Skip to content

Commit de019a9

Browse files
yishaihdledford
yishaih
authored and
dledford
committed
IB/uverbs: Introduce RWQ Indirection table
User applications that want to spread traffic on several WQs, need to create an indirection table, by using already created WQs. Adding uverbs API in order to create and destroy this table. Signed-off-by: Yishai Hadas <[email protected]> Signed-off-by: Matan Barak <[email protected]> Reviewed-by: Sagi Grimberg <[email protected]> Signed-off-by: Doug Ledford <[email protected]>
1 parent 6d39786 commit de019a9

File tree

5 files changed

+253
-0
lines changed

5 files changed

+253
-0
lines changed

‎drivers/infiniband/core/uverbs.h

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -186,6 +186,7 @@ extern struct idr ib_uverbs_srq_idr;
186186
extern struct idr ib_uverbs_xrcd_idr;
187187
extern struct idr ib_uverbs_rule_idr;
188188
extern struct idr ib_uverbs_wq_idr;
189+
extern struct idr ib_uverbs_rwq_ind_tbl_idr;
189190

190191
void idr_remove_uobj(struct idr *idp, struct ib_uobject *uobj);
191192

@@ -284,5 +285,7 @@ IB_UVERBS_DECLARE_EX_CMD(create_qp);
284285
IB_UVERBS_DECLARE_EX_CMD(create_wq);
285286
IB_UVERBS_DECLARE_EX_CMD(modify_wq);
286287
IB_UVERBS_DECLARE_EX_CMD(destroy_wq);
288+
IB_UVERBS_DECLARE_EX_CMD(create_rwq_ind_table);
289+
IB_UVERBS_DECLARE_EX_CMD(destroy_rwq_ind_table);
287290

288291
#endif /* UVERBS_H */

‎drivers/infiniband/core/uverbs_cmd.c

Lines changed: 210 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -58,6 +58,7 @@ static struct uverbs_lock_class srq_lock_class = { .name = "SRQ-uobj" };
5858
static struct uverbs_lock_class xrcd_lock_class = { .name = "XRCD-uobj" };
5959
static struct uverbs_lock_class rule_lock_class = { .name = "RULE-uobj" };
6060
static struct uverbs_lock_class wq_lock_class = { .name = "WQ-uobj" };
61+
static struct uverbs_lock_class rwq_ind_table_lock_class = { .name = "IND_TBL-uobj" };
6162

6263
/*
6364
* The ib_uobject locking scheme is as follows:
@@ -338,6 +339,7 @@ ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
338339
INIT_LIST_HEAD(&ucontext->srq_list);
339340
INIT_LIST_HEAD(&ucontext->ah_list);
340341
INIT_LIST_HEAD(&ucontext->wq_list);
342+
INIT_LIST_HEAD(&ucontext->rwq_ind_tbl_list);
341343
INIT_LIST_HEAD(&ucontext->xrcd_list);
342344
INIT_LIST_HEAD(&ucontext->rule_list);
343345
rcu_read_lock();
@@ -3299,6 +3301,214 @@ int ib_uverbs_ex_modify_wq(struct ib_uverbs_file *file,
32993301
return ret;
33003302
}
33013303

3304+
int ib_uverbs_ex_create_rwq_ind_table(struct ib_uverbs_file *file,
3305+
struct ib_device *ib_dev,
3306+
struct ib_udata *ucore,
3307+
struct ib_udata *uhw)
3308+
{
3309+
struct ib_uverbs_ex_create_rwq_ind_table cmd = {};
3310+
struct ib_uverbs_ex_create_rwq_ind_table_resp resp = {};
3311+
struct ib_uobject *uobj;
3312+
int err = 0;
3313+
struct ib_rwq_ind_table_init_attr init_attr = {};
3314+
struct ib_rwq_ind_table *rwq_ind_tbl;
3315+
struct ib_wq **wqs = NULL;
3316+
u32 *wqs_handles = NULL;
3317+
struct ib_wq *wq = NULL;
3318+
int i, j, num_read_wqs;
3319+
u32 num_wq_handles;
3320+
u32 expected_in_size;
3321+
size_t required_cmd_sz_header;
3322+
size_t required_resp_len;
3323+
3324+
required_cmd_sz_header = offsetof(typeof(cmd), log_ind_tbl_size) + sizeof(cmd.log_ind_tbl_size);
3325+
required_resp_len = offsetof(typeof(resp), ind_tbl_num) + sizeof(resp.ind_tbl_num);
3326+
3327+
if (ucore->inlen < required_cmd_sz_header)
3328+
return -EINVAL;
3329+
3330+
if (ucore->outlen < required_resp_len)
3331+
return -ENOSPC;
3332+
3333+
err = ib_copy_from_udata(&cmd, ucore, required_cmd_sz_header);
3334+
if (err)
3335+
return err;
3336+
3337+
ucore->inbuf += required_cmd_sz_header;
3338+
ucore->inlen -= required_cmd_sz_header;
3339+
3340+
if (cmd.comp_mask)
3341+
return -EOPNOTSUPP;
3342+
3343+
if (cmd.log_ind_tbl_size > IB_USER_VERBS_MAX_LOG_IND_TBL_SIZE)
3344+
return -EINVAL;
3345+
3346+
num_wq_handles = 1 << cmd.log_ind_tbl_size;
3347+
expected_in_size = num_wq_handles * sizeof(__u32);
3348+
if (num_wq_handles == 1)
3349+
/* input size for wq handles is u64 aligned */
3350+
expected_in_size += sizeof(__u32);
3351+
3352+
if (ucore->inlen < expected_in_size)
3353+
return -EINVAL;
3354+
3355+
if (ucore->inlen > expected_in_size &&
3356+
!ib_is_udata_cleared(ucore, expected_in_size,
3357+
ucore->inlen - expected_in_size))
3358+
return -EOPNOTSUPP;
3359+
3360+
wqs_handles = kcalloc(num_wq_handles, sizeof(*wqs_handles),
3361+
GFP_KERNEL);
3362+
if (!wqs_handles)
3363+
return -ENOMEM;
3364+
3365+
err = ib_copy_from_udata(wqs_handles, ucore,
3366+
num_wq_handles * sizeof(__u32));
3367+
if (err)
3368+
goto err_free;
3369+
3370+
wqs = kcalloc(num_wq_handles, sizeof(*wqs), GFP_KERNEL);
3371+
if (!wqs) {
3372+
err = -ENOMEM;
3373+
goto err_free;
3374+
}
3375+
3376+
for (num_read_wqs = 0; num_read_wqs < num_wq_handles;
3377+
num_read_wqs++) {
3378+
wq = idr_read_wq(wqs_handles[num_read_wqs], file->ucontext);
3379+
if (!wq) {
3380+
err = -EINVAL;
3381+
goto put_wqs;
3382+
}
3383+
3384+
wqs[num_read_wqs] = wq;
3385+
}
3386+
3387+
uobj = kmalloc(sizeof(*uobj), GFP_KERNEL);
3388+
if (!uobj) {
3389+
err = -ENOMEM;
3390+
goto put_wqs;
3391+
}
3392+
3393+
init_uobj(uobj, 0, file->ucontext, &rwq_ind_table_lock_class);
3394+
down_write(&uobj->mutex);
3395+
init_attr.log_ind_tbl_size = cmd.log_ind_tbl_size;
3396+
init_attr.ind_tbl = wqs;
3397+
rwq_ind_tbl = ib_dev->create_rwq_ind_table(ib_dev, &init_attr, uhw);
3398+
3399+
if (IS_ERR(rwq_ind_tbl)) {
3400+
err = PTR_ERR(rwq_ind_tbl);
3401+
goto err_uobj;
3402+
}
3403+
3404+
rwq_ind_tbl->ind_tbl = wqs;
3405+
rwq_ind_tbl->log_ind_tbl_size = init_attr.log_ind_tbl_size;
3406+
rwq_ind_tbl->uobject = uobj;
3407+
uobj->object = rwq_ind_tbl;
3408+
rwq_ind_tbl->device = ib_dev;
3409+
atomic_set(&rwq_ind_tbl->usecnt, 0);
3410+
3411+
for (i = 0; i < num_wq_handles; i++)
3412+
atomic_inc(&wqs[i]->usecnt);
3413+
3414+
err = idr_add_uobj(&ib_uverbs_rwq_ind_tbl_idr, uobj);
3415+
if (err)
3416+
goto destroy_ind_tbl;
3417+
3418+
resp.ind_tbl_handle = uobj->id;
3419+
resp.ind_tbl_num = rwq_ind_tbl->ind_tbl_num;
3420+
resp.response_length = required_resp_len;
3421+
3422+
err = ib_copy_to_udata(ucore,
3423+
&resp, resp.response_length);
3424+
if (err)
3425+
goto err_copy;
3426+
3427+
kfree(wqs_handles);
3428+
3429+
for (j = 0; j < num_read_wqs; j++)
3430+
put_wq_read(wqs[j]);
3431+
3432+
mutex_lock(&file->mutex);
3433+
list_add_tail(&uobj->list, &file->ucontext->rwq_ind_tbl_list);
3434+
mutex_unlock(&file->mutex);
3435+
3436+
uobj->live = 1;
3437+
3438+
up_write(&uobj->mutex);
3439+
return 0;
3440+
3441+
err_copy:
3442+
idr_remove_uobj(&ib_uverbs_rwq_ind_tbl_idr, uobj);
3443+
destroy_ind_tbl:
3444+
ib_destroy_rwq_ind_table(rwq_ind_tbl);
3445+
err_uobj:
3446+
put_uobj_write(uobj);
3447+
put_wqs:
3448+
for (j = 0; j < num_read_wqs; j++)
3449+
put_wq_read(wqs[j]);
3450+
err_free:
3451+
kfree(wqs_handles);
3452+
kfree(wqs);
3453+
return err;
3454+
}
3455+
3456+
int ib_uverbs_ex_destroy_rwq_ind_table(struct ib_uverbs_file *file,
3457+
struct ib_device *ib_dev,
3458+
struct ib_udata *ucore,
3459+
struct ib_udata *uhw)
3460+
{
3461+
struct ib_uverbs_ex_destroy_rwq_ind_table cmd = {};
3462+
struct ib_rwq_ind_table *rwq_ind_tbl;
3463+
struct ib_uobject *uobj;
3464+
int ret;
3465+
struct ib_wq **ind_tbl;
3466+
size_t required_cmd_sz;
3467+
3468+
required_cmd_sz = offsetof(typeof(cmd), ind_tbl_handle) + sizeof(cmd.ind_tbl_handle);
3469+
3470+
if (ucore->inlen < required_cmd_sz)
3471+
return -EINVAL;
3472+
3473+
if (ucore->inlen > sizeof(cmd) &&
3474+
!ib_is_udata_cleared(ucore, sizeof(cmd),
3475+
ucore->inlen - sizeof(cmd)))
3476+
return -EOPNOTSUPP;
3477+
3478+
ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3479+
if (ret)
3480+
return ret;
3481+
3482+
if (cmd.comp_mask)
3483+
return -EOPNOTSUPP;
3484+
3485+
uobj = idr_write_uobj(&ib_uverbs_rwq_ind_tbl_idr, cmd.ind_tbl_handle,
3486+
file->ucontext);
3487+
if (!uobj)
3488+
return -EINVAL;
3489+
rwq_ind_tbl = uobj->object;
3490+
ind_tbl = rwq_ind_tbl->ind_tbl;
3491+
3492+
ret = ib_destroy_rwq_ind_table(rwq_ind_tbl);
3493+
if (!ret)
3494+
uobj->live = 0;
3495+
3496+
put_uobj_write(uobj);
3497+
3498+
if (ret)
3499+
return ret;
3500+
3501+
idr_remove_uobj(&ib_uverbs_rwq_ind_tbl_idr, uobj);
3502+
3503+
mutex_lock(&file->mutex);
3504+
list_del(&uobj->list);
3505+
mutex_unlock(&file->mutex);
3506+
3507+
put_uobj(uobj);
3508+
kfree(ind_tbl);
3509+
return ret;
3510+
}
3511+
33023512
int ib_uverbs_ex_create_flow(struct ib_uverbs_file *file,
33033513
struct ib_device *ib_dev,
33043514
struct ib_udata *ucore,

‎drivers/infiniband/core/uverbs_main.c

Lines changed: 13 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -77,6 +77,7 @@ DEFINE_IDR(ib_uverbs_srq_idr);
7777
DEFINE_IDR(ib_uverbs_xrcd_idr);
7878
DEFINE_IDR(ib_uverbs_rule_idr);
7979
DEFINE_IDR(ib_uverbs_wq_idr);
80+
DEFINE_IDR(ib_uverbs_rwq_ind_tbl_idr);
8081

8182
static DEFINE_SPINLOCK(map_lock);
8283
static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
@@ -134,6 +135,8 @@ static int (*uverbs_ex_cmd_table[])(struct ib_uverbs_file *file,
134135
[IB_USER_VERBS_EX_CMD_CREATE_WQ] = ib_uverbs_ex_create_wq,
135136
[IB_USER_VERBS_EX_CMD_MODIFY_WQ] = ib_uverbs_ex_modify_wq,
136137
[IB_USER_VERBS_EX_CMD_DESTROY_WQ] = ib_uverbs_ex_destroy_wq,
138+
[IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL] = ib_uverbs_ex_create_rwq_ind_table,
139+
[IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL] = ib_uverbs_ex_destroy_rwq_ind_table,
137140
};
138141

139142
static void ib_uverbs_add_one(struct ib_device *device);
@@ -269,6 +272,16 @@ static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
269272
kfree(uqp);
270273
}
271274

275+
list_for_each_entry_safe(uobj, tmp, &context->rwq_ind_tbl_list, list) {
276+
struct ib_rwq_ind_table *rwq_ind_tbl = uobj->object;
277+
struct ib_wq **ind_tbl = rwq_ind_tbl->ind_tbl;
278+
279+
idr_remove_uobj(&ib_uverbs_rwq_ind_tbl_idr, uobj);
280+
ib_destroy_rwq_ind_table(rwq_ind_tbl);
281+
kfree(ind_tbl);
282+
kfree(uobj);
283+
}
284+
272285
list_for_each_entry_safe(uobj, tmp, &context->wq_list, list) {
273286
struct ib_wq *wq = uobj->object;
274287
struct ib_uwq_object *uwq =

‎include/rdma/ib_verbs.h

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1326,6 +1326,7 @@ struct ib_ucontext {
13261326
struct list_head xrcd_list;
13271327
struct list_head rule_list;
13281328
struct list_head wq_list;
1329+
struct list_head rwq_ind_tbl_list;
13291330
int closing;
13301331

13311332
struct pid *tgid;

‎include/uapi/rdma/ib_user_verbs.h

Lines changed: 26 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -98,6 +98,8 @@ enum {
9898
IB_USER_VERBS_EX_CMD_CREATE_WQ,
9999
IB_USER_VERBS_EX_CMD_MODIFY_WQ,
100100
IB_USER_VERBS_EX_CMD_DESTROY_WQ,
101+
IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL,
102+
IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL
101103
};
102104

103105
/*
@@ -987,4 +989,28 @@ struct ib_uverbs_ex_modify_wq {
987989
__u32 curr_wq_state;
988990
};
989991

992+
/* Prevent memory allocation rather than max expected size */
993+
#define IB_USER_VERBS_MAX_LOG_IND_TBL_SIZE 0x0d
994+
struct ib_uverbs_ex_create_rwq_ind_table {
995+
__u32 comp_mask;
996+
__u32 log_ind_tbl_size;
997+
/* Following are the wq handles according to log_ind_tbl_size
998+
* wq_handle1
999+
* wq_handle2
1000+
*/
1001+
__u32 wq_handles[0];
1002+
};
1003+
1004+
struct ib_uverbs_ex_create_rwq_ind_table_resp {
1005+
__u32 comp_mask;
1006+
__u32 response_length;
1007+
__u32 ind_tbl_handle;
1008+
__u32 ind_tbl_num;
1009+
};
1010+
1011+
struct ib_uverbs_ex_destroy_rwq_ind_table {
1012+
__u32 comp_mask;
1013+
__u32 ind_tbl_handle;
1014+
};
1015+
9901016
#endif /* IB_USER_VERBS_H */

0 commit comments

Comments
 (0)