* Re: [pull request][for-next 0/9] Generic DIM lib for netdev and RDMA
From: Max Gurtovoy @ 2019-06-06 7:19 UTC (permalink / raw)
To: Leon Romanovsky, Saeed Mahameed
Cc: David S. Miller, Jason Gunthorpe, Doug Ledford, Michael Chan,
Andy Gospodarek, Tal Gilboa, linux-rdma@vger.kernel.org,
netdev@vger.kernel.org
In-Reply-To: <20190606071427.GU5261@mtr-leonro.mtl.com>
On 6/6/2019 10:14 AM, Leon Romanovsky wrote:
> On Wed, Jun 05, 2019 at 11:24:31PM +0000, Saeed Mahameed wrote:
>> Hi Dave, Doug & Jason
>>
>> This series improves DIM - Dynamically-tuned Interrupt
>> Moderation- to be generic for netdev and RDMA use-cases.
>>
>> From Tal and Yamin:
>> The first 7 patches provide the necessary refactoring to current net_dim
>> library which affect some net drivers who are using the API.
>>
>> The last 2 patches provide the RDMA implementation for DIM.
>>
>> For more information please see tag log below.
>>
>> Once we are all happy with the series, please pull to net-next and
>> rdma-next trees.
>>
>> Thanks,
>> Saeed.
>>
>> ---
>> The following changes since commit cd6c84d8f0cdc911df435bb075ba22ce3c605b07:
>>
>> Linux 5.2-rc2 (2019-05-26 16:49:19 -0700)
>>
>> are available in the Git repository at:
>>
>> git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux.git tags/dim-updates-2019-06-05
>>
>> for you to fetch changes up to 1ec9974e75e7a58bff1ab17c4fcda17b180ed3bb:
>>
>> RDMA/core: Provide RDMA DIM support for ULPs (2019-06-05 16:09:02 -0700)
>>
>> ----------------------------------------------------------------
>> dim-updates-2019-06-05
>>
>> From: Tal Gilboa
>>
>> Implement net DIM over a generic DIM library
>>
>> net_dim.h lib exposes an implementation of the DIM algorithm for
>> dynamically-tuned interrupt moderation for networking interfaces.
>>
>> We want a similar functionality for other protocols, which might need to
>> optimize interrupts differently. Main motivation here is DIM for NVMf
>> storage protocol.
>>
>> Current DIM implementation prioritizes reducing interrupt overhead over
>> latency. Also, in order to reduce DIM's own overhead, the algorithm might
>> take some time to identify it needs to change profiles. While this is
>> acceptable for networking, it might not work well on other scenarios.
>>
>> Here I propose a new structure to DIM. The idea is to allow a slightly
>> modified functionality without the risk of breaking Net DIM behavior for
>> netdev. I verified there are no degradations in current DIM behavior with
>> the modified solution.
>>
>> Solution:
>> - Common logic is declared in include/linux/dim.h and implemented in
>> lib/dim/dim.c
>> - Net DIM (existing) logic is declared in include/linux/net_dim.h and
>> implemented in lib/dim/net_dim.c, which uses the common logic from dim.h
>> - Any new DIM logic will be declared in "/include/linux/new_dim.h" and
>> implemented in "lib/dim/new_dim.c".
>> - This new implementation will expose modified versions of profiles,
>> dim_step() and dim_decision().
>>
>> Pros for this solution are:
>> - Zero impact on existing net_dim implementation and usage
>> - Relatively more code reuse (compared to two separate solutions)
>> - Increased extensibility
>>
>> ----------------------------------------------------------------
>> Tal Gilboa (6):
>> linux/dim: Move logic to dim.h
>> linux/dim: Remove "net" prefix from internal DIM members
>> linux/dim: Rename externally exposed macros
>> linux/dim: Rename net_dim_sample() to net_dim_update_sample()
>> linux/dim: Rename externally used net_dim members
>> linux/dim: Move implementation to .c files
>>
>> Yamin Friedman (3):
>> linux/dim: Add completions count to dim_sample
>> linux/dim: Implement rdma_dim
>> RDMA/core: Provide RDMA DIM support for ULPs
> Saeed,
>
> No, for the RDMA patches.
> We need to see usage of those APIs before merging.
I've asked Yamin to prepare patches for NVMeoF initiator and target for
review, so I guess he has it on his plate (this is how he tested it..).
It might cause conflict with NVMe/blk branch maintained by Sagi,
Christoph and Jens.
So we need a plan here.
>
> Thanks
^ permalink raw reply
* Re: [pull request][for-next 0/9] Generic DIM lib for netdev and RDMA
From: Leon Romanovsky @ 2019-06-06 7:14 UTC (permalink / raw)
To: Saeed Mahameed
Cc: David S. Miller, Jason Gunthorpe, Doug Ledford, Michael Chan,
Andy Gospodarek, Tal Gilboa, linux-rdma@vger.kernel.org,
netdev@vger.kernel.org
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
On Wed, Jun 05, 2019 at 11:24:31PM +0000, Saeed Mahameed wrote:
> Hi Dave, Doug & Jason
>
> This series improves DIM - Dynamically-tuned Interrupt
> Moderation- to be generic for netdev and RDMA use-cases.
>
> From Tal and Yamin:
> The first 7 patches provide the necessary refactoring to current net_dim
> library which affect some net drivers who are using the API.
>
> The last 2 patches provide the RDMA implementation for DIM.
>
> For more information please see tag log below.
>
> Once we are all happy with the series, please pull to net-next and
> rdma-next trees.
>
> Thanks,
> Saeed.
>
> ---
> The following changes since commit cd6c84d8f0cdc911df435bb075ba22ce3c605b07:
>
> Linux 5.2-rc2 (2019-05-26 16:49:19 -0700)
>
> are available in the Git repository at:
>
> git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux.git tags/dim-updates-2019-06-05
>
> for you to fetch changes up to 1ec9974e75e7a58bff1ab17c4fcda17b180ed3bb:
>
> RDMA/core: Provide RDMA DIM support for ULPs (2019-06-05 16:09:02 -0700)
>
> ----------------------------------------------------------------
> dim-updates-2019-06-05
>
> From: Tal Gilboa
>
> Implement net DIM over a generic DIM library
>
> net_dim.h lib exposes an implementation of the DIM algorithm for
> dynamically-tuned interrupt moderation for networking interfaces.
>
> We want a similar functionality for other protocols, which might need to
> optimize interrupts differently. Main motivation here is DIM for NVMf
> storage protocol.
>
> Current DIM implementation prioritizes reducing interrupt overhead over
> latency. Also, in order to reduce DIM's own overhead, the algorithm might
> take some time to identify it needs to change profiles. While this is
> acceptable for networking, it might not work well on other scenarios.
>
> Here I propose a new structure to DIM. The idea is to allow a slightly
> modified functionality without the risk of breaking Net DIM behavior for
> netdev. I verified there are no degradations in current DIM behavior with
> the modified solution.
>
> Solution:
> - Common logic is declared in include/linux/dim.h and implemented in
> lib/dim/dim.c
> - Net DIM (existing) logic is declared in include/linux/net_dim.h and
> implemented in lib/dim/net_dim.c, which uses the common logic from dim.h
> - Any new DIM logic will be declared in "/include/linux/new_dim.h" and
> implemented in "lib/dim/new_dim.c".
> - This new implementation will expose modified versions of profiles,
> dim_step() and dim_decision().
>
> Pros for this solution are:
> - Zero impact on existing net_dim implementation and usage
> - Relatively more code reuse (compared to two separate solutions)
> - Increased extensibility
>
> ----------------------------------------------------------------
> Tal Gilboa (6):
> linux/dim: Move logic to dim.h
> linux/dim: Remove "net" prefix from internal DIM members
> linux/dim: Rename externally exposed macros
> linux/dim: Rename net_dim_sample() to net_dim_update_sample()
> linux/dim: Rename externally used net_dim members
> linux/dim: Move implementation to .c files
>
> Yamin Friedman (3):
> linux/dim: Add completions count to dim_sample
> linux/dim: Implement rdma_dim
> RDMA/core: Provide RDMA DIM support for ULPs
Saeed,
No, for the RDMA patches.
We need to see usage of those APIs before merging.
Thanks
^ permalink raw reply
* Re: [PATCH 10/13] megaraid_sas: set virt_boundary_mask in the scsi host
From: Christoph Hellwig @ 2019-06-06 6:41 UTC (permalink / raw)
To: Hannes Reinecke
Cc: Christoph Hellwig, Jens Axboe, Sebastian Ott, Sagi Grimberg,
Max Gurtovoy, Bart Van Assche, Ulf Hansson, Alan Stern,
Oliver Neukum, linux-block, linux-rdma, linux-mmc, linux-nvme,
linux-scsi, megaraidlinux.pdl, MPT-FusionLinux.pdl, linux-hyperv,
linux-usb, usb-storage, linux-kernel
In-Reply-To: <345c3931-0940-7d59-ebc6-fa1ea56c60ac@suse.de>
On Thu, Jun 06, 2019 at 08:02:07AM +0200, Hannes Reinecke wrote:
> > scsi_change_queue_depth(sdev, device_qd);
> >
> What happened to the NOMERGES queue flag?
Quote from the patch description:
"Also remove the bogus nomerges flag, merges do take the virt_boundary
into account."
^ permalink raw reply
* Re: [PATCH 08/13] IB/iser: set virt_boundary_mask in the scsi host
From: Christoph Hellwig @ 2019-06-06 6:24 UTC (permalink / raw)
To: Jason Gunthorpe
Cc: Christoph Hellwig, Jens Axboe, Sebastian Ott, Sagi Grimberg,
Max Gurtovoy, Bart Van Assche, Ulf Hansson, Alan Stern,
Oliver Neukum, linux-block, linux-rdma, linux-mmc, linux-nvme,
linux-scsi, megaraidlinux.pdl, MPT-FusionLinux.pdl, linux-hyperv,
linux-usb, usb-storage, linux-kernel
In-Reply-To: <20190605202235.GC3273@ziepe.ca>
On Wed, Jun 05, 2019 at 05:22:35PM -0300, Jason Gunthorpe wrote:
> On Wed, Jun 05, 2019 at 09:08:31PM +0200, Christoph Hellwig wrote:
> > This ensures all proper DMA layer handling is taken care of by the
> > SCSI midlayer.
>
> Maybe not entirely related to this series, but it looks like the SCSI
> layer is changing the device global dma_set_max_seg_size() - at least
> in RDMA the dma device is being shared between many users, so we
> really don't want SCSI to make this value smaller.
>
> Can we do something about this?
We could do something about it as outlined in my mail - pass the
dma_params explicitly to the dma_map_sg call. But that isn't really
suitable for a short term fix and will take a little more time.
Until we've sorted that out the device paramter needs to be set to
the smallest value supported.
^ permalink raw reply
* Re: [PATCH 10/13] megaraid_sas: set virt_boundary_mask in the scsi host
From: Hannes Reinecke @ 2019-06-06 6:02 UTC (permalink / raw)
To: Christoph Hellwig, Jens Axboe
Cc: Sebastian Ott, Sagi Grimberg, Max Gurtovoy, Bart Van Assche,
Ulf Hansson, Alan Stern, Oliver Neukum, linux-block, linux-rdma,
linux-mmc, linux-nvme, linux-scsi, megaraidlinux.pdl,
MPT-FusionLinux.pdl, linux-hyperv, linux-usb, usb-storage,
linux-kernel
In-Reply-To: <20190605190836.32354-11-hch@lst.de>
On 6/5/19 9:08 PM, Christoph Hellwig wrote:
> This ensures all proper DMA layer handling is taken care of by the
> SCSI midlayer. Note that the effect is global, as the IOMMU merging
> is based off a paramters in struct device. We could still turn if off
> if no PCIe devices are present, but I don't know how to find that out.
>
> Also remove the bogus nomerges flag, merges do take the virt_boundary
> into account.
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
> drivers/scsi/megaraid/megaraid_sas_base.c | 46 +++++----------------
> drivers/scsi/megaraid/megaraid_sas_fusion.c | 7 ++++
> 2 files changed, 18 insertions(+), 35 deletions(-)
>
> diff --git a/drivers/scsi/megaraid/megaraid_sas_base.c b/drivers/scsi/megaraid/megaraid_sas_base.c
> index 3dd1df472dc6..20b3b3f8bc16 100644
> --- a/drivers/scsi/megaraid/megaraid_sas_base.c
> +++ b/drivers/scsi/megaraid/megaraid_sas_base.c
> @@ -1870,39 +1870,6 @@ void megasas_set_dynamic_target_properties(struct scsi_device *sdev,
> }
> }
>
> -/*
> - * megasas_set_nvme_device_properties -
> - * set nomerges=2
> - * set virtual page boundary = 4K (current mr_nvme_pg_size is 4K).
> - * set maximum io transfer = MDTS of NVME device provided by MR firmware.
> - *
> - * MR firmware provides value in KB. Caller of this function converts
> - * kb into bytes.
> - *
> - * e.a MDTS=5 means 2^5 * nvme page size. (In case of 4K page size,
> - * MR firmware provides value 128 as (32 * 4K) = 128K.
> - *
> - * @sdev: scsi device
> - * @max_io_size: maximum io transfer size
> - *
> - */
> -static inline void
> -megasas_set_nvme_device_properties(struct scsi_device *sdev, u32 max_io_size)
> -{
> - struct megasas_instance *instance;
> - u32 mr_nvme_pg_size;
> -
> - instance = (struct megasas_instance *)sdev->host->hostdata;
> - mr_nvme_pg_size = max_t(u32, instance->nvme_page_size,
> - MR_DEFAULT_NVME_PAGE_SIZE);
> -
> - blk_queue_max_hw_sectors(sdev->request_queue, (max_io_size / 512));
> -
> - blk_queue_flag_set(QUEUE_FLAG_NOMERGES, sdev->request_queue);
> - blk_queue_virt_boundary(sdev->request_queue, mr_nvme_pg_size - 1);
> -}
> -
> -
> /*
> * megasas_set_static_target_properties -
> * Device property set by driver are static and it is not required to be
> @@ -1961,8 +1928,10 @@ static void megasas_set_static_target_properties(struct scsi_device *sdev,
> max_io_size_kb = le32_to_cpu(instance->tgt_prop->max_io_size_kb);
> }
>
> - if (instance->nvme_page_size && max_io_size_kb)
> - megasas_set_nvme_device_properties(sdev, (max_io_size_kb << 10));
> + if (instance->nvme_page_size && max_io_size_kb) {
> + blk_queue_max_hw_sectors(sdev->request_queue,
> + (max_io_size_kb << 10) / 512);
> + }
>
> scsi_change_queue_depth(sdev, device_qd);
>
What happened to the NOMERGES queue flag?
Cheers,
Hannes
--
Dr. Hannes Reinecke Teamlead Storage & Networking
hare@suse.de +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Felix Imendörffer, Mary Higgins, Sri Rasiah
HRB 21284 (AG Nürnberg)
^ permalink raw reply
* Re: [PATCH 1/1] net: rds: fix memory leak when unload rds_rdma
From: David Miller @ 2019-06-06 0:08 UTC (permalink / raw)
To: yanjun.zhu; +Cc: santosh.shilimkar, netdev, linux-rdma, rds-devel
In-Reply-To: <1559566099-30289-1-git-send-email-yanjun.zhu@oracle.com>
From: Zhu Yanjun <yanjun.zhu@oracle.com>
Date: Mon, 3 Jun 2019 08:48:19 -0400
> When KASAN is enabled, after several rds connections are
> created, then "rmmod rds_rdma" is run. The following will
> appear.
...
> This is rds connection memory leak. The root cause is:
> When "rmmod rds_rdma" is run, rds_ib_remove_one will call
> rds_ib_dev_shutdown to drop the rds connections.
> rds_ib_dev_shutdown will call rds_conn_drop to drop rds
> connections as below.
> "
> rds_conn_path_drop(&conn->c_path[0], false);
> "
> In the above, destroy is set to false.
...
> In the above function, destroy is set to false. rds_destroy_pending
> is called. This does not move rds connections to ib_nodev_conns.
> So destroy is set to true to move rds connections to ib_nodev_conns.
> In rds_ib_unregister_client, flush_workqueue is called to make rds_wq
> finsh shutdown rds connections. The function rds_ib_destroy_nodev_conns
> is called to shutdown rds connections finally.
> Then rds_ib_recv_exit is called to destroy slab.
...
> Suggested-by: Håkon Bugge <haakon.bugge@oracle.com>
> Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Applied.
^ permalink raw reply
* Re: [PATCHv2 1/1] net: rds: add per rds connection cache statistics
From: David Miller @ 2019-06-06 0:07 UTC (permalink / raw)
To: yanjun.zhu; +Cc: santosh.shilimkar, netdev, linux-rdma, rds-devel
In-Reply-To: <1559536081-25401-1-git-send-email-yanjun.zhu@oracle.com>
From: Zhu Yanjun <yanjun.zhu@oracle.com>
Date: Mon, 3 Jun 2019 00:28:01 -0400
> The variable cache_allocs is to indicate how many frags (KiB) are in one
> rds connection frag cache.
> The command "rds-info -Iv" will output the rds connection cache
> statistics as below:
> "
> RDS IB Connections:
> LocalAddr RemoteAddr Tos SL LocalDev RemoteDev
> 1.1.1.14 1.1.1.14 58 255 fe80::2:c903:a:7a31 fe80::2:c903:a:7a31
> send_wr=256, recv_wr=1024, send_sge=8, rdma_mr_max=4096,
> rdma_mr_size=257, cache_allocs=12
> "
> This means that there are about 12KiB frag in this rds connection frag
> cache.
> Since rds.h in rds-tools is not related with the kernel rds.h, the change
> in kernel rds.h does not affect rds-tools.
> rds-info in rds-tools 2.0.5 and 2.0.6 is tested with this commit. It works
> well.
>
> Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
> ---
> V1->V2: RDS CI is removed.
Applied to net-next.
^ permalink raw reply
* Re: [PATCH 08/13] IB/iser: set virt_boundary_mask in the scsi host
From: Sagi Grimberg @ 2019-06-05 23:35 UTC (permalink / raw)
To: Jason Gunthorpe, Christoph Hellwig
Cc: Jens Axboe, Sebastian Ott, Max Gurtovoy, Bart Van Assche,
Ulf Hansson, Alan Stern, Oliver Neukum, linux-block, linux-rdma,
linux-mmc, linux-nvme, linux-scsi, megaraidlinux.pdl,
MPT-FusionLinux.pdl, linux-hyperv, linux-usb, usb-storage,
linux-kernel
In-Reply-To: <20190605202235.GC3273@ziepe.ca>
>> This ensures all proper DMA layer handling is taken care of by the
>> SCSI midlayer.
>
> Maybe not entirely related to this series, but it looks like the SCSI
> layer is changing the device global dma_set_max_seg_size() - at least
> in RDMA the dma device is being shared between many users, so we
> really don't want SCSI to make this value smaller.
>
> Can we do something about this?
srp seems to do the right thing:
target_host->max_segment_size = ib_dma_max_seg_size(ibdev);
But iser does not, which means that scsi limits it to:
BLK_MAX_SEGMENT_SIZE (64k)
I can send a fix to iser.
> Wondering about other values too, and the interaction with the new
> combining stuff in umem.c
The only other values AFAICT is the dma_boundary that rdma llds don't
set...
^ permalink raw reply
* [for-next 9/9] RDMA/core: Provide RDMA DIM support for ULPs
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Yamin Friedman, Max Gurtovoy, Saeed Mahameed
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
From: Yamin Friedman <yaminf@mellanox.com>
Added the interface in the infiniband driver that applies the rdma_dim
adaptive moderation. There is now a special function for allocating an
ib_cq that uses rdma_dim.
Performance improvement (ConnectX-5 100GbE, x86) running FIO benchmark over
NVMf between two equal end-hosts with 56 cores across a Mellanox switch
using null_blk device:
READS without DIM:
blk size | BW | IOPS | 99th percentile latency | 99.99th latency
512B | 3.8GiB/s | 7.7M | 1401 usec | 2442 usec
4k | 7.0GiB/s | 1.8M | 4817 usec | 6587 usec
64k | 10.7GiB/s| 175k | 9896 usec | 10028 usec
IO WRITES without DIM:
blk size | BW | IOPS | 99th percentile latency | 99.99th latency
512B | 3.6GiB/s | 7.5M | 1434 usec | 2474 usec
4k | 6.3GiB/s | 1.6M | 938 usec | 1221 usec
64k | 10.7GiB/s| 175k | 8979 usec | 12780 usec
IO READS with DIM:
blk size | BW | IOPS | 99th percentile latency | 99.99th latency
512B | 4GiB/s | 8.2M | 816 usec | 889 usec
4k | 10.1GiB/s| 2.65M| 3359 usec | 5080 usec
64k | 10.7GiB/s| 175k | 9896 usec | 10028 usec
IO WRITES with DIM:
blk size | BW | IOPS | 99th percentile latency | 99.99th latency
512B | 3.9GiB/s | 8.1M | 799 usec | 922 usec
4k | 9.6GiB/s | 2.5M | 717 usec | 1004 usec
64k | 10.7GiB/s| 176k | 8586 usec | 12256 usec
The rdma_dim algorithm was designed to measure the effectiveness of
moderation on the flow in a general way and thus should be appropriate
for all RDMA storage protocols.
Signed-off-by: Yamin Friedman <yaminf@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/infiniband/core/cq.c | 78 ++++++++++++++++++-
drivers/net/ethernet/mellanox/mlx4/Kconfig | 1 +
.../net/ethernet/mellanox/mlx5/core/Kconfig | 1 +
include/rdma/ib_verbs.h | 27 ++++++-
4 files changed, 102 insertions(+), 5 deletions(-)
diff --git a/drivers/infiniband/core/cq.c b/drivers/infiniband/core/cq.c
index a4c81992267c..326d928d2763 100644
--- a/drivers/infiniband/core/cq.c
+++ b/drivers/infiniband/core/cq.c
@@ -14,6 +14,7 @@
#include <linux/err.h>
#include <linux/slab.h>
#include <rdma/ib_verbs.h>
+#include <linux/rdma_dim.h>
/* # of WCs to poll for with a single call to ib_poll_cq */
#define IB_POLL_BATCH 16
@@ -26,6 +27,32 @@
#define IB_POLL_FLAGS \
(IB_CQ_NEXT_COMP | IB_CQ_REPORT_MISSED_EVENTS)
+static void ib_cq_rdma_dim_work(struct work_struct *w)
+{
+ struct dim *dim = container_of(w, struct dim, work);
+ struct ib_cq *cq = container_of(dim, struct ib_cq, dim);
+
+ u16 usec = rdma_dim_prof[dim->profile_ix].usec;
+ u16 comps = rdma_dim_prof[dim->profile_ix].comps;
+
+ dim->state = DIM_START_MEASURE;
+
+ cq->device->ops.modify_cq(cq, comps, usec);
+}
+
+static bool rdma_dim_init(struct dim *dim, struct ib_cq *cq)
+{
+ if (!cq->device->ops.modify_cq)
+ return false;
+
+ memset(dim, 0, sizeof(*dim));
+ dim->state = DIM_START_MEASURE;
+ dim->tune_state = DIM_GOING_RIGHT;
+ dim->profile_ix = RDMA_DIM_START_PROFILE;
+
+ return true;
+}
+
static int __ib_process_cq(struct ib_cq *cq, int budget, struct ib_wc *wcs,
int batch)
{
@@ -98,6 +125,24 @@ static int ib_poll_handler(struct irq_poll *iop, int budget)
return completed;
}
+static int ib_poll_dim_handler(struct irq_poll *iop, int budget)
+{
+ struct ib_cq *cq = container_of(iop, struct ib_cq, iop);
+ int completed;
+ struct dim *dim = &cq->dim;
+
+ completed = __ib_process_cq(cq, budget, cq->wc, IB_POLL_BATCH);
+ if (completed < budget) {
+ irq_poll_complete(&cq->iop);
+ if (ib_req_notify_cq(cq, IB_POLL_FLAGS) > 0)
+ irq_poll_sched(&cq->iop);
+ }
+
+ rdma_dim(dim, completed);
+
+ return completed;
+}
+
static void ib_cq_completion_softirq(struct ib_cq *cq, void *private)
{
irq_poll_sched(&cq->iop);
@@ -105,14 +150,18 @@ static void ib_cq_completion_softirq(struct ib_cq *cq, void *private)
static void ib_cq_poll_work(struct work_struct *work)
{
- struct ib_cq *cq = container_of(work, struct ib_cq, work);
+ struct ib_cq *cq = container_of(work, struct ib_cq,
+ work);
int completed;
completed = __ib_process_cq(cq, IB_POLL_BUDGET_WORKQUEUE, cq->wc,
IB_POLL_BATCH);
+
if (completed >= IB_POLL_BUDGET_WORKQUEUE ||
ib_req_notify_cq(cq, IB_POLL_FLAGS) > 0)
queue_work(cq->comp_wq, &cq->work);
+ else if (cq->dim_used)
+ rdma_dim(&cq->dim, completed);
}
static void ib_cq_completion_workqueue(struct ib_cq *cq, void *private)
@@ -129,6 +178,7 @@ static void ib_cq_completion_workqueue(struct ib_cq *cq, void *private)
* @poll_ctx: context to poll the CQ from.
* @caller: module owner name.
* @udata: Valid user data or NULL for kernel object
+ * @use_dim: use dynamic interrupt moderation
*
* This is the proper interface to allocate a CQ for in-kernel users. A
* CQ allocated with this interface will automatically be polled from the
@@ -138,7 +188,8 @@ static void ib_cq_completion_workqueue(struct ib_cq *cq, void *private)
struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
int nr_cqe, int comp_vector,
enum ib_poll_context poll_ctx,
- const char *caller, struct ib_udata *udata)
+ const char *caller, struct ib_udata *udata,
+ bool use_dim)
{
struct ib_cq_init_attr cq_attr = {
.cqe = nr_cqe,
@@ -173,13 +224,30 @@ struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
case IB_POLL_SOFTIRQ:
cq->comp_handler = ib_cq_completion_softirq;
- irq_poll_init(&cq->iop, IB_POLL_BUDGET_IRQ, ib_poll_handler);
+ if (use_dim)
+ cq->dim_used = rdma_dim_init(&cq->dim, cq);
+
+ if (cq->dim_used) {
+ irq_poll_init(&cq->iop, IB_POLL_BUDGET_IRQ,
+ ib_poll_dim_handler);
+ INIT_WORK(&cq->dim.work, ib_cq_rdma_dim_work);
+ } else {
+ irq_poll_init(&cq->iop, IB_POLL_BUDGET_IRQ,
+ ib_poll_handler);
+ }
+
ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
break;
case IB_POLL_WORKQUEUE:
case IB_POLL_UNBOUND_WORKQUEUE:
cq->comp_handler = ib_cq_completion_workqueue;
INIT_WORK(&cq->work, ib_cq_poll_work);
+ if (use_dim)
+ cq->dim_used = rdma_dim_init(&cq->dim, cq);
+
+ if (cq->dim_used)
+ INIT_WORK(&cq->dim.work, ib_cq_rdma_dim_work);
+
ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
cq->comp_wq = (cq->poll_ctx == IB_POLL_WORKQUEUE) ?
ib_comp_wq : ib_comp_unbound_wq;
@@ -217,10 +285,14 @@ void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata)
break;
case IB_POLL_SOFTIRQ:
irq_poll_disable(&cq->iop);
+ if (cq->dim_used)
+ cancel_work_sync(&cq->dim.work);
break;
case IB_POLL_WORKQUEUE:
case IB_POLL_UNBOUND_WORKQUEUE:
cancel_work_sync(&cq->work);
+ if (cq->dim_used)
+ cancel_work_sync(&cq->dim.work);
break;
default:
WARN_ON_ONCE(1);
diff --git a/drivers/net/ethernet/mellanox/mlx4/Kconfig b/drivers/net/ethernet/mellanox/mlx4/Kconfig
index e69c3c31e701..93cd25997b24 100644
--- a/drivers/net/ethernet/mellanox/mlx4/Kconfig
+++ b/drivers/net/ethernet/mellanox/mlx4/Kconfig
@@ -28,6 +28,7 @@ config MLX4_CORE
tristate
depends on PCI
select NET_DEVLINK
+ select DIMLIB
default n
config MLX4_DEBUG
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/Kconfig b/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
index 7845aa5bf6be..ef292fbb53c9 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
+++ b/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
@@ -7,6 +7,7 @@ config MLX5_CORE
tristate "Mellanox 5th generation network adapters (ConnectX series) core driver"
depends on PCI
select NET_DEVLINK
+ select DIMLIB
imply PTP_1588_CLOCK
imply VXLAN
imply MLXFW
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 0742095355f2..7b03fb3e4f0b 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -65,6 +65,7 @@
#include <rdma/restrack.h>
#include <uapi/rdma/rdma_user_ioctl.h>
#include <uapi/rdma/ib_user_ioctl_verbs.h>
+#include <linux/dim.h>
#define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
@@ -1638,6 +1639,8 @@ struct ib_cq {
* Implementation details of the RDMA core, don't use in drivers:
*/
struct rdma_restrack_entry res;
+ struct dim dim;
+ bool dim_used;
};
struct ib_srq {
@@ -3746,7 +3749,8 @@ static inline int ib_post_recv(struct ib_qp *qp,
struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
int nr_cqe, int comp_vector,
enum ib_poll_context poll_ctx,
- const char *caller, struct ib_udata *udata);
+ const char *caller, struct ib_udata *udata,
+ bool use_dim);
/**
* ib_alloc_cq_user: Allocate kernel/user CQ
@@ -3764,7 +3768,7 @@ static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
struct ib_udata *udata)
{
return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
- KBUILD_MODNAME, udata);
+ KBUILD_MODNAME, udata, false);
}
/**
@@ -3785,6 +3789,25 @@ static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
NULL);
}
+/**
+ * ib_alloc_cq_dim: Allocate kernel CQ with dynamic interrupt moderation
+ * @dev: The IB device
+ * @private: Private data attached to the CQE
+ * @nr_cqe: Number of CQEs in the CQ
+ * @comp_vector: Completion vector used for the IRQs
+ * @poll_ctx: Context used for polling the CQ
+ *
+ * NOTE: for user cq use ib_alloc_cq_user with valid udata!
+ */
+static inline struct ib_cq *ib_alloc_cq_dim(struct ib_device *dev,
+ void *private, int nr_cqe,
+ int comp_vector,
+ enum ib_poll_context poll_ctx)
+{
+ return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
+ KBUILD_MODNAME, NULL, true);
+}
+
/**
* ib_free_cq_user - Free kernel/user CQ
* @cq: The CQ to free
--
2.21.0
^ permalink raw reply related
* [for-next 8/9] linux/dim: Implement rdma_dim
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Yamin Friedman, Max Gurtovoy, Saeed Mahameed
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
From: Yamin Friedman <yaminf@mellanox.com>
rdma_dim implements a different algorithm than net_dim and is based on
completions which is how we can implement interrupt moderation in RDMA.
The algorithm optimizes for number of completions and ratio between
completions and events.
It also has a feature for fast reduction of moderation level when the
traffic changes in such a way as to no longer require high moderation in
order to avoid long latencies.
rdma_dim.h will be called from the ib_core module.
Signed-off-by: Yamin Friedman <yaminf@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
MAINTAINERS | 1 +
include/linux/rdma_dim.h | 28 +++++++
lib/dim/Makefile | 7 +-
lib/dim/rdma_dim.c | 162 +++++++++++++++++++++++++++++++++++++++
4 files changed, 197 insertions(+), 1 deletion(-)
create mode 100644 include/linux/rdma_dim.h
create mode 100644 lib/dim/rdma_dim.c
diff --git a/MAINTAINERS b/MAINTAINERS
index cb621d5cf223..86e4698ab390 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -5589,6 +5589,7 @@ DYNAMIC INTERRUPT MODERATION
M: Tal Gilboa <talgi@mellanox.com>
S: Maintained
F: include/linux/net_dim.h
+F: include/linux/rdma_dim.h
F: include/linux/dim.h
F: lib/dim/
diff --git a/include/linux/rdma_dim.h b/include/linux/rdma_dim.h
new file mode 100644
index 000000000000..0623ea5a1e78
--- /dev/null
+++ b/include/linux/rdma_dim.h
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
+/* Copyright (c) 2019 Mellanox Technologies. */
+
+#ifndef RDMA_DIM_H
+#define RDMA_DIM_H
+
+#include <linux/module.h>
+#include <linux/dim.h>
+
+#define RDMA_DIM_PARAMS_NUM_PROFILES 9
+#define RDMA_DIM_START_PROFILE 0
+
+static const struct dim_cq_moder
+rdma_dim_prof[RDMA_DIM_PARAMS_NUM_PROFILES] = {
+ {1, 0, 1, 0},
+ {1, 0, 4, 0},
+ {2, 0, 4, 0},
+ {2, 0, 8, 0},
+ {4, 0, 8, 0},
+ {16, 0, 8, 0},
+ {16, 0, 16, 0},
+ {32, 0, 16, 0},
+ {32, 0, 32, 0},
+};
+
+void rdma_dim(struct dim *dim, u64 completions);
+
+#endif /* RDMA_DIM_H */
diff --git a/lib/dim/Makefile b/lib/dim/Makefile
index 160afe288df0..73ddd0c64661 100644
--- a/lib/dim/Makefile
+++ b/lib/dim/Makefile
@@ -2,8 +2,13 @@
# DIM Dynamic Interrupt Moderation library
#
-obj-$(CONFIG_DIMLIB) = net_dim.o
+obj-$(CONFIG_DIMLIB) += net_dim.o
+obj-$(CONFIG_DIMLIB) += rdma_dim.o
net_dim-y = \
dim.o \
net_dim.o
+
+rdma_dim-y = \
+ dim.o \
+ rdma_dim.o
diff --git a/lib/dim/rdma_dim.c b/lib/dim/rdma_dim.c
new file mode 100644
index 000000000000..503881ec5614
--- /dev/null
+++ b/lib/dim/rdma_dim.c
@@ -0,0 +1,162 @@
+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+/*
+ * Copyright (c) 2019, Mellanox Technologies inc. All rights reserved.
+ */
+
+#include <linux/rdma_dim.h>
+
+/**
+ ** rdma_dim_step: - Moves the moderation profile one step.
+ ** @dim: The moderation struct.
+ **
+ ** Description: Moves the moderation profile of @dim by one step. If we
+ ** are at the edge of the profile range returns DIM_ON_EDGE without
+ ** moving.
+ **/
+static int rdma_dim_step(struct dim *dim)
+{
+ switch (dim->tune_state) {
+ case DIM_PARKING_ON_TOP:
+ /* fall through */
+ case DIM_PARKING_TIRED:
+ break;
+ case DIM_GOING_RIGHT:
+ if (dim->profile_ix == (RDMA_DIM_PARAMS_NUM_PROFILES - 1))
+ return DIM_ON_EDGE;
+ dim->profile_ix++;
+ dim->steps_right++;
+ break;
+ case DIM_GOING_LEFT:
+ if (dim->profile_ix == 0)
+ return DIM_ON_EDGE;
+ dim->profile_ix--;
+ dim->steps_left++;
+ break;
+ }
+
+ return DIM_STEPPED;
+}
+
+/**
+ ** rdma_dim_stats_compare - Compares the current stats to the previous stats.
+ ** @curr: The current dim stats.
+ ** @prev: The previous dim stats.
+ **
+ ** Description: Checks to see if we have significantly more or less
+ ** completions.
+ ** If the completions are not greatly changed checks if the completion to
+ ** event ratio has significantly changed.
+ **/
+static int rdma_dim_stats_compare(struct dim_stats *curr,
+ struct dim_stats *prev)
+{
+ /* first stat */
+ if (!prev->cpms)
+ return DIM_STATS_SAME;
+
+ if (IS_SIGNIFICANT_DIFF(curr->cpms, prev->cpms))
+ return (curr->cpms > prev->cpms) ? DIM_STATS_BETTER :
+ DIM_STATS_WORSE;
+
+ if (IS_SIGNIFICANT_DIFF(curr->cpe_ratio, prev->cpe_ratio))
+ return (curr->cpe_ratio > prev->cpe_ratio) ? DIM_STATS_BETTER :
+ DIM_STATS_WORSE;
+
+ return DIM_STATS_SAME;
+}
+
+/**
+ ** rdma_dim_decision - Decides the next moderation level.
+ ** @curr_stats: The current dim stats.
+ ** @dim: The moderation struct.
+ **
+ ** Description: Uses rdma_dim_stats_compare to decide what the next moderation
+ ** level should be. If the completion to event ratio is low compared to the
+ ** current level we reset the moderation to keep latency low.
+ **/
+static bool rdma_dim_decision(struct dim_stats *curr_stats, struct dim *dim)
+{
+ int prev_ix = dim->profile_ix;
+ int stats_res;
+ int step_res;
+
+ switch (dim->tune_state) {
+ case DIM_PARKING_ON_TOP:
+ /* fall through */
+ case DIM_PARKING_TIRED:
+ break;
+ case DIM_GOING_RIGHT:
+ /* fall through */
+ case DIM_GOING_LEFT:
+ stats_res = rdma_dim_stats_compare(curr_stats,
+ &dim->prev_stats);
+
+ switch (stats_res) {
+ case DIM_STATS_SAME:
+ if (curr_stats->cpe_ratio <= 50 * prev_ix)
+ dim->profile_ix = 0;
+ break;
+ case DIM_STATS_WORSE:
+ dim_turn(dim);
+ /* fall through */
+ case DIM_STATS_BETTER:
+ step_res = rdma_dim_step(dim);
+ if (step_res == DIM_ON_EDGE)
+ dim_turn(dim);
+ break;
+ }
+ break;
+ }
+
+ dim->prev_stats = *curr_stats;
+
+ return dim->profile_ix != prev_ix;
+}
+
+/**
+ ** rdma_dim - Runs the adaptive moderation.
+ ** @dim: The moderation struct.
+ ** @completions: The number of completions collected in this round.
+ **
+ ** Description: Each call to rdma_dim takes the latest amount of
+ ** completions that have been collected and counts them as a new event.
+ ** Once enough events have been collected the algorithm decides a new
+ ** moderation level.
+ **/
+void rdma_dim(struct dim *dim, u64 completions)
+{
+ struct dim_stats curr_stats;
+ u32 nevents;
+ struct dim_sample *curr_sample = &dim->measuring_sample;
+
+ dim_update_sample_with_comps(curr_sample->event_ctr + 1,
+ curr_sample->pkt_ctr,
+ curr_sample->byte_ctr,
+ curr_sample->comp_ctr + completions,
+ &dim->measuring_sample);
+
+ switch (dim->state) {
+ case DIM_MEASURE_IN_PROGRESS:
+ nevents = curr_sample->event_ctr - dim->start_sample.event_ctr;
+ if (nevents < DIM_NEVENTS)
+ break;
+ dim_calc_stats(&dim->start_sample, curr_sample, &curr_stats);
+ if (rdma_dim_decision(&curr_stats, dim)) {
+ dim->state = DIM_APPLY_NEW_PROFILE;
+ schedule_work(&dim->work);
+ break;
+ }
+ /* fall through */
+ case DIM_START_MEASURE:
+ dim->state = DIM_MEASURE_IN_PROGRESS;
+ dim_update_sample_with_comps(curr_sample->event_ctr,
+ curr_sample->pkt_ctr,
+ curr_sample->byte_ctr,
+ curr_sample->comp_ctr,
+ &dim->start_sample);
+ break;
+ case DIM_APPLY_NEW_PROFILE:
+ break;
+ }
+}
+EXPORT_SYMBOL(rdma_dim);
--
2.21.0
^ permalink raw reply related
* [for-next 7/9] linux/dim: Add completions count to dim_sample
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Yamin Friedman, Max Gurtovoy, Saeed Mahameed
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
From: Yamin Friedman <yaminf@mellanox.com>
Added a measurement of completions per/msec to allow for completion based
dim algorithms.
In order to use dynamic interrupt moderation with RDMA we need to have a
different measurment than packets per second. This change is meant to
prepare for adding a new DIM method.
All drivers that use net_dim and thus do not need a completion count will
have the completions set to 0.
Signed-off-by: Yamin Friedman <yaminf@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
include/linux/dim.h | 28 +++++++++++++++++++++++++---
lib/dim/dim.c | 9 +++++++++
2 files changed, 34 insertions(+), 3 deletions(-)
diff --git a/include/linux/dim.h b/include/linux/dim.h
index a6bf84f61dfb..e7b3f8f5544d 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -37,6 +37,7 @@
struct dim_cq_moder {
u16 usec;
u16 pkts;
+ u16 comps;
u8 cq_period_mode;
};
@@ -54,6 +55,7 @@ struct dim_sample {
u32 pkt_ctr;
u32 byte_ctr;
u16 event_ctr;
+ u32 comp_ctr;
};
/**
@@ -65,9 +67,11 @@ struct dim_sample {
* @epms: Events per msec
*/
struct dim_stats {
- int ppms;
- int bpms;
- int epms;
+ int ppms; /* packets per msec */
+ int bpms; /* bytes per msec */
+ int epms; /* events per msec */
+ int cpms; /* completions per msec */
+ int cpe_ratio; /* ratio of completions to events */
};
/**
@@ -89,6 +93,7 @@ struct dim {
u8 state;
struct dim_stats prev_stats;
struct dim_sample start_sample;
+ struct dim_sample measuring_sample;
struct work_struct work;
u8 profile_ix;
u8 mode;
@@ -246,4 +251,21 @@ dim_update_sample(u16 event_ctr, u64 packets, u64 bytes, struct dim_sample *s)
s->event_ctr = event_ctr;
}
+/**
+ * dim_update_sample_with_comps - set a sample's fields with given
+ * values including the completion parameter
+ * @event_ctr: number of events to set
+ * @packets: number of packets to set
+ * @bytes: number of bytes to set
+ * @comps: number of completions to set
+ * @s: DIM sample
+ */
+static inline void
+dim_update_sample_with_comps(u16 event_ctr, u64 packets, u64 bytes, u64 comps,
+ struct dim_sample *s)
+{
+ dim_update_sample(event_ctr, packets, bytes, s);
+ s->comp_ctr = comps;
+}
+
#endif /* DIM_H */
diff --git a/lib/dim/dim.c b/lib/dim/dim.c
index 17d5236759bd..439d641ec796 100644
--- a/lib/dim/dim.c
+++ b/lib/dim/dim.c
@@ -62,6 +62,8 @@ void dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
u32 npkts = BIT_GAP(BITS_PER_TYPE(u32), end->pkt_ctr, start->pkt_ctr);
u32 nbytes = BIT_GAP(BITS_PER_TYPE(u32), end->byte_ctr,
start->byte_ctr);
+ u32 ncomps = BIT_GAP(BITS_PER_TYPE(u32), end->comp_ctr,
+ start->comp_ctr);
if (!delta_us)
return;
@@ -70,5 +72,12 @@ void dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
curr_stats->bpms = DIV_ROUND_UP(nbytes * USEC_PER_MSEC, delta_us);
curr_stats->epms = DIV_ROUND_UP(DIM_NEVENTS * USEC_PER_MSEC,
delta_us);
+ curr_stats->cpms = DIV_ROUND_UP(ncomps * USEC_PER_MSEC, delta_us);
+ if (curr_stats->epms != 0)
+ curr_stats->cpe_ratio =
+ (curr_stats->cpms * 100) / curr_stats->epms;
+ else
+ curr_stats->cpe_ratio = 0;
+
}
EXPORT_SYMBOL(dim_calc_stats);
--
2.21.0
^ permalink raw reply related
* [for-next 6/9] linux/dim: Move implementation to .c files
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Saeed Mahameed
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
From: Tal Gilboa <talgi@mellanox.com>
Moved all logic from dim.h and net_dim.h to dim.c and net_dim.c.
This is both more structurally appealing and would allow to only
expose externally used functions.
Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
MAINTAINERS | 1 +
drivers/net/ethernet/broadcom/Kconfig | 1 +
.../net/ethernet/mellanox/mlx5/core/Kconfig | 1 +
include/linux/dim.h | 238 ++++++++++++------
include/linux/net_dim.h | 216 +++-------------
lib/Kconfig | 8 +
lib/Makefile | 1 +
lib/dim/Makefile | 9 +
lib/dim/dim.c | 74 ++++++
lib/dim/net_dim.c | 191 ++++++++++++++
10 files changed, 491 insertions(+), 249 deletions(-)
create mode 100644 lib/dim/Makefile
create mode 100644 lib/dim/dim.c
create mode 100644 lib/dim/net_dim.c
diff --git a/MAINTAINERS b/MAINTAINERS
index 5d4b852d9d39..cb621d5cf223 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -5590,6 +5590,7 @@ M: Tal Gilboa <talgi@mellanox.com>
S: Maintained
F: include/linux/net_dim.h
F: include/linux/dim.h
+F: lib/dim/
DZ DECSTATION DZ11 SERIAL DRIVER
M: "Maciej W. Rozycki" <macro@linux-mips.org>
diff --git a/drivers/net/ethernet/broadcom/Kconfig b/drivers/net/ethernet/broadcom/Kconfig
index b123509d385f..2e4a8c7237ef 100644
--- a/drivers/net/ethernet/broadcom/Kconfig
+++ b/drivers/net/ethernet/broadcom/Kconfig
@@ -8,6 +8,7 @@ config NET_VENDOR_BROADCOM
default y
depends on (SSB_POSSIBLE && HAS_DMA) || PCI || BCM63XX || \
SIBYTE_SB1xxx_SOC
+ select DIMLIB
---help---
If you have a network (Ethernet) chipset belonging to this class,
say Y.
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/Kconfig b/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
index 2391e3cfb56b..7845aa5bf6be 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
+++ b/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
@@ -34,6 +34,7 @@ config MLX5_CORE_EN
depends on NETDEVICES && ETHERNET && INET && PCI && MLX5_CORE
depends on IPV6=y || IPV6=n || MLX5_CORE=m
select PAGE_POOL
+ select DIMLIB
default n
---help---
Ethernet support in Mellanox Technologies ConnectX-4 NIC.
diff --git a/include/linux/dim.h b/include/linux/dim.h
index 60e5074a7cc0..a6bf84f61dfb 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -6,20 +6,49 @@
#include <linux/module.h>
+/**
+ * Number of events between DIM iterations.
+ * Causes a moderation of the algorithm run.
+ */
#define DIM_NEVENTS 64
-/* more than 10% difference */
+/**
+ * Is a difference between values justifies taking an action.
+ * We consider 10% difference as significant.
+ */
#define IS_SIGNIFICANT_DIFF(val, ref) \
(((100UL * abs((val) - (ref))) / (ref)) > 10)
-#define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) \
-& (BIT_ULL(bits) - 1))
+/**
+ * Calculate the gap between two values.
+ * Take wrap-around and variable size into consideration.
+ */
+#define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) \
+ & (BIT_ULL(bits) - 1))
+
+/**
+ * Structure for CQ moderation values.
+ * Used for communications between DIM and its consumer.
+ *
+ * @usec: CQ timer suggestion (by DIM)
+ * @pkts: CQ packet counter suggestion (by DIM)
+ * @cq_period_mode: CQ priod count mode (from CQE/EQE)
+ */
struct dim_cq_moder {
u16 usec;
u16 pkts;
u8 cq_period_mode;
};
+/**
+ * Structure for DIM sample data.
+ * Used for communications between DIM and its consumer.
+ *
+ * @time: Sample timestamp
+ * @pkt_ctr: Number of packets
+ * @byte_ctr: Number of bytes
+ * @event_ctr: Number of events
+ */
struct dim_sample {
ktime_t time;
u32 pkt_ctr;
@@ -27,13 +56,36 @@ struct dim_sample {
u16 event_ctr;
};
+/**
+ * Structure for DIM stats.
+ * Used for holding current measured rates.
+ *
+ * @ppms: Packets per msec
+ * @bpms: Bytes per msec
+ * @epms: Events per msec
+ */
struct dim_stats {
- int ppms; /* packets per msec */
- int bpms; /* bytes per msec */
- int epms; /* events per msec */
+ int ppms;
+ int bpms;
+ int epms;
};
-struct dim { /* Dynamic Interrupt Moderation */
+/**
+ * Main structure for dynamic interrupt moderation (DIM).
+ * Used for holding all information about a specific DIM instance.
+ *
+ * @state: Algorithm state (see below)
+ * @prev_stats: Measured rates from previous iteration (for comparison)
+ * @start_sample: Sampled data at start of current iteration
+ * @work: Work to perform on action required
+ * @profile_ix: Current moderation profile
+ * @mode: CQ period count mode
+ * @tune_state: Algorithm tuning state (see below)
+ * @steps_right: Number of steps taken towards higher moderation
+ * @steps_left: Number of steps taken towards lower moderation
+ * @tired: Parking depth counter
+ */
+struct dim {
u8 state;
struct dim_stats prev_stats;
struct dim_sample start_sample;
@@ -46,18 +98,49 @@ struct dim { /* Dynamic Interrupt Moderation */
u8 tired;
};
+/**
+ * enum dim_cq_period_mode
+ *
+ * These are the modes for CQ period count.
+ *
+ * @DIM_CQ_PERIOD_MODE_START_FROM_EQE: Start counting from EQE
+ * @DIM_CQ_PERIOD_MODE_START_FROM_CQE: Start counting from CQE (implies timer reset)
+ * @DIM_CQ_PERIOD_NUM_MODES: Number of modes
+ */
enum {
DIM_CQ_PERIOD_MODE_START_FROM_EQE = 0x0,
DIM_CQ_PERIOD_MODE_START_FROM_CQE = 0x1,
DIM_CQ_PERIOD_NUM_MODES
};
+/**
+ * enum dim_state
+ *
+ * These are the DIM algorithm states.
+ * These will determine if the algorithm is in a valid state to start an iteration.
+ *
+ * @DIM_START_MEASURE: This is the first iteration (also after applying a new profile)
+ * @DIM_MEASURE_IN_PROGRESS: Algorithm is already in progress - check if
+ * need to perform an action
+ * @DIM_APPLY_NEW_PROFILE: DIM consumer is currently applying a profile - no need to measure
+ */
enum {
DIM_START_MEASURE,
DIM_MEASURE_IN_PROGRESS,
DIM_APPLY_NEW_PROFILE,
};
+/**
+ * enum dim_tune_state
+ *
+ * These are the DIM algorithm tune states.
+ * These will determine which action the algorithm should perform.
+ *
+ * @DIM_PARKING_ON_TOP: Algorithm found a local top point - exit on significant difference
+ * @DIM_PARKING_TIRED: Algorithm found a deep top point - don't exit if tired > 0
+ * @DIM_GOING_RIGHT: Algorithm is currently trying higher moderation levels
+ * @DIM_GOING_LEFT: Algorithm is currently trying lower moderation levels
+ */
enum {
DIM_PARKING_ON_TOP,
DIM_PARKING_TIRED,
@@ -65,63 +148,95 @@ enum {
DIM_GOING_LEFT,
};
+/**
+ * enum dim_stats_state
+ *
+ * These are the DIM algorithm statistics states.
+ * These will determine the verdict of current iteration.
+ *
+ * @DIM_STATS_WORSE: Current iteration shows worse performance than before
+ * @DIM_STATS_WORSE: Current iteration shows same performance than before
+ * @DIM_STATS_WORSE: Current iteration shows better performance than before
+ */
enum {
DIM_STATS_WORSE,
DIM_STATS_SAME,
DIM_STATS_BETTER,
};
+/**
+ * enum dim_step_result
+ *
+ * These are the DIM algorithm step results.
+ * These describe the result of a step.
+ *
+ * @DIM_STEPPED: Performed a regular step
+ * @DIM_TOO_TIRED: Same kind of step was done multiple times - should go to
+ * tired parking
+ * @DIM_ON_EDGE: Stepped to the most left/right profile
+ */
enum {
DIM_STEPPED,
DIM_TOO_TIRED,
DIM_ON_EDGE,
};
-static inline bool dim_on_top(struct dim *dim)
-{
- switch (dim->tune_state) {
- case DIM_PARKING_ON_TOP:
- case DIM_PARKING_TIRED:
- return true;
- case DIM_GOING_RIGHT:
- return (dim->steps_left > 1) && (dim->steps_right == 1);
- default: /* DIM_GOING_LEFT */
- return (dim->steps_right > 1) && (dim->steps_left == 1);
- }
-}
-
-static inline void dim_turn(struct dim *dim)
-{
- switch (dim->tune_state) {
- case DIM_PARKING_ON_TOP:
- case DIM_PARKING_TIRED:
- break;
- case DIM_GOING_RIGHT:
- dim->tune_state = DIM_GOING_LEFT;
- dim->steps_left = 0;
- break;
- case DIM_GOING_LEFT:
- dim->tune_state = DIM_GOING_RIGHT;
- dim->steps_right = 0;
- break;
- }
-}
-
-static inline void dim_park_on_top(struct dim *dim)
-{
- dim->steps_right = 0;
- dim->steps_left = 0;
- dim->tired = 0;
- dim->tune_state = DIM_PARKING_ON_TOP;
-}
-
-static inline void dim_park_tired(struct dim *dim)
-{
- dim->steps_right = 0;
- dim->steps_left = 0;
- dim->tune_state = DIM_PARKING_TIRED;
-}
-
+/**
+ * dim_on_top - check if current state is a good place to stop (top location)
+ * @dim: DIM context
+ *
+ * Check if current profile is a good place to park at.
+ * This will result in reducing the DIM checks frequency as we assume we
+ * shouldn't probably change profiles, unless traffic pattern wasn't changed.
+ */
+bool dim_on_top(struct dim *dim);
+
+/**
+ * dim_turn - change profile alterning direction
+ * @dim: DIM context
+ *
+ * Go left if we were going right and vice-versa.
+ * Do nothing if currently parking.
+ */
+void dim_turn(struct dim *dim);
+
+/**
+ * dim_park_on_top - enter a parking state on a top location
+ * @dim: DIM context
+ *
+ * Enter parking state.
+ * Clear all movement history.
+ */
+void dim_park_on_top(struct dim *dim);
+
+/**
+ * dim_park_tired - enter a tired parking state
+ * @dim: DIM context
+ *
+ * Enter parking state.
+ * Clear all movement history and cause DIM checks frequency to reduce.
+ */
+void dim_park_tired(struct dim *dim);
+
+/**
+ * dim_calc_stats - calculate the difference between two samples
+ * @start: start sample
+ * @end: end sample
+ * @curr_stats: delta between samples
+ *
+ * Calculate the delta between two samples (in data rates).
+ * Takes into consideration counter wrap-around.
+ */
+void dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
+ struct dim_stats *curr_stats);
+
+/**
+ * dim_update_sample - set a sample's fields with give values
+ * @event_ctr: number of events to set
+ * @packets: number of packets to set
+ * @bytes: number of bytes to set
+ * @s: DIM sample
+ */
static inline void
dim_update_sample(u16 event_ctr, u64 packets, u64 bytes, struct dim_sample *s)
{
@@ -131,23 +246,4 @@ dim_update_sample(u16 event_ctr, u64 packets, u64 bytes, struct dim_sample *s)
s->event_ctr = event_ctr;
}
-static inline void
-dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
- struct dim_stats *curr_stats)
-{
- /* u32 holds up to 71 minutes, should be enough */
- u32 delta_us = ktime_us_delta(end->time, start->time);
- u32 npkts = BIT_GAP(BITS_PER_TYPE(u32), end->pkt_ctr, start->pkt_ctr);
- u32 nbytes = BIT_GAP(BITS_PER_TYPE(u32), end->byte_ctr,
- start->byte_ctr);
-
- if (!delta_us)
- return;
-
- curr_stats->ppms = DIV_ROUND_UP(npkts * USEC_PER_MSEC, delta_us);
- curr_stats->bpms = DIV_ROUND_UP(nbytes * USEC_PER_MSEC, delta_us);
- curr_stats->epms = DIV_ROUND_UP(DIM_NEVENTS * USEC_PER_MSEC,
- delta_us);
-}
-
#endif /* DIM_H */
diff --git a/include/linux/net_dim.h b/include/linux/net_dim.h
index 4e009ec193ef..e3278cb8c392 100644
--- a/include/linux/net_dim.h
+++ b/include/linux/net_dim.h
@@ -37,14 +37,18 @@
#include <linux/module.h>
#include <linux/dim.h>
+/*
+ * Net DIM profiles:
+ * There are different set of profiles for each CQ period mode.
+ * There are different set of profiles for RX/TX CQs.
+ * Each profile size must be of NET_DIM_PARAMS_NUM_PROFILES
+ */
#define NET_DIM_PARAMS_NUM_PROFILES 5
-/* Netdev dynamic interrupt moderation profiles */
#define NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE 256
#define NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE 128
#define NET_DIM_DEF_PROFILE_CQE 1
#define NET_DIM_DEF_PROFILE_EQE 1
-/* All profiles sizes must be NET_PARAMS_DIM_NUM_PROFILES */
#define NET_DIM_RX_EQE_PROFILES { \
{1, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
{8, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
@@ -89,185 +93,41 @@ tx_profile[DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
NET_DIM_TX_CQE_PROFILES,
};
-static inline struct dim_cq_moder
-net_dim_get_rx_moderation(u8 cq_period_mode, int ix)
-{
- struct dim_cq_moder cq_moder = rx_profile[cq_period_mode][ix];
-
- cq_moder.cq_period_mode = cq_period_mode;
- return cq_moder;
-}
-
-static inline struct dim_cq_moder
-net_dim_get_def_rx_moderation(u8 cq_period_mode)
-{
- u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
- NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
-
- return net_dim_get_rx_moderation(cq_period_mode, profile_ix);
-}
-
-static inline struct dim_cq_moder
-net_dim_get_tx_moderation(u8 cq_period_mode, int ix)
-{
- struct dim_cq_moder cq_moder = tx_profile[cq_period_mode][ix];
-
- cq_moder.cq_period_mode = cq_period_mode;
- return cq_moder;
-}
-
-static inline struct dim_cq_moder
-net_dim_get_def_tx_moderation(u8 cq_period_mode)
-{
- u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
- NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
-
- return net_dim_get_tx_moderation(cq_period_mode, profile_ix);
-}
-
-static inline int net_dim_step(struct dim *dim)
-{
- if (dim->tired == (NET_DIM_PARAMS_NUM_PROFILES * 2))
- return DIM_TOO_TIRED;
-
- switch (dim->tune_state) {
- case DIM_PARKING_ON_TOP:
- case DIM_PARKING_TIRED:
- break;
- case DIM_GOING_RIGHT:
- if (dim->profile_ix == (NET_DIM_PARAMS_NUM_PROFILES - 1))
- return DIM_ON_EDGE;
- dim->profile_ix++;
- dim->steps_right++;
- break;
- case DIM_GOING_LEFT:
- if (dim->profile_ix == 0)
- return DIM_ON_EDGE;
- dim->profile_ix--;
- dim->steps_left++;
- break;
- }
-
- dim->tired++;
- return DIM_STEPPED;
-}
-
-static inline void net_dim_exit_parking(struct dim *dim)
-{
- dim->tune_state = dim->profile_ix ? DIM_GOING_LEFT :
- DIM_GOING_RIGHT;
- net_dim_step(dim);
-}
-
-static inline int net_dim_stats_compare(struct dim_stats *curr,
- struct dim_stats *prev)
-{
- if (!prev->bpms)
- return curr->bpms ? DIM_STATS_BETTER :
- DIM_STATS_SAME;
-
- if (IS_SIGNIFICANT_DIFF(curr->bpms, prev->bpms))
- return (curr->bpms > prev->bpms) ? DIM_STATS_BETTER :
- DIM_STATS_WORSE;
-
- if (!prev->ppms)
- return curr->ppms ? DIM_STATS_BETTER :
- DIM_STATS_SAME;
-
- if (IS_SIGNIFICANT_DIFF(curr->ppms, prev->ppms))
- return (curr->ppms > prev->ppms) ? DIM_STATS_BETTER :
- DIM_STATS_WORSE;
-
- if (!prev->epms)
- return DIM_STATS_SAME;
-
- if (IS_SIGNIFICANT_DIFF(curr->epms, prev->epms))
- return (curr->epms < prev->epms) ? DIM_STATS_BETTER :
- DIM_STATS_WORSE;
-
- return DIM_STATS_SAME;
-}
-
-static inline bool net_dim_decision(struct dim_stats *curr_stats,
- struct dim *dim)
-{
- int prev_state = dim->tune_state;
- int prev_ix = dim->profile_ix;
- int stats_res;
- int step_res;
-
- switch (dim->tune_state) {
- case DIM_PARKING_ON_TOP:
- stats_res = net_dim_stats_compare(curr_stats, &dim->prev_stats);
- if (stats_res != DIM_STATS_SAME)
- net_dim_exit_parking(dim);
- break;
-
- case DIM_PARKING_TIRED:
- dim->tired--;
- if (!dim->tired)
- net_dim_exit_parking(dim);
- break;
-
- case DIM_GOING_RIGHT:
- case DIM_GOING_LEFT:
- stats_res = net_dim_stats_compare(curr_stats, &dim->prev_stats);
- if (stats_res != DIM_STATS_BETTER)
- dim_turn(dim);
-
- if (dim_on_top(dim)) {
- dim_park_on_top(dim);
- break;
- }
-
- step_res = net_dim_step(dim);
- switch (step_res) {
- case DIM_ON_EDGE:
- dim_park_on_top(dim);
- break;
- case DIM_TOO_TIRED:
- dim_park_tired(dim);
- break;
- }
-
- break;
- }
+/**
+ * net_dim_get_rx_moderation - provide a CQ moderation object for the given RX profile
+ * @cq_period_mode: CQ period mode
+ * @ix: Profile index
+ */
+struct dim_cq_moder net_dim_get_rx_moderation(u8 cq_period_mode, int ix);
- if (prev_state != DIM_PARKING_ON_TOP ||
- dim->tune_state != DIM_PARKING_ON_TOP)
- dim->prev_stats = *curr_stats;
+/**
+ * net_dim_get_def_rx_moderation - provide the default RX moderation
+ * @cq_period_mode: CQ period mode
+ */
+struct dim_cq_moder net_dim_get_def_rx_moderation(u8 cq_period_mode);
- return dim->profile_ix != prev_ix;
-}
+/**
+ * net_dim_get_tx_moderation - provide a CQ moderation object for the given TX profile
+ * @cq_period_mode: CQ period mode
+ * @ix: Profile index
+ */
+struct dim_cq_moder net_dim_get_tx_moderation(u8 cq_period_mode, int ix);
-static inline void net_dim(struct dim *dim,
- struct dim_sample end_sample)
-{
- struct dim_stats curr_stats;
- u16 nevents;
+/**
+ * net_dim_get_def_tx_moderation - provide the default TX moderation
+ * @cq_period_mode: CQ period mode
+ */
+struct dim_cq_moder net_dim_get_def_tx_moderation(u8 cq_period_mode);
- switch (dim->state) {
- case DIM_MEASURE_IN_PROGRESS:
- nevents = BIT_GAP(BITS_PER_TYPE(u16),
- end_sample.event_ctr,
- dim->start_sample.event_ctr);
- if (nevents < DIM_NEVENTS)
- break;
- dim_calc_stats(&dim->start_sample, &end_sample, &curr_stats);
- if (net_dim_decision(&curr_stats, dim)) {
- dim->state = DIM_APPLY_NEW_PROFILE;
- schedule_work(&dim->work);
- break;
- }
- /* fall through */
- case DIM_START_MEASURE:
- dim_update_sample(end_sample.event_ctr, end_sample.pkt_ctr,
- end_sample.byte_ctr, &dim->start_sample);
- dim->state = DIM_MEASURE_IN_PROGRESS;
- break;
- case DIM_APPLY_NEW_PROFILE:
- break;
- }
-}
+/**
+ * net_dim - main DIM algorithm entry point
+ * @dim: DIM instance information
+ * @end_sample: Current data measurement
+ *
+ * Called by the consumer.
+ * This is the main logic of the algorithm, where data is processed in order to decide on next
+ * required action.
+ */
+void net_dim(struct dim *dim, struct dim_sample end_sample);
#endif /* NET_DIM_H */
diff --git a/lib/Kconfig b/lib/Kconfig
index 90623a0e1942..78ddb9526b62 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -562,6 +562,14 @@ config SIGNATURE
Digital signature verification. Currently only RSA is supported.
Implementation is done using GnuPG MPI library
+config DIMLIB
+ bool "DIM library"
+ default y
+ help
+ Dynamic Interrupt Moderation library.
+ Implements an algorithm for dynamically change CQ modertion values
+ according to run time performance.
+
#
# libfdt files, only selected if needed.
#
diff --git a/lib/Makefile b/lib/Makefile
index fb7697031a79..dcb558c7554d 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -202,6 +202,7 @@ obj-$(CONFIG_GLOB) += glob.o
obj-$(CONFIG_GLOB_SELFTEST) += globtest.o
obj-$(CONFIG_MPILIB) += mpi/
+obj-$(CONFIG_DIMLIB) += dim/
obj-$(CONFIG_SIGNATURE) += digsig.o
lib-$(CONFIG_CLZ_TAB) += clz_tab.o
diff --git a/lib/dim/Makefile b/lib/dim/Makefile
new file mode 100644
index 000000000000..160afe288df0
--- /dev/null
+++ b/lib/dim/Makefile
@@ -0,0 +1,9 @@
+#
+# DIM Dynamic Interrupt Moderation library
+#
+
+obj-$(CONFIG_DIMLIB) = net_dim.o
+
+net_dim-y = \
+ dim.o \
+ net_dim.o
diff --git a/lib/dim/dim.c b/lib/dim/dim.c
new file mode 100644
index 000000000000..17d5236759bd
--- /dev/null
+++ b/lib/dim/dim.c
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+/*
+ * Copyright (c) 2019, Mellanox Technologies inc. All rights reserved.
+ */
+
+#include <linux/dim.h>
+
+bool dim_on_top(struct dim *dim)
+{
+ switch (dim->tune_state) {
+ case DIM_PARKING_ON_TOP:
+ case DIM_PARKING_TIRED:
+ return true;
+ case DIM_GOING_RIGHT:
+ return (dim->steps_left > 1) && (dim->steps_right == 1);
+ default: /* DIM_GOING_LEFT */
+ return (dim->steps_right > 1) && (dim->steps_left == 1);
+ }
+}
+EXPORT_SYMBOL(dim_on_top);
+
+void dim_turn(struct dim *dim)
+{
+ switch (dim->tune_state) {
+ case DIM_PARKING_ON_TOP:
+ case DIM_PARKING_TIRED:
+ break;
+ case DIM_GOING_RIGHT:
+ dim->tune_state = DIM_GOING_LEFT;
+ dim->steps_left = 0;
+ break;
+ case DIM_GOING_LEFT:
+ dim->tune_state = DIM_GOING_RIGHT;
+ dim->steps_right = 0;
+ break;
+ }
+}
+EXPORT_SYMBOL(dim_turn);
+
+void dim_park_on_top(struct dim *dim)
+{
+ dim->steps_right = 0;
+ dim->steps_left = 0;
+ dim->tired = 0;
+ dim->tune_state = DIM_PARKING_ON_TOP;
+}
+EXPORT_SYMBOL(dim_park_on_top);
+
+void dim_park_tired(struct dim *dim)
+{
+ dim->steps_right = 0;
+ dim->steps_left = 0;
+ dim->tune_state = DIM_PARKING_TIRED;
+}
+EXPORT_SYMBOL(dim_park_tired);
+
+void dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
+ struct dim_stats *curr_stats)
+{
+ /* u32 holds up to 71 minutes, should be enough */
+ u32 delta_us = ktime_us_delta(end->time, start->time);
+ u32 npkts = BIT_GAP(BITS_PER_TYPE(u32), end->pkt_ctr, start->pkt_ctr);
+ u32 nbytes = BIT_GAP(BITS_PER_TYPE(u32), end->byte_ctr,
+ start->byte_ctr);
+
+ if (!delta_us)
+ return;
+
+ curr_stats->ppms = DIV_ROUND_UP(npkts * USEC_PER_MSEC, delta_us);
+ curr_stats->bpms = DIV_ROUND_UP(nbytes * USEC_PER_MSEC, delta_us);
+ curr_stats->epms = DIV_ROUND_UP(DIM_NEVENTS * USEC_PER_MSEC,
+ delta_us);
+}
+EXPORT_SYMBOL(dim_calc_stats);
diff --git a/lib/dim/net_dim.c b/lib/dim/net_dim.c
new file mode 100644
index 000000000000..a988de317773
--- /dev/null
+++ b/lib/dim/net_dim.c
@@ -0,0 +1,191 @@
+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+/*
+ * Copyright (c) 2018, Mellanox Technologies inc. All rights reserved.
+ */
+
+#include <linux/net_dim.h>
+#include <linux/dim.h>
+
+struct dim_cq_moder
+net_dim_get_rx_moderation(u8 cq_period_mode, int ix)
+{
+ struct dim_cq_moder cq_moder = rx_profile[cq_period_mode][ix];
+
+ cq_moder.cq_period_mode = cq_period_mode;
+ return cq_moder;
+}
+EXPORT_SYMBOL(net_dim_get_rx_moderation);
+
+struct dim_cq_moder
+net_dim_get_def_rx_moderation(u8 cq_period_mode)
+{
+ u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
+ NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
+
+ return net_dim_get_rx_moderation(cq_period_mode, profile_ix);
+}
+EXPORT_SYMBOL(net_dim_get_def_rx_moderation);
+
+struct dim_cq_moder
+net_dim_get_tx_moderation(u8 cq_period_mode, int ix)
+{
+ struct dim_cq_moder cq_moder = tx_profile[cq_period_mode][ix];
+
+ cq_moder.cq_period_mode = cq_period_mode;
+ return cq_moder;
+}
+EXPORT_SYMBOL(net_dim_get_tx_moderation);
+
+struct dim_cq_moder
+net_dim_get_def_tx_moderation(u8 cq_period_mode)
+{
+ u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
+ NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
+
+ return net_dim_get_tx_moderation(cq_period_mode, profile_ix);
+}
+EXPORT_SYMBOL(net_dim_get_def_tx_moderation);
+
+static int net_dim_step(struct dim *dim)
+{
+ if (dim->tired == (NET_DIM_PARAMS_NUM_PROFILES * 2))
+ return DIM_TOO_TIRED;
+
+ switch (dim->tune_state) {
+ case DIM_PARKING_ON_TOP:
+ case DIM_PARKING_TIRED:
+ break;
+ case DIM_GOING_RIGHT:
+ if (dim->profile_ix == (NET_DIM_PARAMS_NUM_PROFILES - 1))
+ return DIM_ON_EDGE;
+ dim->profile_ix++;
+ dim->steps_right++;
+ break;
+ case DIM_GOING_LEFT:
+ if (dim->profile_ix == 0)
+ return DIM_ON_EDGE;
+ dim->profile_ix--;
+ dim->steps_left++;
+ break;
+ }
+
+ dim->tired++;
+ return DIM_STEPPED;
+}
+
+static void net_dim_exit_parking(struct dim *dim)
+{
+ dim->tune_state = dim->profile_ix ? DIM_GOING_LEFT : DIM_GOING_RIGHT;
+ net_dim_step(dim);
+}
+
+static int net_dim_stats_compare(struct dim_stats *curr,
+ struct dim_stats *prev)
+{
+ if (!prev->bpms)
+ return curr->bpms ? DIM_STATS_BETTER : DIM_STATS_SAME;
+
+ if (IS_SIGNIFICANT_DIFF(curr->bpms, prev->bpms))
+ return (curr->bpms > prev->bpms) ? DIM_STATS_BETTER :
+ DIM_STATS_WORSE;
+
+ if (!prev->ppms)
+ return curr->ppms ? DIM_STATS_BETTER :
+ DIM_STATS_SAME;
+
+ if (IS_SIGNIFICANT_DIFF(curr->ppms, prev->ppms))
+ return (curr->ppms > prev->ppms) ? DIM_STATS_BETTER :
+ DIM_STATS_WORSE;
+
+ if (!prev->epms)
+ return DIM_STATS_SAME;
+
+ if (IS_SIGNIFICANT_DIFF(curr->epms, prev->epms))
+ return (curr->epms < prev->epms) ? DIM_STATS_BETTER :
+ DIM_STATS_WORSE;
+
+ return DIM_STATS_SAME;
+}
+
+static bool net_dim_decision(struct dim_stats *curr_stats, struct dim *dim)
+{
+ int prev_state = dim->tune_state;
+ int prev_ix = dim->profile_ix;
+ int stats_res;
+ int step_res;
+
+ switch (dim->tune_state) {
+ case DIM_PARKING_ON_TOP:
+ stats_res = net_dim_stats_compare(curr_stats,
+ &dim->prev_stats);
+ if (stats_res != DIM_STATS_SAME)
+ net_dim_exit_parking(dim);
+ break;
+
+ case DIM_PARKING_TIRED:
+ dim->tired--;
+ if (!dim->tired)
+ net_dim_exit_parking(dim);
+ break;
+
+ case DIM_GOING_RIGHT:
+ case DIM_GOING_LEFT:
+ stats_res = net_dim_stats_compare(curr_stats,
+ &dim->prev_stats);
+ if (stats_res != DIM_STATS_BETTER)
+ dim_turn(dim);
+
+ if (dim_on_top(dim)) {
+ dim_park_on_top(dim);
+ break;
+ }
+
+ step_res = net_dim_step(dim);
+ switch (step_res) {
+ case DIM_ON_EDGE:
+ dim_park_on_top(dim);
+ break;
+ case DIM_TOO_TIRED:
+ dim_park_tired(dim);
+ break;
+ }
+
+ break;
+ }
+
+ if (prev_state != DIM_PARKING_ON_TOP ||
+ dim->tune_state != DIM_PARKING_ON_TOP)
+ dim->prev_stats = *curr_stats;
+
+ return dim->profile_ix != prev_ix;
+}
+
+void net_dim(struct dim *dim, struct dim_sample end_sample)
+{
+ struct dim_stats curr_stats;
+ u16 nevents;
+
+ switch (dim->state) {
+ case DIM_MEASURE_IN_PROGRESS:
+ nevents = BIT_GAP(BITS_PER_TYPE(u16),
+ end_sample.event_ctr,
+ dim->start_sample.event_ctr);
+ if (nevents < DIM_NEVENTS)
+ break;
+ dim_calc_stats(&dim->start_sample, &end_sample, &curr_stats);
+ if (net_dim_decision(&curr_stats, dim)) {
+ dim->state = DIM_APPLY_NEW_PROFILE;
+ schedule_work(&dim->work);
+ break;
+ }
+ /* fall through */
+ case DIM_START_MEASURE:
+ dim_update_sample(end_sample.event_ctr, end_sample.pkt_ctr,
+ end_sample.byte_ctr, &dim->start_sample);
+ dim->state = DIM_MEASURE_IN_PROGRESS;
+ break;
+ case DIM_APPLY_NEW_PROFILE:
+ break;
+ }
+}
+EXPORT_SYMBOL(net_dim);
--
2.21.0
^ permalink raw reply related
* [for-next 5/9] linux/dim: Rename externally used net_dim members
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Saeed Mahameed
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
From: Tal Gilboa <talgi@mellanox.com>
Removed 'net' prefix from functions and structs used by external drivers.
Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/broadcom/bcmsysport.c | 16 +++++-----
drivers/net/ethernet/broadcom/bcmsysport.h | 2 +-
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 10 +++----
drivers/net/ethernet/broadcom/bnxt/bnxt.h | 2 +-
.../net/ethernet/broadcom/bnxt/bnxt_debugfs.c | 4 +--
drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c | 5 ++--
.../net/ethernet/broadcom/genet/bcmgenet.c | 14 ++++-----
.../net/ethernet/broadcom/genet/bcmgenet.h | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/en.h | 8 ++---
.../net/ethernet/mellanox/mlx5/core/en_dim.c | 10 +++----
.../ethernet/mellanox/mlx5/core/en_ethtool.c | 4 +--
.../net/ethernet/mellanox/mlx5/core/en_main.c | 12 ++++----
.../net/ethernet/mellanox/mlx5/core/en_txrx.c | 8 ++---
include/linux/dim.h | 21 +++++++------
include/linux/net_dim.h | 30 +++++++++----------
15 files changed, 73 insertions(+), 75 deletions(-)
diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c
index faaf8ade15e5..c1247b2948ff 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.c
+++ b/drivers/net/ethernet/broadcom/bcmsysport.c
@@ -612,7 +612,7 @@ static int bcm_sysport_set_coalesce(struct net_device *dev,
struct ethtool_coalesce *ec)
{
struct bcm_sysport_priv *priv = netdev_priv(dev);
- struct net_dim_cq_moder moder;
+ struct dim_cq_moder moder;
u32 usecs, pkts;
unsigned int i;
@@ -995,7 +995,7 @@ static int bcm_sysport_poll(struct napi_struct *napi, int budget)
{
struct bcm_sysport_priv *priv =
container_of(napi, struct bcm_sysport_priv, napi);
- struct net_dim_sample dim_sample;
+ struct dim_sample dim_sample;
unsigned int work_done = 0;
work_done = bcm_sysport_desc_rx(priv, budget);
@@ -1019,8 +1019,8 @@ static int bcm_sysport_poll(struct napi_struct *napi, int budget)
}
if (priv->dim.use_dim) {
- net_dim_update_sample(priv->dim.event_ctr, priv->dim.packets,
- priv->dim.bytes, &dim_sample);
+ dim_update_sample(priv->dim.event_ctr, priv->dim.packets,
+ priv->dim.bytes, &dim_sample);
net_dim(&priv->dim.dim, dim_sample);
}
@@ -1090,13 +1090,13 @@ static void bcm_sysport_resume_from_wol(struct bcm_sysport_priv *priv)
static void bcm_sysport_dim_work(struct work_struct *work)
{
- struct net_dim *dim = container_of(work, struct net_dim, work);
+ struct dim *dim = container_of(work, struct dim, work);
struct bcm_sysport_net_dim *ndim =
container_of(dim, struct bcm_sysport_net_dim, dim);
struct bcm_sysport_priv *priv =
container_of(ndim, struct bcm_sysport_priv, dim);
- struct net_dim_cq_moder cur_profile =
- net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
+ struct dim_cq_moder cur_profile = net_dim_get_rx_moderation(dim->mode,
+ dim->profile_ix);
bcm_sysport_set_rx_coalesce(priv, cur_profile.usec, cur_profile.pkts);
dim->state = DIM_START_MEASURE;
@@ -1449,7 +1449,7 @@ static void bcm_sysport_init_dim(struct bcm_sysport_priv *priv,
static void bcm_sysport_init_rx_coalesce(struct bcm_sysport_priv *priv)
{
struct bcm_sysport_net_dim *dim = &priv->dim;
- struct net_dim_cq_moder moder;
+ struct dim_cq_moder moder;
u32 usecs, pkts;
usecs = priv->rx_coalesce_usecs;
diff --git a/drivers/net/ethernet/broadcom/bcmsysport.h b/drivers/net/ethernet/broadcom/bcmsysport.h
index 6f3141c86436..cbe6d559d964 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.h
+++ b/drivers/net/ethernet/broadcom/bcmsysport.h
@@ -705,7 +705,7 @@ struct bcm_sysport_net_dim {
u16 event_ctr;
unsigned long packets;
unsigned long bytes;
- struct net_dim dim;
+ struct dim dim;
};
/* Software view of the TX ring */
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index eaec949c367a..c54668004600 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -2128,12 +2128,12 @@ static int bnxt_poll(struct napi_struct *napi, int budget)
}
}
if (bp->flags & BNXT_FLAG_DIM) {
- struct net_dim_sample dim_sample;
+ struct dim_sample dim_sample;
- net_dim_update_sample(cpr->event_ctr,
- cpr->rx_packets,
- cpr->rx_bytes,
- &dim_sample);
+ dim_update_sample(cpr->event_ctr,
+ cpr->rx_packets,
+ cpr->rx_bytes,
+ &dim_sample);
net_dim(&cpr->dim, dim_sample);
}
return work_done;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index eca36dd6b751..a552c5539cc9 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -809,7 +809,7 @@ struct bnxt_cp_ring_info {
u64 rx_bytes;
u64 event_ctr;
- struct net_dim dim;
+ struct dim dim;
union {
struct tx_cmp *cp_desc_ring[MAX_CP_PAGES];
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_debugfs.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_debugfs.c
index 94e208e9789f..3d1d53fbb135 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_debugfs.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_debugfs.c
@@ -21,7 +21,7 @@ static ssize_t debugfs_dim_read(struct file *filep,
char __user *buffer,
size_t count, loff_t *ppos)
{
- struct net_dim *dim = filep->private_data;
+ struct dim *dim = filep->private_data;
int len;
char *buf;
@@ -61,7 +61,7 @@ static const struct file_operations debugfs_dim_fops = {
.read = debugfs_dim_read,
};
-static struct dentry *debugfs_dim_ring_init(struct net_dim *dim, int ring_idx,
+static struct dentry *debugfs_dim_ring_init(struct dim *dim, int ring_idx,
struct dentry *dd)
{
static char qname[16];
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c
index 16a4588709d1..11605f9fa61e 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c
@@ -13,15 +13,14 @@
void bnxt_dim_work(struct work_struct *work)
{
- struct net_dim *dim = container_of(work, struct net_dim,
- work);
+ struct dim *dim = container_of(work, struct dim, work);
struct bnxt_cp_ring_info *cpr = container_of(dim,
struct bnxt_cp_ring_info,
dim);
struct bnxt_napi *bnapi = container_of(cpr,
struct bnxt_napi,
cp_ring);
- struct net_dim_cq_moder cur_moder =
+ struct dim_cq_moder cur_moder =
net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
cpr->rx_ring_coal.coal_ticks = cur_moder.usec;
diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.c b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
index 297ae786ffed..b7f8f4f1088f 100644
--- a/drivers/net/ethernet/broadcom/genet/bcmgenet.c
+++ b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
@@ -643,7 +643,7 @@ static void bcmgenet_set_rx_coalesce(struct bcmgenet_rx_ring *ring,
static void bcmgenet_set_ring_rx_coalesce(struct bcmgenet_rx_ring *ring,
struct ethtool_coalesce *ec)
{
- struct net_dim_cq_moder moder;
+ struct dim_cq_moder moder;
u32 usecs, pkts;
ring->rx_coalesce_usecs = ec->rx_coalesce_usecs;
@@ -1898,7 +1898,7 @@ static int bcmgenet_rx_poll(struct napi_struct *napi, int budget)
{
struct bcmgenet_rx_ring *ring = container_of(napi,
struct bcmgenet_rx_ring, napi);
- struct net_dim_sample dim_sample;
+ struct dim_sample dim_sample;
unsigned int work_done;
work_done = bcmgenet_desc_rx(ring, budget);
@@ -1909,8 +1909,8 @@ static int bcmgenet_rx_poll(struct napi_struct *napi, int budget)
}
if (ring->dim.use_dim) {
- net_dim_update_sample(ring->dim.event_ctr, ring->dim.packets,
- ring->dim.bytes, &dim_sample);
+ dim_update_sample(ring->dim.event_ctr, ring->dim.packets,
+ ring->dim.bytes, &dim_sample);
net_dim(&ring->dim.dim, dim_sample);
}
@@ -1919,12 +1919,12 @@ static int bcmgenet_rx_poll(struct napi_struct *napi, int budget)
static void bcmgenet_dim_work(struct work_struct *work)
{
- struct net_dim *dim = container_of(work, struct net_dim, work);
+ struct dim *dim = container_of(work, struct dim, work);
struct bcmgenet_net_dim *ndim =
container_of(dim, struct bcmgenet_net_dim, dim);
struct bcmgenet_rx_ring *ring =
container_of(ndim, struct bcmgenet_rx_ring, dim);
- struct net_dim_cq_moder cur_profile =
+ struct dim_cq_moder cur_profile =
net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
bcmgenet_set_rx_coalesce(ring, cur_profile.usec, cur_profile.pkts);
@@ -2094,7 +2094,7 @@ static void bcmgenet_init_dim(struct bcmgenet_rx_ring *ring,
static void bcmgenet_init_rx_coalesce(struct bcmgenet_rx_ring *ring)
{
struct bcmgenet_net_dim *dim = &ring->dim;
- struct net_dim_cq_moder moder;
+ struct dim_cq_moder moder;
u32 usecs, pkts;
usecs = ring->rx_coalesce_usecs;
diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.h b/drivers/net/ethernet/broadcom/genet/bcmgenet.h
index 14b49612aa86..6e418d9c3706 100644
--- a/drivers/net/ethernet/broadcom/genet/bcmgenet.h
+++ b/drivers/net/ethernet/broadcom/genet/bcmgenet.h
@@ -581,7 +581,7 @@ struct bcmgenet_net_dim {
u16 event_ctr;
unsigned long packets;
unsigned long bytes;
- struct net_dim dim;
+ struct dim dim;
};
struct bcmgenet_rx_ring {
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 3a183d690e23..11efd6e4bdc3 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -238,9 +238,9 @@ struct mlx5e_params {
u16 num_channels;
u8 num_tc;
bool rx_cqe_compress_def;
- struct net_dim_cq_moder rx_cq_moderation;
- struct net_dim_cq_moder tx_cq_moderation;
bool tunneled_offload_en;
+ struct dim_cq_moder rx_cq_moderation;
+ struct dim_cq_moder tx_cq_moderation;
bool lro_en;
u8 tx_min_inline_mode;
bool vlan_strip_disable;
@@ -356,7 +356,7 @@ struct mlx5e_txqsq {
/* dirtied @completion */
u16 cc;
u32 dma_fifo_cc;
- struct net_dim dim; /* Adaptive Moderation */
+ struct dim dim; /* Adaptive Moderation */
/* dirtied @xmit */
u16 pc ____cacheline_aligned_in_smp;
@@ -595,7 +595,7 @@ struct mlx5e_rq {
int ix;
unsigned int hw_mtu;
- struct net_dim dim; /* Dynamic Interrupt Moderation */
+ struct dim dim; /* Dynamic Interrupt Moderation */
/* XDP */
struct bpf_prog *xdp_prog;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c b/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
index a80303add7c0..ba3c1be9f2d3 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
@@ -34,7 +34,7 @@
#include "en.h"
static void
-mlx5e_complete_dim_work(struct net_dim *dim, struct net_dim_cq_moder moder,
+mlx5e_complete_dim_work(struct dim *dim, struct dim_cq_moder moder,
struct mlx5_core_dev *mdev, struct mlx5_core_cq *mcq)
{
mlx5_core_modify_cq_moderation(mdev, mcq, moder.usec, moder.pkts);
@@ -43,9 +43,9 @@ mlx5e_complete_dim_work(struct net_dim *dim, struct net_dim_cq_moder moder,
void mlx5e_rx_dim_work(struct work_struct *work)
{
- struct net_dim *dim = container_of(work, struct net_dim, work);
+ struct dim *dim = container_of(work, struct dim, work);
struct mlx5e_rq *rq = container_of(dim, struct mlx5e_rq, dim);
- struct net_dim_cq_moder cur_moder =
+ struct dim_cq_moder cur_moder =
net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
mlx5e_complete_dim_work(dim, cur_moder, rq->mdev, &rq->cq.mcq);
@@ -53,9 +53,9 @@ void mlx5e_rx_dim_work(struct work_struct *work)
void mlx5e_tx_dim_work(struct work_struct *work)
{
- struct net_dim *dim = container_of(work, struct net_dim, work);
+ struct dim *dim = container_of(work, struct dim, work);
struct mlx5e_txqsq *sq = container_of(dim, struct mlx5e_txqsq, dim);
- struct net_dim_cq_moder cur_moder =
+ struct dim_cq_moder cur_moder =
net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
mlx5e_complete_dim_work(dim, cur_moder, sq->cq.mdev, &sq->cq.mcq);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
index dd764e0471f2..c853b657739c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
@@ -466,7 +466,7 @@ static int mlx5e_set_channels(struct net_device *dev,
int mlx5e_ethtool_get_coalesce(struct mlx5e_priv *priv,
struct ethtool_coalesce *coal)
{
- struct net_dim_cq_moder *rx_moder, *tx_moder;
+ struct dim_cq_moder *rx_moder, *tx_moder;
if (!MLX5_CAP_GEN(priv->mdev, cq_moderation))
return -EOPNOTSUPP;
@@ -521,7 +521,7 @@ mlx5e_set_priv_channels_coalesce(struct mlx5e_priv *priv, struct ethtool_coalesc
int mlx5e_ethtool_set_coalesce(struct mlx5e_priv *priv,
struct ethtool_coalesce *coal)
{
- struct net_dim_cq_moder *rx_moder, *tx_moder;
+ struct dim_cq_moder *rx_moder, *tx_moder;
struct mlx5_core_dev *mdev = priv->mdev;
struct mlx5e_channels new_channels = {};
int err = 0;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 5b89e992e482..9705101c0235 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -1569,7 +1569,7 @@ static void mlx5e_destroy_cq(struct mlx5e_cq *cq)
}
static int mlx5e_open_cq(struct mlx5e_channel *c,
- struct net_dim_cq_moder moder,
+ struct dim_cq_moder moder,
struct mlx5e_cq_param *param,
struct mlx5e_cq *cq)
{
@@ -1774,7 +1774,7 @@ static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
struct mlx5e_channel **cp)
{
int cpu = cpumask_first(mlx5_comp_irq_get_affinity_mask(priv->mdev, ix));
- struct net_dim_cq_moder icocq_moder = {0, 0};
+ struct dim_cq_moder icocq_moder = {0, 0};
struct net_device *netdev = priv->netdev;
struct mlx5e_channel *c;
unsigned int irq;
@@ -4411,9 +4411,9 @@ static bool slow_pci_heuristic(struct mlx5_core_dev *mdev)
link_speed > MLX5E_SLOW_PCI_RATIO * pci_bw;
}
-static struct net_dim_cq_moder mlx5e_get_def_tx_moderation(u8 cq_period_mode)
+static struct dim_cq_moder mlx5e_get_def_tx_moderation(u8 cq_period_mode)
{
- struct net_dim_cq_moder moder;
+ struct dim_cq_moder moder;
moder.cq_period_mode = cq_period_mode;
moder.pkts = MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
@@ -4424,9 +4424,9 @@ static struct net_dim_cq_moder mlx5e_get_def_tx_moderation(u8 cq_period_mode)
return moder;
}
-static struct net_dim_cq_moder mlx5e_get_def_rx_moderation(u8 cq_period_mode)
+static struct dim_cq_moder mlx5e_get_def_rx_moderation(u8 cq_period_mode)
{
- struct net_dim_cq_moder moder;
+ struct dim_cq_moder moder;
moder.cq_period_mode = cq_period_mode;
moder.pkts = MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
index 07432e6428cf..e6c434efbd46 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
@@ -48,24 +48,24 @@ static inline bool mlx5e_channel_no_affinity_change(struct mlx5e_channel *c)
static void mlx5e_handle_tx_dim(struct mlx5e_txqsq *sq)
{
struct mlx5e_sq_stats *stats = sq->stats;
- struct net_dim_sample dim_sample;
+ struct dim_sample dim_sample;
if (unlikely(!test_bit(MLX5E_SQ_STATE_AM, &sq->state)))
return;
- net_dim_update_sample(sq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
+ dim_update_sample(sq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
net_dim(&sq->dim, dim_sample);
}
static void mlx5e_handle_rx_dim(struct mlx5e_rq *rq)
{
struct mlx5e_rq_stats *stats = rq->stats;
- struct net_dim_sample dim_sample;
+ struct dim_sample dim_sample;
if (unlikely(!test_bit(MLX5E_RQ_STATE_AM, &rq->state)))
return;
- net_dim_update_sample(rq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
+ dim_update_sample(rq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
net_dim(&rq->dim, dim_sample);
}
diff --git a/include/linux/dim.h b/include/linux/dim.h
index f0f20ed25497..60e5074a7cc0 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -14,13 +14,13 @@
#define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) \
& (BIT_ULL(bits) - 1))
-struct net_dim_cq_moder {
+struct dim_cq_moder {
u16 usec;
u16 pkts;
u8 cq_period_mode;
};
-struct net_dim_sample {
+struct dim_sample {
ktime_t time;
u32 pkt_ctr;
u32 byte_ctr;
@@ -33,10 +33,10 @@ struct dim_stats {
int epms; /* events per msec */
};
-struct net_dim { /* Dynamic Interrupt Moderation */
+struct dim { /* Dynamic Interrupt Moderation */
u8 state;
struct dim_stats prev_stats;
- struct net_dim_sample start_sample;
+ struct dim_sample start_sample;
struct work_struct work;
u8 profile_ix;
u8 mode;
@@ -77,7 +77,7 @@ enum {
DIM_ON_EDGE,
};
-static inline bool dim_on_top(struct net_dim *dim)
+static inline bool dim_on_top(struct dim *dim)
{
switch (dim->tune_state) {
case DIM_PARKING_ON_TOP:
@@ -90,7 +90,7 @@ static inline bool dim_on_top(struct net_dim *dim)
}
}
-static inline void dim_turn(struct net_dim *dim)
+static inline void dim_turn(struct dim *dim)
{
switch (dim->tune_state) {
case DIM_PARKING_ON_TOP:
@@ -107,7 +107,7 @@ static inline void dim_turn(struct net_dim *dim)
}
}
-static inline void dim_park_on_top(struct net_dim *dim)
+static inline void dim_park_on_top(struct dim *dim)
{
dim->steps_right = 0;
dim->steps_left = 0;
@@ -115,7 +115,7 @@ static inline void dim_park_on_top(struct net_dim *dim)
dim->tune_state = DIM_PARKING_ON_TOP;
}
-static inline void dim_park_tired(struct net_dim *dim)
+static inline void dim_park_tired(struct dim *dim)
{
dim->steps_right = 0;
dim->steps_left = 0;
@@ -123,8 +123,7 @@ static inline void dim_park_tired(struct net_dim *dim)
}
static inline void
-net_dim_update_sample(u16 event_ctr, u64 packets, u64 bytes,
- struct net_dim_sample *s)
+dim_update_sample(u16 event_ctr, u64 packets, u64 bytes, struct dim_sample *s)
{
s->time = ktime_get();
s->pkt_ctr = packets;
@@ -133,7 +132,7 @@ net_dim_update_sample(u16 event_ctr, u64 packets, u64 bytes,
}
static inline void
-dim_calc_stats(struct net_dim_sample *start, struct net_dim_sample *end,
+dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
struct dim_stats *curr_stats)
{
/* u32 holds up to 71 minutes, should be enough */
diff --git a/include/linux/net_dim.h b/include/linux/net_dim.h
index d4b40adc7fa1..4e009ec193ef 100644
--- a/include/linux/net_dim.h
+++ b/include/linux/net_dim.h
@@ -77,28 +77,28 @@
{64, 32} \
}
-static const struct net_dim_cq_moder
+static const struct dim_cq_moder
rx_profile[DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
NET_DIM_RX_EQE_PROFILES,
NET_DIM_RX_CQE_PROFILES,
};
-static const struct net_dim_cq_moder
+static const struct dim_cq_moder
tx_profile[DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
NET_DIM_TX_EQE_PROFILES,
NET_DIM_TX_CQE_PROFILES,
};
-static inline struct net_dim_cq_moder
+static inline struct dim_cq_moder
net_dim_get_rx_moderation(u8 cq_period_mode, int ix)
{
- struct net_dim_cq_moder cq_moder = rx_profile[cq_period_mode][ix];
+ struct dim_cq_moder cq_moder = rx_profile[cq_period_mode][ix];
cq_moder.cq_period_mode = cq_period_mode;
return cq_moder;
}
-static inline struct net_dim_cq_moder
+static inline struct dim_cq_moder
net_dim_get_def_rx_moderation(u8 cq_period_mode)
{
u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
@@ -107,16 +107,16 @@ net_dim_get_def_rx_moderation(u8 cq_period_mode)
return net_dim_get_rx_moderation(cq_period_mode, profile_ix);
}
-static inline struct net_dim_cq_moder
+static inline struct dim_cq_moder
net_dim_get_tx_moderation(u8 cq_period_mode, int ix)
{
- struct net_dim_cq_moder cq_moder = tx_profile[cq_period_mode][ix];
+ struct dim_cq_moder cq_moder = tx_profile[cq_period_mode][ix];
cq_moder.cq_period_mode = cq_period_mode;
return cq_moder;
}
-static inline struct net_dim_cq_moder
+static inline struct dim_cq_moder
net_dim_get_def_tx_moderation(u8 cq_period_mode)
{
u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
@@ -125,7 +125,7 @@ net_dim_get_def_tx_moderation(u8 cq_period_mode)
return net_dim_get_tx_moderation(cq_period_mode, profile_ix);
}
-static inline int net_dim_step(struct net_dim *dim)
+static inline int net_dim_step(struct dim *dim)
{
if (dim->tired == (NET_DIM_PARAMS_NUM_PROFILES * 2))
return DIM_TOO_TIRED;
@@ -152,7 +152,7 @@ static inline int net_dim_step(struct net_dim *dim)
return DIM_STEPPED;
}
-static inline void net_dim_exit_parking(struct net_dim *dim)
+static inline void net_dim_exit_parking(struct dim *dim)
{
dim->tune_state = dim->profile_ix ? DIM_GOING_LEFT :
DIM_GOING_RIGHT;
@@ -189,7 +189,7 @@ static inline int net_dim_stats_compare(struct dim_stats *curr,
}
static inline bool net_dim_decision(struct dim_stats *curr_stats,
- struct net_dim *dim)
+ struct dim *dim)
{
int prev_state = dim->tune_state;
int prev_ix = dim->profile_ix;
@@ -240,8 +240,8 @@ static inline bool net_dim_decision(struct dim_stats *curr_stats,
return dim->profile_ix != prev_ix;
}
-static inline void net_dim(struct net_dim *dim,
- struct net_dim_sample end_sample)
+static inline void net_dim(struct dim *dim,
+ struct dim_sample end_sample)
{
struct dim_stats curr_stats;
u16 nevents;
@@ -261,8 +261,8 @@ static inline void net_dim(struct net_dim *dim,
}
/* fall through */
case DIM_START_MEASURE:
- net_dim_update_sample(end_sample.event_ctr, end_sample.pkt_ctr,
- end_sample.byte_ctr, &dim->start_sample);
+ dim_update_sample(end_sample.event_ctr, end_sample.pkt_ctr,
+ end_sample.byte_ctr, &dim->start_sample);
dim->state = DIM_MEASURE_IN_PROGRESS;
break;
case DIM_APPLY_NEW_PROFILE:
--
2.21.0
^ permalink raw reply related
* [for-next 4/9] linux/dim: Rename net_dim_sample() to net_dim_update_sample()
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Saeed Mahameed
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
From: Tal Gilboa <talgi@mellanox.com>
In order to avoid confusion between the function and the similarly
named struct.
In preparation for removing the 'net' prefix from dim members.
Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/broadcom/bcmsysport.c | 4 ++--
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 8 ++++----
drivers/net/ethernet/broadcom/genet/bcmgenet.c | 4 ++--
drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c | 6 ++----
include/linux/dim.h | 3 ++-
include/linux/net_dim.h | 4 ++--
6 files changed, 14 insertions(+), 15 deletions(-)
diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c
index b5e2f9d2cb71..faaf8ade15e5 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.c
+++ b/drivers/net/ethernet/broadcom/bcmsysport.c
@@ -1019,8 +1019,8 @@ static int bcm_sysport_poll(struct napi_struct *napi, int budget)
}
if (priv->dim.use_dim) {
- net_dim_sample(priv->dim.event_ctr, priv->dim.packets,
- priv->dim.bytes, &dim_sample);
+ net_dim_update_sample(priv->dim.event_ctr, priv->dim.packets,
+ priv->dim.bytes, &dim_sample);
net_dim(&priv->dim.dim, dim_sample);
}
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 49de873043c0..eaec949c367a 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -2130,10 +2130,10 @@ static int bnxt_poll(struct napi_struct *napi, int budget)
if (bp->flags & BNXT_FLAG_DIM) {
struct net_dim_sample dim_sample;
- net_dim_sample(cpr->event_ctr,
- cpr->rx_packets,
- cpr->rx_bytes,
- &dim_sample);
+ net_dim_update_sample(cpr->event_ctr,
+ cpr->rx_packets,
+ cpr->rx_bytes,
+ &dim_sample);
net_dim(&cpr->dim, dim_sample);
}
return work_done;
diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.c b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
index 5286a46ecfb0..297ae786ffed 100644
--- a/drivers/net/ethernet/broadcom/genet/bcmgenet.c
+++ b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
@@ -1909,8 +1909,8 @@ static int bcmgenet_rx_poll(struct napi_struct *napi, int budget)
}
if (ring->dim.use_dim) {
- net_dim_sample(ring->dim.event_ctr, ring->dim.packets,
- ring->dim.bytes, &dim_sample);
+ net_dim_update_sample(ring->dim.event_ctr, ring->dim.packets,
+ ring->dim.bytes, &dim_sample);
net_dim(&ring->dim.dim, dim_sample);
}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
index f9862bf75491..07432e6428cf 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
@@ -53,8 +53,7 @@ static void mlx5e_handle_tx_dim(struct mlx5e_txqsq *sq)
if (unlikely(!test_bit(MLX5E_SQ_STATE_AM, &sq->state)))
return;
- net_dim_sample(sq->cq.event_ctr, stats->packets, stats->bytes,
- &dim_sample);
+ net_dim_update_sample(sq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
net_dim(&sq->dim, dim_sample);
}
@@ -66,8 +65,7 @@ static void mlx5e_handle_rx_dim(struct mlx5e_rq *rq)
if (unlikely(!test_bit(MLX5E_RQ_STATE_AM, &rq->state)))
return;
- net_dim_sample(rq->cq.event_ctr, stats->packets, stats->bytes,
- &dim_sample);
+ net_dim_update_sample(rq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
net_dim(&rq->dim, dim_sample);
}
diff --git a/include/linux/dim.h b/include/linux/dim.h
index 989dbbdf9d45..f0f20ed25497 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -123,7 +123,8 @@ static inline void dim_park_tired(struct net_dim *dim)
}
static inline void
-net_dim_sample(u16 event_ctr, u64 packets, u64 bytes, struct net_dim_sample *s)
+net_dim_update_sample(u16 event_ctr, u64 packets, u64 bytes,
+ struct net_dim_sample *s)
{
s->time = ktime_get();
s->pkt_ctr = packets;
diff --git a/include/linux/net_dim.h b/include/linux/net_dim.h
index e0c97f824dd0..d4b40adc7fa1 100644
--- a/include/linux/net_dim.h
+++ b/include/linux/net_dim.h
@@ -261,8 +261,8 @@ static inline void net_dim(struct net_dim *dim,
}
/* fall through */
case DIM_START_MEASURE:
- net_dim_sample(end_sample.event_ctr, end_sample.pkt_ctr, end_sample.byte_ctr,
- &dim->start_sample);
+ net_dim_update_sample(end_sample.event_ctr, end_sample.pkt_ctr,
+ end_sample.byte_ctr, &dim->start_sample);
dim->state = DIM_MEASURE_IN_PROGRESS;
break;
case DIM_APPLY_NEW_PROFILE:
--
2.21.0
^ permalink raw reply related
* [for-next 3/9] linux/dim: Rename externally exposed macros
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Saeed Mahameed
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
From: Tal Gilboa <talgi@mellanox.com>
Renamed macros in use by external drivers.
Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/broadcom/bcmsysport.c | 4 ++--
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 2 +-
drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c | 2 +-
drivers/net/ethernet/broadcom/genet/bcmgenet.c | 4 ++--
.../net/ethernet/mellanox/mlx5/core/en_dim.c | 2 +-
.../net/ethernet/mellanox/mlx5/core/en_main.c | 10 +++++-----
include/linux/dim.h | 12 ++++++------
include/linux/net_dim.h | 18 +++++++++---------
8 files changed, 27 insertions(+), 27 deletions(-)
diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c
index c623896e3ccb..b5e2f9d2cb71 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.c
+++ b/drivers/net/ethernet/broadcom/bcmsysport.c
@@ -1099,7 +1099,7 @@ static void bcm_sysport_dim_work(struct work_struct *work)
net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
bcm_sysport_set_rx_coalesce(priv, cur_profile.usec, cur_profile.pkts);
- dim->state = NET_DIM_START_MEASURE;
+ dim->state = DIM_START_MEASURE;
}
/* RX and misc interrupt routine */
@@ -1440,7 +1440,7 @@ static void bcm_sysport_init_dim(struct bcm_sysport_priv *priv,
struct bcm_sysport_net_dim *dim = &priv->dim;
INIT_WORK(&dim->dim.work, cb);
- dim->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
+ dim->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
dim->event_ctr = 0;
dim->packets = 0;
dim->bytes = 0;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 8314c00d7537..49de873043c0 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -7810,7 +7810,7 @@ static void bnxt_enable_napi(struct bnxt *bp)
if (bp->bnapi[i]->rx_ring) {
INIT_WORK(&cpr->dim.work, bnxt_dim_work);
- cpr->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
+ cpr->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
}
napi_enable(&bp->bnapi[i]->napi);
}
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c
index afa97c8bb081..16a4588709d1 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c
@@ -28,5 +28,5 @@ void bnxt_dim_work(struct work_struct *work)
cpr->rx_ring_coal.coal_bufs = cur_moder.pkts;
bnxt_hwrm_set_ring_coal(bnapi->bp, bnapi);
- dim->state = NET_DIM_START_MEASURE;
+ dim->state = DIM_START_MEASURE;
}
diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.c b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
index 374b9ff05c88..5286a46ecfb0 100644
--- a/drivers/net/ethernet/broadcom/genet/bcmgenet.c
+++ b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
@@ -1928,7 +1928,7 @@ static void bcmgenet_dim_work(struct work_struct *work)
net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
bcmgenet_set_rx_coalesce(ring, cur_profile.usec, cur_profile.pkts);
- dim->state = NET_DIM_START_MEASURE;
+ dim->state = DIM_START_MEASURE;
}
/* Assign skb to RX DMA descriptor. */
@@ -2085,7 +2085,7 @@ static void bcmgenet_init_dim(struct bcmgenet_rx_ring *ring,
struct bcmgenet_net_dim *dim = &ring->dim;
INIT_WORK(&dim->dim.work, cb);
- dim->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
+ dim->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
dim->event_ctr = 0;
dim->packets = 0;
dim->bytes = 0;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c b/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
index d67adf70a97b..a80303add7c0 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
@@ -38,7 +38,7 @@ mlx5e_complete_dim_work(struct net_dim *dim, struct net_dim_cq_moder moder,
struct mlx5_core_dev *mdev, struct mlx5_core_cq *mcq)
{
mlx5_core_modify_cq_moderation(mdev, mcq, moder.usec, moder.pkts);
- dim->state = NET_DIM_START_MEASURE;
+ dim->state = DIM_START_MEASURE;
}
void mlx5e_rx_dim_work(struct work_struct *work)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 457cc39423f2..5b89e992e482 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -584,11 +584,11 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
switch (params->rx_cq_moderation.cq_period_mode) {
case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
- rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
+ rq->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_CQE;
break;
case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
default:
- rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
+ rq->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
}
rq->page_cache.head = 0;
@@ -2151,7 +2151,7 @@ static void mlx5e_build_ico_cq_param(struct mlx5e_priv *priv,
mlx5e_build_common_cq_param(priv, param);
- param->cq_period_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
+ param->cq_period_mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
}
static void mlx5e_build_icosq_param(struct mlx5e_priv *priv,
@@ -4440,8 +4440,8 @@ static struct net_dim_cq_moder mlx5e_get_def_rx_moderation(u8 cq_period_mode)
static u8 mlx5_to_net_dim_cq_period_mode(u8 cq_period_mode)
{
return cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE ?
- NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE :
- NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
+ DIM_CQ_PERIOD_MODE_START_FROM_CQE :
+ DIM_CQ_PERIOD_MODE_START_FROM_EQE;
}
void mlx5e_set_tx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
diff --git a/include/linux/dim.h b/include/linux/dim.h
index 6ee991681d62..989dbbdf9d45 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -47,15 +47,15 @@ struct net_dim { /* Dynamic Interrupt Moderation */
};
enum {
- NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE = 0x0,
- NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE = 0x1,
- NET_DIM_CQ_PERIOD_NUM_MODES
+ DIM_CQ_PERIOD_MODE_START_FROM_EQE = 0x0,
+ DIM_CQ_PERIOD_MODE_START_FROM_CQE = 0x1,
+ DIM_CQ_PERIOD_NUM_MODES
};
enum {
- NET_DIM_START_MEASURE,
- NET_DIM_MEASURE_IN_PROGRESS,
- NET_DIM_APPLY_NEW_PROFILE,
+ DIM_START_MEASURE,
+ DIM_MEASURE_IN_PROGRESS,
+ DIM_APPLY_NEW_PROFILE,
};
enum {
diff --git a/include/linux/net_dim.h b/include/linux/net_dim.h
index f89fa4fdfb46..e0c97f824dd0 100644
--- a/include/linux/net_dim.h
+++ b/include/linux/net_dim.h
@@ -78,13 +78,13 @@
}
static const struct net_dim_cq_moder
-rx_profile[NET_DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
+rx_profile[DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
NET_DIM_RX_EQE_PROFILES,
NET_DIM_RX_CQE_PROFILES,
};
static const struct net_dim_cq_moder
-tx_profile[NET_DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
+tx_profile[DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
NET_DIM_TX_EQE_PROFILES,
NET_DIM_TX_CQE_PROFILES,
};
@@ -101,7 +101,7 @@ net_dim_get_rx_moderation(u8 cq_period_mode, int ix)
static inline struct net_dim_cq_moder
net_dim_get_def_rx_moderation(u8 cq_period_mode)
{
- u8 profile_ix = cq_period_mode == NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
+ u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
return net_dim_get_rx_moderation(cq_period_mode, profile_ix);
@@ -119,7 +119,7 @@ net_dim_get_tx_moderation(u8 cq_period_mode, int ix)
static inline struct net_dim_cq_moder
net_dim_get_def_tx_moderation(u8 cq_period_mode)
{
- u8 profile_ix = cq_period_mode == NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
+ u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
return net_dim_get_tx_moderation(cq_period_mode, profile_ix);
@@ -247,7 +247,7 @@ static inline void net_dim(struct net_dim *dim,
u16 nevents;
switch (dim->state) {
- case NET_DIM_MEASURE_IN_PROGRESS:
+ case DIM_MEASURE_IN_PROGRESS:
nevents = BIT_GAP(BITS_PER_TYPE(u16),
end_sample.event_ctr,
dim->start_sample.event_ctr);
@@ -255,17 +255,17 @@ static inline void net_dim(struct net_dim *dim,
break;
dim_calc_stats(&dim->start_sample, &end_sample, &curr_stats);
if (net_dim_decision(&curr_stats, dim)) {
- dim->state = NET_DIM_APPLY_NEW_PROFILE;
+ dim->state = DIM_APPLY_NEW_PROFILE;
schedule_work(&dim->work);
break;
}
/* fall through */
- case NET_DIM_START_MEASURE:
+ case DIM_START_MEASURE:
net_dim_sample(end_sample.event_ctr, end_sample.pkt_ctr, end_sample.byte_ctr,
&dim->start_sample);
- dim->state = NET_DIM_MEASURE_IN_PROGRESS;
+ dim->state = DIM_MEASURE_IN_PROGRESS;
break;
- case NET_DIM_APPLY_NEW_PROFILE:
+ case DIM_APPLY_NEW_PROFILE:
break;
}
}
--
2.21.0
^ permalink raw reply related
* [for-next 2/9] linux/dim: Remove "net" prefix from internal DIM members
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Saeed Mahameed
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
From: Tal Gilboa <talgi@mellanox.com>
Only renaming functions and structs which aren't used by an external code.
Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
include/linux/dim.h | 86 ++++++++++++++++++++--------------------
include/linux/net_dim.h | 87 ++++++++++++++++++++---------------------
2 files changed, 86 insertions(+), 87 deletions(-)
diff --git a/include/linux/dim.h b/include/linux/dim.h
index 67d7ca40f3dd..6ee991681d62 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -6,7 +6,7 @@
#include <linux/module.h>
-#define NET_DIM_NEVENTS 64
+#define DIM_NEVENTS 64
/* more than 10% difference */
#define IS_SIGNIFICANT_DIFF(val, ref) \
@@ -27,7 +27,7 @@ struct net_dim_sample {
u16 event_ctr;
};
-struct net_dim_stats {
+struct dim_stats {
int ppms; /* packets per msec */
int bpms; /* bytes per msec */
int epms; /* events per msec */
@@ -35,7 +35,7 @@ struct net_dim_stats {
struct net_dim { /* Dynamic Interrupt Moderation */
u8 state;
- struct net_dim_stats prev_stats;
+ struct dim_stats prev_stats;
struct net_dim_sample start_sample;
struct work_struct work;
u8 profile_ix;
@@ -59,67 +59,67 @@ enum {
};
enum {
- NET_DIM_PARKING_ON_TOP,
- NET_DIM_PARKING_TIRED,
- NET_DIM_GOING_RIGHT,
- NET_DIM_GOING_LEFT,
+ DIM_PARKING_ON_TOP,
+ DIM_PARKING_TIRED,
+ DIM_GOING_RIGHT,
+ DIM_GOING_LEFT,
};
enum {
- NET_DIM_STATS_WORSE,
- NET_DIM_STATS_SAME,
- NET_DIM_STATS_BETTER,
+ DIM_STATS_WORSE,
+ DIM_STATS_SAME,
+ DIM_STATS_BETTER,
};
enum {
- NET_DIM_STEPPED,
- NET_DIM_TOO_TIRED,
- NET_DIM_ON_EDGE,
+ DIM_STEPPED,
+ DIM_TOO_TIRED,
+ DIM_ON_EDGE,
};
-static inline bool net_dim_on_top(struct net_dim *net_dim)
+static inline bool dim_on_top(struct net_dim *dim)
{
- switch (net_dim->tune_state) {
- case NET_DIM_PARKING_ON_TOP:
- case NET_DIM_PARKING_TIRED:
+ switch (dim->tune_state) {
+ case DIM_PARKING_ON_TOP:
+ case DIM_PARKING_TIRED:
return true;
- case NET_DIM_GOING_RIGHT:
- return (net_dim->steps_left > 1) && (net_dim->steps_right == 1);
- default: /* NET_DIM_GOING_LEFT */
- return (net_dim->steps_right > 1) && (net_dim->steps_left == 1);
+ case DIM_GOING_RIGHT:
+ return (dim->steps_left > 1) && (dim->steps_right == 1);
+ default: /* DIM_GOING_LEFT */
+ return (dim->steps_right > 1) && (dim->steps_left == 1);
}
}
-static inline void net_dim_turn(struct net_dim *net_dim)
+static inline void dim_turn(struct net_dim *dim)
{
- switch (net_dim->tune_state) {
- case NET_DIM_PARKING_ON_TOP:
- case NET_DIM_PARKING_TIRED:
+ switch (dim->tune_state) {
+ case DIM_PARKING_ON_TOP:
+ case DIM_PARKING_TIRED:
break;
- case NET_DIM_GOING_RIGHT:
- net_dim->tune_state = NET_DIM_GOING_LEFT;
- net_dim->steps_left = 0;
+ case DIM_GOING_RIGHT:
+ dim->tune_state = DIM_GOING_LEFT;
+ dim->steps_left = 0;
break;
- case NET_DIM_GOING_LEFT:
- net_dim->tune_state = NET_DIM_GOING_RIGHT;
- net_dim->steps_right = 0;
+ case DIM_GOING_LEFT:
+ dim->tune_state = DIM_GOING_RIGHT;
+ dim->steps_right = 0;
break;
}
}
-static inline void net_dim_park_on_top(struct net_dim *net_dim)
+static inline void dim_park_on_top(struct net_dim *dim)
{
- net_dim->steps_right = 0;
- net_dim->steps_left = 0;
- net_dim->tired = 0;
- net_dim->tune_state = NET_DIM_PARKING_ON_TOP;
+ dim->steps_right = 0;
+ dim->steps_left = 0;
+ dim->tired = 0;
+ dim->tune_state = DIM_PARKING_ON_TOP;
}
-static inline void net_dim_park_tired(struct net_dim *net_dim)
+static inline void dim_park_tired(struct net_dim *dim)
{
- net_dim->steps_right = 0;
- net_dim->steps_left = 0;
- net_dim->tune_state = NET_DIM_PARKING_TIRED;
+ dim->steps_right = 0;
+ dim->steps_left = 0;
+ dim->tune_state = DIM_PARKING_TIRED;
}
static inline void
@@ -132,8 +132,8 @@ net_dim_sample(u16 event_ctr, u64 packets, u64 bytes, struct net_dim_sample *s)
}
static inline void
-net_dim_calc_stats(struct net_dim_sample *start, struct net_dim_sample *end,
- struct net_dim_stats *curr_stats)
+dim_calc_stats(struct net_dim_sample *start, struct net_dim_sample *end,
+ struct dim_stats *curr_stats)
{
/* u32 holds up to 71 minutes, should be enough */
u32 delta_us = ktime_us_delta(end->time, start->time);
@@ -146,7 +146,7 @@ net_dim_calc_stats(struct net_dim_sample *start, struct net_dim_sample *end,
curr_stats->ppms = DIV_ROUND_UP(npkts * USEC_PER_MSEC, delta_us);
curr_stats->bpms = DIV_ROUND_UP(nbytes * USEC_PER_MSEC, delta_us);
- curr_stats->epms = DIV_ROUND_UP(NET_DIM_NEVENTS * USEC_PER_MSEC,
+ curr_stats->epms = DIV_ROUND_UP(DIM_NEVENTS * USEC_PER_MSEC,
delta_us);
}
diff --git a/include/linux/net_dim.h b/include/linux/net_dim.h
index 373cda74b167..f89fa4fdfb46 100644
--- a/include/linux/net_dim.h
+++ b/include/linux/net_dim.h
@@ -128,67 +128,67 @@ net_dim_get_def_tx_moderation(u8 cq_period_mode)
static inline int net_dim_step(struct net_dim *dim)
{
if (dim->tired == (NET_DIM_PARAMS_NUM_PROFILES * 2))
- return NET_DIM_TOO_TIRED;
+ return DIM_TOO_TIRED;
switch (dim->tune_state) {
- case NET_DIM_PARKING_ON_TOP:
- case NET_DIM_PARKING_TIRED:
+ case DIM_PARKING_ON_TOP:
+ case DIM_PARKING_TIRED:
break;
- case NET_DIM_GOING_RIGHT:
+ case DIM_GOING_RIGHT:
if (dim->profile_ix == (NET_DIM_PARAMS_NUM_PROFILES - 1))
- return NET_DIM_ON_EDGE;
+ return DIM_ON_EDGE;
dim->profile_ix++;
dim->steps_right++;
break;
- case NET_DIM_GOING_LEFT:
+ case DIM_GOING_LEFT:
if (dim->profile_ix == 0)
- return NET_DIM_ON_EDGE;
+ return DIM_ON_EDGE;
dim->profile_ix--;
dim->steps_left++;
break;
}
dim->tired++;
- return NET_DIM_STEPPED;
+ return DIM_STEPPED;
}
static inline void net_dim_exit_parking(struct net_dim *dim)
{
- dim->tune_state = dim->profile_ix ? NET_DIM_GOING_LEFT :
- NET_DIM_GOING_RIGHT;
+ dim->tune_state = dim->profile_ix ? DIM_GOING_LEFT :
+ DIM_GOING_RIGHT;
net_dim_step(dim);
}
-static inline int net_dim_stats_compare(struct net_dim_stats *curr,
- struct net_dim_stats *prev)
+static inline int net_dim_stats_compare(struct dim_stats *curr,
+ struct dim_stats *prev)
{
if (!prev->bpms)
- return curr->bpms ? NET_DIM_STATS_BETTER :
- NET_DIM_STATS_SAME;
+ return curr->bpms ? DIM_STATS_BETTER :
+ DIM_STATS_SAME;
if (IS_SIGNIFICANT_DIFF(curr->bpms, prev->bpms))
- return (curr->bpms > prev->bpms) ? NET_DIM_STATS_BETTER :
- NET_DIM_STATS_WORSE;
+ return (curr->bpms > prev->bpms) ? DIM_STATS_BETTER :
+ DIM_STATS_WORSE;
if (!prev->ppms)
- return curr->ppms ? NET_DIM_STATS_BETTER :
- NET_DIM_STATS_SAME;
+ return curr->ppms ? DIM_STATS_BETTER :
+ DIM_STATS_SAME;
if (IS_SIGNIFICANT_DIFF(curr->ppms, prev->ppms))
- return (curr->ppms > prev->ppms) ? NET_DIM_STATS_BETTER :
- NET_DIM_STATS_WORSE;
+ return (curr->ppms > prev->ppms) ? DIM_STATS_BETTER :
+ DIM_STATS_WORSE;
if (!prev->epms)
- return NET_DIM_STATS_SAME;
+ return DIM_STATS_SAME;
if (IS_SIGNIFICANT_DIFF(curr->epms, prev->epms))
- return (curr->epms < prev->epms) ? NET_DIM_STATS_BETTER :
- NET_DIM_STATS_WORSE;
+ return (curr->epms < prev->epms) ? DIM_STATS_BETTER :
+ DIM_STATS_WORSE;
- return NET_DIM_STATS_SAME;
+ return DIM_STATS_SAME;
}
-static inline bool net_dim_decision(struct net_dim_stats *curr_stats,
+static inline bool net_dim_decision(struct dim_stats *curr_stats,
struct net_dim *dim)
{
int prev_state = dim->tune_state;
@@ -197,44 +197,44 @@ static inline bool net_dim_decision(struct net_dim_stats *curr_stats,
int step_res;
switch (dim->tune_state) {
- case NET_DIM_PARKING_ON_TOP:
+ case DIM_PARKING_ON_TOP:
stats_res = net_dim_stats_compare(curr_stats, &dim->prev_stats);
- if (stats_res != NET_DIM_STATS_SAME)
+ if (stats_res != DIM_STATS_SAME)
net_dim_exit_parking(dim);
break;
- case NET_DIM_PARKING_TIRED:
+ case DIM_PARKING_TIRED:
dim->tired--;
if (!dim->tired)
net_dim_exit_parking(dim);
break;
- case NET_DIM_GOING_RIGHT:
- case NET_DIM_GOING_LEFT:
+ case DIM_GOING_RIGHT:
+ case DIM_GOING_LEFT:
stats_res = net_dim_stats_compare(curr_stats, &dim->prev_stats);
- if (stats_res != NET_DIM_STATS_BETTER)
- net_dim_turn(dim);
+ if (stats_res != DIM_STATS_BETTER)
+ dim_turn(dim);
- if (net_dim_on_top(dim)) {
- net_dim_park_on_top(dim);
+ if (dim_on_top(dim)) {
+ dim_park_on_top(dim);
break;
}
step_res = net_dim_step(dim);
switch (step_res) {
- case NET_DIM_ON_EDGE:
- net_dim_park_on_top(dim);
+ case DIM_ON_EDGE:
+ dim_park_on_top(dim);
break;
- case NET_DIM_TOO_TIRED:
- net_dim_park_tired(dim);
+ case DIM_TOO_TIRED:
+ dim_park_tired(dim);
break;
}
break;
}
- if ((prev_state != NET_DIM_PARKING_ON_TOP) ||
- (dim->tune_state != NET_DIM_PARKING_ON_TOP))
+ if (prev_state != DIM_PARKING_ON_TOP ||
+ dim->tune_state != DIM_PARKING_ON_TOP)
dim->prev_stats = *curr_stats;
return dim->profile_ix != prev_ix;
@@ -243,7 +243,7 @@ static inline bool net_dim_decision(struct net_dim_stats *curr_stats,
static inline void net_dim(struct net_dim *dim,
struct net_dim_sample end_sample)
{
- struct net_dim_stats curr_stats;
+ struct dim_stats curr_stats;
u16 nevents;
switch (dim->state) {
@@ -251,10 +251,9 @@ static inline void net_dim(struct net_dim *dim,
nevents = BIT_GAP(BITS_PER_TYPE(u16),
end_sample.event_ctr,
dim->start_sample.event_ctr);
- if (nevents < NET_DIM_NEVENTS)
+ if (nevents < DIM_NEVENTS)
break;
- net_dim_calc_stats(&dim->start_sample, &end_sample,
- &curr_stats);
+ dim_calc_stats(&dim->start_sample, &end_sample, &curr_stats);
if (net_dim_decision(&curr_stats, dim)) {
dim->state = NET_DIM_APPLY_NEW_PROFILE;
schedule_work(&dim->work);
--
2.21.0
^ permalink raw reply related
* [for-next 1/9] linux/dim: Move logic to dim.h
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Saeed Mahameed
In-Reply-To: <20190605232348.6452-1-saeedm@mellanox.com>
From: Tal Gilboa <talgi@mellanox.com>
In preparation for supporting more implementations of the DIM
algorithm, I'm moving what would become common logic to a common
library. Downstream DIM implementations will use the common lib
for their implementation.
Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
MAINTAINERS | 1 +
include/linux/dim.h | 153 ++++++++++++++++++++++++++++++++++++++++
include/linux/net_dim.h | 148 +-------------------------------------
3 files changed, 156 insertions(+), 146 deletions(-)
create mode 100644 include/linux/dim.h
diff --git a/MAINTAINERS b/MAINTAINERS
index 429c6c624861..5d4b852d9d39 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -5589,6 +5589,7 @@ DYNAMIC INTERRUPT MODERATION
M: Tal Gilboa <talgi@mellanox.com>
S: Maintained
F: include/linux/net_dim.h
+F: include/linux/dim.h
DZ DECSTATION DZ11 SERIAL DRIVER
M: "Maciej W. Rozycki" <macro@linux-mips.org>
diff --git a/include/linux/dim.h b/include/linux/dim.h
new file mode 100644
index 000000000000..67d7ca40f3dd
--- /dev/null
+++ b/include/linux/dim.h
@@ -0,0 +1,153 @@
+/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
+/* Copyright (c) 2019 Mellanox Technologies. */
+
+#ifndef DIM_H
+#define DIM_H
+
+#include <linux/module.h>
+
+#define NET_DIM_NEVENTS 64
+
+/* more than 10% difference */
+#define IS_SIGNIFICANT_DIFF(val, ref) \
+ (((100UL * abs((val) - (ref))) / (ref)) > 10)
+#define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) \
+& (BIT_ULL(bits) - 1))
+
+struct net_dim_cq_moder {
+ u16 usec;
+ u16 pkts;
+ u8 cq_period_mode;
+};
+
+struct net_dim_sample {
+ ktime_t time;
+ u32 pkt_ctr;
+ u32 byte_ctr;
+ u16 event_ctr;
+};
+
+struct net_dim_stats {
+ int ppms; /* packets per msec */
+ int bpms; /* bytes per msec */
+ int epms; /* events per msec */
+};
+
+struct net_dim { /* Dynamic Interrupt Moderation */
+ u8 state;
+ struct net_dim_stats prev_stats;
+ struct net_dim_sample start_sample;
+ struct work_struct work;
+ u8 profile_ix;
+ u8 mode;
+ u8 tune_state;
+ u8 steps_right;
+ u8 steps_left;
+ u8 tired;
+};
+
+enum {
+ NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE = 0x0,
+ NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE = 0x1,
+ NET_DIM_CQ_PERIOD_NUM_MODES
+};
+
+enum {
+ NET_DIM_START_MEASURE,
+ NET_DIM_MEASURE_IN_PROGRESS,
+ NET_DIM_APPLY_NEW_PROFILE,
+};
+
+enum {
+ NET_DIM_PARKING_ON_TOP,
+ NET_DIM_PARKING_TIRED,
+ NET_DIM_GOING_RIGHT,
+ NET_DIM_GOING_LEFT,
+};
+
+enum {
+ NET_DIM_STATS_WORSE,
+ NET_DIM_STATS_SAME,
+ NET_DIM_STATS_BETTER,
+};
+
+enum {
+ NET_DIM_STEPPED,
+ NET_DIM_TOO_TIRED,
+ NET_DIM_ON_EDGE,
+};
+
+static inline bool net_dim_on_top(struct net_dim *net_dim)
+{
+ switch (net_dim->tune_state) {
+ case NET_DIM_PARKING_ON_TOP:
+ case NET_DIM_PARKING_TIRED:
+ return true;
+ case NET_DIM_GOING_RIGHT:
+ return (net_dim->steps_left > 1) && (net_dim->steps_right == 1);
+ default: /* NET_DIM_GOING_LEFT */
+ return (net_dim->steps_right > 1) && (net_dim->steps_left == 1);
+ }
+}
+
+static inline void net_dim_turn(struct net_dim *net_dim)
+{
+ switch (net_dim->tune_state) {
+ case NET_DIM_PARKING_ON_TOP:
+ case NET_DIM_PARKING_TIRED:
+ break;
+ case NET_DIM_GOING_RIGHT:
+ net_dim->tune_state = NET_DIM_GOING_LEFT;
+ net_dim->steps_left = 0;
+ break;
+ case NET_DIM_GOING_LEFT:
+ net_dim->tune_state = NET_DIM_GOING_RIGHT;
+ net_dim->steps_right = 0;
+ break;
+ }
+}
+
+static inline void net_dim_park_on_top(struct net_dim *net_dim)
+{
+ net_dim->steps_right = 0;
+ net_dim->steps_left = 0;
+ net_dim->tired = 0;
+ net_dim->tune_state = NET_DIM_PARKING_ON_TOP;
+}
+
+static inline void net_dim_park_tired(struct net_dim *net_dim)
+{
+ net_dim->steps_right = 0;
+ net_dim->steps_left = 0;
+ net_dim->tune_state = NET_DIM_PARKING_TIRED;
+}
+
+static inline void
+net_dim_sample(u16 event_ctr, u64 packets, u64 bytes, struct net_dim_sample *s)
+{
+ s->time = ktime_get();
+ s->pkt_ctr = packets;
+ s->byte_ctr = bytes;
+ s->event_ctr = event_ctr;
+}
+
+static inline void
+net_dim_calc_stats(struct net_dim_sample *start, struct net_dim_sample *end,
+ struct net_dim_stats *curr_stats)
+{
+ /* u32 holds up to 71 minutes, should be enough */
+ u32 delta_us = ktime_us_delta(end->time, start->time);
+ u32 npkts = BIT_GAP(BITS_PER_TYPE(u32), end->pkt_ctr, start->pkt_ctr);
+ u32 nbytes = BIT_GAP(BITS_PER_TYPE(u32), end->byte_ctr,
+ start->byte_ctr);
+
+ if (!delta_us)
+ return;
+
+ curr_stats->ppms = DIV_ROUND_UP(npkts * USEC_PER_MSEC, delta_us);
+ curr_stats->bpms = DIV_ROUND_UP(nbytes * USEC_PER_MSEC, delta_us);
+ curr_stats->epms = DIV_ROUND_UP(NET_DIM_NEVENTS * USEC_PER_MSEC,
+ delta_us);
+}
+
+#endif /* DIM_H */
diff --git a/include/linux/net_dim.h b/include/linux/net_dim.h
index fd458389f7d1..373cda74b167 100644
--- a/include/linux/net_dim.h
+++ b/include/linux/net_dim.h
@@ -35,73 +35,10 @@
#define NET_DIM_H
#include <linux/module.h>
-
-struct net_dim_cq_moder {
- u16 usec;
- u16 pkts;
- u8 cq_period_mode;
-};
-
-struct net_dim_sample {
- ktime_t time;
- u32 pkt_ctr;
- u32 byte_ctr;
- u16 event_ctr;
-};
-
-struct net_dim_stats {
- int ppms; /* packets per msec */
- int bpms; /* bytes per msec */
- int epms; /* events per msec */
-};
-
-struct net_dim { /* Adaptive Moderation */
- u8 state;
- struct net_dim_stats prev_stats;
- struct net_dim_sample start_sample;
- struct work_struct work;
- u8 profile_ix;
- u8 mode;
- u8 tune_state;
- u8 steps_right;
- u8 steps_left;
- u8 tired;
-};
-
-enum {
- NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE = 0x0,
- NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE = 0x1,
- NET_DIM_CQ_PERIOD_NUM_MODES
-};
-
-/* Adaptive moderation logic */
-enum {
- NET_DIM_START_MEASURE,
- NET_DIM_MEASURE_IN_PROGRESS,
- NET_DIM_APPLY_NEW_PROFILE,
-};
-
-enum {
- NET_DIM_PARKING_ON_TOP,
- NET_DIM_PARKING_TIRED,
- NET_DIM_GOING_RIGHT,
- NET_DIM_GOING_LEFT,
-};
-
-enum {
- NET_DIM_STATS_WORSE,
- NET_DIM_STATS_SAME,
- NET_DIM_STATS_BETTER,
-};
-
-enum {
- NET_DIM_STEPPED,
- NET_DIM_TOO_TIRED,
- NET_DIM_ON_EDGE,
-};
+#include <linux/dim.h>
#define NET_DIM_PARAMS_NUM_PROFILES 5
-/* Adaptive moderation profiles */
+/* Netdev dynamic interrupt moderation profiles */
#define NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE 256
#define NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE 128
#define NET_DIM_DEF_PROFILE_CQE 1
@@ -188,36 +125,6 @@ net_dim_get_def_tx_moderation(u8 cq_period_mode)
return net_dim_get_tx_moderation(cq_period_mode, profile_ix);
}
-static inline bool net_dim_on_top(struct net_dim *dim)
-{
- switch (dim->tune_state) {
- case NET_DIM_PARKING_ON_TOP:
- case NET_DIM_PARKING_TIRED:
- return true;
- case NET_DIM_GOING_RIGHT:
- return (dim->steps_left > 1) && (dim->steps_right == 1);
- default: /* NET_DIM_GOING_LEFT */
- return (dim->steps_right > 1) && (dim->steps_left == 1);
- }
-}
-
-static inline void net_dim_turn(struct net_dim *dim)
-{
- switch (dim->tune_state) {
- case NET_DIM_PARKING_ON_TOP:
- case NET_DIM_PARKING_TIRED:
- break;
- case NET_DIM_GOING_RIGHT:
- dim->tune_state = NET_DIM_GOING_LEFT;
- dim->steps_left = 0;
- break;
- case NET_DIM_GOING_LEFT:
- dim->tune_state = NET_DIM_GOING_RIGHT;
- dim->steps_right = 0;
- break;
- }
-}
-
static inline int net_dim_step(struct net_dim *dim)
{
if (dim->tired == (NET_DIM_PARAMS_NUM_PROFILES * 2))
@@ -245,21 +152,6 @@ static inline int net_dim_step(struct net_dim *dim)
return NET_DIM_STEPPED;
}
-static inline void net_dim_park_on_top(struct net_dim *dim)
-{
- dim->steps_right = 0;
- dim->steps_left = 0;
- dim->tired = 0;
- dim->tune_state = NET_DIM_PARKING_ON_TOP;
-}
-
-static inline void net_dim_park_tired(struct net_dim *dim)
-{
- dim->steps_right = 0;
- dim->steps_left = 0;
- dim->tune_state = NET_DIM_PARKING_TIRED;
-}
-
static inline void net_dim_exit_parking(struct net_dim *dim)
{
dim->tune_state = dim->profile_ix ? NET_DIM_GOING_LEFT :
@@ -267,9 +159,6 @@ static inline void net_dim_exit_parking(struct net_dim *dim)
net_dim_step(dim);
}
-#define IS_SIGNIFICANT_DIFF(val, ref) \
- (((100UL * abs((val) - (ref))) / (ref)) > 10) /* more than 10% difference */
-
static inline int net_dim_stats_compare(struct net_dim_stats *curr,
struct net_dim_stats *prev)
{
@@ -351,39 +240,6 @@ static inline bool net_dim_decision(struct net_dim_stats *curr_stats,
return dim->profile_ix != prev_ix;
}
-static inline void net_dim_sample(u16 event_ctr,
- u64 packets,
- u64 bytes,
- struct net_dim_sample *s)
-{
- s->time = ktime_get();
- s->pkt_ctr = packets;
- s->byte_ctr = bytes;
- s->event_ctr = event_ctr;
-}
-
-#define NET_DIM_NEVENTS 64
-#define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) & (BIT_ULL(bits) - 1))
-
-static inline void net_dim_calc_stats(struct net_dim_sample *start,
- struct net_dim_sample *end,
- struct net_dim_stats *curr_stats)
-{
- /* u32 holds up to 71 minutes, should be enough */
- u32 delta_us = ktime_us_delta(end->time, start->time);
- u32 npkts = BIT_GAP(BITS_PER_TYPE(u32), end->pkt_ctr, start->pkt_ctr);
- u32 nbytes = BIT_GAP(BITS_PER_TYPE(u32), end->byte_ctr,
- start->byte_ctr);
-
- if (!delta_us)
- return;
-
- curr_stats->ppms = DIV_ROUND_UP(npkts * USEC_PER_MSEC, delta_us);
- curr_stats->bpms = DIV_ROUND_UP(nbytes * USEC_PER_MSEC, delta_us);
- curr_stats->epms = DIV_ROUND_UP(NET_DIM_NEVENTS * USEC_PER_MSEC,
- delta_us);
-}
-
static inline void net_dim(struct net_dim *dim,
struct net_dim_sample end_sample)
{
--
2.21.0
^ permalink raw reply related
* [pull request][for-next 0/9] Generic DIM lib for netdev and RDMA
From: Saeed Mahameed @ 2019-06-05 23:24 UTC (permalink / raw)
To: David S. Miller, Jason Gunthorpe, Doug Ledford
Cc: Michael Chan, Andy Gospodarek, Tal Gilboa,
linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
Saeed Mahameed
Hi Dave, Doug & Jason
This series improves DIM - Dynamically-tuned Interrupt
Moderation- to be generic for netdev and RDMA use-cases.
From Tal and Yamin:
The first 7 patches provide the necessary refactoring to current net_dim
library which affect some net drivers who are using the API.
The last 2 patches provide the RDMA implementation for DIM.
For more information please see tag log below.
Once we are all happy with the series, please pull to net-next and
rdma-next trees.
Thanks,
Saeed.
---
The following changes since commit cd6c84d8f0cdc911df435bb075ba22ce3c605b07:
Linux 5.2-rc2 (2019-05-26 16:49:19 -0700)
are available in the Git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux.git tags/dim-updates-2019-06-05
for you to fetch changes up to 1ec9974e75e7a58bff1ab17c4fcda17b180ed3bb:
RDMA/core: Provide RDMA DIM support for ULPs (2019-06-05 16:09:02 -0700)
----------------------------------------------------------------
dim-updates-2019-06-05
From: Tal Gilboa
Implement net DIM over a generic DIM library
net_dim.h lib exposes an implementation of the DIM algorithm for
dynamically-tuned interrupt moderation for networking interfaces.
We want a similar functionality for other protocols, which might need to
optimize interrupts differently. Main motivation here is DIM for NVMf
storage protocol.
Current DIM implementation prioritizes reducing interrupt overhead over
latency. Also, in order to reduce DIM's own overhead, the algorithm might
take some time to identify it needs to change profiles. While this is
acceptable for networking, it might not work well on other scenarios.
Here I propose a new structure to DIM. The idea is to allow a slightly
modified functionality without the risk of breaking Net DIM behavior for
netdev. I verified there are no degradations in current DIM behavior with
the modified solution.
Solution:
- Common logic is declared in include/linux/dim.h and implemented in
lib/dim/dim.c
- Net DIM (existing) logic is declared in include/linux/net_dim.h and
implemented in lib/dim/net_dim.c, which uses the common logic from dim.h
- Any new DIM logic will be declared in "/include/linux/new_dim.h" and
implemented in "lib/dim/new_dim.c".
- This new implementation will expose modified versions of profiles,
dim_step() and dim_decision().
Pros for this solution are:
- Zero impact on existing net_dim implementation and usage
- Relatively more code reuse (compared to two separate solutions)
- Increased extensibility
----------------------------------------------------------------
Tal Gilboa (6):
linux/dim: Move logic to dim.h
linux/dim: Remove "net" prefix from internal DIM members
linux/dim: Rename externally exposed macros
linux/dim: Rename net_dim_sample() to net_dim_update_sample()
linux/dim: Rename externally used net_dim members
linux/dim: Move implementation to .c files
Yamin Friedman (3):
linux/dim: Add completions count to dim_sample
linux/dim: Implement rdma_dim
RDMA/core: Provide RDMA DIM support for ULPs
MAINTAINERS | 3 +
drivers/infiniband/core/cq.c | 78 ++++-
drivers/net/ethernet/broadcom/Kconfig | 1 +
drivers/net/ethernet/broadcom/bcmsysport.c | 20 +-
drivers/net/ethernet/broadcom/bcmsysport.h | 2 +-
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 12 +-
drivers/net/ethernet/broadcom/bnxt/bnxt.h | 2 +-
drivers/net/ethernet/broadcom/bnxt/bnxt_debugfs.c | 4 +-
drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c | 7 +-
drivers/net/ethernet/broadcom/genet/bcmgenet.c | 18 +-
drivers/net/ethernet/broadcom/genet/bcmgenet.h | 2 +-
drivers/net/ethernet/mellanox/mlx4/Kconfig | 1 +
drivers/net/ethernet/mellanox/mlx5/core/Kconfig | 2 +
drivers/net/ethernet/mellanox/mlx5/core/en.h | 8 +-
drivers/net/ethernet/mellanox/mlx5/core/en_dim.c | 12 +-
.../net/ethernet/mellanox/mlx5/core/en_ethtool.c | 4 +-
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 22 +-
drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c | 10 +-
include/linux/dim.h | 271 +++++++++++++++
include/linux/net_dim.h | 371 +++------------------
include/linux/rdma_dim.h | 28 ++
include/rdma/ib_verbs.h | 27 +-
lib/Kconfig | 8 +
lib/Makefile | 1 +
lib/dim/Makefile | 14 +
lib/dim/dim.c | 83 +++++
lib/dim/net_dim.c | 191 +++++++++++
lib/dim/rdma_dim.c | 162 +++++++++
28 files changed, 968 insertions(+), 396 deletions(-)
create mode 100644 include/linux/dim.h
create mode 100644 include/linux/rdma_dim.h
create mode 100644 lib/dim/Makefile
create mode 100644 lib/dim/dim.c
create mode 100644 lib/dim/net_dim.c
create mode 100644 lib/dim/rdma_dim.c
^ permalink raw reply
* Re: [PATCH 08/13] IB/iser: set virt_boundary_mask in the scsi host
From: Jason Gunthorpe @ 2019-06-05 20:22 UTC (permalink / raw)
To: Christoph Hellwig
Cc: Jens Axboe, Sebastian Ott, Sagi Grimberg, Max Gurtovoy,
Bart Van Assche, Ulf Hansson, Alan Stern, Oliver Neukum,
linux-block, linux-rdma, linux-mmc, linux-nvme, linux-scsi,
megaraidlinux.pdl, MPT-FusionLinux.pdl, linux-hyperv, linux-usb,
usb-storage, linux-kernel
In-Reply-To: <20190605190836.32354-9-hch@lst.de>
On Wed, Jun 05, 2019 at 09:08:31PM +0200, Christoph Hellwig wrote:
> This ensures all proper DMA layer handling is taken care of by the
> SCSI midlayer.
Maybe not entirely related to this series, but it looks like the SCSI
layer is changing the device global dma_set_max_seg_size() - at least
in RDMA the dma device is being shared between many users, so we
really don't want SCSI to make this value smaller.
Can we do something about this?
Wondering about other values too, and the interaction with the new
combining stuff in umem.c
Thanks,
Jason
^ permalink raw reply
* Re: properly communicate queue limits to the DMA layer
From: Christoph Hellwig @ 2019-06-05 19:24 UTC (permalink / raw)
To: Jens Axboe
Cc: Christoph Hellwig, Sebastian Ott, Sagi Grimberg, Max Gurtovoy,
Bart Van Assche, Ulf Hansson, Alan Stern, Oliver Neukum,
linux-block, linux-rdma, linux-mmc, linux-nvme, linux-scsi,
megaraidlinux.pdl, MPT-FusionLinux.pdl, linux-hyperv, linux-usb,
usb-storage, linux-kernel
In-Reply-To: <591cfa1e-fecb-7d00-c855-3b9eb8eb8a2a@kernel.dk>
On Wed, Jun 05, 2019 at 01:17:15PM -0600, Jens Axboe wrote:
> Since I'm heading out shortly and since I think this should make
> the next -rc, I'll tentatively queue this up.
The SCSI bits will need a bit more review, and possibly tweaking
fo megaraid and mpt3sas. But they are really independent of the
other patches, so maybe skip them for now and leave them for Martin
to deal with.
^ permalink raw reply
* Re: properly communicate queue limits to the DMA layer
From: Jens Axboe @ 2019-06-05 19:17 UTC (permalink / raw)
To: Christoph Hellwig
Cc: Sebastian Ott, Sagi Grimberg, Max Gurtovoy, Bart Van Assche,
Ulf Hansson, Alan Stern, Oliver Neukum, linux-block, linux-rdma,
linux-mmc, linux-nvme, linux-scsi, megaraidlinux.pdl,
MPT-FusionLinux.pdl, linux-hyperv, linux-usb, usb-storage,
linux-kernel
In-Reply-To: <20190605190836.32354-1-hch@lst.de>
On 6/5/19 1:08 PM, Christoph Hellwig wrote:
> Hi Jens,
>
> we've always had a bit of a problem communicating the block layer
> queue limits to the DMA layer, which needs to respect them when
> an IOMMU that could merge segments is used. Unfortunately most
> drivers don't get this right. Oddly enough we've been mostly
> getting away with it, although lately dma-debug has been catching
> a few of those issues.
>
> The segment merging fix for devices with PRP-like structures seems
> to have escalated this a bit. The first patch fixes the actual
> report from Sebastian, while the rest fix every drivers that appears
> to have the problem based on a code audit looking for drivers using
> blk_queue_max_segment_size, blk_queue_segment_boundary or
> blk_queue_virt_boundary and calling dma_map_sg eventually. Note
> that for SCSI drivers I've taken the blk_queue_virt_boundary setting
> to the SCSI core, similar to how we did it for the other two settings
> a while ago. This also deals with the fact that the DMA layer
> settings are on a per-device granularity, so the per-device settings
> in a few SCSI drivers can't actually work in an IOMMU environment.
>
> It would be nice to eventually pass these limits as arguments to
> dma_map_sg, but that is a far too big series for the 5.2 merge
> window.
Since I'm heading out shortly and since I think this should make
the next -rc, I'll tentatively queue this up.
--
Jens Axboe
^ permalink raw reply
* [PATCH 13/13] uas: set virt_boundary_mask in the scsi host
From: Christoph Hellwig @ 2019-06-05 19:08 UTC (permalink / raw)
To: Jens Axboe
Cc: Sebastian Ott, Sagi Grimberg, Max Gurtovoy, Bart Van Assche,
Ulf Hansson, Alan Stern, Oliver Neukum, linux-block, linux-rdma,
linux-mmc, linux-nvme, linux-scsi, megaraidlinux.pdl,
MPT-FusionLinux.pdl, linux-hyperv, linux-usb, usb-storage,
linux-kernel
In-Reply-To: <20190605190836.32354-1-hch@lst.de>
This ensures all proper DMA layer handling is taken care of by the
SCSI midlayer.
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
drivers/usb/storage/uas.c | 36 ++++++++++++++++--------------------
1 file changed, 16 insertions(+), 20 deletions(-)
diff --git a/drivers/usb/storage/uas.c b/drivers/usb/storage/uas.c
index 047c5922618f..d20919e7bbf4 100644
--- a/drivers/usb/storage/uas.c
+++ b/drivers/usb/storage/uas.c
@@ -789,29 +789,9 @@ static int uas_slave_alloc(struct scsi_device *sdev)
{
struct uas_dev_info *devinfo =
(struct uas_dev_info *)sdev->host->hostdata;
- int maxp;
sdev->hostdata = devinfo;
- /*
- * We have two requirements here. We must satisfy the requirements
- * of the physical HC and the demands of the protocol, as we
- * definitely want no additional memory allocation in this path
- * ruling out using bounce buffers.
- *
- * For a transmission on USB to continue we must never send
- * a package that is smaller than maxpacket. Hence the length of each
- * scatterlist element except the last must be divisible by the
- * Bulk maxpacket value.
- * If the HC does not ensure that through SG,
- * the upper layer must do that. We must assume nothing
- * about the capabilities off the HC, so we use the most
- * pessimistic requirement.
- */
-
- maxp = usb_maxpacket(devinfo->udev, devinfo->data_in_pipe, 0);
- blk_queue_virt_boundary(sdev->request_queue, maxp - 1);
-
/*
* The protocol has no requirements on alignment in the strict sense.
* Controllers may or may not have alignment restrictions.
@@ -1004,6 +984,22 @@ static int uas_probe(struct usb_interface *intf, const struct usb_device_id *id)
*/
shost->can_queue = devinfo->qdepth - 2;
+ /*
+ * We have two requirements here. We must satisfy the requirements of
+ * the physical HC and the demands of the protocol, as we definitely
+ * want no additional memory allocation in this path ruling out using
+ * bounce buffers.
+ *
+ * For a transmission on USB to continue we must never send a package
+ * that is smaller than maxpacket. Hence the length of each scatterlist
+ * element except the last must be divisible by the Bulk maxpacket
+ * value. If the HC does not ensure that through SG, the upper layer
+ * must do that. We must assume nothing about the capabilities off the
+ * HC, so we use the most pessimistic requirement.
+ */
+ shost->virt_boundary_mask =
+ usb_maxpacket(udev, devinfo->data_in_pipe, 0) - 1;
+
usb_set_intfdata(intf, shost);
result = scsi_add_host(shost, &intf->dev);
if (result)
--
2.20.1
^ permalink raw reply related
* [PATCH 12/13] usb-storage: set virt_boundary_mask in the scsi host
From: Christoph Hellwig @ 2019-06-05 19:08 UTC (permalink / raw)
To: Jens Axboe
Cc: Sebastian Ott, Sagi Grimberg, Max Gurtovoy, Bart Van Assche,
Ulf Hansson, Alan Stern, Oliver Neukum, linux-block, linux-rdma,
linux-mmc, linux-nvme, linux-scsi, megaraidlinux.pdl,
MPT-FusionLinux.pdl, linux-hyperv, linux-usb, usb-storage,
linux-kernel
In-Reply-To: <20190605190836.32354-1-hch@lst.de>
This ensures all proper DMA layer handling is taken care of by the
SCSI midlayer.
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
drivers/usb/storage/scsiglue.c | 10 ----------
drivers/usb/storage/usb.c | 10 ++++++++++
2 files changed, 10 insertions(+), 10 deletions(-)
diff --git a/drivers/usb/storage/scsiglue.c b/drivers/usb/storage/scsiglue.c
index 59190d88fa9f..02c3b66b3f78 100644
--- a/drivers/usb/storage/scsiglue.c
+++ b/drivers/usb/storage/scsiglue.c
@@ -65,7 +65,6 @@ static const char* host_info(struct Scsi_Host *host)
static int slave_alloc (struct scsi_device *sdev)
{
struct us_data *us = host_to_us(sdev->host);
- int maxp;
/*
* Set the INQUIRY transfer length to 36. We don't use any of
@@ -74,15 +73,6 @@ static int slave_alloc (struct scsi_device *sdev)
*/
sdev->inquiry_len = 36;
- /*
- * USB has unusual scatter-gather requirements: the length of each
- * scatterlist element except the last must be divisible by the
- * Bulk maxpacket value. Fortunately this value is always a
- * power of 2. Inform the block layer about this requirement.
- */
- maxp = usb_maxpacket(us->pusb_dev, us->recv_bulk_pipe, 0);
- blk_queue_virt_boundary(sdev->request_queue, maxp - 1);
-
/*
* Some host controllers may have alignment requirements.
* We'll play it safe by requiring 512-byte alignment always.
diff --git a/drivers/usb/storage/usb.c b/drivers/usb/storage/usb.c
index 9a79cd9762f3..b0f23f4f58e3 100644
--- a/drivers/usb/storage/usb.c
+++ b/drivers/usb/storage/usb.c
@@ -1050,6 +1050,16 @@ int usb_stor_probe2(struct us_data *us)
usb_autopm_get_interface_no_resume(us->pusb_intf);
snprintf(us->scsi_name, sizeof(us->scsi_name), "usb-storage %s",
dev_name(&us->pusb_intf->dev));
+
+ /*
+ * USB has unusual scatter-gather requirements: the length of each
+ * scatterlist element except the last must be divisible by the
+ * Bulk maxpacket value. Fortunately this value is always a
+ * power of 2. Inform the block layer about this requirement.
+ */
+ us_to_host(us)->virt_boundary_mask =
+ usb_maxpacket(us->pusb_dev, us->recv_bulk_pipe, 0) - 1;
+
result = scsi_add_host(us_to_host(us), dev);
if (result) {
dev_warn(dev,
--
2.20.1
^ permalink raw reply related
* [PATCH 11/13] mpt3sas: set virt_boundary_mask in the scsi host
From: Christoph Hellwig @ 2019-06-05 19:08 UTC (permalink / raw)
To: Jens Axboe
Cc: Sebastian Ott, Sagi Grimberg, Max Gurtovoy, Bart Van Assche,
Ulf Hansson, Alan Stern, Oliver Neukum, linux-block, linux-rdma,
linux-mmc, linux-nvme, linux-scsi, megaraidlinux.pdl,
MPT-FusionLinux.pdl, linux-hyperv, linux-usb, usb-storage,
linux-kernel
In-Reply-To: <20190605190836.32354-1-hch@lst.de>
This ensures all proper DMA layer handling is taken care of by the
SCSI midlayer. Note that the effect is global, as the IOMMU merging
is based off a paramters in struct device. We could still turn if off
if no PCIe devices are present, but I don't know how to find that out.
Also remove the bogus nomerges flag, merges do take the virt_boundary
into account.
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
drivers/scsi/mpt3sas/mpt3sas_scsih.c | 11 +++--------
1 file changed, 3 insertions(+), 8 deletions(-)
diff --git a/drivers/scsi/mpt3sas/mpt3sas_scsih.c b/drivers/scsi/mpt3sas/mpt3sas_scsih.c
index 1ccfbc7eebe0..03a0df2a3379 100644
--- a/drivers/scsi/mpt3sas/mpt3sas_scsih.c
+++ b/drivers/scsi/mpt3sas/mpt3sas_scsih.c
@@ -2361,14 +2361,6 @@ scsih_slave_configure(struct scsi_device *sdev)
pcie_device_put(pcie_device);
spin_unlock_irqrestore(&ioc->pcie_device_lock, flags);
scsih_change_queue_depth(sdev, qdepth);
- /* Enable QUEUE_FLAG_NOMERGES flag, so that IOs won't be
- ** merged and can eliminate holes created during merging
- ** operation.
- **/
- blk_queue_flag_set(QUEUE_FLAG_NOMERGES,
- sdev->request_queue);
- blk_queue_virt_boundary(sdev->request_queue,
- ioc->page_size - 1);
return 0;
}
@@ -10472,6 +10464,9 @@ _scsih_probe(struct pci_dev *pdev, const struct pci_device_id *id)
shost->transportt = mpt3sas_transport_template;
shost->unique_id = ioc->id;
+ /* XXX: only strictly needed if NVMe devices are attached */
+ shost->virt_boundary_mask = ioc->page_size - 1;
+
if (ioc->is_mcpu_endpoint) {
/* mCPU MPI support 64K max IO */
shost->max_sectors = 128;
--
2.20.1
^ permalink raw reply related
* [PATCH 10/13] megaraid_sas: set virt_boundary_mask in the scsi host
From: Christoph Hellwig @ 2019-06-05 19:08 UTC (permalink / raw)
To: Jens Axboe
Cc: Sebastian Ott, Sagi Grimberg, Max Gurtovoy, Bart Van Assche,
Ulf Hansson, Alan Stern, Oliver Neukum, linux-block, linux-rdma,
linux-mmc, linux-nvme, linux-scsi, megaraidlinux.pdl,
MPT-FusionLinux.pdl, linux-hyperv, linux-usb, usb-storage,
linux-kernel
In-Reply-To: <20190605190836.32354-1-hch@lst.de>
This ensures all proper DMA layer handling is taken care of by the
SCSI midlayer. Note that the effect is global, as the IOMMU merging
is based off a paramters in struct device. We could still turn if off
if no PCIe devices are present, but I don't know how to find that out.
Also remove the bogus nomerges flag, merges do take the virt_boundary
into account.
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
drivers/scsi/megaraid/megaraid_sas_base.c | 46 +++++----------------
drivers/scsi/megaraid/megaraid_sas_fusion.c | 7 ++++
2 files changed, 18 insertions(+), 35 deletions(-)
diff --git a/drivers/scsi/megaraid/megaraid_sas_base.c b/drivers/scsi/megaraid/megaraid_sas_base.c
index 3dd1df472dc6..20b3b3f8bc16 100644
--- a/drivers/scsi/megaraid/megaraid_sas_base.c
+++ b/drivers/scsi/megaraid/megaraid_sas_base.c
@@ -1870,39 +1870,6 @@ void megasas_set_dynamic_target_properties(struct scsi_device *sdev,
}
}
-/*
- * megasas_set_nvme_device_properties -
- * set nomerges=2
- * set virtual page boundary = 4K (current mr_nvme_pg_size is 4K).
- * set maximum io transfer = MDTS of NVME device provided by MR firmware.
- *
- * MR firmware provides value in KB. Caller of this function converts
- * kb into bytes.
- *
- * e.a MDTS=5 means 2^5 * nvme page size. (In case of 4K page size,
- * MR firmware provides value 128 as (32 * 4K) = 128K.
- *
- * @sdev: scsi device
- * @max_io_size: maximum io transfer size
- *
- */
-static inline void
-megasas_set_nvme_device_properties(struct scsi_device *sdev, u32 max_io_size)
-{
- struct megasas_instance *instance;
- u32 mr_nvme_pg_size;
-
- instance = (struct megasas_instance *)sdev->host->hostdata;
- mr_nvme_pg_size = max_t(u32, instance->nvme_page_size,
- MR_DEFAULT_NVME_PAGE_SIZE);
-
- blk_queue_max_hw_sectors(sdev->request_queue, (max_io_size / 512));
-
- blk_queue_flag_set(QUEUE_FLAG_NOMERGES, sdev->request_queue);
- blk_queue_virt_boundary(sdev->request_queue, mr_nvme_pg_size - 1);
-}
-
-
/*
* megasas_set_static_target_properties -
* Device property set by driver are static and it is not required to be
@@ -1961,8 +1928,10 @@ static void megasas_set_static_target_properties(struct scsi_device *sdev,
max_io_size_kb = le32_to_cpu(instance->tgt_prop->max_io_size_kb);
}
- if (instance->nvme_page_size && max_io_size_kb)
- megasas_set_nvme_device_properties(sdev, (max_io_size_kb << 10));
+ if (instance->nvme_page_size && max_io_size_kb) {
+ blk_queue_max_hw_sectors(sdev->request_queue,
+ (max_io_size_kb << 10) / 512);
+ }
scsi_change_queue_depth(sdev, device_qd);
@@ -6258,6 +6227,7 @@ static int megasas_start_aen(struct megasas_instance *instance)
static int megasas_io_attach(struct megasas_instance *instance)
{
struct Scsi_Host *host = instance->host;
+ u32 nvme_page_size = instance->nvme_page_size;
/*
* Export parameters required by SCSI mid-layer
@@ -6298,6 +6268,12 @@ static int megasas_io_attach(struct megasas_instance *instance)
host->max_lun = MEGASAS_MAX_LUN;
host->max_cmd_len = 16;
+ if (nvme_page_size) {
+ if (nvme_page_size > MR_DEFAULT_NVME_PAGE_SIZE)
+ nvme_page_size = MR_DEFAULT_NVME_PAGE_SIZE;
+ host->virt_boundary_mask = nvme_page_size - 1;
+ }
+
/*
* Notify the mid-layer about the new controller
*/
diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.c b/drivers/scsi/megaraid/megaraid_sas_fusion.c
index 4dfa0685a86c..a9ff3a648e7b 100644
--- a/drivers/scsi/megaraid/megaraid_sas_fusion.c
+++ b/drivers/scsi/megaraid/megaraid_sas_fusion.c
@@ -1935,6 +1935,13 @@ megasas_is_prp_possible(struct megasas_instance *instance,
build_prp = true;
}
+/*
+ * XXX: All the code following should go away. The block layer guarantees
+ * merging according to the virt boundary. And while we might have had some
+ * issues with that in the past we fixed them, and any new bug should be fixed
+ * in the core code as well.
+ */
+
/*
* Below code detects gaps/holes in IO data buffers.
* What does holes/gaps mean?
--
2.20.1
^ permalink raw reply related
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox