diff options
| author | Linus Torvalds <torvalds@linux-foundation.org> | 2023-02-24 15:11:03 -0800 |
|---|---|---|
| committer | Linus Torvalds <torvalds@linux-foundation.org> | 2023-02-24 15:11:03 -0800 |
| commit | 8cbd92339db08b19b93d1637e5799ff2a8dddfd2 (patch) | |
| tree | 7e62d961f32e8a2a96271029b376f1e8bbd70a7c /drivers/infiniband/hw | |
| parent | 143c7bc6496c886ce5db2a2f9cec580494690170 (diff) | |
| parent | 66fb1d5df6ace316a4a6e2c31e13fc123ea2b644 (diff) | |
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Pull rdma updates from Jason Gunthorpe:
"Quite a small cycle this time, even with the rc8. I suppose everyone
went to sleep over xmas.
- Minor driver updates for hfi1, cxgb4, erdma, hns, irdma, mlx5, siw,
mana
- inline CQE support for hns
- Have mlx5 display device error codes
- Pinned DMABUF support for irdma
- Continued rxe cleanups, particularly converting the MRs to use
xarray
- Improvements to what can be cached in the mlx5 mkey cache"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (61 commits)
IB/mlx5: Extend debug control for CC parameters
IB/hfi1: Fix sdma.h tx->num_descs off-by-one errors
IB/hfi1: Fix math bugs in hfi1_can_pin_pages()
RDMA/irdma: Add support for dmabuf pin memory regions
RDMA/mlx5: Use query_special_contexts for mkeys
net/mlx5e: Use query_special_contexts for mkeys
net/mlx5: Change define name for 0x100 lkey value
net/mlx5: Expose bits for querying special mkeys
RDMA/rxe: Fix missing memory barriers in rxe_queue.h
RDMA/mana_ib: Fix a bug when the PF indicates more entries for registering memory on first packet
RDMA/rxe: Remove rxe_alloc()
RDMA/cma: Distinguish between sockaddr_in and sockaddr_in6 by size
Subject: RDMA/rxe: Handle zero length rdma
iw_cxgb4: Fix potential NULL dereference in c4iw_fill_res_cm_id_entry()
RDMA/mlx5: Use rdma_umem_for_each_dma_block()
RDMA/umem: Remove unused 'work' member from struct ib_umem
RDMA/irdma: Cap MSIX used to online CPUs + 1
RDMA/mlx5: Check reg_create() create for errors
RDMA/restrack: Correct spelling
RDMA/cxgb4: Fix potential null-ptr-deref in pass_establish()
...
Diffstat (limited to 'drivers/infiniband/hw')
39 files changed, 1236 insertions, 737 deletions
diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c index 499a425a3379..ced615b5ea09 100644 --- a/drivers/infiniband/hw/cxgb4/cm.c +++ b/drivers/infiniband/hw/cxgb4/cm.c @@ -2676,6 +2676,9 @@ static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb) u16 tcp_opt = ntohs(req->tcp_opt); ep = get_ep_from_tid(dev, tid); + if (!ep) + return 0; + pr_debug("ep %p tid %u\n", ep, ep->hwtid); ep->snd_seq = be32_to_cpu(req->snd_isn); ep->rcv_seq = be32_to_cpu(req->rcv_isn); @@ -4144,6 +4147,10 @@ static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb) if (neigh->dev->flags & IFF_LOOPBACK) { pdev = ip_dev_find(&init_net, iph->daddr); + if (!pdev) { + pr_err("%s - failed to find device!\n", __func__); + goto free_dst; + } e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, pdev, 0); pi = (struct port_info *)netdev_priv(pdev); diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c index c7e8d7b3baa1..7e2835dcbc1c 100644 --- a/drivers/infiniband/hw/cxgb4/cq.c +++ b/drivers/infiniband/hw/cxgb4/cq.c @@ -767,7 +767,7 @@ static int __c4iw_poll_cq_one(struct c4iw_cq *chp, struct c4iw_qp *qhp, goto out; wc->wr_id = cookie; - wc->qp = qhp ? &qhp->ibqp : NULL; + wc->qp = &qhp->ibqp; wc->vendor_err = CQE_STATUS(&cqe); wc->wc_flags = 0; diff --git a/drivers/infiniband/hw/cxgb4/restrack.c b/drivers/infiniband/hw/cxgb4/restrack.c index ff645b955a08..fd22c85d35f4 100644 --- a/drivers/infiniband/hw/cxgb4/restrack.c +++ b/drivers/infiniband/hw/cxgb4/restrack.c @@ -238,7 +238,7 @@ int c4iw_fill_res_cm_id_entry(struct sk_buff *msg, if (rdma_nl_put_driver_u64_hex(msg, "history", epcp->history)) goto err_cancel_table; - if (epcp->state == LISTEN) { + if (listen_ep) { if (rdma_nl_put_driver_u32(msg, "stid", listen_ep->stid)) goto err_cancel_table; if (rdma_nl_put_driver_u32(msg, "backlog", listen_ep->backlog)) diff --git a/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h index a2f5e29ef226..1f79537fc8d1 100644 --- a/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h +++ b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h @@ -122,9 +122,7 @@ struct fw_ri_dsgl { __be16 nsge; __be32 len0; __be64 addr0; -#ifndef C99_NOT_SUPPORTED struct fw_ri_dsge_pair sge[]; -#endif }; struct fw_ri_sge { @@ -138,9 +136,7 @@ struct fw_ri_isgl { __u8 r1; __be16 nsge; __be32 r2; -#ifndef C99_NOT_SUPPORTED struct fw_ri_sge sge[]; -#endif }; struct fw_ri_immd { @@ -148,9 +144,7 @@ struct fw_ri_immd { __u8 r1; __be16 r2; __be32 immdlen; -#ifndef C99_NOT_SUPPORTED __u8 data[]; -#endif }; struct fw_ri_tpte { @@ -320,9 +314,7 @@ struct fw_ri_res_wr { __be32 op_nres; __be32 len16_pkd; __u64 cookie; -#ifndef C99_NOT_SUPPORTED struct fw_ri_res res[]; -#endif }; #define FW_RI_RES_WR_NRES_S 0 @@ -562,12 +554,10 @@ struct fw_ri_rdma_write_wr { __be32 plen; __be32 stag_sink; __be64 to_sink; -#ifndef C99_NOT_SUPPORTED union { - struct fw_ri_immd immd_src[0]; - struct fw_ri_isgl isgl_src[0]; + DECLARE_FLEX_ARRAY(struct fw_ri_immd, immd_src); + DECLARE_FLEX_ARRAY(struct fw_ri_isgl, isgl_src); } u; -#endif }; struct fw_ri_send_wr { @@ -581,12 +571,10 @@ struct fw_ri_send_wr { __be32 plen; __be32 r3; __be64 r4; -#ifndef C99_NOT_SUPPORTED union { - struct fw_ri_immd immd_src[0]; - struct fw_ri_isgl isgl_src[0]; + DECLARE_FLEX_ARRAY(struct fw_ri_immd, immd_src); + DECLARE_FLEX_ARRAY(struct fw_ri_isgl, isgl_src); } u; -#endif }; #define FW_RI_SEND_WR_SENDOP_S 0 @@ -618,12 +606,10 @@ struct fw_ri_rdma_write_cmpl_wr { struct fw_ri_isgl isgl_src; } u_cmpl; __be64 r3; -#ifndef C99_NOT_SUPPORTED union fw_ri_write { - struct fw_ri_immd immd_src[0]; - struct fw_ri_isgl isgl_src[0]; + DECLARE_FLEX_ARRAY(struct fw_ri_immd, immd_src); + DECLARE_FLEX_ARRAY(struct fw_ri_isgl, isgl_src); } u; -#endif }; struct fw_ri_rdma_read_wr { diff --git a/drivers/infiniband/hw/erdma/erdma_hw.h b/drivers/infiniband/hw/erdma/erdma_hw.h index ab371fec610c..4c38d99c73f1 100644 --- a/drivers/infiniband/hw/erdma/erdma_hw.h +++ b/drivers/infiniband/hw/erdma/erdma_hw.h @@ -397,7 +397,7 @@ struct erdma_write_sqe { __le32 rsvd; - struct erdma_sge sgl[0]; + struct erdma_sge sgl[]; }; struct erdma_send_sqe { @@ -408,7 +408,7 @@ struct erdma_send_sqe { }; __le32 length; - struct erdma_sge sgl[0]; + struct erdma_sge sgl[]; }; struct erdma_readreq_sqe { diff --git a/drivers/infiniband/hw/erdma/erdma_verbs.c b/drivers/infiniband/hw/erdma/erdma_verbs.c index 5dab1e87975b..9c30d78730aa 100644 --- a/drivers/infiniband/hw/erdma/erdma_verbs.c +++ b/drivers/infiniband/hw/erdma/erdma_verbs.c @@ -1110,12 +1110,14 @@ int erdma_mmap(struct ib_ucontext *ctx, struct vm_area_struct *vma) prot = pgprot_device(vma->vm_page_prot); break; default: - return -EINVAL; + err = -EINVAL; + goto put_entry; } err = rdma_user_mmap_io(ctx, vma, PFN_DOWN(entry->address), PAGE_SIZE, prot, rdma_entry); +put_entry: rdma_user_mmap_entry_put(rdma_entry); return err; } diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c index ebe970f76232..90b672feed83 100644 --- a/drivers/infiniband/hw/hfi1/chip.c +++ b/drivers/infiniband/hw/hfi1/chip.c @@ -1056,7 +1056,7 @@ static void read_link_down_reason(struct hfi1_devdata *dd, u8 *ldr); static void handle_temp_err(struct hfi1_devdata *dd); static void dc_shutdown(struct hfi1_devdata *dd); static void dc_start(struct hfi1_devdata *dd); -static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp, +static int qos_rmt_entries(unsigned int n_krcv_queues, unsigned int *mp, unsigned int *np); static void clear_full_mgmt_pkey(struct hfi1_pportdata *ppd); static int wait_link_transfer_active(struct hfi1_devdata *dd, int wait_ms); @@ -13362,7 +13362,6 @@ static int set_up_context_variables(struct hfi1_devdata *dd) int ret; unsigned ngroups; int rmt_count; - int user_rmt_reduced; u32 n_usr_ctxts; u32 send_contexts = chip_send_contexts(dd); u32 rcv_contexts = chip_rcv_contexts(dd); @@ -13421,28 +13420,34 @@ static int set_up_context_variables(struct hfi1_devdata *dd) (num_kernel_contexts + n_usr_ctxts), &node_affinity.real_cpu_mask); /* - * The RMT entries are currently allocated as shown below: - * 1. QOS (0 to 128 entries); - * 2. FECN (num_kernel_context - 1 + num_user_contexts + - * num_netdev_contexts); - * 3. netdev (num_netdev_contexts). - * It should be noted that FECN oversubscribe num_netdev_contexts - * entries of RMT because both netdev and PSM could allocate any receive - * context between dd->first_dyn_alloc_text and dd->num_rcv_contexts, - * and PSM FECN must reserve an RMT entry for each possible PSM receive - * context. + * RMT entries are allocated as follows: + * 1. QOS (0 to 128 entries) + * 2. FECN (num_kernel_context - 1 [a] + num_user_contexts + + * num_netdev_contexts [b]) + * 3. netdev (NUM_NETDEV_MAP_ENTRIES) + * + * Notes: + * [a] Kernel contexts (except control) are included in FECN if kernel + * TID_RDMA is active. + * [b] Netdev and user contexts are randomly allocated from the same + * context pool, so FECN must cover all contexts in the pool. */ - rmt_count = qos_rmt_entries(dd, NULL, NULL) + (num_netdev_contexts * 2); - if (HFI1_CAP_IS_KSET(TID_RDMA)) - rmt_count += num_kernel_contexts - 1; - if (rmt_count + n_usr_ctxts > NUM_MAP_ENTRIES) { - user_rmt_reduced = NUM_MAP_ENTRIES - rmt_count; - dd_dev_err(dd, - "RMT size is reducing the number of user receive contexts from %u to %d\n", - n_usr_ctxts, - user_rmt_reduced); - /* recalculate */ - n_usr_ctxts = user_rmt_reduced; + rmt_count = qos_rmt_entries(num_kernel_contexts - 1, NULL, NULL) + + (HFI1_CAP_IS_KSET(TID_RDMA) ? num_kernel_contexts - 1 + : 0) + + n_usr_ctxts + + num_netdev_contexts + + NUM_NETDEV_MAP_ENTRIES; + if (rmt_count > NUM_MAP_ENTRIES) { + int over = rmt_count - NUM_MAP_ENTRIES; + /* try to squish user contexts, minimum of 1 */ + if (over >= n_usr_ctxts) { + dd_dev_err(dd, "RMT overflow: reduce the requested number of contexts\n"); + return -EINVAL; + } + dd_dev_err(dd, "RMT overflow: reducing # user contexts from %u to %u\n", + n_usr_ctxts, n_usr_ctxts - over); + n_usr_ctxts -= over; } /* the first N are kernel contexts, the rest are user/netdev contexts */ @@ -14299,15 +14304,15 @@ static void clear_rsm_rule(struct hfi1_devdata *dd, u8 rule_index) } /* return the number of RSM map table entries that will be used for QOS */ -static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp, +static int qos_rmt_entries(unsigned int n_krcv_queues, unsigned int *mp, unsigned int *np) { int i; unsigned int m, n; - u8 max_by_vl = 0; + uint max_by_vl = 0; /* is QOS active at all? */ - if (dd->n_krcv_queues <= MIN_KERNEL_KCTXTS || + if (n_krcv_queues < MIN_KERNEL_KCTXTS || num_vls == 1 || krcvqsset <= 1) goto no_qos; @@ -14365,7 +14370,7 @@ static void init_qos(struct hfi1_devdata *dd, struct rsm_map_table *rmt) if (!rmt) goto bail; - rmt_entries = qos_rmt_entries(dd, &m, &n); + rmt_entries = qos_rmt_entries(dd->n_krcv_queues - 1, &m, &n); if (rmt_entries == 0) goto bail; qpns_per_vl = 1 << m; diff --git a/drivers/infiniband/hw/hfi1/exp_rcv.h b/drivers/infiniband/hw/hfi1/exp_rcv.h index c6291bbf723c..41f7fe5d1839 100644 --- a/drivers/infiniband/hw/hfi1/exp_rcv.h +++ b/drivers/infiniband/hw/hfi1/exp_rcv.h @@ -133,12 +133,13 @@ static inline struct tid_group *tid_group_pop(struct exp_tid_set *set) return grp; } -static inline u32 rcventry2tidinfo(u32 rcventry) +static inline u32 create_tid(u32 rcventry, u32 npages) { u32 pair = rcventry & ~0x1; return EXP_TID_SET(IDX, pair >> 1) | - EXP_TID_SET(CTRL, 1 << (rcventry - pair)); + EXP_TID_SET(CTRL, 1 << (rcventry - pair)) | + EXP_TID_SET(LEN, npages); } /** diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c index 47e4c8084207..b1d6ca7e9708 100644 --- a/drivers/infiniband/hw/hfi1/file_ops.c +++ b/drivers/infiniband/hw/hfi1/file_ops.c @@ -306,6 +306,17 @@ static ssize_t hfi1_write_iter(struct kiocb *kiocb, struct iov_iter *from) return reqs; } +static inline void mmap_cdbg(u16 ctxt, u8 subctxt, u8 type, u8 mapio, u8 vmf, + u64 memaddr, void *memvirt, dma_addr_t memdma, + ssize_t memlen, struct vm_area_struct *vma) +{ + hfi1_cdbg(PROC, + "%u:%u type:%u io/vf/dma:%d/%d/%d, addr:0x%llx, len:%lu(%lu), flags:0x%lx", + ctxt, subctxt, type, mapio, vmf, !!memdma, + memaddr ?: (u64)memvirt, memlen, + vma->vm_end - vma->vm_start, vma->vm_flags); +} + static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) { struct hfi1_filedata *fd = fp->private_data; @@ -315,6 +326,7 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) u64 token = vma->vm_pgoff << PAGE_SHIFT, memaddr = 0; void *memvirt = NULL; + dma_addr_t memdma = 0; u8 subctxt, mapio = 0, vmf = 0, type; ssize_t memlen = 0; int ret = 0; @@ -334,6 +346,11 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) goto done; } + /* + * vm_pgoff is used as a buffer selector cookie. Always mmap from + * the beginning. + */ + vma->vm_pgoff = 0; flags = vma->vm_flags; switch (type) { @@ -355,7 +372,8 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot); mapio = 1; break; - case PIO_CRED: + case PIO_CRED: { + u64 cr_page_offset; if (flags & VM_WRITE) { ret = -EPERM; goto done; @@ -365,10 +383,11 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) * second or third page allocated for credit returns (if number * of enabled contexts > 64 and 128 respectively). */ - memvirt = dd->cr_base[uctxt->numa_id].va; - memaddr = virt_to_phys(memvirt) + - (((u64)uctxt->sc->hw_free - - (u64)dd->cr_base[uctxt->numa_id].va) & PAGE_MASK); + cr_page_offset = ((u64)uctxt->sc->hw_free - + (u64)dd->cr_base[uctxt->numa_id].va) & + PAGE_MASK; + memvirt = dd->cr_base[uctxt->numa_id].va + cr_page_offset; + memdma = dd->cr_base[uctxt->numa_id].dma + cr_page_offset; memlen = PAGE_SIZE; flags &= ~VM_MAYWRITE; flags |= VM_DONTCOPY | VM_DONTEXPAND; @@ -378,14 +397,16 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) * memory been flagged as non-cached? */ /* vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); */ - mapio = 1; break; + } case RCV_HDRQ: memlen = rcvhdrq_size(uctxt); memvirt = uctxt->rcvhdrq; + memdma = uctxt->rcvhdrq_dma; break; case RCV_EGRBUF: { - unsigned long addr; + unsigned long vm_start_save; + unsigned long vm_end_save; int i; /* * The RcvEgr buffer need to be handled differently @@ -404,24 +425,34 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) goto done; } vm_flags_clear(vma, VM_MAYWRITE); - addr = vma->vm_start; + /* + * Mmap multiple separate allocations into a single vma. From + * here, dma_mmap_coherent() calls dma_direct_mmap(), which + * requires the mmap to exactly fill the vma starting at + * vma_start. Adjust the vma start and end for each eager + * buffer segment mapped. Restore the originals when done. + */ + vm_start_save = vma->vm_start; + vm_end_save = vma->vm_end; + vma->vm_end = vma->vm_start; for (i = 0 ; i < uctxt->egrbufs.numbufs; i++) { memlen = uctxt->egrbufs.buffers[i].len; memvirt = uctxt->egrbufs.buffers[i].addr; - ret = remap_pfn_range( - vma, addr, - /* - * virt_to_pfn() does the same, but - * it's not available on x86_64 - * when CONFIG_MMU is enabled. - */ - PFN_DOWN(__pa(memvirt)), - memlen, - vma->vm_page_prot); - if (ret < 0) + memdma = uctxt->egrbufs.buffers[i].dma; + vma->vm_end += memlen; + mmap_cdbg(ctxt, subctxt, type, mapio, vmf, memaddr, + memvirt, memdma, memlen, vma); + ret = dma_mmap_coherent(&dd->pcidev->dev, vma, + memvirt, memdma, memlen); + if (ret < 0) { + vma->vm_start = vm_start_save; + vma->vm_end = vm_end_save; goto done; - addr += memlen; + } + vma->vm_start += memlen; } + vma->vm_start = vm_start_save; + vma->vm_end = vm_end_save; ret = 0; goto done; } @@ -481,6 +512,7 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) } memlen = PAGE_SIZE; memvirt = (void *)hfi1_rcvhdrtail_kvaddr(uctxt); + memdma = uctxt->rcvhdrqtailaddr_dma; flags &= ~VM_MAYWRITE; break; case SUBCTXT_UREGS: @@ -529,14 +561,15 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) } vm_flags_reset(vma, flags); - hfi1_cdbg(PROC, - "%u:%u type:%u io/vf:%d/%d, addr:0x%llx, len:%lu(%lu), flags:0x%lx\n", - ctxt, subctxt, type, mapio, vmf, memaddr, memlen, - vma->vm_end - vma->vm_start, vma->vm_flags); + mmap_cdbg(ctxt, subctxt, type, mapio, vmf, memaddr, memvirt, memdma, + memlen, vma); if (vmf) { vma->vm_pgoff = PFN_DOWN(memaddr); vma->vm_ops = &vm_ops; ret = 0; + } else if (memdma) { + ret = dma_mmap_coherent(&dd->pcidev->dev, vma, + memvirt, memdma, memlen); } else if (mapio) { ret = io_remap_pfn_range(vma, vma->vm_start, PFN_DOWN(memaddr), diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c index 24c0f0d257fc..62b6c5020039 100644 --- a/drivers/infiniband/hw/hfi1/init.c +++ b/drivers/infiniband/hw/hfi1/init.c @@ -464,7 +464,7 @@ bail: * * This wrapper is the free function that matches hfi1_create_ctxtdata(). * When a context is done being used (kernel or user), this function is called - * for the "final" put to match the kref init from hf1i_create_ctxtdata(). + * for the "final" put to match the kref init from hfi1_create_ctxtdata(). * Other users of the context do a get/put sequence to make sure that the * structure isn't removed while in use. */ diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c index a95b654f5254..8ed20392e9f0 100644 --- a/drivers/infiniband/hw/hfi1/sdma.c +++ b/drivers/infiniband/hw/hfi1/sdma.c @@ -3160,8 +3160,7 @@ int _pad_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx) { int rval = 0; - tx->num_desc++; - if ((unlikely(tx->num_desc == tx->desc_limit))) { + if ((unlikely(tx->num_desc + 1 == tx->desc_limit))) { rval = _extend_sdma_tx_descs(dd, tx); if (rval) { __sdma_txclean(dd, tx); @@ -3174,6 +3173,7 @@ int _pad_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx) SDMA_MAP_NONE, dd->sdma_pad_phys, sizeof(u32) - (tx->packet_len & (sizeof(u32) - 1))); + tx->num_desc++; _sdma_close_tx(dd, tx); return rval; } diff --git a/drivers/infiniband/hw/hfi1/sdma.h b/drivers/infiniband/hw/hfi1/sdma.h index d8170fcbfbdd..b023fc461bd5 100644 --- a/drivers/infiniband/hw/hfi1/sdma.h +++ b/drivers/infiniband/hw/hfi1/sdma.h @@ -631,14 +631,13 @@ static inline void sdma_txclean(struct hfi1_devdata *dd, struct sdma_txreq *tx) static inline void _sdma_close_tx(struct hfi1_devdata *dd, struct sdma_txreq *tx) { - tx->descp[tx->num_desc].qw[0] |= - SDMA_DESC0_LAST_DESC_FLAG; - tx->descp[tx->num_desc].qw[1] |= - dd->default_desc1; + u16 last_desc = tx->num_desc - 1; + + tx->descp[last_desc].qw[0] |= SDMA_DESC0_LAST_DESC_FLAG; + tx->descp[last_desc].qw[1] |= dd->default_desc1; if (tx->flags & SDMA_TXREQ_F_URGENT) - tx->descp[tx->num_desc].qw[1] |= - (SDMA_DESC1_HEAD_TO_HOST_FLAG | - SDMA_DESC1_INT_REQ_FLAG); + tx->descp[last_desc].qw[1] |= (SDMA_DESC1_HEAD_TO_HOST_FLAG | + SDMA_DESC1_INT_REQ_FLAG); } static inline int _sdma_txadd_daddr( @@ -655,6 +654,7 @@ static inline int _sdma_txadd_daddr( type, addr, len); WARN_ON(len > tx->tlen); + tx->num_desc++; tx->tlen -= len; /* special cases for last */ if (!tx->tlen) { @@ -666,7 +666,6 @@ static inline int _sdma_txadd_daddr( _sdma_close_tx(dd, tx); } } - tx->num_desc++; return rval; } diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.c b/drivers/infiniband/hw/hfi1/user_exp_rcv.c index 350884d5f089..96058baf36ed 100644 --- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c +++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c @@ -27,8 +27,7 @@ static bool tid_cover_invalidate(struct mmu_interval_notifier *mni, const struct mmu_notifier_range *range, unsigned long cur_seq); static int program_rcvarray(struct hfi1_filedata *fd, struct tid_user_buf *, - struct tid_group *grp, - unsigned int start, u16 count, + struct tid_group *grp, u16 count, u32 *tidlist, unsigned int *tididx, unsigned int *pmapped); static int unprogram_rcvarray(struct hfi1_filedata *fd, u32 tidinfo); @@ -250,7 +249,7 @@ int hfi1_user_exp_rcv_setup(struct hfi1_filedata *fd, int ret = 0, need_group = 0, pinned; struct hfi1_ctxtdata *uctxt = fd->uctxt; struct hfi1_devdata *dd = uctxt->dd; - unsigned int ngroups, pageidx = 0, pageset_count, + unsigned int ngroups, pageset_count, tididx = 0, mapped, mapped_pages = 0; u32 *tidlist = NULL; struct tid_user_buf *tidbuf; @@ -332,7 +331,7 @@ int hfi1_user_exp_rcv_setup(struct hfi1_filedata *fd, tid_group_pop(&uctxt->tid_group_list); ret = program_rcvarray(fd, tidbuf, grp, - pageidx, dd->rcv_entries.group_size, + dd->rcv_entries.group_size, tidlist, &tididx, &mapped); /* * If there was a failure to program the RcvArray @@ -348,11 +347,10 @@ int hfi1_user_exp_rcv_setup(struct hfi1_filedata *fd, tid_group_add_tail(grp, &uctxt->tid_full_list); ngroups--; - pageidx += ret; mapped_pages += mapped; } - while (pageidx < pageset_count) { + while (tididx < pageset_count) { struct tid_group *grp, *ptr; /* * If we don't have any partially used tid groups, check @@ -374,11 +372,11 @@ int hfi1_user_exp_rcv_setup(struct hfi1_filedata *fd, */ list_for_each_entry_safe(grp, ptr, &uctxt->tid_used_list.list, list) { - unsigned use = min_t(unsigned, pageset_count - pageidx, + unsigned use = min_t(unsigned, pageset_count - tididx, grp->size - grp->used); ret = program_rcvarray(fd, tidbuf, grp, - pageidx, use, tidlist, + use, tidlist, &tididx, &mapped); if (ret < 0) { hfi1_cdbg(TID, @@ -390,11 +388,10 @@ int hfi1_user_exp_rcv_setup(struct hfi1_filedata *fd, tid_group_move(grp, &uctxt->tid_used_list, &uctxt->tid_full_list); - pageidx += ret; mapped_pages += mapped; need_group = 0; /* Check if we are done so we break out early */ - if (pageidx >= pageset_count) + if (tididx >= pageset_count) break; } else if (WARN_ON(ret == 0)) { /* @@ -638,7 +635,6 @@ static u32 find_phys_blocks(struct tid_user_buf *tidbuf, unsigned int npages) * struct tid_pageset holding information on physically contiguous * chunks from the user buffer), and other fields. * @grp: RcvArray group - * @start: starting index into sets array * @count: number of struct tid_pageset's to program * @tidlist: the array of u32 elements when the information about the * programmed RcvArray entries is to be encoded. @@ -658,14 +654,14 @@ static u32 find_phys_blocks(struct tid_user_buf *tidbuf, unsigned int npages) * number of RcvArray entries programmed. */ static int program_rcvarray(struct hfi1_filedata *fd, struct tid_user_buf *tbuf, - struct tid_group *grp, - unsigned int start, u16 count, + struct tid_group *grp, u16 count, u32 *tidlist, unsigned int *tididx, unsigned int *pmapped) { struct hfi1_ctxtdata *uctxt = fd->uctxt; struct hfi1_devdata *dd = uctxt->dd; u16 idx; + unsigned int start = *tididx; u32 tidinfo = 0, rcventry, useidx = 0; int mapped = 0; @@ -710,8 +706,7 @@ static int program_rcvarray(struct hfi1_filedata *fd, struct tid_user_buf *tbuf, return ret; mapped += npages; - tidinfo = rcventry2tidinfo(rcventry - uctxt->expected_base) | - EXP_TID_SET(LEN, npages); + tidinfo = create_tid(rcventry - uctxt->expected_base, npages); tidlist[(*tididx)++] = tidinfo; grp->used++; grp->map |= 1 << useidx++; @@ -795,20 +790,20 @@ static int unprogram_rcvarray(struct hfi1_filedata *fd, u32 tidinfo) struct hfi1_ctxtdata *uctxt = fd->uctxt; struct hfi1_devdata *dd = uctxt->dd; struct tid_rb_node *node; - u8 tidctrl = EXP_TID_GET(tidinfo, CTRL); + u32 tidctrl = EXP_TID_GET(tidinfo, CTRL); u32 tididx = EXP_TID_GET(tidinfo, IDX) << 1, rcventry; - if (tididx >= uctxt->expected_count) { - dd_dev_err(dd, "Invalid RcvArray entry (%u) index for ctxt %u\n", - tididx, uctxt->ctxt); - return -EINVAL; - } - - if (tidctrl == 0x3) + if (tidctrl == 0x3 || tidctrl == 0x0) return -EINVAL; rcventry = tididx + (tidctrl - 1); + if (rcventry >= uctxt->expected_count) { + dd_dev_err(dd, "Invalid RcvArray entry (%u) index for ctxt %u\n", + rcventry, uctxt->ctxt); + return -EINVAL; + } + node = fd->entry_to_rb[rcventry]; if (!node || node->rcventry != (uctxt->expected_base + rcventry)) return -EBADF; @@ -920,9 +915,8 @@ static bool tid_rb_invalidate(struct mmu_interval_notifier *mni, spin_lock(&fdata->invalid_lock); if (fdata->invalid_tid_idx < uctxt->expected_count) { fdata->invalid_tids[fdata->invalid_tid_idx] = - rcventry2tidinfo(node->rcventry - uctxt->expected_base); - fdata->invalid_tids[fdata->invalid_tid_idx] |= - EXP_TID_SET(LEN, node->npages); + create_tid(node->rcventry - uctxt->expected_base, + node->npages); if (!fdata->invalid_tid_idx) { unsigned long *ev; diff --git a/drivers/infiniband/hw/hfi1/user_pages.c b/drivers/infiniband/hw/hfi1/user_pages.c index 7bce963e2ae6..36aaedc65145 100644 --- a/drivers/infiniband/hw/hfi1/user_pages.c +++ b/drivers/infiniband/hw/hfi1/user_pages.c @@ -29,33 +29,52 @@ MODULE_PARM_DESC(cache_size, "Send and receive side cache size limit (in MB)"); bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm, u32 nlocked, u32 npages) { - unsigned long ulimit = rlimit(RLIMIT_MEMLOCK), pinned, cache_limit, - size = (cache_size * (1UL << 20)); /* convert to bytes */ - unsigned int usr_ctxts = - dd->num_rcv_contexts - dd->first_dyn_alloc_ctxt; - bool can_lock = capable(CAP_IPC_LOCK); + unsigned long ulimit_pages; + unsigned long cache_limit_pages; + unsigned int usr_ctxts; /* - * Calculate per-cache size. The calculation below uses only a quarter - * of the available per-context limit. This leaves space for other - * pinning. Should we worry about shared ctxts? + * Perform RLIMIT_MEMLOCK based checks unless CAP_IPC_LOCK is present. */ - cache_limit = (ulimit / usr_ctxts) / 4; - - /* If ulimit isn't set to "unlimited" and is smaller than cache_size. */ - if (ulimit != (-1UL) && size > cache_limit) - size = cache_limit; - - /* Convert to number of pages */ - size = DIV_ROUND_UP(size, PAGE_SIZE); - - pinned = atomic64_read(&mm->pinned_vm); + if (!capable(CAP_IPC_LOCK)) { + ulimit_pages = + DIV_ROUND_DOWN_ULL(rlimit(RLIMIT_MEMLOCK), PAGE_SIZE); + + /* + * Pinning these pages would exceed this process's locked memory + * limit. + */ + if (atomic64_read(&mm->pinned_vm) + npages > ulimit_pages) + return false; + + /* + * Only allow 1/4 of the user's RLIMIT_MEMLOCK to be used for HFI + * caches. This fraction is then equally distributed among all + * existing user contexts. Note that if RLIMIT_MEMLOCK is + * 'unlimited' (-1), the value of this limit will be > 2^42 pages + * (2^64 / 2^12 / 2^8 / 2^2). + * + * The effectiveness of this check may be reduced if I/O occurs on + * some user contexts before all user contexts are created. This + * check assumes that this process is the only one using this + * context (e.g., the corresponding fd was not passed to another + * process for concurrent access) as there is no per-context, + * per-process tracking of pinned pages. It also assumes that each + * user context has only one cache to limit. + */ + usr_ctxts = dd->num_rcv_contexts - dd->first_dyn_alloc_ctxt; + if (nlocked + npages > (ulimit_pages / usr_ctxts / 4)) + return false; + } - /* First, check the absolute limit against all pinned pages. */ - if (pinned + npages >= ulimit && !can_lock) + /* + * Pinning these pages would exceed the size limit for this cache. + */ + cache_limit_pages = cache_size * (1024 * 1024) / PAGE_SIZE; + if (nlocked + npages > cache_limit_pages) return false; - return ((nlocked + npages) <= size) || can_lock; + return true; } int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr, size_t npages, diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c index e6e17984553c..7f6d7fc7951d 100644 --- a/drivers/infiniband/hw/hfi1/verbs.c +++ b/drivers/infiniband/hw/hfi1/verbs.c @@ -1598,13 +1598,11 @@ static const char * const driver_cntr_names[] = { "DRIVER_EgrHdrFull" }; -static DEFINE_MUTEX(cntr_names_lock); /* protects the *_cntr_names bufers */ static struct rdma_stat_desc *dev_cntr_descs; static struct rdma_stat_desc *port_cntr_descs; int num_driver_cntrs = ARRAY_SIZE(driver_cntr_names); static int num_dev_cntrs; static int num_port_cntrs; -static int cntr_names_initialized; /* * Convert a list of names separated by '\n' into an array of NULL terminated @@ -1615,8 +1613,8 @@ static int init_cntr_names(const char *names_in, const size_t names_len, int num_extra_names, int *num_cntrs, struct rdma_stat_desc **cntr_descs) { - struct rdma_stat_desc *q; - char *names_out, *p; + struct rdma_stat_desc *names_out; + char *p; int i, n; n = 0; @@ -1624,65 +1622,45 @@ static int init_cntr_names(const char *names_in, const size_t names_len, if (names_in[i] == '\n') n++; - names_out = - kzalloc((n + num_extra_names) * sizeof(*q) + names_len, - GFP_KERNEL); + names_out = kzalloc((n + num_extra_names) * sizeof(*names_out) + + names_len, + GFP_KERNEL); if (!names_out) { *num_cntrs = 0; *cntr_descs = NULL; return -ENOMEM; } - p = names_out + (n + num_extra_names) * sizeof(*q); + p = (char *)&names_out[n + num_extra_names]; memcpy(p, names_in, names_len); - q = (struct rdma_stat_desc *)names_out; for (i = 0; i < n; i++) { - q[i].name = p; + names_out[i].name = p; p = strchr(p, '\n'); *p++ = '\0'; } *num_cntrs = n; - *cntr_descs = (struct rdma_stat_desc *)names_out; + *cntr_descs = names_out; return 0; } -static int init_counters(struct ib_device *ibdev) -{ - struct hfi1_devdata *dd = dd_from_ibdev(ibdev); - int i, err = 0; - - mutex_lock(&cntr_names_lock); - if (cntr_names_initialized) - goto out_unlock; - - err = init_cntr_names(dd->cntrnames, dd->cntrnameslen, num_driver_cntrs, - &num_dev_cntrs, &dev_cntr_descs); - if (err) - goto out_unlock; - - for (i = 0; i < num_driver_cntrs; i++) - dev_cntr_descs[num_dev_cntrs + i].name = driver_cntr_names[i]; - - err = init_cntr_names(dd->portcntrnames, dd->portcntrnameslen, 0, - &num_port_cntrs, &port_cntr_descs); - if (err) { - kfree(dev_cntr_descs); - dev_cntr_descs = NULL; - goto out_unlock; - } - cntr_names_initialized = 1; - -out_unlock: - mutex_unlock(&cntr_names_lock); - return err; -} - static struct rdma_hw_stats *hfi1_alloc_hw_device_stats(struct ib_device *ibdev) { - if (init_counters(ibdev)) - return NULL; + if (!dev_cntr_descs) { + struct hfi1_devdata *dd = dd_from_ibdev(ibdev); + int i, err; + + err = init_cntr_names(dd->cntrnames, dd->cntrnameslen, + num_driver_cntrs, + &num_dev_cntrs, &dev_cntr_descs); + if (err) + return NULL; + + for (i = 0; i < num_driver_cntrs; i++) + dev_cntr_descs[num_dev_cntrs + i].name = + driver_cntr_names[i]; + } return rdma_alloc_hw_stats_struct(dev_cntr_descs, num_dev_cntrs + num_driver_cntrs, RDMA_HW_STATS_DEFAULT_LIFESPAN); @@ -1691,8 +1669,16 @@ static struct rdma_hw_stats *hfi1_alloc_hw_device_stats(struct ib_device *ibdev) static struct rdma_hw_stats *hfi_alloc_hw_port_stats(struct ib_device *ibdev, u32 port_num) { - if (init_counters(ibdev)) - return NULL; + if (!port_cntr_descs) { + struct hfi1_devdata *dd = dd_from_ibdev(ibdev); + int err; + + err = init_cntr_names(dd->portcntrnames, dd->portcntrnameslen, + 0, + &num_port_cntrs, &port_cntr_descs); + if (err) + return NULL; + } return rdma_alloc_hw_stats_struct(port_cntr_descs, num_port_cntrs, RDMA_HW_STATS_DEFAULT_LIFESPAN); } @@ -1917,13 +1903,10 @@ void hfi1_unregister_ib_device(struct hfi1_devdata *dd) del_timer_sync(&dev->mem_timer); verbs_txreq_exit(dev); - mutex_lock(&cntr_names_lock); kfree(dev_cntr_descs); kfree(port_cntr_descs); dev_cntr_descs = NULL; port_cntr_descs = NULL; - cntr_names_initialized = 0; - mutex_unlock(&cntr_names_lock); } void hfi1_cnp_rcv(struct hfi1_packet *packet) diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h index f701cc86896b..84239b907de2 100644 --- a/drivers/infiniband/hw/hns/hns_roce_device.h +++ b/drivers/infiniband/hw/hns/hns_roce_device.h @@ -144,6 +144,7 @@ enum { HNS_ROCE_CAP_FLAG_DIRECT_WQE = BIT(12), HNS_ROCE_CAP_FLAG_SDI_MODE = BIT(14), HNS_ROCE_CAP_FLAG_STASH = BIT(17), + HNS_ROCE_CAP_FLAG_CQE_INLINE = BIT(19), }; #define HNS_ROCE_DB_TYPE_COUNT 2 @@ -567,21 +568,6 @@ struct hns_roce_mbox_msg { struct hns_roce_dev; -struct hns_roce_rinl_sge { - void *addr; - u32 len; -}; - -struct hns_roce_rinl_wqe { - struct hns_roce_rinl_sge *sg_list; - u32 sge_cnt; -}; - -struct hns_roce_rinl_buf { - struct hns_roce_rinl_wqe *wqe_list; - u32 wqe_cnt; -}; - enum { HNS_ROCE_FLUSH_FLAG = 0, }; @@ -632,7 +618,6 @@ struct hns_roce_qp { /* 0: flush needed, 1: unneeded */ unsigned long flush_flag; struct hns_roce_work flush_work; - struct hns_roce_rinl_buf rq_inl_buf; struct list_head node; /* all qps are on a list */ struct list_head rq_node; /* all recv qps are on a list */ struct list_head sq_node; /* all send qps are on a list */ @@ -887,7 +872,7 @@ struct hns_roce_hw { u32 step_idx); int (*modify_qp)(struct ib_qp *ibqp, const struct ib_qp_attr *attr, int attr_mask, enum ib_qp_state cur_state, - enum ib_qp_state new_state); + enum ib_qp_state new_state, struct ib_udata *udata); int (*qp_flow_control_init)(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp); void (*dereg_mr)(struct hns_roce_dev *hr_dev); diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index b2421883993b..dbf97fe5948f 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -821,22 +821,10 @@ static void fill_recv_sge_to_wqe(const struct ib_recv_wr *wr, void *wqe, static void fill_rq_wqe(struct hns_roce_qp *hr_qp, const struct ib_recv_wr *wr, u32 wqe_idx, u32 max_sge) { - struct hns_roce_rinl_sge *sge_list; void *wqe = NULL; - u32 i; wqe = hns_roce_get_recv_wqe(hr_qp, wqe_idx); fill_recv_sge_to_wqe(wr, wqe, max_sge, hr_qp->rq.rsv_sge); - - /* rq support inline data */ - if (hr_qp->rq_inl_buf.wqe_cnt) { - sge_list = hr_qp->rq_inl_buf.wqe_list[wqe_idx].sg_list; - hr_qp->rq_inl_buf.wqe_list[wqe_idx].sge_cnt = (u32)wr->num_sge; - for (i = 0; i < wr->num_sge; i++) { - sge_list[i].addr = (void *)(u64)wr->sg_list[i].addr; - sge_list[i].len = wr->sg_list[i].length; - } - } } static int hns_roce_v2_post_recv(struct ib_qp *ibqp, @@ -2849,7 +2837,7 @@ static int free_mr_modify_rsv_qp(struct hns_roce_dev *hr_dev, attr->port_num = 1; attr->qp_access_flags = IB_ACCESS_REMOTE_WRITE; ret = hr_dev->hw->modify_qp(&hr_qp->ibqp, attr, mask, IB_QPS_INIT, - IB_QPS_INIT); + IB_QPS_INIT, NULL); if (ret) { ibdev_err(ibdev, "failed to modify qp to init, ret = %d.\n", ret); @@ -2871,7 +2859,7 @@ static int free_mr_modify_rsv_qp(struct hns_roce_dev *hr_dev, rdma_ah_set_sl(&attr->ah_attr, (u8)sl_num); ret = hr_dev->hw->modify_qp(&hr_qp->ibqp, attr, mask, IB_QPS_INIT, - IB_QPS_RTR); + IB_QPS_RTR, NULL); hr_dev->loop_idc = loopback; if (ret) { ibdev_err(ibdev, "failed to modify qp to rtr, ret = %d.\n", @@ -2886,7 +2874,7 @@ static int free_mr_modify_rsv_qp(struct hns_roce_dev *hr_dev, attr->retry_cnt = HNS_ROCE_FREE_MR_USED_QP_RETRY_CNT; attr->timeout = HNS_ROCE_FREE_MR_USED_QP_TIMEOUT; ret = hr_dev->hw->modify_qp(&hr_qp->ibqp, attr, mask, IB_QPS_RTR, - IB_QPS_RTS); + IB_QPS_RTS, NULL); if (ret) ibdev_err(ibdev, "failed to modify qp to rts, ret = %d.\n", ret); @@ -3730,39 +3718,6 @@ static int hns_roce_v2_req_notify_cq(struct ib_cq *ibcq, return 0; } -static int hns_roce_handle_recv_inl_wqe(struct hns_roce_v2_cqe *cqe, - struct hns_roce_qp *qp, - struct ib_wc *wc) -{ - struct hns_roce_rinl_sge *sge_list; - u32 wr_num, wr_cnt, sge_num; - u32 sge_cnt, data_len, size; - void *wqe_buf; - - wr_num = hr_reg_read(cqe, CQE_WQE_IDX); - wr_cnt = wr_num & (qp->rq.wqe_cnt - 1); - - sge_list = qp->rq_inl_buf.wqe_list[wr_cnt].sg_list; - sge_num = qp->rq_inl_buf.wqe_list[wr_cnt].sge_cnt; - wqe_buf = hns_roce_get_recv_wqe(qp, wr_cnt); - data_len = wc->byte_len; - - for (sge_cnt = 0; (sge_cnt < sge_num) && (data_len); sge_cnt++) { - size = min(sge_list[sge_cnt].len, data_len); - memcpy((void *)sge_list[sge_cnt].addr, wqe_buf, size); - - data_len -= size; - wqe_buf += size; - } - - if (unlikely(data_len)) { - wc->status = IB_WC_LOC_LEN_ERR; - return -EAGAIN; - } - - return 0; -} - static int sw_comp(struct hns_roce_qp *hr_qp, struct hns_roce_wq *wq, int num_entries, struct ib_wc *wc) { @@ -3974,22 +3929,10 @@ static void fill_send_wc(struct ib_wc *wc, struct hns_roce_v2_cqe *cqe) wc->opcode = ib_opcode; } -static inline bool is_rq_inl_enabled(struct ib_wc *wc, u32 hr_opcode, - struct hns_roce_v2_cqe *cqe) -{ - return wc->qp->qp_type != IB_QPT_UD && wc->qp->qp_type != IB_QPT_GSI && - (hr_opcode == HNS_ROCE_V2_OPCODE_SEND || - hr_opcode == HNS_ROCE_V2_OPCODE_SEND_WITH_IMM || - hr_opcode == HNS_ROCE_V2_OPCODE_SEND_WITH_INV) && - hr_reg_read(cqe, CQE_RQ_INLINE); -} - static int fill_recv_wc(struct ib_wc *wc, struct hns_roce_v2_cqe *cqe) { - struct hns_roce_qp *qp = to_hr_qp(wc->qp); u32 hr_opcode; int ib_opcode; - int ret; wc->byte_len = le32_to_cpu(cqe->byte_cnt); @@ -4014,12 +3957,6 @@ static int fill_recv_wc(struct ib_wc *wc, struct hns_roce_v2_cqe *cqe) else wc->opcode = ib_opcode; - if (is_rq_inl_enabled(wc, hr_opcode, cqe)) { - ret = hns_roce_handle_recv_inl_wqe(cqe, qp, wc); - if (unlikely(ret)) - return ret; - } - wc->sl = hr_reg_read(cqe, CQE_SL); wc->src_qp = hr_reg_read(cqe, CQE_RMT_QPN); wc->slid = 0; @@ -4445,10 +4382,6 @@ static void modify_qp_reset_to_init(struct ib_qp *ibqp, hr_reg_write(context, QPC_RQ_DB_RECORD_ADDR_H, upper_32_bits(hr_qp->rdb.dma)); - if (ibqp->qp_type != IB_QPT_UD && ibqp->qp_type != IB_QPT_GSI) - hr_reg_write_bool(context, QPC_RQIE, - hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RQ_INLINE); - hr_reg_write(context, QPC_RX_CQN, get_cqn(ibqp->recv_cq)); if (ibqp->srq) { @@ -4639,8 +4572,11 @@ static inline enum ib_mtu get_mtu(struct ib_qp *ibqp, static int modify_qp_init_to_rtr(struct ib_qp *ibqp, const struct ib_qp_attr *attr, int attr_mask, struct hns_roce_v2_qp_context *context, - struct hns_roce_v2_qp_context *qpc_mask) + struct hns_roce_v2_qp_context *qpc_mask, + struct ib_udata *udata) { + struct hns_roce_ucontext *uctx = rdma_udata_to_drv_context(udata, + struct hns_roce_ucontext, ibucontext); struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device); struct hns_roce_qp *hr_qp = to_hr_qp(ibqp); struct ib_device *ibdev = &hr_dev->ib_dev; @@ -4760,6 +4696,26 @@ static int modify_qp_init_to_rtr(struct ib_qp *ibqp, hr_reg_write(context, QPC_LP_SGEN_INI, 3); hr_reg_clear(qpc_mask, QPC_LP_SGEN_INI); + if (udata && ibqp->qp_type == IB_QPT_RC && + (uctx->config & HNS_ROCE_RQ_INLINE_FLAGS)) { + hr_reg_write_bool(context, QPC_RQIE, + hr_dev->caps.flags & + HNS_ROCE_CAP_FLAG_RQ_INLINE); + hr_reg_clear(qpc_mask, QPC_RQIE); + } + + if (udata && + (ibqp->qp_type == IB_QPT_RC || ibqp->qp_type == IB_QPT_XRC_TGT) && + (uctx->config & HNS_ROCE_CQE_INLINE_FLAGS)) { + hr_reg_write_bool(context, QPC_CQEIE, + hr_dev->caps.flags & + HNS_ROCE_CAP_FLAG_CQE_INLINE); + hr_reg_clear(qpc_mask, QPC_CQEIE); + + hr_reg_write(context, QPC_CQEIS, 0); + hr_reg_clear(qpc_mask, QPC_CQEIS); + } + return 0; } @@ -5107,7 +5063,8 @@ static int hns_roce_v2_set_abs_fields(struct ib_qp *ibqp, enum ib_qp_state cur_state, enum ib_qp_state new_state, struct hns_roce_v2_qp_context *context, - struct hns_roce_v2_qp_context *qpc_mask) + struct hns_roce_v2_qp_context *qpc_mask, + struct ib_udata *udata) { struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device); int ret = 0; @@ -5124,7 +5081,7 @@ static int hns_roce_v2_set_abs_fields(struct ib_qp *ibqp, modify_qp_init_to_init(ibqp, attr, context, qpc_mask); } else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) { ret = modify_qp_init_to_rtr(ibqp, attr, attr_mask, context, - qpc_mask); + qpc_mask, udata); } else if (cur_state == IB_QPS_RTR && new_state == IB_QPS_RTS) { ret = modify_qp_rtr_to_rts(ibqp, attr, attr_mask, context, qpc_mask); @@ -5329,7 +5286,7 @@ static void v2_set_flushed_fields(struct ib_qp *ibqp, static int hns_roce_v2_modify_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr, int attr_mask, enum ib_qp_state cur_state, - enum ib_qp_state new_state) + enum ib_qp_state new_state, struct ib_udata *udata) { struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device); struct hns_roce_qp *hr_qp = to_hr_qp(ibqp); @@ -5352,7 +5309,7 @@ static int hns_roce_v2_modify_qp(struct ib_qp *ibqp, memset(qpc_mask, 0xff, hr_dev->caps.qpc_sz); ret = hns_roce_v2_set_abs_fields(ibqp, attr, attr_mask, cur_state, - new_state, context, qpc_mask); + new_state, context, qpc_mask, udata); if (ret) goto out; @@ -5555,7 +5512,7 @@ static int hns_roce_v2_destroy_qp_common(struct hns_roce_dev *hr_dev, if (modify_qp_is_ok(hr_qp)) { /* Modify qp to reset before destroying qp */ ret = hns_roce_v2_modify_qp(&hr_qp->ibqp, NULL, 0, - hr_qp->state, IB_QPS_RESET); + hr_qp->state, IB_QPS_RESET, udata); if (ret) ibdev_err(ibdev, "failed to modify QP to RST, ret = %d.\n", diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h index b1b3e1e0b84e..af9d00225cdf 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h @@ -531,7 +531,8 @@ struct hns_roce_v2_qp_context { #define QPC_RQ_RTY_TX_ERR QPC_FIELD_LOC(607, 607) #define QPC_RX_CQN QPC_FIELD_LOC(631, 608) #define QPC_XRC_QP_TYPE QPC_FIELD_LOC(632, 632) -#define QPC_RSV3 QPC_FIELD_LOC(634, 633) +#define QPC_CQEIE QPC_FIELD_LOC(633, 633) +#define QPC_CQEIS QPC_FIELD_LOC(634, 634) #define QPC_MIN_RNR_TIME QPC_FIELD_LOC(639, 635) #define QPC_RQ_PRODUCER_IDX QPC_FIELD_LOC(655, 640) #define QPC_RQ_CONSUMER_IDX QPC_FIELD_LOC(671, 656) diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c index 8ba68ac12388..485e110ca433 100644 --- a/drivers/infiniband/hw/hns/hns_roce_main.c +++ b/drivers/infiniband/hw/hns/hns_roce_main.c @@ -379,6 +379,18 @@ static int hns_roce_alloc_ucontext(struct ib_ucontext *uctx, resp.max_inline_data = hr_dev->caps.max_sq_inline; } + if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RQ_INLINE) { + context->config |= ucmd.config & HNS_ROCE_RQ_INLINE_FLAGS; + if (context->config & HNS_ROCE_RQ_INLINE_FLAGS) + resp.config |= HNS_ROCE_RSP_RQ_INLINE_FLAGS; + } + + if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQE_INLINE) { + context->config |= ucmd.config & HNS_ROCE_CQE_INLINE_FLAGS; + if (context->config & HNS_ROCE_CQE_INLINE_FLAGS) + resp.config |= HNS_ROCE_RSP_CQE_INLINE_FLAGS; + } + ret = hns_roce_uar_alloc(hr_dev, &context->uar); if (ret) goto error_fail_uar_alloc; @@ -443,14 +455,15 @@ static int hns_roce_mmap(struct ib_ucontext *uctx, struct vm_area_struct *vma) prot = pgprot_device(vma->vm_page_prot); break; default: - return -EINVAL; + ret = -EINVAL; + goto out; } ret = rdma_user_mmap_io(uctx, vma, pfn, rdma_entry->npages * PAGE_SIZE, prot, rdma_entry); +out: rdma_user_mmap_entry_put(rdma_entry); - return ret; } diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c index 0ae335fb205c..d855a917f4cf 100644 --- a/drivers/infiniband/hw/hns/hns_roce_qp.c +++ b/drivers/infiniband/hw/hns/hns_roce_qp.c @@ -433,7 +433,6 @@ static int set_rq_size(struct hns_roce_dev *hr_dev, struct ib_qp_cap *cap, if (!has_rq) { hr_qp->rq.wqe_cnt = 0; hr_qp->rq.max_gs = 0; - hr_qp->rq_inl_buf.wqe_cnt = 0; cap->max_recv_wr = 0; cap->max_recv_sge = 0; @@ -463,12 +462,6 @@ static int set_rq_size(struct hns_roce_dev *hr_dev, struct ib_qp_cap *cap, hr_qp->rq.max_gs); hr_qp->rq.wqe_cnt = cnt; - if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RQ_INLINE && - hr_qp->ibqp.qp_type != IB_QPT_UD && - hr_qp->ibqp.qp_type != IB_QPT_GSI) - hr_qp->rq_inl_buf.wqe_cnt = cnt; - else - hr_qp->rq_inl_buf.wqe_cnt = 0; cap->max_recv_wr = cnt; cap->max_recv_sge = hr_qp->rq.max_gs - hr_qp->rq.rsv_sge; @@ -732,49 +725,6 @@ static int hns_roce_qp_has_rq(struct ib_qp_init_attr *attr) return 1; } -static int alloc_rq_inline_buf(struct hns_roce_qp *hr_qp, - struct ib_qp_init_attr *init_attr) -{ - u32 max_recv_sge = init_attr->cap.max_recv_sge; - u32 wqe_cnt = hr_qp->rq_inl_buf.wqe_cnt; - struct hns_roce_rinl_wqe *wqe_list; - int i; - - /* allocate recv inline buf */ - wqe_list = kcalloc(wqe_cnt, sizeof(struct hns_roce_rinl_wqe), - GFP_KERNEL); - if (!wqe_list) - goto err; - - /* Allocate a continuous buffer for all inline sge we need */ - wqe_list[0].sg_list = kcalloc(wqe_cnt, (max_recv_sge * - sizeof(struct hns_roce_rinl_sge)), - GFP_KERNEL); - if (!wqe_list[0].sg_list) - goto err_wqe_list; - - /* Assign buffers of sg_list to each inline wqe */ - for (i = 1; i < wqe_cnt; i++) - wqe_list[i].sg_list = &wqe_list[0].sg_list[i * max_recv_sge]; - - hr_qp->rq_inl_buf.wqe_list = wqe_list; - - return 0; - -err_wqe_list: - kfree(wqe_list); - -err: - return -ENOMEM; -} - -static void free_rq_inline_buf(struct hns_roce_qp *hr_qp) -{ - if (hr_qp->rq_inl_buf.wqe_list) - kfree(hr_qp->rq_inl_buf.wqe_list[0].sg_list); - kfree(hr_qp->rq_inl_buf.wqe_list); -} - static int alloc_qp_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp, struct ib_qp_init_attr *init_attr, struct ib_udata *udata, unsigned long addr) @@ -783,18 +733,6 @@ static int alloc_qp_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp, struct hns_roce_buf_attr buf_attr = {}; int ret; - if (!udata && hr_qp->rq_inl_buf.wqe_cnt) { - ret = alloc_rq_inline_buf(hr_qp, init_attr); - if (ret) { - ibdev_err(ibdev, - "failed to alloc inline buf, ret = %d.\n", - ret); - return ret; - } - } else { - hr_qp->rq_inl_buf.wqe_list = NULL; - } - ret = set_wqe_buf_attr(hr_dev, hr_qp, &buf_attr); if (ret) { ibdev_err(ibdev, "failed to split WQE buf, ret = %d.\n", ret); @@ -814,7 +752,6 @@ static int alloc_qp_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp, return 0; err_inline: - free_rq_inline_buf(hr_qp); return ret; } @@ -822,7 +759,6 @@ err_inline: static void free_qp_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp) { hns_roce_mtr_destroy(hr_dev, &hr_qp->mtr); - free_rq_inline_buf(hr_qp); } static inline bool user_qp_has_sdb(struct hns_roce_dev *hr_dev, @@ -1410,7 +1346,7 @@ int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, goto out; ret = hr_dev->hw->modify_qp(ibqp, attr, attr_mask, cur_state, - new_state); + new_state, udata); out: mutex_unlock(&hr_qp->mutex); diff --git a/drivers/infiniband/hw/irdma/hw.c b/drivers/infiniband/hw/irdma/hw.c index ab246447520b..2e1e2bad0401 100644 --- a/drivers/infiniband/hw/irdma/hw.c +++ b/drivers/infiniband/hw/irdma/hw.c @@ -483,6 +483,8 @@ static int irdma_save_msix_info(struct irdma_pci_f *rf) iw_qvlist->num_vectors = rf->msix_count; if (rf->msix_count <= num_online_cpus()) rf->msix_shared = true; + else if (rf->msix_count > num_online_cpus() + 1) + rf->msix_count = num_online_cpus() + 1; pmsix = rf->msix_entries; for (i = 0, ceq_idx = 0; i < rf->msix_count; i++, iw_qvinfo++) { diff --git a/drivers/infiniband/hw/irdma/verbs.c b/drivers/infiniband/hw/irdma/verbs.c index f6973ea55eda..1b2e3e800c9a 100644 --- a/drivers/infiniband/hw/irdma/verbs.c +++ b/drivers/infiniband/hw/irdma/verbs.c @@ -2745,6 +2745,162 @@ static int irdma_hwreg_mr(struct irdma_device *iwdev, struct irdma_mr *iwmr, return ret; } +static int irdma_reg_user_mr_type_mem(struct irdma_mr *iwmr, int access) +{ + struct irdma_device *iwdev = to_iwdev(iwmr->ibmr.device); + struct irdma_pbl *iwpbl = &iwmr->iwpbl; + bool use_pbles; + u32 stag; + int err; + + use_pbles = iwmr->page_cnt != 1; + + err = irdma_setup_pbles(iwdev->rf, iwmr, use_pbles, false); + if (err) + return err; + + if (use_pbles) { + err = irdma_check_mr_contiguous(&iwpbl->pble_alloc, + iwmr->page_size); + if (err) { + irdma_free_pble(iwdev->rf->pble_rsrc, &iwpbl->pble_alloc); + iwpbl->pbl_allocated = false; + } + } + + stag = irdma_create_stag(iwdev); + if (!stag) { + err = -ENOMEM; + goto free_pble; + } + + iwmr->stag = stag; + iwmr->ibmr.rkey = stag; + iwmr->ibmr.lkey = stag; + err = irdma_hwreg_mr(iwdev, iwmr, access); + if (err) + goto err_hwreg; + + return 0; + +err_hwreg: + irdma_free_stag(iwdev, stag); + +free_pble: + if (iwpbl->pble_alloc.level != PBLE_LEVEL_0 && iwpbl->pbl_allocated) + irdma_free_pble(iwdev->rf->pble_rsrc, &iwpbl->pble_alloc); + + return err; +} + +static struct irdma_mr *irdma_alloc_iwmr(struct ib_umem *region, + struct ib_pd *pd, u64 virt, + enum irdma_memreg_type reg_type) +{ + struct irdma_device *iwdev = to_iwdev(pd->device); + struct irdma_pbl *iwpbl = NULL; + struct irdma_mr *iwmr = NULL; + unsigned long pgsz_bitmap; + + iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL); + if (!iwmr) + return ERR_PTR(-ENOMEM); + + iwpbl = &iwmr->iwpbl; + iwpbl->iwmr = iwmr; + iwmr->region = region; + iwmr->ibmr.pd = pd; + iwmr->ibmr.device = pd->device; + iwmr->ibmr.iova = virt; + iwmr->type = reg_type; + + pgsz_bitmap = (reg_type == IRDMA_MEMREG_TYPE_MEM) ? + iwdev->rf->sc_dev.hw_attrs.page_size_cap : PAGE_SIZE; + + iwmr->page_size = ib_umem_find_best_pgsz(region, pgsz_bitmap, virt); + if (unlikely(!iwmr->page_size)) { + kfree(iwmr); + return ERR_PTR(-EOPNOTSUPP); + } + + iwmr->len = region->length; + iwpbl->user_base = virt; + iwmr->page_cnt = ib_umem_num_dma_blocks(region, iwmr->page_size); + + return iwmr; +} + +static void irdma_free_iwmr(struct irdma_mr *iwmr) +{ + kfree(iwmr); +} + +static int irdma_reg_user_mr_type_qp(struct irdma_mem_reg_req req, + struct ib_udata *udata, + struct irdma_mr *iwmr) +{ + struct irdma_device *iwdev = to_iwdev(iwmr->ibmr.device); + struct irdma_pbl *iwpbl = &iwmr->iwpbl; + struct irdma_ucontext *ucontext = NULL; + unsigned long flags; + bool use_pbles; + u32 total; + int err; + + total = req.sq_pages + req.rq_pages + 1; + if (total > iwmr->page_cnt) + return -EINVAL; + + total = req.sq_pages + req.rq_pages; + use_pbles = total > 2; + err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles); + if (err) + return err; + + ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext, + ibucontext); + spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags); + list_add_tail(&iwpbl->list, &ucontext->qp_reg_mem_list); + iwpbl->on_list = true; + spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags); + + return 0; +} + +static int irdma_reg_user_mr_type_cq(struct irdma_mem_reg_req req, + struct ib_udata *udata, + struct irdma_mr *iwmr) +{ + struct irdma_device *iwdev = to_iwdev(iwmr->ibmr.device); + struct irdma_pbl *iwpbl = &iwmr->iwpbl; + struct irdma_ucontext *ucontext = NULL; + u8 shadow_pgcnt = 1; + unsigned long flags; + bool use_pbles; + u32 total; + int err; + + if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_CQ_RESIZE) + shadow_pgcnt = 0; + total = req.cq_pages + shadow_pgcnt; + if (total > iwmr->page_cnt) + return -EINVAL; + + use_pbles = req.cq_pages > 1; + err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles); + if (err) + return err; + + ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext, + ibucontext); + spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags); + list_add_tail(&iwpbl->list, &ucontext->cq_reg_mem_list); + iwpbl->on_list = true; + spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags); + + return 0; +} + /** * irdma_reg_user_mr - Register a user memory region * @pd: ptr of pd @@ -2760,18 +2916,10 @@ static struct ib_mr *irdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 len, { #define IRDMA_MEM_REG_MIN_REQ_LEN offsetofend(struct irdma_mem_reg_req, sq_pages) struct irdma_device *iwdev = to_iwdev(pd->device); - struct irdma_ucontext *ucontext; - struct irdma_pble_alloc *palloc; - struct irdma_pbl *iwpbl; - struct irdma_mr *iwmr; - struct ib_umem *region; - struct irdma_mem_reg_req req; - u32 total, stag = 0; - u8 shadow_pgcnt = 1; - bool use_pbles = false; - unsigned long flags; - int err = -EINVAL; - int ret; + struct irdma_mem_reg_req req = {}; + struct ib_umem *region = NULL; + struct irdma_mr *iwmr = NULL; + int err; if (len > iwdev->rf->sc_dev.hw_attrs.max_mr_size) return ERR_PTR(-EINVAL); @@ -2792,122 +2940,80 @@ static struct ib_mr *irdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 len, return ERR_PTR(-EFAULT); } - iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL); - if (!iwmr) { + iwmr = irdma_alloc_iwmr(region, pd, virt, req.reg_type); + if (IS_ERR(iwmr)) { ib_umem_release(region); - return ERR_PTR(-ENOMEM); - } - - iwpbl = &iwmr->iwpbl; - iwpbl->iwmr = iwmr; - iwmr->region = region; - iwmr->ibmr.pd = pd; - iwmr->ibmr.device = pd->device; - iwmr->ibmr.iova = virt; - iwmr->page_size = PAGE_SIZE; - - if (req.reg_type == IRDMA_MEMREG_TYPE_MEM) { - iwmr->page_size = ib_umem_find_best_pgsz(region, - iwdev->rf->sc_dev.hw_attrs.page_size_cap, - virt); - if (unlikely(!iwmr->page_size)) { - kfree(iwmr); - ib_umem_release(region); - return ERR_PTR(-EOPNOTSUPP); - } + return (struct ib_mr *)iwmr; } - iwmr->len = region->length; - iwpbl->user_base = virt; - palloc = &iwpbl->pble_alloc; - iwmr->type = req.reg_type; - iwmr->page_cnt = ib_umem_num_dma_blocks(region, iwmr->page_size); switch (req.reg_type) { case IRDMA_MEMREG_TYPE_QP: - total = req.sq_pages + req.rq_pages + shadow_pgcnt; - if (total > iwmr->page_cnt) { - err = -EINVAL; - goto error; - } - total = req.sq_pages + req.rq_pages; - use_pbles = (total > 2); - err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles); + err = irdma_reg_user_mr_type_qp(req, udata, iwmr); if (err) goto error; - ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext, - ibucontext); - spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags); - list_add_tail(&iwpbl->list, &ucontext->qp_reg_mem_list); - iwpbl->on_list = true; - spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags); break; case IRDMA_MEMREG_TYPE_CQ: - if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_CQ_RESIZE) - shadow_pgcnt = 0; - total = req.cq_pages + shadow_pgcnt; - if (total > iwmr->page_cnt) { - err = -EINVAL; - goto error; - } - - use_pbles = (req.cq_pages > 1); - err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles); + err = irdma_reg_user_mr_type_cq(req, udata, iwmr); if (err) goto error; - - ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext, - ibucontext); - spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags); - list_add_tail(&iwpbl->list, &ucontext->cq_reg_mem_list); - iwpbl->on_list = true; - spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags); break; case IRDMA_MEMREG_TYPE_MEM: - use_pbles = (iwmr->page_cnt != 1); - - err = irdma_setup_pbles(iwdev->rf, iwmr, use_pbles, false); + err = irdma_reg_user_mr_type_mem(iwmr, access); if (err) goto error; - if (use_pbles) { - ret = irdma_check_mr_contiguous(palloc, - iwmr->page_size); - if (ret) { - irdma_free_pble(iwdev->rf->pble_rsrc, palloc); - iwpbl->pbl_allocated = false; - } - } - - stag = irdma_create_stag(iwdev); - if (!stag) { - err = -ENOMEM; - goto error; - } - - iwmr->stag = stag; - iwmr->ibmr.rkey = stag; - iwmr->ibmr.lkey = stag; - err = irdma_hwreg_mr(iwdev, iwmr, access); - if (err) { - irdma_free_stag(iwdev, stag); - goto error; - } - break; default: + err = -EINVAL; goto error; } - iwmr->type = req.reg_type; - return &iwmr->ibmr; - error: - if (palloc->level != PBLE_LEVEL_0 && iwpbl->pbl_allocated) - irdma_free_pble(iwdev->rf->pble_rsrc, palloc); ib_umem_release(region); - kfree(iwmr); + irdma_free_iwmr(iwmr); + + return ERR_PTR(err); +} + +static struct ib_mr *irdma_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start, + u64 len, u64 virt, + int fd, int access, + struct ib_udata *udata) +{ + struct irdma_device *iwdev = to_iwdev(pd->device); + struct ib_umem_dmabuf *umem_dmabuf; + struct irdma_mr *iwmr; + int err; + + if (len > iwdev->rf->sc_dev.hw_attrs.max_mr_size) + return ERR_PTR(-EINVAL); + + umem_dmabuf = ib_umem_dmabuf_get_pinned(pd->device, start, len, fd, access); + if (IS_ERR(umem_dmabuf)) { + err = PTR_ERR(umem_dmabuf); + ibdev_dbg(&iwdev->ibdev, "Failed to get dmabuf umem[%d]\n", err); + return ERR_PTR(err); + } + + iwmr = irdma_alloc_iwmr(&umem_dmabuf->umem, pd, virt, IRDMA_MEMREG_TYPE_MEM); + if (IS_ERR(iwmr)) { + err = PTR_ERR(iwmr); + goto err_release; + } + + err = irdma_reg_user_mr_type_mem(iwmr, access); + if (err) + goto err_iwmr; + + return &iwmr->ibmr; + +err_iwmr: + irdma_free_iwmr(iwmr); + +err_release: + ib_umem_release(&umem_dmabuf->umem); return ERR_PTR(err); } @@ -4418,6 +4524,7 @@ static const struct ib_device_ops irdma_dev_ops = { .query_port = irdma_query_port, .query_qp = irdma_query_qp, .reg_user_mr = irdma_reg_user_mr, + .reg_user_mr_dmabuf = irdma_reg_user_mr_dmabuf, .req_notify_cq = irdma_req_notify_cq, .resize_cq = irdma_resize_cq, INIT_RDMA_OBJ_SIZE(ib_pd, irdma_pd, ibpd), diff --git a/drivers/infiniband/hw/mana/main.c b/drivers/infiniband/hw/mana/main.c index 8b3bc302d6f3..7be4c3adb4e2 100644 --- a/drivers/infiniband/hw/mana/main.c +++ b/drivers/infiniband/hw/mana/main.c @@ -249,7 +249,8 @@ static int mana_ib_gd_first_dma_region(struct mana_ib_dev *dev, struct gdma_context *gc, struct gdma_create_dma_region_req *create_req, - size_t num_pages, mana_handle_t *gdma_region) + size_t num_pages, mana_handle_t *gdma_region, + u32 expected_status) { struct gdma_create_dma_region_resp create_resp = {}; unsigned int create_req_msg_size; @@ -261,7 +262,7 @@ mana_ib_gd_first_dma_region(struct mana_ib_dev *dev, err = mana_gd_send_request(gc, create_req_msg_size, create_req, sizeof(create_resp), &create_resp); - if (err || create_resp.hdr.status) { + if (err || create_resp.hdr.status != expected_status) { ibdev_dbg(&dev->ib_dev, "Failed to create DMA region: %d, 0x%x\n", err, create_resp.hdr.status); @@ -372,14 +373,21 @@ int mana_ib_gd_create_dma_region(struct mana_ib_dev *dev, struct ib_umem *umem, page_addr_list = create_req->page_addr_list; rdma_umem_for_each_dma_block(umem, &biter, page_sz) { + u32 expected_status = 0; + page_addr_list[tail++] = rdma_block_iter_dma_address(&biter); if (tail < num_pages_to_handle) continue; + if (num_pages_processed + num_pages_to_handle < + num_pages_total) + expected_status = GDMA_STATUS_MORE_ENTRIES; + if (!num_pages_processed) { /* First create message */ err = mana_ib_gd_first_dma_region(dev, gc, create_req, - tail, gdma_region); + tail, gdma_region, + expected_status); if (err) goto out; @@ -392,14 +400,8 @@ int mana_ib_gd_create_dma_region(struct mana_ib_dev *dev, struct ib_umem *umem, page_addr_list = add_req->page_addr_list; } else { /* Subsequent create messages */ - u32 expected_s = 0; - - if (num_pages_processed + num_pages_to_handle < - num_pages_total) - expected_s = GDMA_STATUS_MORE_ENTRIES; - err = mana_ib_gd_add_dma_region(dev, gc, add_req, tail, - expected_s); + expected_status); if (err) break; } diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index dceebcd885bb..b18e9f2adc82 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -3303,6 +3303,10 @@ static int __init mlx4_ib_init(void) if (!wq) return -ENOMEM; + err = mlx4_ib_qp_event_init(); + if (err) + goto clean_qp_event; + err = mlx4_ib_cm_init(); if (err) goto clean_wq; @@ -3324,6 +3328,9 @@ clean_cm: mlx4_ib_cm_destroy(); clean_wq: + mlx4_ib_qp_event_cleanup(); + +clean_qp_event: destroy_workqueue(wq); return err; } @@ -3333,6 +3340,7 @@ static void __exit mlx4_ib_cleanup(void) mlx4_unregister_interface(&mlx4_ib_interface); mlx4_ib_mcg_destroy(); mlx4_ib_cm_destroy(); + mlx4_ib_qp_event_cleanup(); destroy_workqueue(wq); } diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index 6a3b0f121045..17fee1e73a45 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -940,4 +940,7 @@ int mlx4_ib_umem_calc_optimal_mtt_size(struct ib_umem *umem, u64 start_va, int mlx4_ib_cm_init(void); void mlx4_ib_cm_destroy(void); +int mlx4_ib_qp_event_init(void); +void mlx4_ib_qp_event_cleanup(void); + #endif /* MLX4_IB_H */ diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index b17d6ebc5b70..884825b2e5f7 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c @@ -102,6 +102,14 @@ enum mlx4_ib_source_type { MLX4_IB_RWQ_SRC = 1, }; +struct mlx4_ib_qp_event_work { + struct work_struct work; + struct mlx4_qp *qp; + enum mlx4_event type; +}; + +static struct workqueue_struct *mlx4_ib_qp_event_wq; + static int is_tunnel_qp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp) { if (!mlx4_is_master(dev->dev)) @@ -200,50 +208,77 @@ static void stamp_send_wqe(struct mlx4_ib_qp *qp, int n) } } +static void mlx4_ib_handle_qp_event(struct work_struct *_work) +{ + struct mlx4_ib_qp_event_work *qpe_work = + container_of(_work, struct mlx4_ib_qp_event_work, work); + struct ib_qp *ibqp = &to_mibqp(qpe_work->qp)->ibqp; + struct ib_event event = {}; + + event.device = ibqp->device; + event.element.qp = ibqp; + + switch (qpe_work->type) { + case MLX4_EVENT_TYPE_PATH_MIG: + event.event = IB_EVENT_PATH_MIG; + break; + case MLX4_EVENT_TYPE_COMM_EST: + event.event = IB_EVENT_COMM_EST; + break; + case MLX4_EVENT_TYPE_SQ_DRAINED: + event.event = IB_EVENT_SQ_DRAINED; + break; + case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE: + event.event = IB_EVENT_QP_LAST_WQE_REACHED; + break; + case MLX4_EVENT_TYPE_WQ_CATAS_ERROR: + event.event = IB_EVENT_QP_FATAL; + break; + case MLX4_EVENT_TYPE_PATH_MIG_FAILED: + event.event = IB_EVENT_PATH_MIG_ERR; + break; + case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR: + event.event = IB_EVENT_QP_REQ_ERR; + break; + case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR: + event.event = IB_EVENT_QP_ACCESS_ERR; + break; + default: + pr_warn("Unexpected event type %d on QP %06x\n", + qpe_work->type, qpe_work->qp->qpn); + goto out; + } + + ibqp->event_handler(&event, ibqp->qp_context); + +out: + mlx4_put_qp(qpe_work->qp); + kfree(qpe_work); +} + static void mlx4_ib_qp_event(struct mlx4_qp *qp, enum mlx4_event type) { - struct ib_event event; struct ib_qp *ibqp = &to_mibqp(qp)->ibqp; + struct mlx4_ib_qp_event_work *qpe_work; if (type == MLX4_EVENT_TYPE_PATH_MIG) to_mibqp(qp)->port = to_mibqp(qp)->alt_port; - if (ibqp->event_handler) { - event.device = ibqp->device; - event.element.qp = ibqp; - switch (type) { - case MLX4_EVENT_TYPE_PATH_MIG: - event.event = IB_EVENT_PATH_MIG; - break; - case MLX4_EVENT_TYPE_COMM_EST: - event.event = IB_EVENT_COMM_EST; - break; - case MLX4_EVENT_TYPE_SQ_DRAINED: - event.event = IB_EVENT_SQ_DRAINED; - break; - case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE: - event.event = IB_EVENT_QP_LAST_WQE_REACHED; - break; - case MLX4_EVENT_TYPE_WQ_CATAS_ERROR: - event.event = IB_EVENT_QP_FATAL; - break; - case MLX4_EVENT_TYPE_PATH_MIG_FAILED: - event.event = IB_EVENT_PATH_MIG_ERR; - break; - case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR: - event.event = IB_EVENT_QP_REQ_ERR; - break; - case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR: - event.event = IB_EVENT_QP_ACCESS_ERR; - break; - default: - pr_warn("Unexpected event type %d " - "on QP %06x\n", type, qp->qpn); - return; - } + if (!ibqp->event_handler) + goto out_no_handler; - ibqp->event_handler(&event, ibqp->qp_context); - } + qpe_work = kzalloc(sizeof(*qpe_work), GFP_ATOMIC); + if (!qpe_work) + goto out_no_handler; + + qpe_work->qp = qp; + qpe_work->type = type; + INIT_WORK(&qpe_work->work, mlx4_ib_handle_qp_event); + queue_work(mlx4_ib_qp_event_wq, &qpe_work->work); + return; + +out_no_handler: + mlx4_put_qp(qp); } static void mlx4_ib_wq_event(struct mlx4_qp *qp, enum mlx4_event type) @@ -4468,3 +4503,17 @@ void mlx4_ib_drain_rq(struct ib_qp *qp) handle_drain_completion(cq, &rdrain, dev); } + +int mlx4_ib_qp_event_init(void) +{ + mlx4_ib_qp_event_wq = alloc_ordered_workqueue("mlx4_ib_qp_event_wq", 0); + if (!mlx4_ib_qp_event_wq) + return -ENOMEM; + + return 0; +} + +void mlx4_ib_qp_event_cleanup(void) +{ + destroy_workqueue(mlx4_ib_qp_event_wq); +} diff --git a/drivers/infiniband/hw/mlx5/cmd.c b/drivers/infiniband/hw/mlx5/cmd.c index ff3742b0460a..1d0c8d5e745b 100644 --- a/drivers/infiniband/hw/mlx5/cmd.c +++ b/drivers/infiniband/hw/mlx5/cmd.c @@ -5,34 +5,41 @@ #include "cmd.h" -int mlx5_cmd_dump_fill_mkey(struct mlx5_core_dev *dev, u32 *mkey) +int mlx5r_cmd_query_special_mkeys(struct mlx5_ib_dev *dev) { u32 out[MLX5_ST_SZ_DW(query_special_contexts_out)] = {}; u32 in[MLX5_ST_SZ_DW(query_special_contexts_in)] = {}; + bool is_terminate, is_dump, is_null; int err; - MLX5_SET(query_special_contexts_in, in, opcode, - MLX5_CMD_OP_QUERY_SPECIAL_CONTEXTS); - err = mlx5_cmd_exec_inout(dev, query_special_contexts, in, out); - if (!err) - *mkey = MLX5_GET(query_special_contexts_out, out, - dump_fill_mkey); - return err; -} + is_terminate = MLX5_CAP_GEN(dev->mdev, terminate_scatter_list_mkey); + is_dump = MLX5_CAP_GEN(dev->mdev, dump_fill_mkey); + is_null = MLX5_CAP_GEN(dev->mdev, null_mkey); -int mlx5_cmd_null_mkey(struct mlx5_core_dev *dev, u32 *null_mkey) -{ - u32 out[MLX5_ST_SZ_DW(query_special_contexts_out)] = {}; - u32 in[MLX5_ST_SZ_DW(query_special_contexts_in)] = {}; - int err; + dev->mkeys.terminate_scatter_list_mkey = MLX5_TERMINATE_SCATTER_LIST_LKEY; + if (!is_terminate && !is_dump && !is_null) + return 0; MLX5_SET(query_special_contexts_in, in, opcode, MLX5_CMD_OP_QUERY_SPECIAL_CONTEXTS); - err = mlx5_cmd_exec_inout(dev, query_special_contexts, in, out); - if (!err) - *null_mkey = MLX5_GET(query_special_contexts_out, out, - null_mkey); - return err; + err = mlx5_cmd_exec_inout(dev->mdev, query_special_contexts, in, out); + if (err) + return err; + + if (is_dump) + dev->mkeys.dump_fill_mkey = MLX5_GET(query_special_contexts_out, + out, dump_fill_mkey); + + if (is_null) + dev->mkeys.null_mkey = cpu_to_be32( + MLX5_GET(query_special_contexts_out, out, null_mkey)); + + if (is_terminate) + dev->mkeys.terminate_scatter_list_mkey = + cpu_to_be32(MLX5_GET(query_special_contexts_out, out, + terminate_scatter_list_mkey)); + + return 0; } int mlx5_cmd_query_cong_params(struct mlx5_core_dev *dev, int cong_point, diff --git a/drivers/infiniband/hw/mlx5/cmd.h b/drivers/infiniband/hw/mlx5/cmd.h index ee46638db5de..93a971a40d11 100644 --- a/drivers/infiniband/hw/mlx5/cmd.h +++ b/drivers/infiniband/hw/mlx5/cmd.h @@ -37,8 +37,7 @@ #include <linux/kernel.h> #include <linux/mlx5/driver.h> -int mlx5_cmd_dump_fill_mkey(struct mlx5_core_dev *dev, u32 *mkey); -int mlx5_cmd_null_mkey(struct mlx5_core_dev *dev, u32 *null_mkey); +int mlx5r_cmd_query_special_mkeys(struct mlx5_ib_dev *dev); int mlx5_cmd_query_cong_params(struct mlx5_core_dev *dev, int cong_point, void *out); int mlx5_cmd_dealloc_pd(struct mlx5_core_dev *dev, u32 pdn, u16 uid); diff --git a/drivers/infiniband/hw/mlx5/cong.c b/drivers/infiniband/hw/mlx5/cong.c index 290ea8ac3838..f87531318feb 100644 --- a/drivers/infiniband/hw/mlx5/cong.c +++ b/drivers/infiniband/hw/mlx5/cong.c @@ -38,6 +38,7 @@ enum mlx5_ib_cong_node_type { MLX5_IB_RROCE_ECN_RP = 1, MLX5_IB_RROCE_ECN_NP = 2, + MLX5_IB_RROCE_GENERAL = 3, }; static const char * const mlx5_ib_dbg_cc_name[] = { @@ -61,6 +62,8 @@ static const char * const mlx5_ib_dbg_cc_name[] = { "np_cnp_dscp", "np_cnp_prio_mode", "np_cnp_prio", + "rtt_resp_dscp_valid", + "rtt_resp_dscp", }; #define MLX5_IB_RP_CLAMP_TGT_RATE_ATTR BIT(1) @@ -84,14 +87,18 @@ static const char * const mlx5_ib_dbg_cc_name[] = { #define MLX5_IB_NP_CNP_DSCP_ATTR BIT(3) #define MLX5_IB_NP_CNP_PRIO_MODE_ATTR BIT(4) +#define MLX5_IB_GENERAL_RTT_RESP_DSCP_ATTR BIT(0) + static enum mlx5_ib_cong_node_type mlx5_ib_param_to_node(enum mlx5_ib_dbg_cc_types param_offset) { - if (param_offset >= MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE && - param_offset <= MLX5_IB_DBG_CC_RP_GD) + if (param_offset <= MLX5_IB_DBG_CC_RP_GD) return MLX5_IB_RROCE_ECN_RP; - else + + if (param_offset <= MLX5_IB_DBG_CC_NP_CNP_PRIO) return MLX5_IB_RROCE_ECN_NP; + + return MLX5_IB_RROCE_GENERAL; } static u32 mlx5_get_cc_param_val(void *field, int offset) @@ -157,6 +164,12 @@ static u32 mlx5_get_cc_param_val(void *field, int offset) case MLX5_IB_DBG_CC_NP_CNP_PRIO: return MLX5_GET(cong_control_r_roce_ecn_np, field, cnp_802p_prio); + case MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP_VALID: + return MLX5_GET(cong_control_r_roce_general, field, + rtt_resp_dscp_valid); + case MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP: + return MLX5_GET(cong_control_r_roce_general, field, + rtt_resp_dscp); default: return 0; } @@ -264,6 +277,15 @@ static void mlx5_ib_set_cc_param_mask_val(void *field, int offset, MLX5_SET(cong_control_r_roce_ecn_np, field, cnp_prio_mode, 0); MLX5_SET(cong_control_r_roce_ecn_np, field, cnp_802p_prio, var); break; + case MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP_VALID: + *attr_mask |= MLX5_IB_GENERAL_RTT_RESP_DSCP_ATTR; + MLX5_SET(cong_control_r_roce_general, field, rtt_resp_dscp_valid, var); + break; + case MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP: + *attr_mask |= MLX5_IB_GENERAL_RTT_RESP_DSCP_ATTR; + MLX5_SET(cong_control_r_roce_general, field, rtt_resp_dscp_valid, 1); + MLX5_SET(cong_control_r_roce_general, field, rtt_resp_dscp, var); + break; } } diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 471c3455dfeb..5b988db66b8f 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -1756,13 +1756,9 @@ static int set_ucontext_resp(struct ib_ucontext *uctx, struct mlx5_ib_dev *dev = to_mdev(ibdev); struct mlx5_ib_ucontext *context = to_mucontext(uctx); struct mlx5_bfreg_info *bfregi = &context->bfregi; - int err; if (MLX5_CAP_GEN(dev->mdev, dump_fill_mkey)) { - err = mlx5_cmd_dump_fill_mkey(dev->mdev, - &resp->dump_fill_mkey); - if (err) - return err; + resp->dump_fill_mkey = dev->mkeys.dump_fill_mkey; resp->comp_mask |= MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_DUMP_FILL_MKEY; } @@ -3666,6 +3662,10 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev) dev->port[i].roce.last_port_state = IB_PORT_DOWN; } + err = mlx5r_cmd_query_special_mkeys(dev); + if (err) + return err; + err = mlx5_ib_init_multiport_master(dev); if (err) return err; @@ -4030,12 +4030,7 @@ static int mlx5_ib_stage_ib_reg_init(struct mlx5_ib_dev *dev) static void mlx5_ib_stage_pre_ib_reg_umr_cleanup(struct mlx5_ib_dev *dev) { - int err; - - err = mlx5_mkey_cache_cleanup(dev); - if (err) - mlx5_ib_warn(dev, "mr cache cleanup failed\n"); - + mlx5_mkey_cache_cleanup(dev); mlx5r_umr_resource_cleanup(dev); } @@ -4433,6 +4428,10 @@ static int __init mlx5_ib_init(void) return -ENOMEM; } + ret = mlx5_ib_qp_event_init(); + if (ret) + goto qp_event_err; + mlx5_ib_odp_init(); ret = mlx5r_rep_init(); if (ret) @@ -4450,6 +4449,8 @@ drv_err: mp_err: mlx5r_rep_cleanup(); rep_err: + mlx5_ib_qp_event_cleanup(); +qp_event_err: destroy_workqueue(mlx5_ib_event_wq); free_page((unsigned long)xlt_emergency_page); return ret; @@ -4461,6 +4462,7 @@ static void __exit mlx5_ib_cleanup(void) auxiliary_driver_unregister(&mlx5r_mp_driver); mlx5r_rep_cleanup(); + mlx5_ib_qp_event_cleanup(); destroy_workqueue(mlx5_ib_event_wq); free_page((unsigned long)xlt_emergency_page); } diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index 7394e7f36ba7..efa4dc6e7dee 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -617,12 +617,21 @@ enum mlx5_mkey_type { MLX5_MKEY_INDIRECT_DEVX, }; +struct mlx5r_cache_rb_key { + u8 ats:1; + unsigned int access_mode; + unsigned int access_flags; + unsigned int ndescs; +}; + struct mlx5_ib_mkey { u32 key; enum mlx5_mkey_type type; unsigned int ndescs; struct wait_queue_head wait; refcount_t usecount; + /* User Mkey must hold either a rb_key or a cache_ent. */ + struct mlx5r_cache_rb_key rb_key; struct mlx5_cache_ent *cache_ent; }; @@ -737,11 +746,11 @@ struct mlx5_cache_ent { unsigned long reserved; char name[4]; - u32 order; - u32 access_mode; - u32 page; - unsigned int ndescs; + struct rb_node node; + struct mlx5r_cache_rb_key rb_key; + + u8 is_tmp:1; u8 disabled:1; u8 fill_to_high_water:1; @@ -771,9 +780,11 @@ struct mlx5r_async_create_mkey { struct mlx5_mkey_cache { struct workqueue_struct *wq; - struct mlx5_cache_ent ent[MAX_MKEY_CACHE_ENTRIES]; - struct dentry *root; + struct rb_root rb_root; + struct mutex rb_lock; + struct dentry *fs_root; unsigned long last_add; + struct delayed_work remove_ent_dwork; }; struct mlx5_ib_port_resources { @@ -877,6 +888,8 @@ enum mlx5_ib_dbg_cc_types { MLX5_IB_DBG_CC_NP_CNP_DSCP, MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE, MLX5_IB_DBG_CC_NP_CNP_PRIO, + MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP_VALID, + MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP, MLX5_IB_DBG_CC_MAX, }; @@ -1054,6 +1067,13 @@ struct mlx5_port_caps { u8 ext_port_cap; }; + +struct mlx5_special_mkeys { + u32 dump_fill_mkey; + __be32 null_mkey; + __be32 terminate_scatter_list_mkey; +}; + struct mlx5_ib_dev { struct ib_device ib_dev; struct mlx5_core_dev *mdev; @@ -1084,7 +1104,6 @@ struct mlx5_ib_dev { struct xarray odp_mkeys; - u32 null_mkey; struct mlx5_ib_flow_db *flow_db; /* protect resources needed as part of reset flow */ spinlock_t reset_flow_resource_lock; @@ -1113,6 +1132,7 @@ struct mlx5_ib_dev { struct mlx5_port_caps port_caps[MLX5_MAX_PORTS]; u16 pkey_table_len; u8 lag_ports; + struct mlx5_special_mkeys mkeys; }; static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq) @@ -1319,11 +1339,15 @@ void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas, void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num); int mlx5_ib_get_cqe_size(struct ib_cq *ibcq); int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev); -int mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev); +void mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev); +struct mlx5_cache_ent * +mlx5r_cache_create_ent_locked(struct mlx5_ib_dev *dev, + struct mlx5r_cache_rb_key rb_key, + bool persistent_entry); struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, - struct mlx5_cache_ent *ent, - int access_flags); + int access_flags, int access_mode, + int ndescs); int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask, struct ib_mr_status *mr_status); @@ -1347,7 +1371,7 @@ int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq); void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev); int __init mlx5_ib_odp_init(void); void mlx5_ib_odp_cleanup(void); -void mlx5_odp_init_mkey_cache_entry(struct mlx5_cache_ent *ent); +int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev); void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, struct mlx5_ib_mr *mr, int flags); @@ -1366,7 +1390,10 @@ static inline int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, static inline void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev) {} static inline int mlx5_ib_odp_init(void) { return 0; } static inline void mlx5_ib_odp_cleanup(void) {} -static inline void mlx5_odp_init_mkey_cache_entry(struct mlx5_cache_ent *ent) {} +static inline int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev) +{ + return 0; +} static inline void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, struct mlx5_ib_mr *mr, int flags) {} diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index 053fe946e45a..67356f515261 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -140,19 +140,16 @@ static void create_mkey_warn(struct mlx5_ib_dev *dev, int status, void *out) mlx5_cmd_out_err(dev->mdev, MLX5_CMD_OP_CREATE_MKEY, 0, out); } - -static int push_mkey(struct mlx5_cache_ent *ent, bool limit_pendings, - void *to_store) +static int push_mkey_locked(struct mlx5_cache_ent *ent, bool limit_pendings, + void *to_store) { XA_STATE(xas, &ent->mkeys, 0); void *curr; - xa_lock_irq(&ent->mkeys); if (limit_pendings && - (ent->reserved - ent->stored) > MAX_PENDING_REG_MR) { - xa_unlock_irq(&ent->mkeys); + (ent->reserved - ent->stored) > MAX_PENDING_REG_MR) return -EAGAIN; - } + while (1) { /* * This is cmpxchg (NULL, XA_ZERO_ENTRY) however this version @@ -191,6 +188,7 @@ static int push_mkey(struct mlx5_cache_ent *ent, bool limit_pendings, break; xa_lock_irq(&ent->mkeys); } + xa_lock_irq(&ent->mkeys); if (xas_error(&xas)) return xas_error(&xas); if (WARN_ON(curr)) @@ -198,6 +196,17 @@ static int push_mkey(struct mlx5_cache_ent *ent, bool limit_pendings, return 0; } +static int push_mkey(struct mlx5_cache_ent *ent, bool limit_pendings, + void *to_store) +{ + int ret; + + xa_lock_irq(&ent->mkeys); + ret = push_mkey_locked(ent, limit_pendings, to_store); + xa_unlock_irq(&ent->mkeys); + return ret; +} + static void undo_push_reserve_mkey(struct mlx5_cache_ent *ent) { void *old; @@ -292,12 +301,14 @@ static void set_cache_mkc(struct mlx5_cache_ent *ent, void *mkc) set_mkc_access_pd_addr_fields(mkc, 0, 0, ent->dev->umrc.pd); MLX5_SET(mkc, mkc, free, 1); MLX5_SET(mkc, mkc, umr_en, 1); - MLX5_SET(mkc, mkc, access_mode_1_0, ent->access_mode & 0x3); - MLX5_SET(mkc, mkc, access_mode_4_2, (ent->access_mode >> 2) & 0x7); + MLX5_SET(mkc, mkc, access_mode_1_0, ent->rb_key.access_mode & 0x3); + MLX5_SET(mkc, mkc, access_mode_4_2, + (ent->rb_key.access_mode >> 2) & 0x7); MLX5_SET(mkc, mkc, translations_octword_size, - get_mkc_octo_size(ent->access_mode, ent->ndescs)); - MLX5_SET(mkc, mkc, log_page_size, ent->page); + get_mkc_octo_size(ent->rb_key.access_mode, + ent->rb_key.ndescs)); + MLX5_SET(mkc, mkc, log_page_size, PAGE_SHIFT); } /* Asynchronously schedule new MRs to be populated in the cache. */ @@ -515,18 +526,22 @@ static const struct file_operations limit_fops = { static bool someone_adding(struct mlx5_mkey_cache *cache) { - unsigned int i; - - for (i = 0; i < MAX_MKEY_CACHE_ENTRIES; i++) { - struct mlx5_cache_ent *ent = &cache->ent[i]; - bool ret; + struct mlx5_cache_ent *ent; + struct rb_node *node; + bool ret; + mutex_lock(&cache->rb_lock); + for (node = rb_first(&cache->rb_root); node; node = rb_next(node)) { + ent = rb_entry(node, struct mlx5_cache_ent, node); xa_lock_irq(&ent->mkeys); ret = ent->stored < ent->limit; xa_unlock_irq(&ent->mkeys); - if (ret) + if (ret) { + mutex_unlock(&cache->rb_lock); return true; + } } + mutex_unlock(&cache->rb_lock); return false; } @@ -539,7 +554,7 @@ static void queue_adjust_cache_locked(struct mlx5_cache_ent *ent) { lockdep_assert_held(&ent->mkeys.xa_lock); - if (ent->disabled || READ_ONCE(ent->dev->fill_delay)) + if (ent->disabled || READ_ONCE(ent->dev->fill_delay) || ent->is_tmp) return; if (ent->stored < ent->limit) { ent->fill_to_high_water = true; @@ -590,8 +605,8 @@ static void __cache_work_func(struct mlx5_cache_ent *ent) if (err != -EAGAIN) { mlx5_ib_warn( dev, - "command failed order %d, err %d\n", - ent->order, err); + "add keys command failed, err %d\n", + err); queue_delayed_work(cache->wq, &ent->dwork, msecs_to_jiffies(1000)); } @@ -637,16 +652,99 @@ static void delayed_cache_work_func(struct work_struct *work) __cache_work_func(ent); } -struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, - struct mlx5_cache_ent *ent, - int access_flags) +static int cache_ent_key_cmp(struct mlx5r_cache_rb_key key1, + struct mlx5r_cache_rb_key key2) +{ + int res; + + res = key1.ats - key2.ats; + if (res) + return res; + + res = key1.access_mode - key2.access_mode; + if (res) + return res; + + res = key1.access_flags - key2.access_flags; + if (res) + return res; + + /* + * keep ndescs the last in the compare table since the find function + * searches for an exact match on all properties and only closest + * match in size. + */ + return key1.ndescs - key2.ndescs; +} + +static int mlx5_cache_ent_insert(struct mlx5_mkey_cache *cache, + struct mlx5_cache_ent *ent) +{ + struct rb_node **new = &cache->rb_root.rb_node, *parent = NULL; + struct mlx5_cache_ent *cur; + int cmp; + + /* Figure out where to put new node */ + while (*new) { + cur = rb_entry(*new, struct mlx5_cache_ent, node); + parent = *new; + cmp = cache_ent_key_cmp(cur->rb_key, ent->rb_key); + if (cmp > 0) + new = &((*new)->rb_left); + if (cmp < 0) + new = &((*new)->rb_right); + if (cmp == 0) { + mutex_unlock(&cache->rb_lock); + return -EEXIST; + } + } + + /* Add new node and rebalance tree. */ + rb_link_node(&ent->node, parent, new); + rb_insert_color(&ent->node, &cache->rb_root); + + return 0; +} + +static struct mlx5_cache_ent * +mkey_cache_ent_from_rb_key(struct mlx5_ib_dev *dev, + struct mlx5r_cache_rb_key rb_key) +{ + struct rb_node *node = dev->cache.rb_root.rb_node; + struct mlx5_cache_ent *cur, *smallest = NULL; + int cmp; + + /* + * Find the smallest ent with order >= requested_order. + */ + while (node) { + cur = rb_entry(node, struct mlx5_cache_ent, node); + cmp = cache_ent_key_cmp(cur->rb_key, rb_key); + if (cmp > 0) { + smallest = cur; + node = node->rb_left; + } + if (cmp < 0) + node = node->rb_right; + if (cmp == 0) + return cur; + } + + return (smallest && + smallest->rb_key.access_mode == rb_key.access_mode && + smallest->rb_key.access_flags == rb_key.access_flags && + smallest->rb_key.ats == rb_key.ats) ? + smallest : + NULL; +} + +static struct mlx5_ib_mr *_mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, + struct mlx5_cache_ent *ent, + int access_flags) { struct mlx5_ib_mr *mr; int err; - if (!mlx5r_umr_can_reconfig(dev, 0, access_flags)) - return ERR_PTR(-EOPNOTSUPP); - mr = kzalloc(sizeof(*mr), GFP_KERNEL); if (!mr) return ERR_PTR(-ENOMEM); @@ -677,10 +775,48 @@ struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, return mr; } -static void clean_keys(struct mlx5_ib_dev *dev, int c) +static int get_unchangeable_access_flags(struct mlx5_ib_dev *dev, + int access_flags) +{ + int ret = 0; + + if ((access_flags & IB_ACCESS_REMOTE_ATOMIC) && + MLX5_CAP_GEN(dev->mdev, atomic) && + MLX5_CAP_GEN(dev->mdev, umr_modify_atomic_disabled)) + ret |= IB_ACCESS_REMOTE_ATOMIC; + + if ((access_flags & IB_ACCESS_RELAXED_ORDERING) && + MLX5_CAP_GEN(dev->mdev, relaxed_ordering_write) && + !MLX5_CAP_GEN(dev->mdev, relaxed_ordering_write_umr)) + ret |= IB_ACCESS_RELAXED_ORDERING; + + if ((access_flags & IB_ACCESS_RELAXED_ORDERING) && + MLX5_CAP_GEN(dev->mdev, relaxed_ordering_read) && + !MLX5_CAP_GEN(dev->mdev, relaxed_ordering_read_umr)) + ret |= IB_ACCESS_RELAXED_ORDERING; + + return ret; +} + +struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, + int access_flags, int access_mode, + int ndescs) +{ + struct mlx5r_cache_rb_key rb_key = { + .ndescs = ndescs, + .access_mode = access_mode, + .access_flags = get_unchangeable_access_flags(dev, access_flags) + }; + struct mlx5_cache_ent *ent = mkey_cache_ent_from_rb_key(dev, rb_key); + + if (!ent) + return ERR_PTR(-EOPNOTSUPP); + + return _mlx5_mr_cache_alloc(dev, ent, access_flags); +} + +static void clean_keys(struct mlx5_ib_dev *dev, struct mlx5_cache_ent *ent) { - struct mlx5_mkey_cache *cache = &dev->cache; - struct mlx5_cache_ent *ent = &cache->ent[c]; u32 mkey; cancel_delayed_work(&ent->dwork); @@ -699,31 +835,39 @@ static void mlx5_mkey_cache_debugfs_cleanup(struct mlx5_ib_dev *dev) if (!mlx5_debugfs_root || dev->is_rep) return; - debugfs_remove_recursive(dev->cache.root); - dev->cache.root = NULL; + debugfs_remove_recursive(dev->cache.fs_root); + dev->cache.fs_root = NULL; } -static void mlx5_mkey_cache_debugfs_init(struct mlx5_ib_dev *dev) +static void mlx5_mkey_cache_debugfs_add_ent(struct mlx5_ib_dev *dev, + struct mlx5_cache_ent *ent) { - struct mlx5_mkey_cache *cache = &dev->cache; - struct mlx5_cache_ent *ent; + int order = order_base_2(ent->rb_key.ndescs); struct dentry *dir; - int i; if (!mlx5_debugfs_root || dev->is_rep) return; - cache->root = debugfs_create_dir("mr_cache", mlx5_debugfs_get_dev_root(dev->mdev)); + if (ent->rb_key.access_mode == MLX5_MKC_ACCESS_MODE_KSM) + order = MLX5_IMR_KSM_CACHE_ENTRY + 2; - for (i = 0; i < MAX_MKEY_CACHE_ENTRIES; i++) { - ent = &cache->ent[i]; - sprintf(ent->name, "%d", ent->order); - dir = debugfs_create_dir(ent->name, cache->root); - debugfs_create_file("size", 0600, dir, ent, &size_fops); - debugfs_create_file("limit", 0600, dir, ent, &limit_fops); - debugfs_create_ulong("cur", 0400, dir, &ent->stored); - debugfs_create_u32("miss", 0600, dir, &ent->miss); - } + sprintf(ent->name, "%d", order); + dir = debugfs_create_dir(ent->name, dev->cache.fs_root); + debugfs_create_file("size", 0600, dir, ent, &size_fops); + debugfs_create_file("limit", 0600, dir, ent, &limit_fops); + debugfs_create_ulong("cur", 0400, dir, &ent->stored); + debugfs_create_u32("miss", 0600, dir, &ent->miss); +} + +static void mlx5_mkey_cache_debugfs_init(struct mlx5_ib_dev *dev) +{ + struct dentry *dbg_root = mlx5_debugfs_get_dev_root(dev->mdev); + struct mlx5_mkey_cache *cache = &dev->cache; + + if (!mlx5_debugfs_root || dev->is_rep) + return; + + cache->fs_root = debugfs_create_dir("mr_cache", dbg_root); } static void delay_time_func(struct timer_list *t) @@ -733,13 +877,100 @@ static void delay_time_func(struct timer_list *t) WRITE_ONCE(dev->fill_delay, 0); } +struct mlx5_cache_ent * +mlx5r_cache_create_ent_locked(struct mlx5_ib_dev *dev, + struct mlx5r_cache_rb_key rb_key, + bool persistent_entry) +{ + struct mlx5_cache_ent *ent; + int order; + int ret; + + ent = kzalloc(sizeof(*ent), GFP_KERNEL); + if (!ent) + return ERR_PTR(-ENOMEM); + + xa_init_flags(&ent->mkeys, XA_FLAGS_LOCK_IRQ); + ent->rb_key = rb_key; + ent->dev = dev; + ent->is_tmp = !persistent_entry; + + INIT_DELAYED_WORK(&ent->dwork, delayed_cache_work_func); + + ret = mlx5_cache_ent_insert(&dev->cache, ent); + if (ret) { + kfree(ent); + return ERR_PTR(ret); + } + + if (persistent_entry) { + if (rb_key.access_mode == MLX5_MKC_ACCESS_MODE_KSM) + order = MLX5_IMR_KSM_CACHE_ENTRY; + else + order = order_base_2(rb_key.ndescs) - 2; + + if ((dev->mdev->profile.mask & MLX5_PROF_MASK_MR_CACHE) && + !dev->is_rep && mlx5_core_is_pf(dev->mdev) && + mlx5r_umr_can_load_pas(dev, 0)) + ent->limit = dev->mdev->profile.mr_cache[order].limit; + else + ent->limit = 0; + + mlx5_mkey_cache_debugfs_add_ent(dev, ent); + } else { + mod_delayed_work(ent->dev->cache.wq, + &ent->dev->cache.remove_ent_dwork, + msecs_to_jiffies(30 * 1000)); + } + + return ent; +} + +static void remove_ent_work_func(struct work_struct *work) +{ + struct mlx5_mkey_cache *cache; + struct mlx5_cache_ent *ent; + struct rb_node *cur; + + cache = container_of(work, struct mlx5_mkey_cache, + remove_ent_dwork.work); + mutex_lock(&cache->rb_lock); + cur = rb_last(&cache->rb_root); + while (cur) { + ent = rb_entry(cur, struct mlx5_cache_ent, node); + cur = rb_prev(cur); + mutex_unlock(&cache->rb_lock); + + xa_lock_irq(&ent->mkeys); + if (!ent->is_tmp) { + xa_unlock_irq(&ent->mkeys); + mutex_lock(&cache->rb_lock); + continue; + } + xa_unlock_irq(&ent->mkeys); + + clean_keys(ent->dev, ent); + mutex_lock(&cache->rb_lock); + } + mutex_unlock(&cache->rb_lock); +} + int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev) { struct mlx5_mkey_cache *cache = &dev->cache; + struct rb_root *root = &dev->cache.rb_root; + struct mlx5r_cache_rb_key rb_key = { + .access_mode = MLX5_MKC_ACCESS_MODE_MTT, + }; struct mlx5_cache_ent *ent; + struct rb_node *node; + int ret; int i; mutex_init(&dev->slow_path_mutex); + mutex_init(&dev->cache.rb_lock); + dev->cache.rb_root = RB_ROOT; + INIT_DELAYED_WORK(&dev->cache.remove_ent_dwork, remove_ent_work_func); cache->wq = alloc_ordered_workqueue("mkey_cache", WQ_MEM_RECLAIM); if (!cache->wq) { mlx5_ib_warn(dev, "failed to create work queue\n"); @@ -748,52 +979,51 @@ int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev) mlx5_cmd_init_async_ctx(dev->mdev, &dev->async_ctx); timer_setup(&dev->delay_timer, delay_time_func, 0); - for (i = 0; i < MAX_MKEY_CACHE_ENTRIES; i++) { - ent = &cache->ent[i]; - xa_init_flags(&ent->mkeys, XA_FLAGS_LOCK_IRQ); - ent->order = i + 2; - ent->dev = dev; - ent->limit = 0; - - INIT_DELAYED_WORK(&ent->dwork, delayed_cache_work_func); - - if (i > MKEY_CACHE_LAST_STD_ENTRY) { - mlx5_odp_init_mkey_cache_entry(ent); - continue; + mlx5_mkey_cache_debugfs_init(dev); + mutex_lock(&cache->rb_lock); + for (i = 0; i <= mkey_cache_max_order(dev); i++) { + rb_key.ndescs = 1 << (i + 2); + ent = mlx5r_cache_create_ent_locked(dev, rb_key, true); + if (IS_ERR(ent)) { + ret = PTR_ERR(ent); + goto err; } + } - if (ent->order > mkey_cache_max_order(dev)) - continue; + ret = mlx5_odp_init_mkey_cache(dev); + if (ret) + goto err; - ent->page = PAGE_SHIFT; - ent->ndescs = 1 << ent->order; - ent->access_mode = MLX5_MKC_ACCESS_MODE_MTT; - if ((dev->mdev->profile.mask & MLX5_PROF_MASK_MR_CACHE) && - !dev->is_rep && mlx5_core_is_pf(dev->mdev) && - mlx5r_umr_can_load_pas(dev, 0)) - ent->limit = dev->mdev->profile.mr_cache[i].limit; - else - ent->limit = 0; + mutex_unlock(&cache->rb_lock); + for (node = rb_first(root); node; node = rb_next(node)) { + ent = rb_entry(node, struct mlx5_cache_ent, node); xa_lock_irq(&ent->mkeys); queue_adjust_cache_locked(ent); xa_unlock_irq(&ent->mkeys); } - mlx5_mkey_cache_debugfs_init(dev); - return 0; + +err: + mutex_unlock(&cache->rb_lock); + mlx5_mkey_cache_debugfs_cleanup(dev); + mlx5_ib_warn(dev, "failed to create mkey cache entry\n"); + return ret; } -int mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev) +void mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev) { - unsigned int i; + struct rb_root *root = &dev->cache.rb_root; + struct mlx5_cache_ent *ent; + struct rb_node *node; if (!dev->cache.wq) - return 0; - - for (i = 0; i < MAX_MKEY_CACHE_ENTRIES; i++) { - struct mlx5_cache_ent *ent = &dev->cache.ent[i]; + return; + cancel_delayed_work_sync(&dev->cache.remove_ent_dwork); + mutex_lock(&dev->cache.rb_lock); + for (node = rb_first(root); node; node = rb_next(node)) { + ent = rb_entry(node, struct mlx5_cache_ent, node); xa_lock_irq(&ent->mkeys); ent->disabled = true; xa_unlock_irq(&ent->mkeys); @@ -803,13 +1033,18 @@ int mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev) mlx5_mkey_cache_debugfs_cleanup(dev); mlx5_cmd_cleanup_async_ctx(&dev->async_ctx); - for (i = 0; i < MAX_MKEY_CACHE_ENTRIES; i++) - clean_keys(dev, i); + node = rb_first(root); + while (node) { + ent = rb_entry(node, struct mlx5_cache_ent, node); + node = rb_next(node); + clean_keys(dev, ent); + rb_erase(&ent->node, root); + kfree(ent); + } + mutex_unlock(&dev->cache.rb_lock); destroy_workqueue(dev->cache.wq); del_timer_sync(&dev->delay_timer); - - return 0; } struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc) @@ -873,23 +1108,10 @@ static int get_octo_len(u64 addr, u64 len, int page_shift) static int mkey_cache_max_order(struct mlx5_ib_dev *dev) { if (MLX5_CAP_GEN(dev->mdev, umr_extended_translation_offset)) - return MKEY_CACHE_LAST_STD_ENTRY + 2; + return MKEY_CACHE_LAST_STD_ENTRY; return MLX5_MAX_UMR_SHIFT; } -static struct mlx5_cache_ent *mkey_cache_ent_from_order(struct mlx5_ib_dev *dev, - unsigned int order) -{ - struct mlx5_mkey_cache *cache = &dev->cache; - - if (order < cache->ent[0].order) - return &cache->ent[0]; - order = order - cache->ent[0].order; - if (order > MKEY_CACHE_LAST_STD_ENTRY) - return NULL; - return &cache->ent[order]; -} - static void set_mr_fields(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr, u64 length, int access_flags, u64 iova) { @@ -916,6 +1138,9 @@ static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, struct ib_umem *umem, u64 iova, int access_flags) { + struct mlx5r_cache_rb_key rb_key = { + .access_mode = MLX5_MKC_ACCESS_MODE_MTT, + }; struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_cache_ent *ent; struct mlx5_ib_mr *mr; @@ -928,22 +1153,26 @@ static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, 0, iova); if (WARN_ON(!page_size)) return ERR_PTR(-EINVAL); - ent = mkey_cache_ent_from_order( - dev, order_base_2(ib_umem_num_dma_blocks(umem, page_size))); + + rb_key.ndescs = ib_umem_num_dma_blocks(umem, page_size); + rb_key.ats = mlx5_umem_needs_ats(dev, umem, access_flags); + rb_key.access_flags = get_unchangeable_access_flags(dev, access_flags); + ent = mkey_cache_ent_from_rb_key(dev, rb_key); /* - * Matches access in alloc_cache_mr(). If the MR can't come from the - * cache then synchronously create an uncached one. + * If the MR can't come from the cache then synchronously create an uncached + * one. */ - if (!ent || ent->limit == 0 || - !mlx5r_umr_can_reconfig(dev, 0, access_flags) || - mlx5_umem_needs_ats(dev, umem, access_flags)) { + if (!ent) { mutex_lock(&dev->slow_path_mutex); mr = reg_create(pd, umem, iova, access_flags, page_size, false); mutex_unlock(&dev->slow_path_mutex); + if (IS_ERR(mr)) + return mr; + mr->mmkey.rb_key = rb_key; return mr; } - mr = mlx5_mr_cache_alloc(dev, ent, access_flags); + mr = _mlx5_mr_cache_alloc(dev, ent, access_flags); if (IS_ERR(mr)) return mr; @@ -1030,6 +1259,7 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, goto err_2; } mr->mmkey.type = MLX5_MKEY_MR; + mr->mmkey.ndescs = get_octo_len(iova, umem->length, mr->page_shift); mr->umem = umem; set_mr_fields(dev, mr, umem->length, access_flags, iova); kvfree(in); @@ -1372,7 +1602,7 @@ static bool can_use_umr_rereg_pas(struct mlx5_ib_mr *mr, mlx5_umem_find_best_pgsz(new_umem, mkc, log_page_size, 0, iova); if (WARN_ON(!*page_size)) return false; - return (1ULL << mr->mmkey.cache_ent->order) >= + return (mr->mmkey.cache_ent->rb_key.ndescs) >= ib_umem_num_dma_blocks(new_umem, *page_size); } @@ -1567,6 +1797,49 @@ mlx5_free_priv_descs(struct mlx5_ib_mr *mr) } } +static int cache_ent_find_and_store(struct mlx5_ib_dev *dev, + struct mlx5_ib_mr *mr) +{ + struct mlx5_mkey_cache *cache = &dev->cache; + struct mlx5_cache_ent *ent; + int ret; + + if (mr->mmkey.cache_ent) { + xa_lock_irq(&mr->mmkey.cache_ent->mkeys); + mr->mmkey.cache_ent->in_use--; + goto end; + } + + mutex_lock(&cache->rb_lock); + ent = mkey_cache_ent_from_rb_key(dev, mr->mmkey.rb_key); + if (ent) { + if (ent->rb_key.ndescs == mr->mmkey.rb_key.ndescs) { + if (ent->disabled) { + mutex_unlock(&cache->rb_lock); + return -EOPNOTSUPP; + } + mr->mmkey.cache_ent = ent; + xa_lock_irq(&mr->mmkey.cache_ent->mkeys); + mutex_unlock(&cache->rb_lock); + goto end; + } + } + + ent = mlx5r_cache_create_ent_locked(dev, mr->mmkey.rb_key, false); + mutex_unlock(&cache->rb_lock); + if (IS_ERR(ent)) + return PTR_ERR(ent); + + mr->mmkey.cache_ent = ent; + xa_lock_irq(&mr->mmkey.cache_ent->mkeys); + +end: + ret = push_mkey_locked(mr->mmkey.cache_ent, false, + xa_mk_value(mr->mmkey.key)); + xa_unlock_irq(&mr->mmkey.cache_ent->mkeys); + return ret; +} + int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) { struct mlx5_ib_mr *mr = to_mmr(ibmr); @@ -1612,16 +1885,11 @@ int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) } /* Stop DMA */ - if (mr->mmkey.cache_ent) { - xa_lock_irq(&mr->mmkey.cache_ent->mkeys); - mr->mmkey.cache_ent->in_use--; - xa_unlock_irq(&mr->mmkey.cache_ent->mkeys); - + if (mr->umem && mlx5r_umr_can_load_pas(dev, mr->umem->length)) if (mlx5r_umr_revoke_mr(mr) || - push_mkey(mr->mmkey.cache_ent, false, - xa_mk_value(mr->mmkey.key))) + cache_ent_find_and_store(dev, mr)) mr->mmkey.cache_ent = NULL; - } + if (!mr->mmkey.cache_ent) { rc = destroy_mkey(to_mdev(mr->ibmr.device), mr); if (rc) diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c index e6e021af6aa9..4a04cbc5b78a 100644 --- a/drivers/infiniband/hw/mlx5/odp.c +++ b/drivers/infiniband/hw/mlx5/odp.c @@ -104,7 +104,7 @@ static void populate_klm(struct mlx5_klm *pklm, size_t idx, size_t nentries, if (flags & MLX5_IB_UPD_XLT_ZAP) { for (; pklm != end; pklm++, idx++) { pklm->bcount = cpu_to_be32(MLX5_IMR_MTT_SIZE); - pklm->key = cpu_to_be32(mr_to_mdev(imr)->null_mkey); + pklm->key = mr_to_mdev(imr)->mkeys.null_mkey; pklm->va = 0; } return; @@ -137,7 +137,7 @@ static void populate_klm(struct mlx5_klm *pklm, size_t idx, size_t nentries, pklm->key = cpu_to_be32(mtt->ibmr.lkey); pklm->va = cpu_to_be64(idx * MLX5_IMR_MTT_SIZE); } else { - pklm->key = cpu_to_be32(mr_to_mdev(imr)->null_mkey); + pklm->key = mr_to_mdev(imr)->mkeys.null_mkey; pklm->va = 0; } } @@ -417,8 +417,9 @@ static struct mlx5_ib_mr *implicit_get_child_mr(struct mlx5_ib_mr *imr, if (IS_ERR(odp)) return ERR_CAST(odp); - mr = mlx5_mr_cache_alloc(dev, &dev->cache.ent[MLX5_IMR_MTT_CACHE_ENTRY], - imr->access_flags); + mr = mlx5_mr_cache_alloc(dev, imr->access_flags, + MLX5_MKC_ACCESS_MODE_MTT, + MLX5_IMR_MTT_ENTRIES); if (IS_ERR(mr)) { ib_umem_odp_release(odp); return mr; @@ -492,9 +493,8 @@ struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, if (IS_ERR(umem_odp)) return ERR_CAST(umem_odp); - imr = mlx5_mr_cache_alloc(dev, - &dev->cache.ent[MLX5_IMR_KSM_CACHE_ENTRY], - access_flags); + imr = mlx5_mr_cache_alloc(dev, access_flags, MLX5_MKC_ACCESS_MODE_KSM, + mlx5_imr_ksm_entries); if (IS_ERR(imr)) { ib_umem_odp_release(umem_odp); return imr; @@ -986,7 +986,7 @@ static int pagefault_data_segments(struct mlx5_ib_dev *dev, { int ret = 0, npages = 0; u64 io_virt; - u32 key; + __be32 key; u32 byte_count; size_t bcnt; int inline_segment; @@ -1000,7 +1000,7 @@ static int pagefault_data_segments(struct mlx5_ib_dev *dev, struct mlx5_wqe_data_seg *dseg = wqe; io_virt = be64_to_cpu(dseg->addr); - key = be32_to_cpu(dseg->lkey); + key = dseg->lkey; byte_count = be32_to_cpu(dseg->byte_count); inline_segment = !!(byte_count & MLX5_INLINE_SEG); bcnt = byte_count & ~MLX5_INLINE_SEG; @@ -1014,7 +1014,8 @@ static int pagefault_data_segments(struct mlx5_ib_dev *dev, } /* receive WQE end of sg list. */ - if (receive_queue && bcnt == 0 && key == MLX5_INVALID_LKEY && + if (receive_queue && bcnt == 0 && + key == dev->mkeys.terminate_scatter_list_mkey && io_virt == 0) break; @@ -1034,7 +1035,7 @@ static int pagefault_data_segments(struct mlx5_ib_dev *dev, continue; } - ret = pagefault_single_data_segment(dev, NULL, key, + ret = pagefault_single_data_segment(dev, NULL, be32_to_cpu(key), io_virt, bcnt, &pfault->bytes_committed, bytes_mapped); @@ -1587,26 +1588,22 @@ mlx5_ib_odp_destroy_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq) return err; } -void mlx5_odp_init_mkey_cache_entry(struct mlx5_cache_ent *ent) +int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev) { - if (!(ent->dev->odp_caps.general_caps & IB_ODP_SUPPORT_IMPLICIT)) - return; + struct mlx5r_cache_rb_key rb_key = { + .access_mode = MLX5_MKC_ACCESS_MODE_KSM, + .ndescs = mlx5_imr_ksm_entries, + }; + struct mlx5_cache_ent *ent; - switch (ent->order - 2) { - case MLX5_IMR_MTT_CACHE_ENTRY: - ent->page = PAGE_SHIFT; - ent->ndescs = MLX5_IMR_MTT_ENTRIES; - ent->access_mode = MLX5_MKC_ACCESS_MODE_MTT; - ent->limit = 0; - break; + if (!(dev->odp_caps.general_caps & IB_ODP_SUPPORT_IMPLICIT)) + return 0; - case MLX5_IMR_KSM_CACHE_ENTRY: - ent->page = MLX5_KSM_PAGE_SHIFT; - ent->ndescs = mlx5_imr_ksm_entries; - ent->access_mode = MLX5_MKC_ACCESS_MODE_KSM; - ent->limit = 0; - break; - } + ent = mlx5r_cache_create_ent_locked(dev, rb_key, true); + if (IS_ERR(ent)) + return PTR_ERR(ent); + + return 0; } static const struct ib_device_ops mlx5_ib_dev_odp_ops = { @@ -1615,25 +1612,15 @@ static const struct ib_device_ops mlx5_ib_dev_odp_ops = { int mlx5_ib_odp_init_one(struct mlx5_ib_dev *dev) { - int ret = 0; - internal_fill_odp_caps(dev); if (!(dev->odp_caps.general_caps & IB_ODP_SUPPORT)) - return ret; + return 0; ib_set_device_ops(&dev->ib_dev, &mlx5_ib_dev_odp_ops); - if (dev->odp_caps.general_caps & IB_ODP_SUPPORT_IMPLICIT) { - ret = mlx5_cmd_null_mkey(dev->mdev, &dev->null_mkey); - if (ret) { - mlx5_ib_err(dev, "Error getting null_mkey %d\n", ret); - return ret; - } - } - mutex_init(&dev->odp_eq_mutex); - return ret; + return 0; } void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *dev) diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c index cf953d23d18d..7cc3b973dec7 100644 --- a/drivers/infiniband/hw/mlx5/qp.c +++ b/drivers/infiniband/hw/mlx5/qp.c @@ -71,6 +71,14 @@ struct mlx5_modify_raw_qp_param { u32 port; }; +struct mlx5_ib_qp_event_work { + struct work_struct work; + struct mlx5_core_qp *qp; + int type; +}; + +static struct workqueue_struct *mlx5_ib_qp_event_wq; + static void get_cqs(enum ib_qp_type qp_type, struct ib_cq *ib_send_cq, struct ib_cq *ib_recv_cq, struct mlx5_ib_cq **send_cq, struct mlx5_ib_cq **recv_cq); @@ -302,51 +310,120 @@ int mlx5_ib_read_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, void *buffer, return mlx5_ib_read_user_wqe_srq(srq, wqe_index, buffer, buflen, bc); } +static void mlx5_ib_qp_err_syndrome(struct ib_qp *ibqp) +{ + struct mlx5_ib_dev *dev = to_mdev(ibqp->device); + int outlen = MLX5_ST_SZ_BYTES(query_qp_out); + struct mlx5_ib_qp *qp = to_mqp(ibqp); + void *pas_ext_union, *err_syn; + u32 *outb; + int err; + + if (!MLX5_CAP_GEN(dev->mdev, qpc_extension) || + !MLX5_CAP_GEN(dev->mdev, qp_error_syndrome)) + return; + + outb = kzalloc(outlen, GFP_KERNEL); + if (!outb) + return; + + err = mlx5_core_qp_query(dev, &qp->trans_qp.base.mqp, outb, outlen, + true); + if (err) + goto out; + + pas_ext_union = + MLX5_ADDR_OF(query_qp_out, outb, qp_pas_or_qpc_ext_and_pas); + err_syn = MLX5_ADDR_OF(qpc_extension_and_pas_list_in, pas_ext_union, + qpc_data_extension.error_syndrome); + + pr_err("%s/%d: QP %d error: %s (0x%x 0x%x 0x%x)\n", + ibqp->device->name, ibqp->port, ibqp->qp_num, + ib_wc_status_msg( + MLX5_GET(cqe_error_syndrome, err_syn, syndrome)), + MLX5_GET(cqe_error_syndrome, err_syn, vendor_error_syndrome), + MLX5_GET(cqe_error_syndrome, err_syn, hw_syndrome_type), + MLX5_GET(cqe_error_syndrome, err_syn, hw_error_syndrome)); +out: + kfree(outb); +} + +static void mlx5_ib_handle_qp_event(struct work_struct *_work) +{ + struct mlx5_ib_qp_event_work *qpe_work = + container_of(_work, struct mlx5_ib_qp_event_work, work); + struct ib_qp *ibqp = &to_mibqp(qpe_work->qp)->ibqp; + struct ib_event event = {}; + + event.device = ibqp->device; + event.element.qp = ibqp; + switch (qpe_work->type) { + case MLX5_EVENT_TYPE_PATH_MIG: + event.event = IB_EVENT_PATH_MIG; + break; + case MLX5_EVENT_TYPE_COMM_EST: + event.event = IB_EVENT_COMM_EST; + break; + case MLX5_EVENT_TYPE_SQ_DRAINED: + event.event = IB_EVENT_SQ_DRAINED; + break; + case MLX5_EVENT_TYPE_SRQ_LAST_WQE: + event.event = IB_EVENT_QP_LAST_WQE_REACHED; + break; + case MLX5_EVENT_TYPE_WQ_CATAS_ERROR: + event.event = IB_EVENT_QP_FATAL; + break; + case MLX5_EVENT_TYPE_PATH_MIG_FAILED: + event.event = IB_EVENT_PATH_MIG_ERR; + break; + case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR: + event.event = IB_EVENT_QP_REQ_ERR; + break; + case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR: + event.event = IB_EVENT_QP_ACCESS_ERR; + break; + default: + pr_warn("mlx5_ib: Unexpected event type %d on QP %06x\n", + qpe_work->type, qpe_work->qp->qpn); + goto out; + } + + if ((event.event == IB_EVENT_QP_FATAL) || + (event.event == IB_EVENT_QP_ACCESS_ERR)) + mlx5_ib_qp_err_syndrome(ibqp); + + ibqp->event_handler(&event, ibqp->qp_context); + +out: + mlx5_core_res_put(&qpe_work->qp->common); + kfree(qpe_work); +} + static void mlx5_ib_qp_event(struct mlx5_core_qp *qp, int type) { struct ib_qp *ibqp = &to_mibqp(qp)->ibqp; - struct ib_event event; + struct mlx5_ib_qp_event_work *qpe_work; if (type == MLX5_EVENT_TYPE_PATH_MIG) { /* This event is only valid for trans_qps */ to_mibqp(qp)->port = to_mibqp(qp)->trans_qp.alt_port; } - if (ibqp->event_handler) { - event.device = ibqp->device; - event.element.qp = ibqp; - switch (type) { - case MLX5_EVENT_TYPE_PATH_MIG: - event.event = IB_EVENT_PATH_MIG; - break; - case MLX5_EVENT_TYPE_COMM_EST: - event.event = IB_EVENT_COMM_EST; - break; - case MLX5_EVENT_TYPE_SQ_DRAINED: - event.event = IB_EVENT_SQ_DRAINED; - break; - case MLX5_EVENT_TYPE_SRQ_LAST_WQE: - event.event = IB_EVENT_QP_LAST_WQE_REACHED; - break; - case MLX5_EVENT_TYPE_WQ_CATAS_ERROR: - event.event = IB_EVENT_QP_FATAL; - break; - case MLX5_EVENT_TYPE_PATH_MIG_FAILED: - event.event = IB_EVENT_PATH_MIG_ERR; - break; - case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR: - event.event = IB_EVENT_QP_REQ_ERR; - break; - case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR: - event.event = IB_EVENT_QP_ACCESS_ERR; - break; - default: - pr_warn("mlx5_ib: Unexpected event type %d on QP %06x\n", type, qp->qpn); - return; - } + if (!ibqp->event_handler) + goto out_no_handler; - ibqp->event_handler(&event, ibqp->qp_context); - } + qpe_work = kzalloc(sizeof(*qpe_work), GFP_ATOMIC); + if (!qpe_work) + goto out_no_handler; + + qpe_work->qp = qp; + qpe_work->type = type; + INIT_WORK(&qpe_work->work, mlx5_ib_handle_qp_event); + queue_work(mlx5_ib_qp_event_wq, &qpe_work->work); + return; + +out_no_handler: + mlx5_core_res_put(&qp->common); } static int set_rq_size(struct mlx5_ib_dev *dev, struct ib_qp_cap *cap, @@ -4827,7 +4904,8 @@ static int query_qp_attr(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp, if (!outb) return -ENOMEM; - err = mlx5_core_qp_query(dev, &qp->trans_qp.base.mqp, outb, outlen); + err = mlx5_core_qp_query(dev, &qp->trans_qp.base.mqp, outb, outlen, + false); if (err) goto out; @@ -5720,3 +5798,17 @@ out: mutex_unlock(&mqp->mutex); return err; } + +int mlx5_ib_qp_event_init(void) +{ + mlx5_ib_qp_event_wq = alloc_ordered_workqueue("mlx5_ib_qp_event_wq", 0); + if (!mlx5_ib_qp_event_wq) + return -ENOMEM; + + return 0; +} + +void mlx5_ib_qp_event_cleanup(void) +{ + destroy_workqueue(mlx5_ib_qp_event_wq); +} diff --git a/drivers/infiniband/hw/mlx5/qp.h b/drivers/infiniband/hw/mlx5/qp.h index 5d4e140db99c..77f9b4a54816 100644 --- a/drivers/infiniband/hw/mlx5/qp.h +++ b/drivers/infiniband/hw/mlx5/qp.h @@ -20,7 +20,7 @@ int mlx5_core_qp_modify(struct mlx5_ib_dev *dev, u16 opcode, u32 opt_param_mask, int mlx5_core_destroy_qp(struct mlx5_ib_dev *dev, struct mlx5_core_qp *qp); int mlx5_core_destroy_dct(struct mlx5_ib_dev *dev, struct mlx5_core_dct *dct); int mlx5_core_qp_query(struct mlx5_ib_dev *dev, struct mlx5_core_qp *qp, - u32 *out, int outlen); + u32 *out, int outlen, bool qpc_ext); int mlx5_core_dct_query(struct mlx5_ib_dev *dev, struct mlx5_core_dct *dct, u32 *out, int outlen); @@ -44,4 +44,6 @@ void mlx5_core_res_put(struct mlx5_core_rsc_common *res); int mlx5_core_xrcd_alloc(struct mlx5_ib_dev *dev, u32 *xrcdn); int mlx5_core_xrcd_dealloc(struct mlx5_ib_dev *dev, u32 xrcdn); int mlx5_ib_qp_set_counter(struct ib_qp *qp, struct rdma_counter *counter); +int mlx5_ib_qp_event_init(void); +void mlx5_ib_qp_event_cleanup(void); #endif /* _MLX5_IB_QP_H */ diff --git a/drivers/infiniband/hw/mlx5/qpc.c b/drivers/infiniband/hw/mlx5/qpc.c index 542e4c63a8de..bae0334d6e7f 100644 --- a/drivers/infiniband/hw/mlx5/qpc.c +++ b/drivers/infiniband/hw/mlx5/qpc.c @@ -135,7 +135,8 @@ static int rsc_event_notifier(struct notifier_block *nb, case MLX5_RES_SQ: qp = (struct mlx5_core_qp *)common; qp->event(qp, event_type); - break; + /* Need to put resource in event handler */ + return NOTIFY_OK; case MLX5_RES_DCT: dct = (struct mlx5_core_dct *)common; if (event_type == MLX5_EVENT_TYPE_DCT_DRAINED) @@ -504,12 +505,14 @@ void mlx5_cleanup_qp_table(struct mlx5_ib_dev *dev) } int mlx5_core_qp_query(struct mlx5_ib_dev *dev, struct mlx5_core_qp *qp, - u32 *out, int outlen) + u32 *out, int outlen, bool qpc_ext) { u32 in[MLX5_ST_SZ_DW(query_qp_in)] = {}; MLX5_SET(query_qp_in, in, opcode, MLX5_CMD_OP_QUERY_QP); MLX5_SET(query_qp_in, in, qpn, qp->qpn); + MLX5_SET(query_qp_in, in, qpc_ext, qpc_ext); + return mlx5_cmd_exec(dev->mdev, in, sizeof(in), out, outlen); } diff --git a/drivers/infiniband/hw/mlx5/srq.c b/drivers/infiniband/hw/mlx5/srq.c index 09b365a98bbf..a056ea835da5 100644 --- a/drivers/infiniband/hw/mlx5/srq.c +++ b/drivers/infiniband/hw/mlx5/srq.c @@ -447,7 +447,7 @@ int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr, if (i < srq->msrq.max_avail_gather) { scat[i].byte_count = 0; - scat[i].lkey = cpu_to_be32(MLX5_INVALID_LKEY); + scat[i].lkey = dev->mkeys.terminate_scatter_list_mkey; scat[i].addr = 0; } } diff --git a/drivers/infiniband/hw/mlx5/umr.c b/drivers/infiniband/hw/mlx5/umr.c index 029e9536ec28..55f4e048d947 100644 --- a/drivers/infiniband/hw/mlx5/umr.c +++ b/drivers/infiniband/hw/mlx5/umr.c @@ -636,9 +636,7 @@ int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) mlx5r_umr_set_update_xlt_data_seg(&wqe.data_seg, &sg); cur_mtt = mtt; - rdma_for_each_block(mr->umem->sgt_append.sgt.sgl, &biter, - mr->umem->sgt_append.sgt.nents, - BIT(mr->page_shift)) { + rdma_umem_for_each_dma_block(mr->umem, &biter, BIT(mr->page_shift)) { if (cur_mtt == (void *)mtt + sg.length) { dma_sync_single_for_device(ddev, sg.addr, sg.length, DMA_TO_DEVICE); diff --git a/drivers/infiniband/hw/mlx5/wr.c b/drivers/infiniband/hw/mlx5/wr.c index 855f3f4fefad..df1d1b0a3ef7 100644 --- a/drivers/infiniband/hw/mlx5/wr.c +++ b/drivers/infiniband/hw/mlx5/wr.c @@ -1252,7 +1252,7 @@ int mlx5_ib_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr, if (i < qp->rq.max_gs) { scat[i].byte_count = 0; - scat[i].lkey = cpu_to_be32(MLX5_INVALID_LKEY); + scat[i].lkey = dev->mkeys.terminate_scatter_list_mkey; scat[i].addr = 0; } |