* [patch] qedr: Unlock on error in qedr_post_send()
From: Dan Carpenter @ 2016-11-15 9:51 UTC (permalink / raw)
To: Doug Ledford, Ram Amrani
Cc: Sean Hefty, Hal Rosenstock, Rajesh Borundia, linux-rdma,
kernel-janitors
If "wr" is NULL then we return without unlocking. I added an unlock
label and changed the direct returns to gotos.
Fixes: afa0e13be754 ("qedr: Add support for data path")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index a615142..01d05b0 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -2973,17 +2973,18 @@ int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
(qp->state == QED_ROCE_QP_STATE_ERR)) {
- spin_unlock_irqrestore(&qp->q_lock, flags);
*bad_wr = wr;
DP_DEBUG(dev, QEDR_MSG_CQ,
"QP in wrong state! QP icid=0x%x state %d\n",
qp->icid, qp->state);
- return -EINVAL;
+ rc = -EINVAL;
+ goto unlock;
}
if (!wr) {
DP_ERR(dev, "Got an empty post send.\n");
- return -EINVAL;
+ rc = -EINVAL;
+ goto unlock;
}
while (wr) {
@@ -3012,6 +3013,7 @@ int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
/* Make sure write sticks */
mmiowb();
+unlock:
spin_unlock_irqrestore(&qp->q_lock, flags);
return rc;
^ permalink raw reply related
* RE: [patch] qedr: Unlock on error in qedr_post_send()
From: Amrani, Ram @ 2016-11-15 10:28 UTC (permalink / raw)
To: Dan Carpenter, Doug Ledford
Cc: Sean Hefty, Hal Rosenstock, Borundia, Rajesh,
linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
kernel-janitors-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
Wei Yongjun
In-Reply-To: <20161115095059.GB15424@mwanda>
> If "wr" is NULL then we return without unlocking. I added an unlock
> label and changed the direct returns to gotos.
>
> Fixes: afa0e13be754 ("qedr: Add support for data path")
> Signed-off-by: Dan Carpenter <dan.carpenter-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>
>
> diff --git a/drivers/infiniband/hw/qedr/verbs.c
> b/drivers/infiniband/hw/qedr/verbs.c
> index a615142..01d05b0 100644
> --- a/drivers/infiniband/hw/qedr/verbs.c
> +++ b/drivers/infiniband/hw/qedr/verbs.c
> @@ -2973,17 +2973,18 @@ int qedr_post_send(struct ib_qp *ibqp, struct
> ib_send_wr *wr,
>
> if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
> (qp->state == QED_ROCE_QP_STATE_ERR)) {
> - spin_unlock_irqrestore(&qp->q_lock, flags);
> *bad_wr = wr;
> DP_DEBUG(dev, QEDR_MSG_CQ,
> "QP in wrong state! QP icid=0x%x state %d\n",
> qp->icid, qp->state);
> - return -EINVAL;
> + rc = -EINVAL;
> + goto unlock;
> }
>
> if (!wr) {
> DP_ERR(dev, "Got an empty post send.\n");
> - return -EINVAL;
> + rc = -EINVAL;
> + goto unlock;
> }
>
> while (wr) {
> @@ -3012,6 +3013,7 @@ int qedr_post_send(struct ib_qp *ibqp, struct
> ib_send_wr *wr,
> /* Make sure write sticks */
> mmiowb();
>
> +unlock:
> spin_unlock_irqrestore(&qp->q_lock, flags);
>
> return rc;
Thanks for catching this. However this was already taken care of by Wei Yongjun in a different manner.
See: https://patchwork.kernel.org/patch/9409053/
Thanks,
Ram
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* RE: [bug report] qedr: Add support for QP verbs
From: Amrani, Ram @ 2016-11-15 10:36 UTC (permalink / raw)
To: Dan Carpenter; +Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
In-Reply-To: <20161114130310.GA14500@mwanda>
> Hello Ram Amrani,
>
> The patch cecbcddf6461: "qedr: Add support for QP verbs" from Oct 10, 2016,
> leads to the following static checker warning:
>
> drivers/infiniband/hw/qedr/verbs.c:2067 qedr_destroy_qp()
> '0x5 | 0x1' has '0x1' set on both sides
>
> drivers/infiniband/hw/qedr/verbs.c
> 2056 int qedr_destroy_qp(struct ib_qp *ibqp)
> 2057 {
> 2058 struct qedr_qp *qp = get_qedr_qp(ibqp);
> 2059 struct qedr_dev *dev = qp->dev;
> 2060 struct ib_qp_attr attr;
> 2061 int attr_mask = 0;
> 2062 int rc = 0;
> 2063
> 2064 DP_DEBUG(dev, QEDR_MSG_QP, "destroy qp: destroying %p, qp
> type=%d\n",
> 2065 qp, qp->qp_type);
> 2066
> 2067 if (qp->state != (QED_ROCE_QP_STATE_RESET |
> QED_ROCE_QP_STATE_ERR |
> 2068 QED_ROCE_QP_STATE_INIT)) {
>
> These aren't bitfields, they're just numbers. This code is pretty suspect. Not
> sure what was intended.
>
> 2069 attr.qp_state = IB_QPS_ERR;
> 2070 attr_mask |= IB_QP_STATE;
> 2071
> 2072 /* Change the QP state to ERROR */
> 2073 qedr_modify_qp(ibqp, &attr, attr_mask, NULL);
> 2074 }
>
> regards,
> dan carpenter
That's a real bug. Thanks!
Will fix.
Ram
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* suspicious rcu_dereference_protected() usage in mlx4 when tearing down an interface
From: Sagi Grimberg @ 2016-11-15 12:30 UTC (permalink / raw)
To: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Yishai Hadas,
Matan Barak, Leon Romanovsky
Hi Guys,
I just hit this with 4.9-rc4 on my test vm.
Didn't find any patches on the list so maybe this
is new?
[ 12.392197] ===============================
[ 12.392927] [ INFO: suspicious RCU usage. ]
[ 12.393625] 4.9.0-rc4-e386eaffebc4-lb #25 Tainted: G OE
[ 12.394633] -------------------------------
[ 12.395317] drivers/net/ethernet/mellanox/mlx4/en_rx.c:543 suspicious
rcu_dereference_protected() usage!
[ 12.396873]
other info that might help us debug this:
[ 12.398134]
rcu_scheduler_active = 1, debug_locks = 0
[ 12.399107] 1 lock held by rmmod/707:
[ 12.399665] #0:
[ 12.399939] (
[ 12.400230] intf_mutex
[ 12.400378] ){+.+.+.}
[ 12.400805] , at:
[ 12.401139] [<ffffffffc015e93d>] mlx4_unregister_interface+0x1d/0x70
[mlx4_core]
[ 12.402275]
stack backtrace:
[ 12.402926] CPU: 3 PID: 707 Comm: rmmod Tainted: G OE
4.9.0-rc4-e386eaffebc4-lb #25
[ 12.404221] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014
[ 12.405619] ffffb5fdc0fcfda0 ffffffffb6450643 ffff951f787b22c0
0000000000000001
[ 12.406793] ffffb5fdc0fcfdd0 ffffffffb60e78f7 ffff951f73966000
0000000000000400
[ 12.407966] ffff951f70430ad8 ffff951f73584800 ffffb5fdc0fcfe08
ffffffffc020f77a
[ 12.409069] Call Trace:
[ 12.409437] [<ffffffffb6450643>] dump_stack+0x85/0xc2
[ 12.410151] [<ffffffffb60e78f7>] lockdep_rcu_suspicious+0xd7/0x110
[ 12.411024] [<ffffffffc020f77a>] mlx4_en_destroy_rx_ring+0xba/0xc0
[mlx4_en]
[ 12.412032] [<ffffffffc0215ec6>] mlx4_en_free_resources+0xb6/0xf0
[mlx4_en]
[ 12.413006] [<ffffffffc021abe2>] mlx4_en_destroy_netdev+0xe2/0x1b0
[mlx4_en]
[ 12.413994] [<ffffffffc020b105>] mlx4_en_remove+0xe5/0xf0 [mlx4_en]
[ 12.414883] [<ffffffffc015e903>] mlx4_remove_device+0x73/0x90
[mlx4_core]
[ 12.415844] [<ffffffffc015e95e>] mlx4_unregister_interface+0x3e/0x70
[mlx4_core]
[ 12.416880] [<ffffffffc021dd6b>] mlx4_en_cleanup+0x10/0x2a5 [mlx4_en]
[ 12.417773] [<ffffffffb613322a>] SyS_delete_module+0x1ba/0x220
[ 12.418545] [<ffffffffb68e3085>] entry_SYSCALL_64_fastpath+0x23/0xc6
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* [PATCH rdma-next 0/4] Add packet pacing support for IB verbs
From: Rony Efraim @ 2016-11-15 14:43 UTC (permalink / raw)
To: Jason Gunthorpe, Leon Romanovsky
Cc: Hefty, Sean, Bodong Wang,
dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org,
linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
In-Reply-To: <DB5PR0501MB19281A8064B283BE35EDE821B0BF0-1FH/Iesddo5/SeJcUcAJq8DSnupUy6xnnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="utf-8", Size: 1237 bytes --]
On Thu, Nov 10, 2016 at 6:47 PM +0200, Jason Gunthorpe wrote:
>On Thu, Nov 10, 2016 at 09:22:42AM +0200, Leon Romanovsky wrote:
>> I'll let to Bodong to elaborate on it more, but as far as I see, the
>> AH attribute is relevant to UD QP only, while the packet pacing is
>> intended for all types of QPs.
>In IB static rate is supposed to work for all QP types.
Indeed we already have a static_rate field, which applies to all IB traffic (including UD).
The problem is:
- This field is only u8, and uses IB standard rate enumerations. For pacing, we need arbitrary rates.
- The field doesn't apply to Raw Ethernet QPs, AH or AV which isn't applicable to Raw Ethernet. It is incorrect to support AV/AH for Raw Ethernet QPs.
The rate_limit configuration came form the application like required BW for the streaming and not from the fabric (SM).
The rate_limit resolution is much higher and it is 1Kb/s.
Both of the fields are limit the rate but coming from a different entities and required separate fields.
The actual limit should be the minimum of both of them.
Rony
>Jason
N‹§²æìr¸›yúèšØb²X¬¶Ç§vØ^–)Þº{.nÇ+‰·¥Š{±ÙšŠ{ayº\x1dʇڙë,j\a¢f£¢·hš‹»öì\x17/oSc¾™Ú³9˜uÀ¦æå‰È&jw¨®\x03(階ŠÝ¢j"ú\x1a¶^[m§ÿïêäz¹Þ–Šàþf£¢·hšˆ§~ˆmš
^ permalink raw reply
* RE: [PATCH for-next 03/11] IB/hns: Optimize the logic of allocating memory using APIs
From: Salil Mehta @ 2016-11-15 15:52 UTC (permalink / raw)
To: Leon Romanovsky
Cc: dledford@redhat.com, Huwei (Xavier), oulijun,
mehta.salil.lnk@gmail.com, linux-rdma@vger.kernel.org,
netdev@vger.kernel.org, linux-kernel@vger.kernel.org, Linuxarm,
Zhangping (ZP)
In-Reply-To: <20161109072130.GH27883@leon.nu>
> -----Original Message-----
> From: Leon Romanovsky [mailto:leon@kernel.org]
> Sent: Wednesday, November 09, 2016 7:22 AM
> To: Salil Mehta
> Cc: dledford@redhat.com; Huwei (Xavier); oulijun;
> mehta.salil.lnk@gmail.com; linux-rdma@vger.kernel.org;
> netdev@vger.kernel.org; linux-kernel@vger.kernel.org; Linuxarm;
> Zhangping (ZP)
> Subject: Re: [PATCH for-next 03/11] IB/hns: Optimize the logic of
> allocating memory using APIs
>
> On Fri, Nov 04, 2016 at 04:36:25PM +0000, Salil Mehta wrote:
> > From: "Wei Hu (Xavier)" <xavier.huwei@huawei.com>
> >
> > This patch modified the logic of allocating memory using APIs in
> > hns RoCE driver. We used kcalloc instead of kmalloc_array and
> > bitmap_zero. And When kcalloc failed, call vzalloc to alloc
> > memory.
> >
> > Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
> > Signed-off-by: Ping Zhang <zhangping5@huawei.com>
> > Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
> > ---
> > drivers/infiniband/hw/hns/hns_roce_mr.c | 15 ++++++++-------
> > 1 file changed, 8 insertions(+), 7 deletions(-)
> >
> > diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c
> b/drivers/infiniband/hw/hns/hns_roce_mr.c
> > index fb87883..d3dfb5f 100644
> > --- a/drivers/infiniband/hw/hns/hns_roce_mr.c
> > +++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
> > @@ -137,11 +137,12 @@ static int hns_roce_buddy_init(struct
> hns_roce_buddy *buddy, int max_order)
> >
> > for (i = 0; i <= buddy->max_order; ++i) {
> > s = BITS_TO_LONGS(1 << (buddy->max_order - i));
> > - buddy->bits[i] = kmalloc_array(s, sizeof(long),
> GFP_KERNEL);
> > - if (!buddy->bits[i])
> > - goto err_out_free;
> > -
> > - bitmap_zero(buddy->bits[i], 1 << (buddy->max_order - i));
> > + buddy->bits[i] = kcalloc(s, sizeof(long), GFP_KERNEL);
> > + if (!buddy->bits[i]) {
> > + buddy->bits[i] = vzalloc(s * sizeof(long));
>
> I wonder, why don't you use directly vzalloc instead of kcalloc
> fallback?
As we know we will have physical contiguous pages if the kcalloc
call succeeds. This will give us a chance to have better performance
over the allocations which are just virtually contiguous through the
function vzalloc(). Therefore, later has only been used as a fallback
when our memory request cannot be entertained through kcalloc.
Are you suggesting that there will not be much performance penalty
if we use just vzalloc ?
>
> > + if (!buddy->bits[i])
> > + goto err_out_free;
> > + }
> > }
^ permalink raw reply
* [PATCH V2 for-next 00/11] Code improvements & fixes for HNS RoCE driver
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford
Cc: salil.mehta, xavier.huwei, oulijun, xushaobo2, mehta.salil.lnk,
lijun_nudt, linux-rdma, netdev, linux-kernel, linuxarm
This patchset introduces some code improvements and fixes
for the identified problems in the HNS RoCE driver.
Lijun Ou (4):
IB/hns: Add the interface for querying QP1
IB/hns: add self loopback for CM
IB/hns: Modify the condition of notifying hardware loopback
IB/hns: Fix the bug for qp state in hns_roce_v1_m_qp()
Salil Mehta (1):
IB/hns: Fix for Checkpatch.pl comment style errors
Shaobo Xu (1):
IB/hns: Implement the add_gid/del_gid and optimize the GIDs
management
Wei Hu (Xavier) (5):
IB/hns: Add code for refreshing CQ CI using TPTR
IB/hns: Optimize the logic of allocating memory using APIs
IB/hns: Modify the macro for the timeout when cmd process
IB/hns: Modify query info named port_num when querying RC QP
IB/hns: Change qpn allocation to round-robin mode.
drivers/infiniband/hw/hns/hns_roce_alloc.c | 11 +-
drivers/infiniband/hw/hns/hns_roce_cmd.c | 8 +-
drivers/infiniband/hw/hns/hns_roce_cmd.h | 7 +-
drivers/infiniband/hw/hns/hns_roce_common.h | 2 -
drivers/infiniband/hw/hns/hns_roce_cq.c | 17 +-
drivers/infiniband/hw/hns/hns_roce_device.h | 45 ++--
drivers/infiniband/hw/hns/hns_roce_eq.c | 6 +-
drivers/infiniband/hw/hns/hns_roce_hem.c | 6 +-
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 267 +++++++++++++++++------
drivers/infiniband/hw/hns/hns_roce_hw_v1.h | 17 +-
drivers/infiniband/hw/hns/hns_roce_main.c | 311 +++++++--------------------
drivers/infiniband/hw/hns/hns_roce_mr.c | 21 +-
drivers/infiniband/hw/hns/hns_roce_pd.c | 5 +-
drivers/infiniband/hw/hns/hns_roce_qp.c | 2 +-
14 files changed, 363 insertions(+), 362 deletions(-)
--
1.7.9.5
^ permalink raw reply
* [PATCH V2 for-next 01/11] IB/hns: Add the interface for querying QP1
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford
Cc: salil.mehta, xavier.huwei, oulijun, xushaobo2, mehta.salil.lnk,
lijun_nudt, linux-rdma, netdev, linux-kernel, linuxarm
In-Reply-To: <20161115181053.399568-1-salil.mehta@huawei.com>
From: Lijun Ou <oulijun@huawei.com>
In old code, It only added the interface for querying non-specific
QP. This patch mainly adds an interface for querying QP1.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
---
Change Log
Patch V2: Addressed the comment provided by Anurup M
Link: https://patchwork.kernel.org/patch/9412855/
Patch V1: Initial Submit
---
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 83 +++++++++++++++++++++++++++-
drivers/infiniband/hw/hns/hns_roce_hw_v1.h | 6 +-
2 files changed, 86 insertions(+), 3 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 71232e5..7485514 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -2630,8 +2630,78 @@ static int hns_roce_v1_query_qpc(struct hns_roce_dev *hr_dev,
return ret;
}
-int hns_roce_v1_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
- int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr)
+static int hns_roce_v1_q_sqp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
+ int qp_attr_mask,
+ struct ib_qp_init_attr *qp_init_attr)
+{
+ struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
+ struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+ struct hns_roce_sqp_context context;
+ u32 addr;
+
+ mutex_lock(&hr_qp->mutex);
+
+ if (hr_qp->state == IB_QPS_RESET) {
+ qp_attr->qp_state = IB_QPS_RESET;
+ goto done;
+ }
+
+ addr = ROCEE_QP1C_CFG0_0_REG +
+ hr_qp->port * sizeof(struct hns_roce_sqp_context);
+ context.qp1c_bytes_4 = roce_read(hr_dev, addr);
+ context.sq_rq_bt_l = roce_read(hr_dev, addr + 1);
+ context.qp1c_bytes_12 = roce_read(hr_dev, addr + 2);
+ context.qp1c_bytes_16 = roce_read(hr_dev, addr + 3);
+ context.qp1c_bytes_20 = roce_read(hr_dev, addr + 4);
+ context.cur_rq_wqe_ba_l = roce_read(hr_dev, addr + 5);
+ context.qp1c_bytes_28 = roce_read(hr_dev, addr + 6);
+ context.qp1c_bytes_32 = roce_read(hr_dev, addr + 7);
+ context.cur_sq_wqe_ba_l = roce_read(hr_dev, addr + 8);
+ context.qp1c_bytes_40 = roce_read(hr_dev, addr + 9);
+
+ hr_qp->state = roce_get_field(context.qp1c_bytes_4,
+ QP1C_BYTES_4_QP_STATE_M,
+ QP1C_BYTES_4_QP_STATE_S);
+ qp_attr->qp_state = hr_qp->state;
+ qp_attr->path_mtu = IB_MTU_256;
+ qp_attr->path_mig_state = IB_MIG_ARMED;
+ qp_attr->qkey = QKEY_VAL;
+ qp_attr->rq_psn = 0;
+ qp_attr->sq_psn = 0;
+ qp_attr->dest_qp_num = 1;
+ qp_attr->qp_access_flags = 6;
+
+ qp_attr->pkey_index = roce_get_field(context.qp1c_bytes_20,
+ QP1C_BYTES_20_PKEY_IDX_M,
+ QP1C_BYTES_20_PKEY_IDX_S);
+ qp_attr->port_num = hr_qp->port + 1;
+ qp_attr->sq_draining = 0;
+ qp_attr->max_rd_atomic = 0;
+ qp_attr->max_dest_rd_atomic = 0;
+ qp_attr->min_rnr_timer = 0;
+ qp_attr->timeout = 0;
+ qp_attr->retry_cnt = 0;
+ qp_attr->rnr_retry = 0;
+ qp_attr->alt_timeout = 0;
+
+done:
+ qp_attr->cur_qp_state = qp_attr->qp_state;
+ qp_attr->cap.max_recv_wr = hr_qp->rq.wqe_cnt;
+ qp_attr->cap.max_recv_sge = hr_qp->rq.max_gs;
+ qp_attr->cap.max_send_wr = hr_qp->sq.wqe_cnt;
+ qp_attr->cap.max_send_sge = hr_qp->sq.max_gs;
+ qp_attr->cap.max_inline_data = 0;
+ qp_init_attr->cap = qp_attr->cap;
+ qp_init_attr->create_flags = 0;
+
+ mutex_unlock(&hr_qp->mutex);
+
+ return 0;
+}
+
+static int hns_roce_v1_q_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
+ int qp_attr_mask,
+ struct ib_qp_init_attr *qp_init_attr)
{
struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
@@ -2767,6 +2837,15 @@ int hns_roce_v1_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
return ret;
}
+int hns_roce_v1_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
+ int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr)
+{
+ struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+
+ return hr_qp->doorbell_qpn <= 1 ?
+ hns_roce_v1_q_sqp(ibqp, qp_attr, qp_attr_mask, qp_init_attr) :
+ hns_roce_v1_q_qp(ibqp, qp_attr, qp_attr_mask, qp_init_attr);
+}
static void hns_roce_v1_destroy_qp_common(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *hr_qp,
int is_user)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.h b/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
index 539b0a3b..2e1878b 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
@@ -480,13 +480,17 @@ struct hns_roce_sqp_context {
u32 qp1c_bytes_12;
u32 qp1c_bytes_16;
u32 qp1c_bytes_20;
- u32 qp1c_bytes_28;
u32 cur_rq_wqe_ba_l;
+ u32 qp1c_bytes_28;
u32 qp1c_bytes_32;
u32 cur_sq_wqe_ba_l;
u32 qp1c_bytes_40;
};
+#define QP1C_BYTES_4_QP_STATE_S 0
+#define QP1C_BYTES_4_QP_STATE_M \
+ (((1UL << 3) - 1) << QP1C_BYTES_4_QP_STATE_S)
+
#define QP1C_BYTES_4_SQ_WQE_SHIFT_S 8
#define QP1C_BYTES_4_SQ_WQE_SHIFT_M \
(((1UL << 4) - 1) << QP1C_BYTES_4_SQ_WQE_SHIFT_S)
--
1.7.9.5
^ permalink raw reply related
* [PATCH V2 for-next 02/11] IB/hns: Add code for refreshing CQ CI using TPTR
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: salil.mehta-hv44wF8Li93QT0dZR+AlfA,
xavier.huwei-hv44wF8Li93QT0dZR+AlfA,
oulijun-hv44wF8Li93QT0dZR+AlfA, xushaobo2-hv44wF8Li93QT0dZR+AlfA,
mehta.salil.lnk-Re5JQEeQqe8AvxtiuMwx3w, lijun_nudt-9Onoh4P/yGk,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linuxarm-hv44wF8Li93QT0dZR+AlfA, Dongdong Huang
In-Reply-To: <20161115181053.399568-1-salil.mehta-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
From: "Wei Hu (Xavier)" <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
This patch added the code for refreshing CQ CI using TPTR in hip06
SoC.
We will send a doorbell to hardware for refreshing CQ CI when user
succeed to poll a cqe. But it will be failed if the doorbell has
been blocked. So hardware will read a special buffer called TPTR
to get the lastest CI value when the cq is almost full.
This patch support the special CI buffer as follows:
a) Alloc the memory for TPTR in the hns_roce_tptr_init function and
free it in hns_roce_tptr_free function, these two functions will
be called in probe function and in the remove function.
b) Add the code for computing offset(every cq need 2 bytes) and
write the dma addr to every cq context to notice hardware in the
function named hns_roce_v1_write_cqc.
c) Add code for mapping TPTR buffer to user space in function named
hns_roce_mmap. The mapping distinguish TPTR and UAR of user mode
by vm_pgoff(0: UAR, 1: TPTR, others:invaild) in hip06.
d) Alloc the code for refreshing CQ CI using TPTR in the function
named hns_roce_v1_poll_cq.
e) Add some variable definitions to the related structure.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Dongdong Huang(Donald) <hdd.huang-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Salil Mehta <salil.mehta-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/hw/hns/hns_roce_common.h | 2 -
drivers/infiniband/hw/hns/hns_roce_cq.c | 9 +++
drivers/infiniband/hw/hns/hns_roce_device.h | 6 +-
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 79 ++++++++++++++++++++++++---
drivers/infiniband/hw/hns/hns_roce_hw_v1.h | 9 +++
drivers/infiniband/hw/hns/hns_roce_main.c | 13 ++++-
6 files changed, 103 insertions(+), 15 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_common.h b/drivers/infiniband/hw/hns/hns_roce_common.h
index 2970161..0dcb620 100644
--- a/drivers/infiniband/hw/hns/hns_roce_common.h
+++ b/drivers/infiniband/hw/hns/hns_roce_common.h
@@ -253,8 +253,6 @@
#define ROCEE_VENDOR_ID_REG 0x0
#define ROCEE_VENDOR_PART_ID_REG 0x4
-#define ROCEE_HW_VERSION_REG 0x8
-
#define ROCEE_SYS_IMAGE_GUID_L_REG 0xC
#define ROCEE_SYS_IMAGE_GUID_H_REG 0x10
diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index 0973659..5dc8d92 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -349,6 +349,15 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
goto err_mtt;
}
+ /*
+ * For the QP created by kernel space, tptr value should be initialized
+ * to zero; For the QP created by user space, it will cause synchronous
+ * problems if tptr is set to zero here, so we initialze it in user
+ * space.
+ */
+ if (!context)
+ *hr_cq->tptr_addr = 0;
+
/* Get created cq handler and carry out event */
hr_cq->comp = hns_roce_ib_cq_comp;
hr_cq->event = hns_roce_ib_cq_event;
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 3417315..7242b14 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -37,6 +37,8 @@
#define DRV_NAME "hns_roce"
+#define HNS_ROCE_HW_VER1 ('h' << 24 | 'i' << 16 | '0' << 8 | '6')
+
#define MAC_ADDR_OCTET_NUM 6
#define HNS_ROCE_MAX_MSG_LEN 0x80000000
@@ -296,7 +298,7 @@ struct hns_roce_cq {
u32 cq_depth;
u32 cons_index;
void __iomem *cq_db_l;
- void __iomem *tptr_addr;
+ u16 *tptr_addr;
unsigned long cqn;
u32 vector;
atomic_t refcount;
@@ -553,6 +555,8 @@ struct hns_roce_dev {
int cmd_mod;
int loop_idc;
+ dma_addr_t tptr_dma_addr; /*only for hw v1*/
+ u32 tptr_size; /*only for hw v1*/
struct hns_roce_hw *hw;
};
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 7485514..959d5ca 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -849,6 +849,45 @@ static void hns_roce_bt_free(struct hns_roce_dev *hr_dev)
priv->bt_table.qpc_buf.buf, priv->bt_table.qpc_buf.map);
}
+static int hns_roce_tptr_init(struct hns_roce_dev *hr_dev)
+{
+ struct device *dev = &hr_dev->pdev->dev;
+ struct hns_roce_buf_list *tptr_buf;
+ struct hns_roce_v1_priv *priv;
+
+ priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+ tptr_buf = &priv->tptr_table.tptr_buf;
+
+ /*
+ * This buffer will be used for CQ's tptr(tail pointer), also
+ * named ci(customer index). Every CQ will use 2 bytes to save
+ * cqe ci in hip06. Hardware will read this area to get new ci
+ * when the queue is almost full.
+ */
+ tptr_buf->buf = dma_alloc_coherent(dev, HNS_ROCE_V1_TPTR_BUF_SIZE,
+ &tptr_buf->map, GFP_KERNEL);
+ if (!tptr_buf->buf)
+ return -ENOMEM;
+
+ hr_dev->tptr_dma_addr = tptr_buf->map;
+ hr_dev->tptr_size = HNS_ROCE_V1_TPTR_BUF_SIZE;
+
+ return 0;
+}
+
+static void hns_roce_tptr_free(struct hns_roce_dev *hr_dev)
+{
+ struct device *dev = &hr_dev->pdev->dev;
+ struct hns_roce_buf_list *tptr_buf;
+ struct hns_roce_v1_priv *priv;
+
+ priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+ tptr_buf = &priv->tptr_table.tptr_buf;
+
+ dma_free_coherent(dev, HNS_ROCE_V1_TPTR_BUF_SIZE,
+ tptr_buf->buf, tptr_buf->map);
+}
+
/**
* hns_roce_v1_reset - reset RoCE
* @hr_dev: RoCE device struct pointer
@@ -906,12 +945,11 @@ void hns_roce_v1_profile(struct hns_roce_dev *hr_dev)
hr_dev->vendor_id = le32_to_cpu(roce_read(hr_dev, ROCEE_VENDOR_ID_REG));
hr_dev->vendor_part_id = le32_to_cpu(roce_read(hr_dev,
ROCEE_VENDOR_PART_ID_REG));
- hr_dev->hw_rev = le32_to_cpu(roce_read(hr_dev, ROCEE_HW_VERSION_REG));
-
hr_dev->sys_image_guid = le32_to_cpu(roce_read(hr_dev,
ROCEE_SYS_IMAGE_GUID_L_REG)) |
((u64)le32_to_cpu(roce_read(hr_dev,
ROCEE_SYS_IMAGE_GUID_H_REG)) << 32);
+ hr_dev->hw_rev = HNS_ROCE_HW_VER1;
caps->num_qps = HNS_ROCE_V1_MAX_QP_NUM;
caps->max_wqes = HNS_ROCE_V1_MAX_WQE_NUM;
@@ -1009,8 +1047,17 @@ int hns_roce_v1_init(struct hns_roce_dev *hr_dev)
goto error_failed_bt_init;
}
+ ret = hns_roce_tptr_init(hr_dev);
+ if (ret) {
+ dev_err(dev, "tptr init failed!\n");
+ goto error_failed_tptr_init;
+ }
+
return 0;
+error_failed_tptr_init:
+ hns_roce_bt_free(hr_dev);
+
error_failed_bt_init:
hns_roce_port_enable(hr_dev, HNS_ROCE_PORT_DOWN);
hns_roce_raq_free(hr_dev);
@@ -1022,6 +1069,7 @@ int hns_roce_v1_init(struct hns_roce_dev *hr_dev)
void hns_roce_v1_exit(struct hns_roce_dev *hr_dev)
{
+ hns_roce_tptr_free(hr_dev);
hns_roce_bt_free(hr_dev);
hns_roce_port_enable(hr_dev, HNS_ROCE_PORT_DOWN);
hns_roce_raq_free(hr_dev);
@@ -1339,14 +1387,21 @@ void hns_roce_v1_write_cqc(struct hns_roce_dev *hr_dev,
dma_addr_t dma_handle, int nent, u32 vector)
{
struct hns_roce_cq_context *cq_context = NULL;
- void __iomem *tptr_addr;
+ struct hns_roce_buf_list *tptr_buf;
+ struct hns_roce_v1_priv *priv;
+ dma_addr_t tptr_dma_addr;
+ int offset;
+
+ priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+ tptr_buf = &priv->tptr_table.tptr_buf;
cq_context = mb_buf;
memset(cq_context, 0, sizeof(*cq_context));
- tptr_addr = 0;
- hr_dev->priv_addr = tptr_addr;
- hr_cq->tptr_addr = tptr_addr;
+ /* Get the tptr for this CQ. */
+ offset = hr_cq->cqn * HNS_ROCE_V1_TPTR_ENTRY_SIZE;
+ tptr_dma_addr = tptr_buf->map + offset;
+ hr_cq->tptr_addr = (u16 *)(tptr_buf->buf + offset);
/* Register cq_context members */
roce_set_field(cq_context->cqc_byte_4,
@@ -1390,10 +1445,10 @@ void hns_roce_v1_write_cqc(struct hns_roce_dev *hr_dev,
roce_set_field(cq_context->cqc_byte_20,
CQ_CONTEXT_CQC_BYTE_20_CQE_TPTR_ADDR_H_M,
CQ_CONTEXT_CQC_BYTE_20_CQE_TPTR_ADDR_H_S,
- (u64)tptr_addr >> 44);
+ tptr_dma_addr >> 44);
cq_context->cqc_byte_20 = cpu_to_le32(cq_context->cqc_byte_20);
- cq_context->cqe_tptr_addr_l = (u32)((u64)tptr_addr >> 12);
+ cq_context->cqe_tptr_addr_l = (u32)(tptr_dma_addr >> 12);
roce_set_field(cq_context->cqc_byte_32,
CQ_CONTEXT_CQC_BYTE_32_CUR_CQE_BA1_H_M,
@@ -1659,8 +1714,14 @@ int hns_roce_v1_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
break;
}
- if (npolled)
+ if (npolled) {
+ *hr_cq->tptr_addr = hr_cq->cons_index &
+ ((hr_cq->cq_depth << 1) - 1);
+
+ /* Memroy barrier */
+ wmb();
hns_roce_v1_cq_set_ci(hr_cq, hr_cq->cons_index);
+ }
spin_unlock_irqrestore(&hr_cq->lock, flags);
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.h b/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
index 2e1878b..6004c7f 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
@@ -104,6 +104,10 @@
#define HNS_ROCE_BT_RSV_BUF_SIZE (1 << 17)
+#define HNS_ROCE_V1_TPTR_ENTRY_SIZE 2
+#define HNS_ROCE_V1_TPTR_BUF_SIZE \
+ (HNS_ROCE_V1_TPTR_ENTRY_SIZE * HNS_ROCE_V1_MAX_CQ_NUM)
+
#define HNS_ROCE_ODB_POLL_MODE 0
#define HNS_ROCE_SDB_NORMAL_MODE 0
@@ -983,10 +987,15 @@ struct hns_roce_bt_table {
struct hns_roce_buf_list cqc_buf;
};
+struct hns_roce_tptr_table {
+ struct hns_roce_buf_list tptr_buf;
+};
+
struct hns_roce_v1_priv {
struct hns_roce_db_table db_table;
struct hns_roce_raq_table raq_table;
struct hns_roce_bt_table bt_table;
+ struct hns_roce_tptr_table tptr_table;
};
int hns_dsaf_roce_reset(struct fwnode_handle *dsaf_fwnode, bool dereset);
diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index 764e35a..6770171 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -549,6 +549,8 @@ static int hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext)
static int hns_roce_mmap(struct ib_ucontext *context,
struct vm_area_struct *vma)
{
+ struct hns_roce_dev *hr_dev = to_hr_dev(context->device);
+
if (((vma->vm_end - vma->vm_start) % PAGE_SIZE) != 0)
return -EINVAL;
@@ -558,10 +560,15 @@ static int hns_roce_mmap(struct ib_ucontext *context,
to_hr_ucontext(context)->uar.pfn,
PAGE_SIZE, vma->vm_page_prot))
return -EAGAIN;
-
- } else {
+ } else if (vma->vm_pgoff == 1 && hr_dev->hw_rev == HNS_ROCE_HW_VER1) {
+ /* vm_pgoff: 1 -- TPTR */
+ if (io_remap_pfn_range(vma, vma->vm_start,
+ hr_dev->tptr_dma_addr >> PAGE_SHIFT,
+ hr_dev->tptr_size,
+ vma->vm_page_prot))
+ return -EAGAIN;
+ } else
return -EINVAL;
- }
return 0;
}
--
1.7.9.5
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related
* [PATCH V2 for-next 03/11] IB/hns: Optimize the logic of allocating memory using APIs
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford
Cc: salil.mehta, xavier.huwei, oulijun, xushaobo2, mehta.salil.lnk,
lijun_nudt, linux-rdma, netdev, linux-kernel, linuxarm,
Ping Zhang
In-Reply-To: <20161115181053.399568-1-salil.mehta@huawei.com>
From: "Wei Hu (Xavier)" <xavier.huwei@huawei.com>
This patch modified the logic of allocating memory using APIs in
hns RoCE driver. We used kcalloc instead of kmalloc_array and
bitmap_zero. And When kcalloc failed, call vzalloc to alloc
memory.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Ping Zhang <zhangping5@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_mr.c | 15 ++++++++-------
1 file changed, 8 insertions(+), 7 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index fb87883..d3dfb5f 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -137,11 +137,12 @@ static int hns_roce_buddy_init(struct hns_roce_buddy *buddy, int max_order)
for (i = 0; i <= buddy->max_order; ++i) {
s = BITS_TO_LONGS(1 << (buddy->max_order - i));
- buddy->bits[i] = kmalloc_array(s, sizeof(long), GFP_KERNEL);
- if (!buddy->bits[i])
- goto err_out_free;
-
- bitmap_zero(buddy->bits[i], 1 << (buddy->max_order - i));
+ buddy->bits[i] = kcalloc(s, sizeof(long), GFP_KERNEL);
+ if (!buddy->bits[i]) {
+ buddy->bits[i] = vzalloc(s * sizeof(long));
+ if (!buddy->bits[i])
+ goto err_out_free;
+ }
}
set_bit(0, buddy->bits[buddy->max_order]);
@@ -151,7 +152,7 @@ static int hns_roce_buddy_init(struct hns_roce_buddy *buddy, int max_order)
err_out_free:
for (i = 0; i <= buddy->max_order; ++i)
- kfree(buddy->bits[i]);
+ kvfree(buddy->bits[i]);
err_out:
kfree(buddy->bits);
@@ -164,7 +165,7 @@ static void hns_roce_buddy_cleanup(struct hns_roce_buddy *buddy)
int i;
for (i = 0; i <= buddy->max_order; ++i)
- kfree(buddy->bits[i]);
+ kvfree(buddy->bits[i]);
kfree(buddy->bits);
kfree(buddy->num_free);
--
1.7.9.5
^ permalink raw reply related
* [PATCH V2 for-next 04/11] IB/hns: add self loopback for CM
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford
Cc: salil.mehta, xavier.huwei, oulijun, xushaobo2, mehta.salil.lnk,
lijun_nudt, linux-rdma, netdev, linux-kernel, linuxarm,
Peter Chen
In-Reply-To: <20161115181053.399568-1-salil.mehta@huawei.com>
From: Lijun Ou <oulijun@huawei.com>
This patch mainly adds self loopback support for CM.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Peter Chen <luck.chen@huawei.com>
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 11 +++++++++++
drivers/infiniband/hw/hns/hns_roce_hw_v1.h | 2 ++
2 files changed, 13 insertions(+)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 959d5ca..e080dd6 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -32,6 +32,7 @@
#include <linux/platform_device.h>
#include <linux/acpi.h>
+#include <linux/etherdevice.h>
#include <rdma/ib_umem.h>
#include "hns_roce_common.h"
#include "hns_roce_device.h"
@@ -72,6 +73,8 @@ int hns_roce_v1_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
int nreq = 0;
u32 ind = 0;
int ret = 0;
+ u8 *smac;
+ int loopback;
if (unlikely(ibqp->qp_type != IB_QPT_GSI &&
ibqp->qp_type != IB_QPT_RC)) {
@@ -129,6 +132,14 @@ int hns_roce_v1_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
UD_SEND_WQE_U32_8_DMAC_5_M,
UD_SEND_WQE_U32_8_DMAC_5_S,
ah->av.mac[5]);
+
+ smac = (u8 *)hr_dev->dev_addr[qp->port];
+ loopback = ether_addr_equal_unaligned(ah->av.mac,
+ smac) ? 1 : 0;
+ roce_set_bit(ud_sq_wqe->u32_8,
+ UD_SEND_WQE_U32_8_LOOPBACK_INDICATOR_S,
+ loopback);
+
roce_set_field(ud_sq_wqe->u32_8,
UD_SEND_WQE_U32_8_OPERATION_TYPE_M,
UD_SEND_WQE_U32_8_OPERATION_TYPE_S,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.h b/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
index 6004c7f..cf28f1b 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
@@ -440,6 +440,8 @@ struct hns_roce_ud_send_wqe {
#define UD_SEND_WQE_U32_8_DMAC_5_M \
(((1UL << 8) - 1) << UD_SEND_WQE_U32_8_DMAC_5_S)
+#define UD_SEND_WQE_U32_8_LOOPBACK_INDICATOR_S 22
+
#define UD_SEND_WQE_U32_8_OPERATION_TYPE_S 16
#define UD_SEND_WQE_U32_8_OPERATION_TYPE_M \
(((1UL << 4) - 1) << UD_SEND_WQE_U32_8_OPERATION_TYPE_S)
--
1.7.9.5
^ permalink raw reply related
* [PATCH V2 for-next 05/11] IB/hns: Modify the condition of notifying hardware loopback
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: salil.mehta-hv44wF8Li93QT0dZR+AlfA,
xavier.huwei-hv44wF8Li93QT0dZR+AlfA,
oulijun-hv44wF8Li93QT0dZR+AlfA, xushaobo2-hv44wF8Li93QT0dZR+AlfA,
mehta.salil.lnk-Re5JQEeQqe8AvxtiuMwx3w, lijun_nudt-9Onoh4P/yGk,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linuxarm-hv44wF8Li93QT0dZR+AlfA
In-Reply-To: <20161115181053.399568-1-salil.mehta-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
From: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
This patch modified the condition of notifying hardware loopback.
In hip06, RoCE Engine has several ports, one QP is related
to one port. hardware only support loopback in the same port,
not in the different ports.
So, If QP related to port N, the dmac in the QP context equals
the smac of the local port N or the loop_idc is 1, we should
set loopback bit in QP context to notify hardware.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Salil Mehta <salil.mehta-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 24 +++++++-----------------
1 file changed, 7 insertions(+), 17 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index e080dd6..643a2ff 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -2244,24 +2244,14 @@ static int hns_roce_v1_m_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
QP_CONTEXT_QPC_BYTE_32_SIGNALING_TYPE_S,
hr_qp->sq_signal_bits);
- for (port = 0; port < hr_dev->caps.num_ports; port++) {
- smac = (u8 *)hr_dev->dev_addr[port];
- dev_dbg(dev, "smac: %2x: %2x: %2x: %2x: %2x: %2x\n",
- smac[0], smac[1], smac[2], smac[3], smac[4],
- smac[5]);
- if ((dmac[0] == smac[0]) && (dmac[1] == smac[1]) &&
- (dmac[2] == smac[2]) && (dmac[3] == smac[3]) &&
- (dmac[4] == smac[4]) && (dmac[5] == smac[5])) {
- roce_set_bit(context->qpc_bytes_32,
- QP_CONTEXT_QPC_BYTE_32_LOOPBACK_INDICATOR_S,
- 1);
- break;
- }
- }
-
- if (hr_dev->loop_idc == 0x1)
+ port = (attr_mask & IB_QP_PORT) ? (attr->port_num - 1) :
+ hr_qp->port;
+ smac = (u8 *)hr_dev->dev_addr[port];
+ /* when dmac equals smac or loop_idc is 1, it should loopback */
+ if (ether_addr_equal_unaligned(dmac, smac) ||
+ hr_dev->loop_idc == 0x1)
roce_set_bit(context->qpc_bytes_32,
- QP_CONTEXT_QPC_BYTE_32_LOOPBACK_INDICATOR_S, 1);
+ QP_CONTEXT_QPC_BYTE_32_LOOPBACK_INDICATOR_S, 1);
roce_set_bit(context->qpc_bytes_32,
QP_CONTEXT_QPC_BYTE_32_GLOBAL_HEADER_S,
--
1.7.9.5
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related
* [PATCH V2 for-next 06/11] IB/hns: Fix the bug for qp state in hns_roce_v1_m_qp()
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford
Cc: salil.mehta, xavier.huwei, oulijun, xushaobo2, mehta.salil.lnk,
lijun_nudt, linux-rdma, netdev, linux-kernel, linuxarm
In-Reply-To: <20161115181053.399568-1-salil.mehta@huawei.com>
From: Lijun Ou <oulijun@huawei.com>
In old code, the value of qp state from qpc was assigned for
attr->qp_state. The value may be an error while attr_mask &
IB_QP_STATE is zero.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 643a2ff..b835a55 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -2571,7 +2571,7 @@ static int hns_roce_v1_m_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
/* Every status migrate must change state */
roce_set_field(context->qpc_bytes_144,
QP_CONTEXT_QPC_BYTES_144_QP_STATE_M,
- QP_CONTEXT_QPC_BYTES_144_QP_STATE_S, attr->qp_state);
+ QP_CONTEXT_QPC_BYTES_144_QP_STATE_S, new_state);
/* SW pass context to HW */
ret = hns_roce_v1_qp_modify(hr_dev, &hr_qp->mtt,
--
1.7.9.5
^ permalink raw reply related
* [PATCH V2 for-next 07/11] IB/hns: Modify the macro for the timeout when cmd process
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: salil.mehta-hv44wF8Li93QT0dZR+AlfA,
xavier.huwei-hv44wF8Li93QT0dZR+AlfA,
oulijun-hv44wF8Li93QT0dZR+AlfA, xushaobo2-hv44wF8Li93QT0dZR+AlfA,
mehta.salil.lnk-Re5JQEeQqe8AvxtiuMwx3w, lijun_nudt-9Onoh4P/yGk,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linuxarm-hv44wF8Li93QT0dZR+AlfA
In-Reply-To: <20161115181053.399568-1-salil.mehta-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
From: "Wei Hu (Xavier)" <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
This patch modified the macro for the timeout when cmd is
processing as follows:
Before modification:
enum {
HNS_ROCE_CMD_TIME_CLASS_A = 10000,
HNS_ROCE_CMD_TIME_CLASS_B = 10000,
HNS_ROCE_CMD_TIME_CLASS_C = 10000,
};
After modification:
#define HNS_ROCE_CMD_TIMEOUT_MSECS 10000
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Salil Mehta <salil.mehta-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/hw/hns/hns_roce_cmd.h | 7 +------
drivers/infiniband/hw/hns/hns_roce_cq.c | 4 ++--
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 8 ++++----
drivers/infiniband/hw/hns/hns_roce_mr.c | 4 ++--
4 files changed, 9 insertions(+), 14 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.h b/drivers/infiniband/hw/hns/hns_roce_cmd.h
index e3997d3..ed14ad3 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cmd.h
+++ b/drivers/infiniband/hw/hns/hns_roce_cmd.h
@@ -34,6 +34,7 @@
#define _HNS_ROCE_CMD_H
#define HNS_ROCE_MAILBOX_SIZE 4096
+#define HNS_ROCE_CMD_TIMEOUT_MSECS 10000
enum {
/* TPT commands */
@@ -57,12 +58,6 @@ enum {
HNS_ROCE_CMD_QUERY_QP = 0x22,
};
-enum {
- HNS_ROCE_CMD_TIME_CLASS_A = 10000,
- HNS_ROCE_CMD_TIME_CLASS_B = 10000,
- HNS_ROCE_CMD_TIME_CLASS_C = 10000,
-};
-
struct hns_roce_cmd_mailbox {
void *buf;
dma_addr_t dma;
diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index 5dc8d92..461a273 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -77,7 +77,7 @@ static int hns_roce_sw2hw_cq(struct hns_roce_dev *dev,
unsigned long cq_num)
{
return hns_roce_cmd_mbox(dev, mailbox->dma, 0, cq_num, 0,
- HNS_ROCE_CMD_SW2HW_CQ, HNS_ROCE_CMD_TIME_CLASS_A);
+ HNS_ROCE_CMD_SW2HW_CQ, HNS_ROCE_CMD_TIMEOUT_MSECS);
}
static int hns_roce_cq_alloc(struct hns_roce_dev *hr_dev, int nent,
@@ -176,7 +176,7 @@ static int hns_roce_hw2sw_cq(struct hns_roce_dev *dev,
{
return hns_roce_cmd_mbox(dev, 0, mailbox ? mailbox->dma : 0, cq_num,
mailbox ? 0 : 1, HNS_ROCE_CMD_HW2SW_CQ,
- HNS_ROCE_CMD_TIME_CLASS_A);
+ HNS_ROCE_CMD_TIMEOUT_MSECS);
}
static void hns_roce_free_cq(struct hns_roce_dev *hr_dev,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index b835a55..509ea75 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -1871,12 +1871,12 @@ static int hns_roce_v1_qp_modify(struct hns_roce_dev *hr_dev,
if (op[cur_state][new_state] == HNS_ROCE_CMD_2RST_QP)
return hns_roce_cmd_mbox(hr_dev, 0, 0, hr_qp->qpn, 2,
HNS_ROCE_CMD_2RST_QP,
- HNS_ROCE_CMD_TIME_CLASS_A);
+ HNS_ROCE_CMD_TIMEOUT_MSECS);
if (op[cur_state][new_state] == HNS_ROCE_CMD_2ERR_QP)
return hns_roce_cmd_mbox(hr_dev, 0, 0, hr_qp->qpn, 2,
HNS_ROCE_CMD_2ERR_QP,
- HNS_ROCE_CMD_TIME_CLASS_A);
+ HNS_ROCE_CMD_TIMEOUT_MSECS);
mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
if (IS_ERR(mailbox))
@@ -1886,7 +1886,7 @@ static int hns_roce_v1_qp_modify(struct hns_roce_dev *hr_dev,
ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0, hr_qp->qpn, 0,
op[cur_state][new_state],
- HNS_ROCE_CMD_TIME_CLASS_C);
+ HNS_ROCE_CMD_TIMEOUT_MSECS);
hns_roce_free_cmd_mailbox(hr_dev, mailbox);
return ret;
@@ -2681,7 +2681,7 @@ static int hns_roce_v1_query_qpc(struct hns_roce_dev *hr_dev,
ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, hr_qp->qpn, 0,
HNS_ROCE_CMD_QUERY_QP,
- HNS_ROCE_CMD_TIME_CLASS_A);
+ HNS_ROCE_CMD_TIMEOUT_MSECS);
if (!ret)
memcpy(hr_context, mailbox->buf, sizeof(*hr_context));
else
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index d3dfb5f..2227962 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -53,7 +53,7 @@ static int hns_roce_sw2hw_mpt(struct hns_roce_dev *hr_dev,
{
return hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0, mpt_index, 0,
HNS_ROCE_CMD_SW2HW_MPT,
- HNS_ROCE_CMD_TIME_CLASS_B);
+ HNS_ROCE_CMD_TIMEOUT_MSECS);
}
static int hns_roce_hw2sw_mpt(struct hns_roce_dev *hr_dev,
@@ -62,7 +62,7 @@ static int hns_roce_hw2sw_mpt(struct hns_roce_dev *hr_dev,
{
return hns_roce_cmd_mbox(hr_dev, 0, mailbox ? mailbox->dma : 0,
mpt_index, !mailbox, HNS_ROCE_CMD_HW2SW_MPT,
- HNS_ROCE_CMD_TIME_CLASS_B);
+ HNS_ROCE_CMD_TIMEOUT_MSECS);
}
static int hns_roce_buddy_alloc(struct hns_roce_buddy *buddy, int order,
--
1.7.9.5
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related
* [PATCH V2 for-next 08/11] IB/hns: Modify query info named port_num when querying RC QP
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: salil.mehta-hv44wF8Li93QT0dZR+AlfA,
xavier.huwei-hv44wF8Li93QT0dZR+AlfA,
oulijun-hv44wF8Li93QT0dZR+AlfA, xushaobo2-hv44wF8Li93QT0dZR+AlfA,
mehta.salil.lnk-Re5JQEeQqe8AvxtiuMwx3w, lijun_nudt-9Onoh4P/yGk,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linuxarm-hv44wF8Li93QT0dZR+AlfA
In-Reply-To: <20161115181053.399568-1-salil.mehta-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
From: "Wei Hu (Xavier)" <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
This patch modified the output query info qp_attr->port_num
to fix bug in hip06.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Salil Mehta <salil.mehta-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 4 +---
1 file changed, 1 insertion(+), 3 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 509ea75..34b7898 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -2857,9 +2857,7 @@ static int hns_roce_v1_q_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
qp_attr->pkey_index = roce_get_field(context->qpc_bytes_12,
QP_CONTEXT_QPC_BYTES_12_P_KEY_INDEX_M,
QP_CONTEXT_QPC_BYTES_12_P_KEY_INDEX_S);
- qp_attr->port_num = (u8)roce_get_field(context->qpc_bytes_156,
- QP_CONTEXT_QPC_BYTES_156_PORT_NUM_M,
- QP_CONTEXT_QPC_BYTES_156_PORT_NUM_S) + 1;
+ qp_attr->port_num = hr_qp->port + 1;
qp_attr->sq_draining = 0;
qp_attr->max_rd_atomic = roce_get_field(context->qpc_bytes_156,
QP_CONTEXT_QPC_BYTES_156_INITIATOR_DEPTH_M,
--
1.7.9.5
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related
* [PATCH V2 for-next 09/11] IB/hns: Change qpn allocation to round-robin mode.
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford
Cc: salil.mehta, xavier.huwei, oulijun, xushaobo2, mehta.salil.lnk,
lijun_nudt, linux-rdma, netdev, linux-kernel, linuxarm
In-Reply-To: <20161115181053.399568-1-salil.mehta@huawei.com>
From: "Wei Hu (Xavier)" <xavier.huwei@huawei.com>
When using CM to establish connections, qp number that was freed
just now will be rejected by ib core. To fix these problem, We
change qpn allocation to round-robin mode. We added the round-robin
mode for allocating resources using bitmap. We use round-robin mode
for qp number and non round-robing mode for other resources like
cq number, pd number etc.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_alloc.c | 11 +++++++----
drivers/infiniband/hw/hns/hns_roce_cq.c | 4 ++--
drivers/infiniband/hw/hns/hns_roce_device.h | 9 +++++++--
drivers/infiniband/hw/hns/hns_roce_mr.c | 2 +-
drivers/infiniband/hw/hns/hns_roce_pd.c | 5 +++--
drivers/infiniband/hw/hns/hns_roce_qp.c | 2 +-
6 files changed, 21 insertions(+), 12 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_alloc.c b/drivers/infiniband/hw/hns/hns_roce_alloc.c
index 863a17a..605962f 100644
--- a/drivers/infiniband/hw/hns/hns_roce_alloc.c
+++ b/drivers/infiniband/hw/hns/hns_roce_alloc.c
@@ -61,9 +61,10 @@ int hns_roce_bitmap_alloc(struct hns_roce_bitmap *bitmap, unsigned long *obj)
return ret;
}
-void hns_roce_bitmap_free(struct hns_roce_bitmap *bitmap, unsigned long obj)
+void hns_roce_bitmap_free(struct hns_roce_bitmap *bitmap, unsigned long obj,
+ int rr)
{
- hns_roce_bitmap_free_range(bitmap, obj, 1);
+ hns_roce_bitmap_free_range(bitmap, obj, 1, rr);
}
int hns_roce_bitmap_alloc_range(struct hns_roce_bitmap *bitmap, int cnt,
@@ -106,7 +107,8 @@ int hns_roce_bitmap_alloc_range(struct hns_roce_bitmap *bitmap, int cnt,
}
void hns_roce_bitmap_free_range(struct hns_roce_bitmap *bitmap,
- unsigned long obj, int cnt)
+ unsigned long obj, int cnt,
+ int rr)
{
int i;
@@ -116,7 +118,8 @@ void hns_roce_bitmap_free_range(struct hns_roce_bitmap *bitmap,
for (i = 0; i < cnt; i++)
clear_bit(obj + i, bitmap->table);
- bitmap->last = min(bitmap->last, obj);
+ if (!rr)
+ bitmap->last = min(bitmap->last, obj);
bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
& bitmap->mask;
spin_unlock(&bitmap->lock);
diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index 461a273..c9f6c3d 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -166,7 +166,7 @@ static int hns_roce_cq_alloc(struct hns_roce_dev *hr_dev, int nent,
hns_roce_table_put(hr_dev, &cq_table->table, hr_cq->cqn);
err_out:
- hns_roce_bitmap_free(&cq_table->bitmap, hr_cq->cqn);
+ hns_roce_bitmap_free(&cq_table->bitmap, hr_cq->cqn, BITMAP_NO_RR);
return ret;
}
@@ -204,7 +204,7 @@ static void hns_roce_free_cq(struct hns_roce_dev *hr_dev,
spin_unlock_irq(&cq_table->lock);
hns_roce_table_put(hr_dev, &cq_table->table, hr_cq->cqn);
- hns_roce_bitmap_free(&cq_table->bitmap, hr_cq->cqn);
+ hns_roce_bitmap_free(&cq_table->bitmap, hr_cq->cqn, BITMAP_NO_RR);
}
static int hns_roce_ib_get_cq_umem(struct hns_roce_dev *hr_dev,
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 7242b14..593a42a 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -72,6 +72,9 @@
#define HNS_ROCE_MAX_GID_NUM 16
#define HNS_ROCE_GID_SIZE 16
+#define BITMAP_NO_RR 0
+#define BITMAP_RR 1
+
#define MR_TYPE_MR 0x00
#define MR_TYPE_DMA 0x03
@@ -661,7 +664,8 @@ int hns_roce_buf_write_mtt(struct hns_roce_dev *hr_dev,
void hns_roce_cleanup_qp_table(struct hns_roce_dev *hr_dev);
int hns_roce_bitmap_alloc(struct hns_roce_bitmap *bitmap, unsigned long *obj);
-void hns_roce_bitmap_free(struct hns_roce_bitmap *bitmap, unsigned long obj);
+void hns_roce_bitmap_free(struct hns_roce_bitmap *bitmap, unsigned long obj,
+ int rr);
int hns_roce_bitmap_init(struct hns_roce_bitmap *bitmap, u32 num, u32 mask,
u32 reserved_bot, u32 resetrved_top);
void hns_roce_bitmap_cleanup(struct hns_roce_bitmap *bitmap);
@@ -669,7 +673,8 @@ int hns_roce_bitmap_init(struct hns_roce_bitmap *bitmap, u32 num, u32 mask,
int hns_roce_bitmap_alloc_range(struct hns_roce_bitmap *bitmap, int cnt,
int align, unsigned long *obj);
void hns_roce_bitmap_free_range(struct hns_roce_bitmap *bitmap,
- unsigned long obj, int cnt);
+ unsigned long obj, int cnt,
+ int rr);
struct ib_ah *hns_roce_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
int hns_roce_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr);
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index 2227962..6396bc5 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -288,7 +288,7 @@ static void hns_roce_mr_free(struct hns_roce_dev *hr_dev,
}
hns_roce_bitmap_free(&hr_dev->mr_table.mtpt_bitmap,
- key_to_hw_index(mr->key));
+ key_to_hw_index(mr->key), BITMAP_NO_RR);
}
static int hns_roce_mr_enable(struct hns_roce_dev *hr_dev,
diff --git a/drivers/infiniband/hw/hns/hns_roce_pd.c b/drivers/infiniband/hw/hns/hns_roce_pd.c
index 05db7d5..a64500f 100644
--- a/drivers/infiniband/hw/hns/hns_roce_pd.c
+++ b/drivers/infiniband/hw/hns/hns_roce_pd.c
@@ -40,7 +40,7 @@ static int hns_roce_pd_alloc(struct hns_roce_dev *hr_dev, unsigned long *pdn)
static void hns_roce_pd_free(struct hns_roce_dev *hr_dev, unsigned long pdn)
{
- hns_roce_bitmap_free(&hr_dev->pd_bitmap, pdn);
+ hns_roce_bitmap_free(&hr_dev->pd_bitmap, pdn, BITMAP_NO_RR);
}
int hns_roce_init_pd_table(struct hns_roce_dev *hr_dev)
@@ -121,7 +121,8 @@ int hns_roce_uar_alloc(struct hns_roce_dev *hr_dev, struct hns_roce_uar *uar)
void hns_roce_uar_free(struct hns_roce_dev *hr_dev, struct hns_roce_uar *uar)
{
- hns_roce_bitmap_free(&hr_dev->uar_table.bitmap, uar->index);
+ hns_roce_bitmap_free(&hr_dev->uar_table.bitmap, uar->index,
+ BITMAP_NO_RR);
}
int hns_roce_init_uar_table(struct hns_roce_dev *hr_dev)
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index e86dd8d..4775b5c 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -250,7 +250,7 @@ void hns_roce_release_range_qp(struct hns_roce_dev *hr_dev, int base_qpn,
if (base_qpn < SQP_NUM)
return;
- hns_roce_bitmap_free_range(&qp_table->bitmap, base_qpn, cnt);
+ hns_roce_bitmap_free_range(&qp_table->bitmap, base_qpn, cnt, BITMAP_RR);
}
static int hns_roce_set_rq_size(struct hns_roce_dev *hr_dev,
--
1.7.9.5
^ permalink raw reply related
* [PATCH V2 for-next 10/11] IB/hns: Implement the add_gid/del_gid and optimize the GIDs management
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford
Cc: salil.mehta, xavier.huwei, oulijun, xushaobo2, mehta.salil.lnk,
lijun_nudt, linux-rdma, netdev, linux-kernel, linuxarm
In-Reply-To: <20161115181053.399568-1-salil.mehta@huawei.com>
From: Shaobo Xu <xushaobo2@huawei.com>
IB core has implemented the calculation of GIDs and the management
of GID tables, and it is now responsible to supply query function
for GIDs. So the calculation of GIDs and the management of GID
tables in the RoCE driver is redundant.
The patch is to implement the add_gid/del_gid to set the GIDs in
the RoCE driver, remove the redundant calculation and management of
GIDs in the notifier call of the net device and the inet, and
update the query_gid.
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_device.h | 2 -
drivers/infiniband/hw/hns/hns_roce_main.c | 270 +++++----------------------
2 files changed, 48 insertions(+), 224 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 593a42a..9ef1cc3 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -429,8 +429,6 @@ struct hns_roce_ib_iboe {
struct net_device *netdevs[HNS_ROCE_MAX_PORTS];
struct notifier_block nb;
struct notifier_block nb_inet;
- /* 16 GID is shared by 6 port in v1 engine. */
- union ib_gid gid_table[HNS_ROCE_MAX_GID_NUM];
u8 phy_port[HNS_ROCE_MAX_PORTS];
};
diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index 6770171..795ef97 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -35,52 +35,13 @@
#include <rdma/ib_addr.h>
#include <rdma/ib_smi.h>
#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_cache.h>
#include "hns_roce_common.h"
#include "hns_roce_device.h"
#include "hns_roce_user.h"
#include "hns_roce_hem.h"
/**
- * hns_roce_addrconf_ifid_eui48 - Get default gid.
- * @eui: eui.
- * @vlan_id: gid
- * @dev: net device
- * Description:
- * MAC convert to GID
- * gid[0..7] = fe80 0000 0000 0000
- * gid[8] = mac[0] ^ 2
- * gid[9] = mac[1]
- * gid[10] = mac[2]
- * gid[11] = ff (VLAN ID high byte (4 MS bits))
- * gid[12] = fe (VLAN ID low byte)
- * gid[13] = mac[3]
- * gid[14] = mac[4]
- * gid[15] = mac[5]
- */
-static void hns_roce_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
- struct net_device *dev)
-{
- memcpy(eui, dev->dev_addr, 3);
- memcpy(eui + 5, dev->dev_addr + 3, 3);
- if (vlan_id < 0x1000) {
- eui[3] = vlan_id >> 8;
- eui[4] = vlan_id & 0xff;
- } else {
- eui[3] = 0xff;
- eui[4] = 0xfe;
- }
- eui[0] ^= 2;
-}
-
-static void hns_roce_make_default_gid(struct net_device *dev, union ib_gid *gid)
-{
- memset(gid, 0, sizeof(*gid));
- gid->raw[0] = 0xFE;
- gid->raw[1] = 0x80;
- hns_roce_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
-}
-
-/**
* hns_get_gid_index - Get gid index.
* @hr_dev: pointer to structure hns_roce_dev.
* @port: port, value range: 0 ~ MAX
@@ -96,30 +57,6 @@ int hns_get_gid_index(struct hns_roce_dev *hr_dev, u8 port, int gid_index)
return gid_index * hr_dev->caps.num_ports + port;
}
-static int hns_roce_set_gid(struct hns_roce_dev *hr_dev, u8 port, int gid_index,
- union ib_gid *gid)
-{
- struct device *dev = &hr_dev->pdev->dev;
- u8 gid_idx = 0;
-
- if (gid_index >= hr_dev->caps.gid_table_len[port]) {
- dev_err(dev, "gid_index %d illegal, port %d gid range: 0~%d\n",
- gid_index, port, hr_dev->caps.gid_table_len[port] - 1);
- return -EINVAL;
- }
-
- gid_idx = hns_get_gid_index(hr_dev, port, gid_index);
-
- if (!memcmp(gid, &hr_dev->iboe.gid_table[gid_idx], sizeof(*gid)))
- return -EINVAL;
-
- memcpy(&hr_dev->iboe.gid_table[gid_idx], gid, sizeof(*gid));
-
- hr_dev->hw->set_gid(hr_dev, port, gid_index, gid);
-
- return 0;
-}
-
static void hns_roce_set_mac(struct hns_roce_dev *hr_dev, u8 port, u8 *addr)
{
u8 phy_port;
@@ -147,15 +84,44 @@ static void hns_roce_set_mtu(struct hns_roce_dev *hr_dev, u8 port, int mtu)
hr_dev->hw->set_mtu(hr_dev, phy_port, tmp);
}
-static void hns_roce_update_gids(struct hns_roce_dev *hr_dev, int port)
+static int hns_roce_add_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, const union ib_gid *gid,
+ const struct ib_gid_attr *attr, void **context)
+{
+ struct hns_roce_dev *hr_dev = to_hr_dev(device);
+ u8 port = port_num - 1;
+ unsigned long flags;
+
+ if (port >= hr_dev->caps.num_ports)
+ return -EINVAL;
+
+ spin_lock_irqsave(&hr_dev->iboe.lock, flags);
+
+ hr_dev->hw->set_gid(hr_dev, port, index, (union ib_gid *)gid);
+
+ spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
+
+ return 0;
+}
+
+static int hns_roce_del_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, void **context)
{
- struct ib_event event;
+ struct hns_roce_dev *hr_dev = to_hr_dev(device);
+ union ib_gid zgid = { {0} };
+ u8 port = port_num - 1;
+ unsigned long flags;
+
+ if (port >= hr_dev->caps.num_ports)
+ return -EINVAL;
- /* Refresh gid in ib_cache */
- event.device = &hr_dev->ib_dev;
- event.element.port_num = port + 1;
- event.event = IB_EVENT_GID_CHANGE;
- ib_dispatch_event(&event);
+ spin_lock_irqsave(&hr_dev->iboe.lock, flags);
+
+ hr_dev->hw->set_gid(hr_dev, port, index, &zgid);
+
+ spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
+
+ return 0;
}
static int handle_en_event(struct hns_roce_dev *hr_dev, u8 port,
@@ -164,8 +130,6 @@ static int handle_en_event(struct hns_roce_dev *hr_dev, u8 port,
struct device *dev = &hr_dev->pdev->dev;
struct net_device *netdev;
unsigned long flags;
- union ib_gid gid;
- int ret = 0;
netdev = hr_dev->iboe.netdevs[port];
if (!netdev) {
@@ -181,10 +145,6 @@ static int handle_en_event(struct hns_roce_dev *hr_dev, u8 port,
case NETDEV_REGISTER:
case NETDEV_CHANGEADDR:
hns_roce_set_mac(hr_dev, port, netdev->dev_addr);
- hns_roce_make_default_gid(netdev, &gid);
- ret = hns_roce_set_gid(hr_dev, port, 0, &gid);
- if (!ret)
- hns_roce_update_gids(hr_dev, port);
break;
case NETDEV_DOWN:
/*
@@ -197,7 +157,7 @@ static int handle_en_event(struct hns_roce_dev *hr_dev, u8 port,
}
spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
- return ret;
+ return 0;
}
static int hns_roce_netdev_event(struct notifier_block *self,
@@ -224,118 +184,17 @@ static int hns_roce_netdev_event(struct notifier_block *self,
return NOTIFY_DONE;
}
-static void hns_roce_addr_event(int event, struct net_device *event_netdev,
- struct hns_roce_dev *hr_dev, union ib_gid *gid)
-{
- struct hns_roce_ib_iboe *iboe = NULL;
- int gid_table_len = 0;
- unsigned long flags;
- union ib_gid zgid;
- u8 gid_idx = 0;
- u8 port = 0;
- int i = 0;
- int free;
- struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
- rdma_vlan_dev_real_dev(event_netdev) :
- event_netdev;
-
- if (event != NETDEV_UP && event != NETDEV_DOWN)
- return;
-
- iboe = &hr_dev->iboe;
- while (port < hr_dev->caps.num_ports) {
- if (real_dev == iboe->netdevs[port])
- break;
- port++;
- }
-
- if (port >= hr_dev->caps.num_ports) {
- dev_dbg(&hr_dev->pdev->dev, "can't find netdev\n");
- return;
- }
-
- memset(zgid.raw, 0, sizeof(zgid.raw));
- free = -1;
- gid_table_len = hr_dev->caps.gid_table_len[port];
-
- spin_lock_irqsave(&hr_dev->iboe.lock, flags);
-
- for (i = 0; i < gid_table_len; i++) {
- gid_idx = hns_get_gid_index(hr_dev, port, i);
- if (!memcmp(gid->raw, iboe->gid_table[gid_idx].raw,
- sizeof(gid->raw)))
- break;
- if (free < 0 && !memcmp(zgid.raw,
- iboe->gid_table[gid_idx].raw, sizeof(zgid.raw)))
- free = i;
- }
-
- if (i >= gid_table_len) {
- if (free < 0) {
- spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
- dev_dbg(&hr_dev->pdev->dev,
- "gid_index overflow, port(%d)\n", port);
- return;
- }
- if (!hns_roce_set_gid(hr_dev, port, free, gid))
- hns_roce_update_gids(hr_dev, port);
- } else if (event == NETDEV_DOWN) {
- if (!hns_roce_set_gid(hr_dev, port, i, &zgid))
- hns_roce_update_gids(hr_dev, port);
- }
-
- spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
-}
-
-static int hns_roce_inet_event(struct notifier_block *self, unsigned long event,
- void *ptr)
-{
- struct in_ifaddr *ifa = ptr;
- struct hns_roce_dev *hr_dev;
- struct net_device *dev = ifa->ifa_dev->dev;
- union ib_gid gid;
-
- ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
-
- hr_dev = container_of(self, struct hns_roce_dev, iboe.nb_inet);
-
- hns_roce_addr_event(event, dev, hr_dev, &gid);
-
- return NOTIFY_DONE;
-}
-
-static int hns_roce_setup_mtu_gids(struct hns_roce_dev *hr_dev)
+static int hns_roce_setup_mtu_mac(struct hns_roce_dev *hr_dev)
{
- struct in_ifaddr *ifa_list = NULL;
- union ib_gid gid = {{0} };
- u32 ipaddr = 0;
- int index = 0;
- int ret = 0;
- u8 i = 0;
+ u8 i;
for (i = 0; i < hr_dev->caps.num_ports; i++) {
hns_roce_set_mtu(hr_dev, i,
ib_mtu_enum_to_int(hr_dev->caps.max_mtu));
hns_roce_set_mac(hr_dev, i, hr_dev->iboe.netdevs[i]->dev_addr);
-
- if (hr_dev->iboe.netdevs[i]->ip_ptr) {
- ifa_list = hr_dev->iboe.netdevs[i]->ip_ptr->ifa_list;
- index = 1;
- while (ifa_list) {
- ipaddr = ifa_list->ifa_address;
- ipv6_addr_set_v4mapped(ipaddr,
- (struct in6_addr *)&gid);
- ret = hns_roce_set_gid(hr_dev, i, index, &gid);
- if (ret)
- break;
- index++;
- ifa_list = ifa_list->ifa_next;
- }
- hns_roce_update_gids(hr_dev, i);
- }
}
- return ret;
+ return 0;
}
static int hns_roce_query_device(struct ib_device *ib_dev,
@@ -444,31 +303,6 @@ static enum rdma_link_layer hns_roce_get_link_layer(struct ib_device *device,
static int hns_roce_query_gid(struct ib_device *ib_dev, u8 port_num, int index,
union ib_gid *gid)
{
- struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
- struct device *dev = &hr_dev->pdev->dev;
- u8 gid_idx = 0;
- u8 port;
-
- if (port_num < 1 || port_num > hr_dev->caps.num_ports ||
- index >= hr_dev->caps.gid_table_len[port_num - 1]) {
- dev_err(dev,
- "port_num %d index %d illegal! correct range: port_num 1~%d index 0~%d!\n",
- port_num, index, hr_dev->caps.num_ports,
- hr_dev->caps.gid_table_len[port_num - 1] - 1);
- return -EINVAL;
- }
-
- port = port_num - 1;
- gid_idx = hns_get_gid_index(hr_dev, port, index);
- if (gid_idx >= HNS_ROCE_MAX_GID_NUM) {
- dev_err(dev, "port_num %d index %d illegal! total gid num %d!\n",
- port_num, index, HNS_ROCE_MAX_GID_NUM);
- return -EINVAL;
- }
-
- memcpy(gid->raw, hr_dev->iboe.gid_table[gid_idx].raw,
- HNS_ROCE_GID_SIZE);
-
return 0;
}
@@ -646,6 +480,8 @@ static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
ib_dev->get_link_layer = hns_roce_get_link_layer;
ib_dev->get_netdev = hns_roce_get_netdev;
ib_dev->query_gid = hns_roce_query_gid;
+ ib_dev->add_gid = hns_roce_add_gid;
+ ib_dev->del_gid = hns_roce_del_gid;
ib_dev->query_pkey = hns_roce_query_pkey;
ib_dev->alloc_ucontext = hns_roce_alloc_ucontext;
ib_dev->dealloc_ucontext = hns_roce_dealloc_ucontext;
@@ -688,32 +524,22 @@ static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
return ret;
}
- ret = hns_roce_setup_mtu_gids(hr_dev);
+ ret = hns_roce_setup_mtu_mac(hr_dev);
if (ret) {
- dev_err(dev, "roce_setup_mtu_gids failed!\n");
- goto error_failed_setup_mtu_gids;
+ dev_err(dev, "setup_mtu_mac failed!\n");
+ goto error_failed_setup_mtu_mac;
}
iboe->nb.notifier_call = hns_roce_netdev_event;
ret = register_netdevice_notifier(&iboe->nb);
if (ret) {
dev_err(dev, "register_netdevice_notifier failed!\n");
- goto error_failed_setup_mtu_gids;
- }
-
- iboe->nb_inet.notifier_call = hns_roce_inet_event;
- ret = register_inetaddr_notifier(&iboe->nb_inet);
- if (ret) {
- dev_err(dev, "register inet addr notifier failed!\n");
- goto error_failed_register_inetaddr_notifier;
+ goto error_failed_setup_mtu_mac;
}
return 0;
-error_failed_register_inetaddr_notifier:
- unregister_netdevice_notifier(&iboe->nb);
-
-error_failed_setup_mtu_gids:
+error_failed_setup_mtu_mac:
ib_unregister_device(ib_dev);
return ret;
--
1.7.9.5
^ permalink raw reply related
* [PATCH V2 for-next 11/11] IB/hns: Fix for Checkpatch.pl comment style errors
From: Salil Mehta @ 2016-11-15 18:10 UTC (permalink / raw)
To: dledford
Cc: salil.mehta, xavier.huwei, oulijun, xushaobo2, mehta.salil.lnk,
lijun_nudt, linux-rdma, netdev, linux-kernel, linuxarm
In-Reply-To: <20161115181053.399568-1-salil.mehta@huawei.com>
This patch correct the comment style errors caught by
checkpatch.pl script
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_cmd.c | 8 ++--
drivers/infiniband/hw/hns/hns_roce_device.h | 28 ++++++-------
drivers/infiniband/hw/hns/hns_roce_eq.c | 6 +--
drivers/infiniband/hw/hns/hns_roce_hem.c | 6 +--
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 58 +++++++++++++--------------
drivers/infiniband/hw/hns/hns_roce_main.c | 28 ++++++-------
6 files changed, 67 insertions(+), 67 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.c b/drivers/infiniband/hw/hns/hns_roce_cmd.c
index 2a0b6c0..8c1f7a6 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cmd.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cmd.c
@@ -216,10 +216,10 @@ static int __hns_roce_cmd_mbox_wait(struct hns_roce_dev *hr_dev, u64 in_param,
goto out;
/*
- * It is timeout when wait_for_completion_timeout return 0
- * The return value is the time limit set in advance
- * how many seconds showing
- */
+ * It is timeout when wait_for_completion_timeout return 0
+ * The return value is the time limit set in advance
+ * how many seconds showing
+ */
if (!wait_for_completion_timeout(&context->done,
msecs_to_jiffies(timeout))) {
dev_err(dev, "[cmd]wait_for_completion_timeout timeout\n");
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 9ef1cc3..e48464d 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -201,9 +201,9 @@ struct hns_roce_bitmap {
/* Order = 0: bitmap is biggest, order = max bitmap is least (only a bit) */
/* Every bit repesent to a partner free/used status in bitmap */
/*
-* Initial, bits of other bitmap are all 0 except that a bit of max_order is 1
-* Bit = 1 represent to idle and available; bit = 0: not available
-*/
+ * Initial, bits of other bitmap are all 0 except that a bit of max_order is 1
+ * Bit = 1 represent to idle and available; bit = 0: not available
+ */
struct hns_roce_buddy {
/* Members point to every order level bitmap */
unsigned long **bits;
@@ -365,25 +365,25 @@ struct hns_roce_cmdq {
struct mutex hcr_mutex;
struct semaphore poll_sem;
/*
- * Event mode: cmd register mutex protection,
- * ensure to not exceed max_cmds and user use limit region
- */
+ * Event mode: cmd register mutex protection,
+ * ensure to not exceed max_cmds and user use limit region
+ */
struct semaphore event_sem;
int max_cmds;
spinlock_t context_lock;
int free_head;
struct hns_roce_cmd_context *context;
/*
- * Result of get integer part
- * which max_comds compute according a power of 2
- */
+ * Result of get integer part
+ * which max_comds compute according a power of 2
+ */
u16 token_mask;
/*
- * Process whether use event mode, init default non-zero
- * After the event queue of cmd event ready,
- * can switch into event mode
- * close device, switch into poll mode(non event mode)
- */
+ * Process whether use event mode, init default non-zero
+ * After the event queue of cmd event ready,
+ * can switch into event mode
+ * close device, switch into poll mode(non event mode)
+ */
u8 use_events;
u8 toggle;
};
diff --git a/drivers/infiniband/hw/hns/hns_roce_eq.c b/drivers/infiniband/hw/hns/hns_roce_eq.c
index 21e21b0..50f8649 100644
--- a/drivers/infiniband/hw/hns/hns_roce_eq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_eq.c
@@ -371,9 +371,9 @@ static int hns_roce_aeq_ovf_int(struct hns_roce_dev *hr_dev,
int i = 0;
/**
- * AEQ overflow ECC mult bit err CEQ overflow alarm
- * must clear interrupt, mask irq, clear irq, cancel mask operation
- */
+ * AEQ overflow ECC mult bit err CEQ overflow alarm
+ * must clear interrupt, mask irq, clear irq, cancel mask operation
+ */
aeshift_val = roce_read(hr_dev, ROCEE_CAEP_AEQC_AEQE_SHIFT_REG);
if (roce_get_bit(aeshift_val,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index 250d8f2..c5104e0 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -80,9 +80,9 @@ struct hns_roce_hem *hns_roce_alloc_hem(struct hns_roce_dev *hr_dev, int npages,
--order;
/*
- * Alloc memory one time. If failed, don't alloc small block
- * memory, directly return fail.
- */
+ * Alloc memory one time. If failed, don't alloc small block
+ * memory, directly return fail.
+ */
mem = &chunk->mem[chunk->npages];
buf = dma_alloc_coherent(&hr_dev->pdev->dev, PAGE_SIZE << order,
&sg_dma_address(mem), gfp_mask);
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 34b7898..125ab90 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -1352,9 +1352,9 @@ static void __hns_roce_v1_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
}
/*
- * Now backwards through the CQ, removing CQ entries
- * that match our QP by overwriting them with next entries.
- */
+ * Now backwards through the CQ, removing CQ entries
+ * that match our QP by overwriting them with next entries.
+ */
while ((int) --prod_index - (int) hr_cq->cons_index >= 0) {
cqe = get_cqe(hr_cq, prod_index & hr_cq->ib_cq.cqe);
if ((roce_get_field(cqe->cqe_byte_16, CQE_BYTE_16_LOCAL_QPN_M,
@@ -1376,9 +1376,9 @@ static void __hns_roce_v1_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
if (nfreed) {
hr_cq->cons_index += nfreed;
/*
- * Make sure update of buffer contents is done before
- * updating consumer index.
- */
+ * Make sure update of buffer contents is done before
+ * updating consumer index.
+ */
wmb();
hns_roce_v1_cq_set_ci(hr_cq, hr_cq->cons_index);
@@ -1473,7 +1473,7 @@ void hns_roce_v1_write_cqc(struct hns_roce_dev *hr_dev,
roce_set_bit(cq_context->cqc_byte_32,
CQ_CQNTEXT_CQC_BYTE_32_TYPE_OF_COMPLETION_NOTIFICATION_S,
0);
- /*The initial value of cq's ci is 0 */
+ /* The initial value of cq's ci is 0 */
roce_set_field(cq_context->cqc_byte_32,
CQ_CONTEXT_CQC_BYTE_32_CQ_CONS_IDX_M,
CQ_CONTEXT_CQC_BYTE_32_CQ_CONS_IDX_S, 0);
@@ -1490,9 +1490,9 @@ int hns_roce_v1_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
notification_flag = (flags & IB_CQ_SOLICITED_MASK) ==
IB_CQ_SOLICITED ? CQ_DB_REQ_NOT : CQ_DB_REQ_NOT_SOL;
/*
- * flags = 0; Notification Flag = 1, next
- * flags = 1; Notification Flag = 0, solocited
- */
+ * flags = 0; Notification Flag = 1, next
+ * flags = 1; Notification Flag = 0, solocited
+ */
doorbell[0] = hr_cq->cons_index & ((hr_cq->cq_depth << 1) - 1);
roce_set_bit(doorbell[1], ROCEE_DB_OTHERS_H_ROCEE_DB_OTH_HW_SYNS_S, 1);
roce_set_field(doorbell[1], ROCEE_DB_OTHERS_H_ROCEE_DB_OTH_CMD_M,
@@ -1647,10 +1647,10 @@ static int hns_roce_v1_poll_one(struct hns_roce_cq *hr_cq,
wq = &(*cur_qp)->sq;
if ((*cur_qp)->sq_signal_bits) {
/*
- * If sg_signal_bit is 1,
- * firstly tail pointer updated to wqe
- * which current cqe correspond to
- */
+ * If sg_signal_bit is 1,
+ * firstly tail pointer updated to wqe
+ * which current cqe correspond to
+ */
wqe_ctr = (u16)roce_get_field(cqe->cqe_byte_4,
CQE_BYTE_4_WQE_INDEX_M,
CQE_BYTE_4_WQE_INDEX_S);
@@ -2072,11 +2072,11 @@ static int hns_roce_v1_m_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
}
/*
- *Reset to init
- * Mandatory param:
- * IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT | IB_QP_ACCESS_FLAGS
- * Optional param: NA
- */
+ * Reset to init
+ * Mandatory param:
+ * IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT | IB_QP_ACCESS_FLAGS
+ * Optional param: NA
+ */
if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
roce_set_field(context->qpc_bytes_4,
QP_CONTEXT_QPC_BYTES_4_TRANSPORT_SERVICE_TYPE_M,
@@ -2584,9 +2584,9 @@ static int hns_roce_v1_m_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
}
/*
- * Use rst2init to instead of init2init with drv,
- * need to hw to flash RQ HEAD by DB again
- */
+ * Use rst2init to instead of init2init with drv,
+ * need to hw to flash RQ HEAD by DB again
+ */
if (cur_state == IB_QPS_INIT && new_state == IB_QPS_INIT) {
/* Memory barrier */
wmb();
@@ -2708,7 +2708,7 @@ static int hns_roce_v1_q_sqp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
goto done;
}
- addr = ROCEE_QP1C_CFG0_0_REG +
+ addr = ROCEE_QP1C_CFG0_0_REG +
hr_qp->port * sizeof(struct hns_roce_sqp_context);
context.qp1c_bytes_4 = roce_read(hr_dev, addr);
context.sq_rq_bt_l = roce_read(hr_dev, addr + 1);
@@ -2921,9 +2921,9 @@ static void hns_roce_v1_destroy_qp_common(struct hns_roce_dev *hr_dev,
if (hr_qp->ibqp.qp_type == IB_QPT_RC) {
if (hr_qp->state != IB_QPS_RESET) {
/*
- * Set qp to ERR,
- * waiting for hw complete processing all dbs
- */
+ * Set qp to ERR,
+ * waiting for hw complete processing all dbs
+ */
if (hns_roce_v1_qp_modify(hr_dev, NULL,
to_hns_roce_state(
(enum ib_qp_state)hr_qp->state),
@@ -2936,9 +2936,9 @@ static void hns_roce_v1_destroy_qp_common(struct hns_roce_dev *hr_dev,
sdbisusepr_val = roce_read(hr_dev,
ROCEE_SDB_ISSUE_PTR_REG);
/*
- * Query db process status,
- * until hw process completely
- */
+ * Query db process status,
+ * until hw process completely
+ */
end = msecs_to_jiffies(
HNS_ROCE_QP_DESTROY_TIMEOUT_MSECS) + jiffies;
do {
diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index 795ef97..914d0ac 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -148,8 +148,8 @@ static int handle_en_event(struct hns_roce_dev *hr_dev, u8 port,
break;
case NETDEV_DOWN:
/*
- * In v1 engine, only support all ports closed together.
- */
+ * In v1 engine, only support all ports closed together.
+ */
break;
default:
dev_dbg(dev, "NETDEV event = 0x%x!\n", (u32)(event));
@@ -773,10 +773,10 @@ static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
}
/**
-* hns_roce_setup_hca - setup host channel adapter
-* @hr_dev: pointer to hns roce device
-* Return : int
-*/
+ * hns_roce_setup_hca - setup host channel adapter
+ * @hr_dev: pointer to hns roce device
+ * Return : int
+ */
static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
{
int ret;
@@ -841,11 +841,11 @@ static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
}
/**
-* hns_roce_probe - RoCE driver entrance
-* @pdev: pointer to platform device
-* Return : int
-*
-*/
+ * hns_roce_probe - RoCE driver entrance
+ * @pdev: pointer to platform device
+ * Return : int
+ *
+ */
static int hns_roce_probe(struct platform_device *pdev)
{
int ret;
@@ -958,9 +958,9 @@ static int hns_roce_probe(struct platform_device *pdev)
}
/**
-* hns_roce_remove - remove RoCE device
-* @pdev: pointer to platform device
-*/
+ * hns_roce_remove - remove RoCE device
+ * @pdev: pointer to platform device
+ */
static int hns_roce_remove(struct platform_device *pdev)
{
struct hns_roce_dev *hr_dev = platform_get_drvdata(pdev);
--
1.7.9.5
^ permalink raw reply related
* NFSD generic R/W API (sendto path) performance results
From: Chuck Lever @ 2016-11-15 18:45 UTC (permalink / raw)
To: Christoph Hellwig, Sagi Grimberg; +Cc: List Linux RDMA Mailing
I've built a prototype conversion of the in-kernel NFS server's sendto
path to use the new generic R/W API. This path handles NFS Replies, so
it is responsible for building and sending RDMA Writes carrying NFS
READ payloads, and for transmitting all NFS Replies.
I've published the prototype (against my for-4.10 server series) here:
http://git.linux-nfs.org/?p=cel/cel-2.6.git;a=shortlog;h=refs/heads/nfsd-rdma-rw-api
It's the very last patch in the series.
"iozone -i0 -i1 -s2g -r1m -I" with NFSv3, sec=sys, CX-3 on both sides,
FDR fabric, share is a tmpfs. This test writes and reads a 2GB file with
1MB direct writes and reads.
The client forms NFS requests with a single 1MB RDMA segment to catch
the NFS READ payload. Before the conversion, the server posts a series
of single Write WRs with 30 pages each, for each RDMA segment written
to the client. After the conversion, the server posts a single chain
of 30-page Write WRs for each RDMA segment written to the client.
Before the API conversion: rdma_stat_post_send = 45097
After the API conversion: rdma_stat_post_send = 16411
That's what I expected to see. This shows the number of ib_post_send
calls is significantly lower after the conversion.
Unfortunately the throughput and latency numbers are worse (ignore
the write/rewrite numbers for now). Output is in kBytes/sec.
Before conversion, one iozone run:
kB reclen write rewrite read reread
2097152 1024 772835 931267 1895922 1927848
READ:
4098 ops (49%)
avg bytes sent per op: 140 avg bytes received per op: 1048704
backlog wait: 0.006345 RTT: 0.321132 total execute time: 0.332113
After conversion:
kB reclen write rewrite read reread
2097152 1024 703850 913824 1561682 1441448
READ:
4098 ops (49%)
avg bytes sent per op: 140 avg bytes received per op: 1048704
backlog wait: 0.010737 RTT: 0.469497 total execute time: 0.488043
That's 140us worse RTT per READ, in this run. The gap between before and
after was roughly the same for all runs.
To partially explain this, I captured traffic on the server using ibdump
during a similar iozone test. This removes fabric and client HCA latencies
from the picture.
This is a QD=1 test, so it's easy to analyze individual NFS READ operations
in each capture. I computed three latency numbers per READ transaction
based on the timestamps in the capture file, which should be accurate to
1 microsecond:
1. Call took: the time between when the server i/f sees the incoming RDMA
Send carrying the NFS READ Call, and when the server i/f sees the outgoing
RDMA Send carrying the NFS READ Reply.
2. Call-to-first-Write: the time between when the server i/f sees the
incoming RDMA Send carrying the NFS READ Call, and when the server i/f
sees the first outgoing RDMA Write request. Roughly how long it takes
the server to prepare and post the RDMA Writes.
3. First-to-last-Write: the time between when the server i/f sees the
first outgoing RDMA Write request, and when the server i/f sees the
last outgoing RDMA Write request. Roughly how long it takes the HCA
to transmit the RDMA Writes.
Averages over 5 NFS READ calls chosen at random, before conversion:
Call took 414us. Call-to-first-Write 85us. First-to-last-Write 327us
Averages over 5 NFS READ calls chosen at random, after conversion:
Call took 521us. Call-to-first-Write 160us. First-to-last-Write 360us
The gap between before and after results was 100% consistent with
the average results across the individual NFS READ operations.
There are two stories here:
1. Call-to-first-Write takes longer. My first guess is that the server
takes longer to build and DMA map a long Write WR chain than it does
to build, map, and post a single Write WR. The HCA can get started
transmitting Writes sooner, and the server continues working on
posting Write WRs in parallel with the on-the-wire activity.
2. First-to-last-Write takes longer. I don't have any explanation
for the HCA taking 10% longer to transmit the full 1MB payload.
--
Chuck Lever
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* 4.9 cxgb4 patches
From: Steve Wise @ 2016-11-15 20:16 UTC (permalink / raw)
To: 'Doug Ledford'; +Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA
Hey Doug,
Please confirm that you plan to merge these for 4.9-rc:
https://patchwork.kernel.org/patch/9383045/
https://patchwork.kernel.org/patch/9411375/
Thanks!
Steve
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* Re: suspicious rcu_dereference_protected() usage in mlx4 when tearing down an interface
From: Or Gerlitz @ 2016-11-15 20:19 UTC (permalink / raw)
To: Sagi Grimberg, Tariq Toukan
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
In-Reply-To: <52285420-40b9-afdb-cb29-7e0c939ece26-NQWnxTmZq1alnMjI0IkVqw@public.gmane.org>
On Tue, Nov 15, 2016 at 2:30 PM, Sagi Grimberg <sagi-NQWnxTmZq1alnMjI0IkVqw@public.gmane.org> wrote:
> I just hit this with 4.9-rc4 on my test vm.
> Didn't find any patches on the list so maybe this is new?
Yep, seems to be new, I stepped on it today as well and sent note to
Tariq (cc-ed) who is the driver maintainer,
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* RE: NFSD generic R/W API (sendto path) performance results
From: Steve Wise @ 2016-11-15 20:35 UTC (permalink / raw)
To: 'Chuck Lever', 'Christoph Hellwig',
'Sagi Grimberg'
Cc: 'List Linux RDMA Mailing'
In-Reply-To: <9170C872-DEE1-4D96-B9D8-E9D2B3F91915-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>
>
> I've built a prototype conversion of the in-kernel NFS server's sendto
> path to use the new generic R/W API. This path handles NFS Replies, so
> it is responsible for building and sending RDMA Writes carrying NFS
> READ payloads, and for transmitting all NFS Replies.
>
> I've published the prototype (against my for-4.10 server series) here:
>
>
http://git.linux-nfs.org/?p=cel/cel-2.6.git;a=shortlog;h=refs/heads/nfsd-rdma-rw
-api
>
> It's the very last patch in the series.
>
>
> "iozone -i0 -i1 -s2g -r1m -I" with NFSv3, sec=sys, CX-3 on both sides,
> FDR fabric, share is a tmpfs. This test writes and reads a 2GB file with
> 1MB direct writes and reads.
>
> The client forms NFS requests with a single 1MB RDMA segment to catch
> the NFS READ payload. Before the conversion, the server posts a series
> of single Write WRs with 30 pages each, for each RDMA segment written
> to the client. After the conversion, the server posts a single chain
> of 30-page Write WRs for each RDMA segment written to the client.
>
> Before the API conversion: rdma_stat_post_send = 45097
>
> After the API conversion: rdma_stat_post_send = 16411
>
> That's what I expected to see. This shows the number of ib_post_send
> calls is significantly lower after the conversion.
>
>
> Unfortunately the throughput and latency numbers are worse (ignore
> the write/rewrite numbers for now). Output is in kBytes/sec.
>
> Before conversion, one iozone run:
>
> kB reclen write rewrite read reread
> 2097152 1024 772835 931267 1895922 1927848
>
> READ:
> 4098 ops (49%)
> avg bytes sent per op: 140 avg bytes received per op: 1048704
> backlog wait: 0.006345 RTT: 0.321132 total execute time: 0.332113
>
> After conversion:
>
> kB reclen write rewrite read reread
> 2097152 1024 703850 913824 1561682 1441448
>
> READ:
> 4098 ops (49%)
> avg bytes sent per op: 140 avg bytes received per op: 1048704
> backlog wait: 0.010737 RTT: 0.469497 total execute time: 0.488043
>
> That's 140us worse RTT per READ, in this run. The gap between before and
> after was roughly the same for all runs.
>
>
> To partially explain this, I captured traffic on the server using ibdump
> during a similar iozone test. This removes fabric and client HCA latencies
> from the picture.
>
> This is a QD=1 test, so it's easy to analyze individual NFS READ operations
> in each capture. I computed three latency numbers per READ transaction
> based on the timestamps in the capture file, which should be accurate to
> 1 microsecond:
>
> 1. Call took: the time between when the server i/f sees the incoming RDMA
> Send carrying the NFS READ Call, and when the server i/f sees the outgoing
> RDMA Send carrying the NFS READ Reply.
>
> 2. Call-to-first-Write: the time between when the server i/f sees the
> incoming RDMA Send carrying the NFS READ Call, and when the server i/f
> sees the first outgoing RDMA Write request. Roughly how long it takes
> the server to prepare and post the RDMA Writes.
>
> 3. First-to-last-Write: the time between when the server i/f sees the
> first outgoing RDMA Write request, and when the server i/f sees the
> last outgoing RDMA Write request. Roughly how long it takes the HCA
> to transmit the RDMA Writes.
>
>
> Averages over 5 NFS READ calls chosen at random, before conversion:
> Call took 414us. Call-to-first-Write 85us. First-to-last-Write 327us
>
> Averages over 5 NFS READ calls chosen at random, after conversion:
> Call took 521us. Call-to-first-Write 160us. First-to-last-Write 360us
>
> The gap between before and after results was 100% consistent with
> the average results across the individual NFS READ operations.
>
>
Good work here!
> There are two stories here:
>
> 1. Call-to-first-Write takes longer. My first guess is that the server
> takes longer to build and DMA map a long Write WR chain than it does
> to build, map, and post a single Write WR. The HCA can get started
> transmitting Writes sooner, and the server continues working on
> posting Write WRs in parallel with the on-the-wire activity.
>
So perhaps the RDMA R/W API can have a threshold where it will dump a list of
WRs once it exceeds the threshold, and continue chunking? That threshold, by
the way, is probably device-specific.
> 2. First-to-last-Write takes longer. I don't have any explanation
> for the HCA taking 10% longer to transmit the full 1MB payload.
>
Perhaps the single WR posts are hitting device's fast-path and lowering latency
vs a long chain post that must be DMAed by the device? I'm not sure exactly how
the MLX devices work, but they do have a fast path that utilizes the CPU's
write-combining logic to send a WR over the bus as a single PCIE transaction.
But your WRs are probably large since they have 30 pages in the SGE. I'm not
sure what the threshold is for this fastpath logic for mlx. For cxgb, its 64B,
so the WR would have to fit in 64B to take advantage.
Steve.
>
> --
> Chuck Lever
>
>
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* LPC RDMA Workshop 2016 Summary
From: Leon Romanovsky @ 2016-11-16 6:19 UTC (permalink / raw)
To: RDMA mailing list; +Cc: Christoph Lameter, Liran Liss
[-- Attachment #1: Type: text/plain, Size: 5031 bytes --]
Hello RDMAers,
First of all, we would like to say thank you for everyone who made the RDMA
workshop and summit happened. This two day event was held at KS/LPC 2016 venue
and offsite, respectively [1].
It wasn’t happened without you, Thank you.
The below summary for the first day came from many attendees and doesn’t try to
cover everything discussed there. We invite you to share your notes with us and
mailing list.
The following list mostly consists from future plans and TODOs.
· Implementing a new Linux RDMA provider by Knut Omang
o Weak documentation for driver developers.
o Unclear development guides, no best practices.
o Need general RDMA test suite.
· New IOCTL ABI by Matan Barak
o Agreed on acceptance track. All core changes in bisectable way and ioctl
parts under EXPERIMENTAL Kconfig option and after that patch-per-command
transition.
o Extensive review was done and it will be addressed in the actual patches.
o This new code is valid candidate to introduce tracepoints into RDMA
subsystem.
· Consolidation of RDMA User-space code by Jason Gunthorpe
o Ongoing work on code quality/modernization, there is need to continue with
sparse annotations, coverity, MMIO accessor.
o There is need for proper systemd integration to autoload modules and proper
vendor upgrade sequence.
o 7 licenses is too much for 76K LOC code, there is clear benefit in
decreasing number of copyrights.
o Ensure that distribution packages are built from rdma-core. RedHat is
already doing it, there is need to continue to work with other distributions.
o Reuse kernel uAPI headers as a source base for libibverbs and providers.
· Debuggability and Tracing by Leon Romanovsky
o Extend standard tools as much as possible, like ss and ethtool.
o Use tracepoints for MAD snooping, new ABI work and memory registrations
flows.
o Reduce redundant printks.
· Integration with other subsystems by Liran Liss
o Challenges to integrate complex RDMA schemes (GPU-to-HCA-to-GPU) where CPU
is not involved. Multiple solutions were sent to mailing lists, but not
accepted yet.
o Best candidate seems to be DMA-BUFs, which don’t map peer BARs to the CPU.
o Associating SELinux security contexts with partition keys and subnet IDs is
an effective policy for isolating users in an IB fabric.
· Containers and RDMA by Liran Liss
o Supporting RoCE UD QPs is subtle because these QPs may only send and
receive from a subset of the GIDs – those that correspond to netdevs within the
process's namespace. Providers must enforce this implicitly, or fail UD QP
creation.
o Consider delivering RoCE CM packets through the network stack for better
enforcement.
o For IB, a process must be constrained to the partitions associated with the
netdev within the process's network namespace.
o RDMA cgroups will expose two types of objects only (instead of exposing
verbs objects): HCA objects and HCA handles, which will be presented in
absolute values and can be limited by the drivers. It will allow native support
of future vendor specific objects.
· New Fabrics by Ira Weiny
o It was open discussion about the scope of RDMA subsystem and the old
requirement to support verbs for every driver in driver/infiniband/, which
doesn’t hold for usnic and PSM.
· Future and Roadmap for RDMA for Doug Ledford
o The interoperability between IB vendors is less important now, since
Mellanox is the only one active vendor and Oracle will be in the future.
o RoCE is the main place for interoperability.
o Revealed UNH testing lab initiative to test and perform debug hackatons for
vendors, in similar way to NFS-RDMA.
o OFED software stack is rising, while upstream continues to shine :)
· User Mode Ethernet Verbs by Tzahi Oved
o Verbs objects suite well OS bypass Eth programming with generic object model
to expend to new I/O offloads. These APIs can support both Eth and IPoIB
offloads.
o Developers can use common API for OS bypass and share objects like memory
registration, CQs, SRQs for both RDMA and user mode Eth.
o New introduced RSS objects present generic approach of breaking QP into
transport and work queues allowing flexible future expansion.
o Object oriented approach taken with CQ can be taken with posting work queue.
o Future vendor specific capabilities can be implemented using vendor
specific new kernel ABI support and user mode vendor API query through
libibverbs.
· Dual Licensing by Susan Coulter
o OFA can’t and won’t enforce dual-license on kernel space.
o User space will continue to be under dual-license.
It was great event and we are looking forward to see you next year.
Christoph and Leon.
[1] http://marc.info/?l=linux-rdma&m=147767504329945&w=2
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply
* libhns: userspace library for hns
From: Leon Romanovsky @ 2016-11-16 7:58 UTC (permalink / raw)
To: Lijun Ou
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA,
linuxarm-hv44wF8Li93QT0dZR+AlfA, Jason Gunthorpe
[-- Attachment #1: Type: text/plain, Size: 765 bytes --]
Hi Lijun,
I realized that didn't send this notice, while accepted your libhns into
rdma-core, so I'm sending it now to be sure that it is not missed by any
future reviewers.
---
We accepted your libhns code and didn't continue to pursue you to
correctly parse DT strings [1], mainly for the following reasons:
1. It is your private library and support of angry customers will be on you.
2. We won't accept hacks and you will be required to implement it correctly
and handle all incompatibility in out-of-tree code.
3. Once your new hns2 arrive, it should be implemented in the scope of existing
libhns. We wouldn't allow to create libhns2 without real reason (different HW).
Hope it clears now.
[1] http://marc.info/?l=linux-rdma&m=147879905628034&w=2
Thanks
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply
* Re: [PATCH for-next 03/11] IB/hns: Optimize the logic of allocating memory using APIs
From: Leon Romanovsky @ 2016-11-16 8:36 UTC (permalink / raw)
To: Salil Mehta
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org, Huwei (Xavier),
oulijun, mehta.salil.lnk-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org,
linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
netdev-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Linuxarm,
Zhangping (ZP)
In-Reply-To: <F4CC6FACFEB3C54C9141D49AD221F7F91A7A2371@lhreml503-mbx>
[-- Attachment #1: Type: text/plain, Size: 3338 bytes --]
On Tue, Nov 15, 2016 at 03:52:46PM +0000, Salil Mehta wrote:
> > -----Original Message-----
> > From: Leon Romanovsky [mailto:leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org]
> > Sent: Wednesday, November 09, 2016 7:22 AM
> > To: Salil Mehta
> > Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org; Huwei (Xavier); oulijun;
> > mehta.salil.lnk-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org; linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org;
> > netdev-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; Linuxarm;
> > Zhangping (ZP)
> > Subject: Re: [PATCH for-next 03/11] IB/hns: Optimize the logic of
> > allocating memory using APIs
> >
> > On Fri, Nov 04, 2016 at 04:36:25PM +0000, Salil Mehta wrote:
> > > From: "Wei Hu (Xavier)" <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
> > >
> > > This patch modified the logic of allocating memory using APIs in
> > > hns RoCE driver. We used kcalloc instead of kmalloc_array and
> > > bitmap_zero. And When kcalloc failed, call vzalloc to alloc
> > > memory.
> > >
> > > Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
> > > Signed-off-by: Ping Zhang <zhangping5-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
> > > Signed-off-by: Salil Mehta <salil.mehta-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
> > > ---
> > > drivers/infiniband/hw/hns/hns_roce_mr.c | 15 ++++++++-------
> > > 1 file changed, 8 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c
> > b/drivers/infiniband/hw/hns/hns_roce_mr.c
> > > index fb87883..d3dfb5f 100644
> > > --- a/drivers/infiniband/hw/hns/hns_roce_mr.c
> > > +++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
> > > @@ -137,11 +137,12 @@ static int hns_roce_buddy_init(struct
> > hns_roce_buddy *buddy, int max_order)
> > >
> > > for (i = 0; i <= buddy->max_order; ++i) {
> > > s = BITS_TO_LONGS(1 << (buddy->max_order - i));
> > > - buddy->bits[i] = kmalloc_array(s, sizeof(long),
> > GFP_KERNEL);
> > > - if (!buddy->bits[i])
> > > - goto err_out_free;
> > > -
> > > - bitmap_zero(buddy->bits[i], 1 << (buddy->max_order - i));
> > > + buddy->bits[i] = kcalloc(s, sizeof(long), GFP_KERNEL);
> > > + if (!buddy->bits[i]) {
> > > + buddy->bits[i] = vzalloc(s * sizeof(long));
> >
> > I wonder, why don't you use directly vzalloc instead of kcalloc
> > fallback?
> As we know we will have physical contiguous pages if the kcalloc
> call succeeds. This will give us a chance to have better performance
> over the allocations which are just virtually contiguous through the
> function vzalloc(). Therefore, later has only been used as a fallback
> when our memory request cannot be entertained through kcalloc.
>
> Are you suggesting that there will not be much performance penalty
> if we use just vzalloc ?
Not exactly,
I asked it, because we have similar code in our drivers and this
construction looks strange to me.
1. If performance is critical, we will use kmalloc.
2. If performance is not critical, we will use vmalloc.
But in this case, such construction shows me that we can live with
vmalloc performance and kmalloc allocation are not really needed.
In your specific case, I'm not sure that kcalloc will ever fail.
Thanks
>
> >
> > > + if (!buddy->bits[i])
> > > + goto err_out_free;
> > > + }
> > > }
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply
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