allocate it per PD allocation.
>>
>> Signed-off-by: Avihai Horon <avihaih@nvidia.com>
>> Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
>> Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
>> ---
>>   drivers/infiniband/core/verbs.c              | 3 ++-
>>   drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c | 1 +
>>   2 files changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
>> index a1782f8a6ca0..9b719f7d6fd5 100644
>> --- a/drivers/infiniband/core/verbs.c
>> +++ b/drivers/infiniband/core/verbs.c
>> @@ -287,7 +287,8 @@ struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
>>       if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
>>           pd->local_dma_lkey = device->local_dma_lkey;
>>       else
>> -        mr_access_flags |= IB_ACCESS_LOCAL_WRITE;
>> +        mr_access_flags |=
>> +            IB_ACCESS_LOCAL_WRITE | IB_ACCESS_RELAXED_ORDERING;
>
> So, do local_dma_lkey's get relaxed ordering unconditionally?
>
>>       if (flags & IB_PD_UNSAFE_GLOBAL_RKEY) {
>>           pr_warn("%s: enabling unsafe global rkey\n", caller);
>> diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c
>> index b3fa783698a0..d74827694f92 100644
>> --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c
>> +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c
>> @@ -66,6 +66,7 @@ struct ib_mr *pvrdma_get_dma_mr(struct ib_pd *pd, int acc)
>>       int ret;
>>         /* Support only LOCAL_WRITE flag for DMA MRs */
>> +    acc &= ~IB_ACCESS_RELAXED_ORDERING;
>>       if (acc & ~IB_ACCESS_LOCAL_WRITE) {
>>           dev_warn(&dev->pdev->dev,
>>                "unsupported dma mr access flags %#x\n", acc);
>
> Why does the pvrdma driver require relaxed ordering to be off?

PVRDMA doesn't support any other flags other than LOCAL_WRITE for
DMA MRs so the MR creation will fail if any new unconditionally added
flag isn't cleared.
Leon Romanovsky April 6, 2021, 6:28 a.m. UTC | #3
On Mon, Apr 05, 2021 at 02:01:16PM -0400, Tom Talpey wrote:
> On 4/5/2021 1:23 AM, Leon Romanovsky wrote:
> > From: Avihai Horon <avihaih@nvidia.com>
> > 
> > Enable Relaxed Ordering in __ib_alloc_pd() allocation of the
> > local_dma_lkey.
> > 
> > This will take effect only for devices that don't pre-allocate the lkey
> > but allocate it per PD allocation.
> > 
> > Signed-off-by: Avihai Horon <avihaih@nvidia.com>
> > Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
> > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > ---
> >   drivers/infiniband/core/verbs.c              | 3 ++-
> >   drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c | 1 +
> >   2 files changed, 3 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
> > index a1782f8a6ca0..9b719f7d6fd5 100644
> > --- a/drivers/infiniband/core/verbs.c
> > +++ b/drivers/infiniband/core/verbs.c
> > @@ -287,7 +287,8 @@ struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
> >   	if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
> >   		pd->local_dma_lkey = device->local_dma_lkey;
> >   	else
> > -		mr_access_flags |= IB_ACCESS_LOCAL_WRITE;
> > +		mr_access_flags |=
> > +			IB_ACCESS_LOCAL_WRITE | IB_ACCESS_RELAXED_ORDERING;
> 
> So, do local_dma_lkey's get relaxed ordering unconditionally?

Yes, in mlx5, this lkey is created on the fly.

Thanks
diff mbox series

Patch

diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index a1782f8a6ca0..9b719f7d6fd5 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -287,7 +287,8 @@  struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
 	if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
 		pd->local_dma_lkey = device->local_dma_lkey;
 	else
-		mr_access_flags |= IB_ACCESS_LOCAL_WRITE;
+		mr_access_flags |=
+			IB_ACCESS_LOCAL_WRITE | IB_ACCESS_RELAXED_ORDERING;
 
 	if (flags & IB_PD_UNSAFE_GLOBAL_RKEY) {
 		pr_warn("%s: enabling unsafe global rkey\n", caller);
diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c
index b3fa783698a0..d74827694f92 100644
--- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c
+++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c
@@ -66,6 +66,7 @@  struct ib_mr *pvrdma_get_dma_mr(struct ib_pd *pd, int acc)
 	int ret;
 
 	/* Support only LOCAL_WRITE flag for DMA MRs */
+	acc &= ~IB_ACCESS_RELAXED_ORDERING;
 	if (acc & ~IB_ACCESS_LOCAL_WRITE) {
 		dev_warn(&dev->pdev->dev,
 			 "unsupported dma mr access flags %#x\n", acc);