iv/linux
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

RDMA/hns: Support getting max QP number from firmware

Browse Source 
All functions of HIP09's ROCEE share on-chip resources for all QPs, the
driver needs configure the resource index and number for each function
during the init stage.

Link: https://lore.kernel.org/r/1622541427-42193-1-git-send-email-liweihang@huawei.com
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
This commit is contained in:
Xi Wang 2021-06-01 17:57:07 +08:00 committed by Jason Gunthorpe
parent 7ce6095e3b
commit 61b460d100
6 changed files with 107 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
drivers/infiniband/hw/hns/hns_roce_device.h
View File

@ -262,8 +262,6 @@ struct hns_roce_hem_table {
u32 type; u32 type;
/* HEM array elment num */ /* HEM array elment num */
unsigned long num_hem; unsigned long num_hem;
/* HEM entry record obj total num */
unsigned long num_obj;
/* Single obj size */ /* Single obj size */
unsigned long obj_size; unsigned long obj_size;
unsigned long table_chunk_size; unsigned long table_chunk_size;
@ -742,6 +740,7 @@ struct hns_roce_caps {
u32 max_rq_sg; u32 max_rq_sg;
u32 max_extend_sg; u32 max_extend_sg;
u32 num_qps; u32 num_qps;
u32 num_pi_qps;
u32 reserved_qps; u32 reserved_qps;
int num_qpc_timer; int num_qpc_timer;
int num_cqc_timer; int num_cqc_timer;
@ -1048,7 +1047,7 @@ static inline void hns_roce_write64_k(__le32 val[2], void __iomem *dest)
static inline struct hns_roce_qp static inline struct hns_roce_qp
*__hns_roce_qp_lookup(struct hns_roce_dev *hr_dev, u32 qpn) *__hns_roce_qp_lookup(struct hns_roce_dev *hr_dev, u32 qpn)
{ {
return xa_load(&hr_dev->qp_table_xa, qpn & (hr_dev->caps.num_qps - 1)); return xa_load(&hr_dev->qp_table_xa, qpn);
} }
static inline void *hns_roce_buf_offset(struct hns_roce_buf *buf, static inline void *hns_roce_buf_offset(struct hns_roce_buf *buf,

25
drivers/infiniband/hw/hns/hns_roce_hem.c
View File

@ -224,8 +224,7 @@ int hns_roce_calc_hem_mhop(struct hns_roce_dev *hr_dev,
chunk_ba_num = mhop->bt_chunk_size / BA_BYTE_LEN; chunk_ba_num = mhop->bt_chunk_size / BA_BYTE_LEN;
chunk_size = table->type < HEM_TYPE_MTT ? mhop->buf_chunk_size : chunk_size = table->type < HEM_TYPE_MTT ? mhop->buf_chunk_size :
mhop->bt_chunk_size; mhop->bt_chunk_size;
table_idx = (*obj & (table->num_obj - 1)) / table_idx = *obj / (chunk_size / table->obj_size);
(chunk_size / table->obj_size);
switch (bt_num) { switch (bt_num) {
case 3: case 3:
mhop->l2_idx = table_idx & (chunk_ba_num - 1); mhop->l2_idx = table_idx & (chunk_ba_num - 1);
@ -578,8 +577,7 @@ int hns_roce_table_get(struct hns_roce_dev *hr_dev,
if (hns_roce_check_whether_mhop(hr_dev, table->type)) if (hns_roce_check_whether_mhop(hr_dev, table->type))
return hns_roce_table_mhop_get(hr_dev, table, obj); return hns_roce_table_mhop_get(hr_dev, table, obj);
i = (obj & (table->num_obj - 1)) / (table->table_chunk_size / i = obj / (table->table_chunk_size / table->obj_size);
table->obj_size);
mutex_lock(&table->mutex); mutex_lock(&table->mutex);
@ -697,8 +695,7 @@ void hns_roce_table_put(struct hns_roce_dev *hr_dev,
return; return;
} }
i = (obj & (table->num_obj - 1)) / i = obj / (table->table_chunk_size / table->obj_size);
(table->table_chunk_size / table->obj_size);
if (!refcount_dec_and_mutex_lock(&table->hem[i]->refcount, if (!refcount_dec_and_mutex_lock(&table->hem[i]->refcount,
&table->mutex)) &table->mutex))
@ -736,8 +733,8 @@ void *hns_roce_table_find(struct hns_roce_dev *hr_dev,
if (!hns_roce_check_whether_mhop(hr_dev, table->type)) { if (!hns_roce_check_whether_mhop(hr_dev, table->type)) {
obj_per_chunk = table->table_chunk_size / table->obj_size; obj_per_chunk = table->table_chunk_size / table->obj_size;
hem = table->hem[(obj & (table->num_obj - 1)) / obj_per_chunk]; hem = table->hem[obj / obj_per_chunk];
idx_offset = (obj & (table->num_obj - 1)) % obj_per_chunk; idx_offset = obj % obj_per_chunk;
dma_offset = offset = idx_offset * table->obj_size; dma_offset = offset = idx_offset * table->obj_size;
} else { } else {
u32 seg_size = 64; /* 8 bytes per BA and 8 BA per segment */ u32 seg_size = 64; /* 8 bytes per BA and 8 BA per segment */
@ -754,8 +751,7 @@ void *hns_roce_table_find(struct hns_roce_dev *hr_dev,
hem_idx = i; hem_idx = i;
hem = table->hem[hem_idx]; hem = table->hem[hem_idx];
dma_offset = offset = (obj & (table->num_obj - 1)) * seg_size % dma_offset = offset = obj * seg_size % mhop.bt_chunk_size;
mhop.bt_chunk_size;
if (mhop.hop_num == 2) if (mhop.hop_num == 2)
dma_offset = offset = 0; dma_offset = offset = 0;
} }
@ -797,7 +793,7 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
if (!hns_roce_check_whether_mhop(hr_dev, type)) { if (!hns_roce_check_whether_mhop(hr_dev, type)) {
table->table_chunk_size = hr_dev->caps.chunk_sz; table->table_chunk_size = hr_dev->caps.chunk_sz;
obj_per_chunk = table->table_chunk_size / obj_size; obj_per_chunk = table->table_chunk_size / obj_size;
num_hem = (nobj + obj_per_chunk - 1) / obj_per_chunk; num_hem = DIV_ROUND_UP(nobj, obj_per_chunk);
table->hem = kcalloc(num_hem, sizeof(*table->hem), GFP_KERNEL); table->hem = kcalloc(num_hem, sizeof(*table->hem), GFP_KERNEL);
if (!table->hem) if (!table->hem)
@ -819,8 +815,9 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
hop_num = mhop.hop_num; hop_num = mhop.hop_num;
obj_per_chunk = buf_chunk_size / obj_size; obj_per_chunk = buf_chunk_size / obj_size;
num_hem = (nobj + obj_per_chunk - 1) / obj_per_chunk; num_hem = DIV_ROUND_UP(nobj, obj_per_chunk);
bt_chunk_num = bt_chunk_size / BA_BYTE_LEN; bt_chunk_num = bt_chunk_size / BA_BYTE_LEN;
if (type >= HEM_TYPE_MTT) if (type >= HEM_TYPE_MTT)
num_bt_l0 = bt_chunk_num; num_bt_l0 = bt_chunk_num;
@ -832,8 +829,7 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
if (check_whether_bt_num_3(type, hop_num)) { if (check_whether_bt_num_3(type, hop_num)) {
unsigned long num_bt_l1; unsigned long num_bt_l1;
num_bt_l1 = (num_hem + bt_chunk_num - 1) / num_bt_l1 = DIV_ROUND_UP(num_hem, bt_chunk_num);
bt_chunk_num;
table->bt_l1 = kcalloc(num_bt_l1, table->bt_l1 = kcalloc(num_bt_l1,
sizeof(*table->bt_l1), sizeof(*table->bt_l1),
GFP_KERNEL); GFP_KERNEL);
@ -865,7 +861,6 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
table->type = type; table->type = type;
table->num_hem = num_hem; table->num_hem = num_hem;
table->num_obj = nobj;
table->obj_size = obj_size; table->obj_size = obj_size;
table->lowmem = use_lowmem; table->lowmem = use_lowmem;
mutex_init(&table->mutex); mutex_init(&table->mutex);

3
drivers/infiniband/hw/hns/hns_roce_hw_v1.c
View File

@ -479,8 +479,7 @@ static int hns_roce_v1_set_hem(struct hns_roce_dev *hr_dev,
long end; long end;
/* Find the HEM(Hardware Entry Memory) entry */ /* Find the HEM(Hardware Entry Memory) entry */
unsigned long i = (obj & (table->num_obj - 1)) / unsigned long i = obj / (table->table_chunk_size / table->obj_size);
(table->table_chunk_size / table->obj_size);
switch (table->type) { switch (table->type) {
case HEM_TYPE_QPC: case HEM_TYPE_QPC:

90
drivers/infiniband/hw/hns/hns_roce_hw_v2.c
View File

@ -1675,6 +1675,29 @@ static int load_func_res_caps(struct hns_roce_dev *hr_dev, bool is_vf)
return 0; return 0;
} }
static int load_ext_cfg_caps(struct hns_roce_dev *hr_dev, bool is_vf)
{
struct hns_roce_cmq_desc desc;
struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
struct hns_roce_caps *caps = &hr_dev->caps;
u32 func_num, qp_num;
int ret;
hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_EXT_CFG, true);
ret = hns_roce_cmq_send(hr_dev, &desc, 1);
if (ret)
return ret;
func_num = is_vf ? 1 : max_t(u32, 1, hr_dev->func_num);
qp_num = hr_reg_read(req, EXT_CFG_QP_PI_NUM) / func_num;
caps->num_pi_qps = round_down(qp_num, HNS_ROCE_QP_BANK_NUM);
qp_num = hr_reg_read(req, EXT_CFG_QP_NUM) / func_num;
caps->num_qps = round_down(qp_num, HNS_ROCE_QP_BANK_NUM);
return 0;
}
static int load_pf_timer_res_caps(struct hns_roce_dev *hr_dev) static int load_pf_timer_res_caps(struct hns_roce_dev *hr_dev)
{ {
struct hns_roce_cmq_desc desc; struct hns_roce_cmq_desc desc;
@ -1695,27 +1718,48 @@ static int load_pf_timer_res_caps(struct hns_roce_dev *hr_dev)
return 0; return 0;
} }
static int query_func_resource_caps(struct hns_roce_dev *hr_dev, bool is_vf)
{
struct device *dev = hr_dev->dev;
int ret;
ret = load_func_res_caps(hr_dev, is_vf);
if (ret) {
dev_err(dev, "failed to load res caps, ret = %d (%s).\n", ret,
is_vf ? "vf" : "pf");
return ret;
}
if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
ret = load_ext_cfg_caps(hr_dev, is_vf);
if (ret)
dev_err(dev, "failed to load ext cfg, ret = %d (%s).\n",
ret, is_vf ? "vf" : "pf");
}
return ret;
}
static int hns_roce_query_pf_resource(struct hns_roce_dev *hr_dev) static int hns_roce_query_pf_resource(struct hns_roce_dev *hr_dev)
{ {
struct device *dev = hr_dev->dev; struct device *dev = hr_dev->dev;
int ret; int ret;
ret = load_func_res_caps(hr_dev, false); ret = query_func_resource_caps(hr_dev, false);
if (ret) { if (ret)
dev_err(dev, "failed to load func caps, ret = %d.\n", ret);
return ret; return ret;
}
ret = load_pf_timer_res_caps(hr_dev); ret = load_pf_timer_res_caps(hr_dev);
if (ret) if (ret)
dev_err(dev, "failed to load timer res, ret = %d.\n", ret); dev_err(dev, "failed to load pf timer resource, ret = %d.\n",
ret);
return ret; return ret;
} }
static int hns_roce_query_vf_resource(struct hns_roce_dev *hr_dev) static int hns_roce_query_vf_resource(struct hns_roce_dev *hr_dev)
{ {
return load_func_res_caps(hr_dev, true); return query_func_resource_caps(hr_dev, true);
} }
static int __hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev, static int __hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev,
@ -1802,6 +1846,24 @@ static int config_vf_hem_resource(struct hns_roce_dev *hr_dev, int vf_id)
return hns_roce_cmq_send(hr_dev, desc, 2); return hns_roce_cmq_send(hr_dev, desc, 2);
} }
static int config_vf_ext_resource(struct hns_roce_dev *hr_dev, u32 vf_id)
{
struct hns_roce_cmq_desc desc;
struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
struct hns_roce_caps *caps = &hr_dev->caps;
hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_EXT_CFG, false);
hr_reg_write(req, EXT_CFG_VF_ID, vf_id);
hr_reg_write(req, EXT_CFG_QP_PI_NUM, caps->num_pi_qps);
hr_reg_write(req, EXT_CFG_QP_PI_IDX, vf_id * caps->num_pi_qps);
hr_reg_write(req, EXT_CFG_QP_NUM, caps->num_qps);
hr_reg_write(req, EXT_CFG_QP_IDX, vf_id * caps->num_qps);
return hns_roce_cmq_send(hr_dev, &desc, 1);
}
static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev) static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev)
{ {
u32 func_num = max_t(u32, 1, hr_dev->func_num); u32 func_num = max_t(u32, 1, hr_dev->func_num);
@ -1810,8 +1872,22 @@ static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev)
for (vf_id = 0; vf_id < func_num; vf_id++) { for (vf_id = 0; vf_id < func_num; vf_id++) {
ret = config_vf_hem_resource(hr_dev, vf_id); ret = config_vf_hem_resource(hr_dev, vf_id);
if (ret) if (ret) {
dev_err(hr_dev->dev,
"failed to config vf-%u hem res, ret = %d.\n",
vf_id, ret);
return ret; return ret;
}
if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
ret = config_vf_ext_resource(hr_dev, vf_id);
if (ret) {
dev_err(hr_dev->dev,
"failed to config vf-%u ext res, ret = %d.\n",
vf_id, ret);
return ret;
}
}
} }
return 0; return 0;

10
drivers/infiniband/hw/hns/hns_roce_hw_v2.h
View File

@ -253,6 +253,7 @@ enum hns_roce_opcode_type {
HNS_ROCE_OPC_QUERY_VF_RES = 0x850e, HNS_ROCE_OPC_QUERY_VF_RES = 0x850e,
HNS_ROCE_OPC_CFG_GMV_TBL = 0x850f, HNS_ROCE_OPC_CFG_GMV_TBL = 0x850f,
HNS_ROCE_OPC_CFG_GMV_BT = 0x8510, HNS_ROCE_OPC_CFG_GMV_BT = 0x8510,
HNS_ROCE_OPC_EXT_CFG = 0x8512,
HNS_SWITCH_PARAMETER_CFG = 0x1033, HNS_SWITCH_PARAMETER_CFG = 0x1033,
}; };
@ -1344,6 +1345,15 @@ struct hns_roce_func_clear {
#define HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_INTERVAL 40 #define HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_INTERVAL 40
#define HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_FAIL_WAIT 20 #define HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_FAIL_WAIT 20
/* Fields of HNS_ROCE_OPC_EXT_CFG */
#define EXT_CFG_VF_ID CMQ_REQ_FIELD_LOC(31, 0)
#define EXT_CFG_QP_PI_IDX CMQ_REQ_FIELD_LOC(45, 32)
#define EXT_CFG_QP_PI_NUM CMQ_REQ_FIELD_LOC(63, 48)
#define EXT_CFG_QP_NUM CMQ_REQ_FIELD_LOC(87, 64)
#define EXT_CFG_QP_IDX CMQ_REQ_FIELD_LOC(119, 96)
#define EXT_CFG_LLM_IDX CMQ_REQ_FIELD_LOC(139, 128)
#define EXT_CFG_LLM_NUM CMQ_REQ_FIELD_LOC(156, 144)
#define CFG_LLM_A_BA_L CMQ_REQ_FIELD_LOC(31, 0) #define CFG_LLM_A_BA_L CMQ_REQ_FIELD_LOC(31, 0)
#define CFG_LLM_A_BA_H CMQ_REQ_FIELD_LOC(63, 32) #define CFG_LLM_A_BA_H CMQ_REQ_FIELD_LOC(63, 32)
#define CFG_LLM_A_DEPTH CMQ_REQ_FIELD_LOC(76, 64) #define CFG_LLM_A_DEPTH CMQ_REQ_FIELD_LOC(76, 64)

2
drivers/infiniband/hw/hns/hns_roce_qp.c
View File

@ -379,7 +379,7 @@ void hns_roce_qp_remove(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
list_del(&hr_qp->rq_node); list_del(&hr_qp->rq_node);
xa_lock_irqsave(xa, flags); xa_lock_irqsave(xa, flags);
__xa_erase(xa, hr_qp->qpn & (hr_dev->caps.num_qps - 1)); __xa_erase(xa, hr_qp->qpn);
xa_unlock_irqrestore(xa, flags); xa_unlock_irqrestore(xa, flags);
} }