kernel/linux.git/net/rds/ib.h, branch linux-7.1.y

net/rds: Optimize rds_ib_laddr_check

2026-04-12T20:33:19+00:00

rds_ib_laddr_check() creates a CM_ID and attempts to bind the address in question to it. This in order to qualify the allegedly local address as a usable IB/RoCE address. In the field, ExaWatcher runs rds-ping to all ports in the fabric from all local ports. This using all active ToS'es. In a full rack system, we have 14 cell servers and eight db servers. Typically, 6 ToS'es are used. This implies 528 rds-ping invocations per ExaWatcher's "RDSinfo" interval. Adding to this, each rds-ping invocation creates eight sockets and binds the local address to them: socket(AF_RDS, SOCK_SEQPACKET, 0) = 3 bind(3, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("192.168.36.2")}, 16) = 0 socket(AF_RDS, SOCK_SEQPACKET, 0) = 4 bind(4, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("192.168.36.2")}, 16) = 0 socket(AF_RDS, SOCK_SEQPACKET, 0) = 5 bind(5, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("192.168.36.2")}, 16) = 0 socket(AF_RDS, SOCK_SEQPACKET, 0) = 6 bind(6, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("192.168.36.2")}, 16) = 0 socket(AF_RDS, SOCK_SEQPACKET, 0) = 7 bind(7, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("192.168.36.2")}, 16) = 0 socket(AF_RDS, SOCK_SEQPACKET, 0) = 8 bind(8, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("192.168.36.2")}, 16) = 0 socket(AF_RDS, SOCK_SEQPACKET, 0) = 9 bind(9, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("192.168.36.2")}, 16) = 0 socket(AF_RDS, SOCK_SEQPACKET, 0) = 10 bind(10, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("192.168.36.2")}, 16) = 0 So, at every interval ExaWatcher executes rds-ping's, 4224 CM_IDs are allocated, considering this full-rack system. After the a CM_ID has been allocated, rdma_bind_addr() is called, with the port number being zero. This implies that the CMA will attempt to search for an un-used ephemeral port. Simplified, the algorithm is to start at a random position in the available port space, and then if needed, iterate until an un-used port is found. The book-keeping of used ports uses the idr system, which again uses slab to allocate new struct idr_layer's. The size is 2092 bytes and slab tries to reduce the wasted space. Hence, it chooses an order:3 allocation, for which 15 idr_layer structs will fit and only 1388 bytes are wasted per the 32KiB order:3 chunk. Although this order:3 allocation seems like a good space/speed trade-off, it does not resonate well with how it used by the CMA. The combination of the randomized starting point in the port space (which has close to zero spatial locality) and the close proximity in time of the 4224 invocations of the rds-ping's, creates a memory hog for order:3 allocations. These costly allocations may need reclaims and/or compaction. At worst, they may fail and produce a stack trace such as (from uek4): [] __inc_zone_page_state+0x35/0x40 [] page_add_file_rmap+0x57/0x60 [] remove_migration_pte+0x3f/0x3c0 [ksplice_6cn872bt_vmlinux_new] [] rmap_walk+0xd8/0x340 [] remove_migration_ptes+0x40/0x50 [] migrate_pages+0x3ec/0x890 [] compact_zone+0x32d/0x9a0 [] compact_zone_order+0x6d/0x90 [] try_to_compact_pages+0x102/0x270 [] __alloc_pages_direct_compact+0x46/0x100 [] __alloc_pages_nodemask+0x74b/0xaa0 [] alloc_pages_current+0x91/0x110 [] new_slab+0x38b/0x480 [] __slab_alloc+0x3b7/0x4a0 [ksplice_s0dk66a8_vmlinux_new] [] kmem_cache_alloc+0x1fb/0x250 [] idr_layer_alloc+0x36/0x90 [] idr_get_empty_slot+0x28c/0x3d0 [] idr_alloc+0x4d/0xf0 [] cma_alloc_port+0x4d/0xa0 [rdma_cm] [] rdma_bind_addr+0x2ae/0x5b0 [rdma_cm] [] rds_ib_laddr_check+0x83/0x2c0 [ksplice_6l2xst5i_rds_rdma_new] [] rds_trans_get_preferred+0x5b/0xa0 [rds] [] rds_bind+0x212/0x280 [rds] [] SYSC_bind+0xe6/0x120 [] SyS_bind+0xe/0x10 [] system_call_fastpath+0x18/0xd4 To avoid these excessive calls to rdma_bind_addr(), we optimize rds_ib_laddr_check() by simply checking if the address in question has been used before. The rds_rdma module keeps track of addresses associated with IB devices, and the function rds_ib_get_device() is used to determine if the address already has been qualified as a valid local address. If not found, we call the legacy rds_ib_laddr_check(), now renamed to rds_ib_laddr_check_cm(). Signed-off-by: Håkon Bugge Signed-off-by: Somasundaram Krishnasamy Signed-off-by: Gerd Rausch Signed-off-by: Allison Henderson Link: https://patch.msgid.link/20260408080420.540032-2-achender@kernel.org Signed-off-by: Jakub Kicinski

RDS: IB: Remove unused declarations

2024-08-01T16:03:28+00:00

Commit f4f943c958a2 ("RDS: IB: ack more receive completions to improve performance") removed rds_ib_recv_tasklet_fn() implementation but not the declaration. And commit ec16227e1414 ("RDS/IB: Infiniband transport") declared but never implemented other functions. Signed-off-by: Yue Haibing Reviewed-by: Simon Horman Reviewed-by: Allison Henderson Link: https://patch.msgid.link/20240731063630.3592046-1-yuehaibing@huawei.com Signed-off-by: Jakub Kicinski

rds: stop using dmapool

2020-11-17T19:22:06+00:00

RDMA ULPs should only perform DMA through the ib_dma_* API instead of using the hidden dma_device directly. In addition using the dma coherent API family that dmapool is a part of can be very ineffcient on plaforms that are not DMA coherent. Switch to use slab allocations and the ib_dma_* APIs instead. Link: https://lore.kernel.org/r/20201106181941.1878556-6-hch@lst.de Signed-off-by: Christoph Hellwig Acked-by: Santosh Shilimkar Signed-off-by: Jason Gunthorpe

RDMA: Lift ibdev_to_node from rds to common code

2020-11-12T17:33:44+00:00

Lift the ibdev_to_node from rds to common code and document it. Link: https://lore.kernel.org/r/20201106181941.1878556-4-hch@lst.de Signed-off-by: Christoph Hellwig Signed-off-by: Jason Gunthorpe

net/rds: NULL pointer de-reference in rds_ib_add_one()

2020-06-15T19:58:59+00:00

The parent field of a struct device may be NULL. The macro ibdev_to_node() should check for that. Signed-off-by: Ka-Cheong Poon Acked-by: Santosh Shilimkar Signed-off-by: David S. Miller

RDMA/rds: Remove FMR support for memory registration

2020-06-02T23:32:53+00:00

Use FRWR method for memory registration by default and remove the ancient and unsafe FMR method. Link: https://lore.kernel.org/r/3-v3-f58e6669d5d3+2cf-fmr_removal_jgg@mellanox.com Signed-off-by: Max Gurtovoy Signed-off-by: Jason Gunthorpe

net/rds: Handle ODP mr registration/unregistration

2020-01-18T09:48:19+00:00

On-Demand-Paging MRs are registered using ib_reg_user_mr and unregistered with ib_dereg_mr. Signed-off-by: Hans Westgaard Ry Acked-by: Santosh Shilimkar Signed-off-by: Leon Romanovsky

net/rds: Use DMA memory pool allocation for rds_header

2019-10-03T19:11:08+00:00

Currently, RDS calls ib_dma_alloc_coherent() to allocate a large piece of contiguous DMA coherent memory to store struct rds_header for sending/receiving packets. The memory allocated is then partitioned into struct rds_header. This is not necessary and can be costly at times when memory is fragmented. Instead, RDS should use the DMA memory pool interface to handle this. The DMA addresses of the pre- allocated headers are stored in an array. At send/receive ring initialization and refill time, this arrary is de-referenced to get the DMA addresses. This array is not accessed at send/receive packet processing. Suggested-by: Håkon Bugge Signed-off-by: Ka-Cheong Poon Acked-by: Santosh Shilimkar Signed-off-by: David S. Miller

net: rds: add service level support in rds-info

2019-08-24T23:55:25+00:00

>From IB specific 7.6.5 SERVICE LEVEL, Service Level (SL) is used to identify different flows within an IBA subnet. It is carried in the local route header of the packet. Before this commit, run "rds-info -I". The outputs are as below: " RDS IB Connections: LocalAddr RemoteAddr Tos SL LocalDev RemoteDev 192.2.95.3 192.2.95.1 2 0 fe80::21:28:1a:39 fe80::21:28:10:b9 192.2.95.3 192.2.95.1 1 0 fe80::21:28:1a:39 fe80::21:28:10:b9 192.2.95.3 192.2.95.1 0 0 fe80::21:28:1a:39 fe80::21:28:10:b9 " After this commit, the output is as below: " RDS IB Connections: LocalAddr RemoteAddr Tos SL LocalDev RemoteDev 192.2.95.3 192.2.95.1 2 2 fe80::21:28:1a:39 fe80::21:28:10:b9 192.2.95.3 192.2.95.1 1 1 fe80::21:28:1a:39 fe80::21:28:10:b9 192.2.95.3 192.2.95.1 0 0 fe80::21:28:1a:39 fe80::21:28:10:b9 " The commit fe3475af3bdf ("net: rds: add per rds connection cache statistics") adds cache_allocs in struct rds_info_rdma_connection as below: struct rds_info_rdma_connection { ... __u32 rdma_mr_max; __u32 rdma_mr_size; __u8 tos; __u32 cache_allocs; }; The peer struct in rds-tools of struct rds_info_rdma_connection is as below: struct rds_info_rdma_connection { ... uint32_t rdma_mr_max; uint32_t rdma_mr_size; uint8_t tos; uint8_t sl; uint32_t cache_allocs; }; The difference between userspace and kernel is the member variable sl. In the kernel struct, the member variable sl is missing. This will introduce risks. So it is necessary to use this commit to avoid this risk. Fixes: fe3475af3bdf ("net: rds: add per rds connection cache statistics") CC: Joe Jin CC: JUNXIAO_BI Suggested-by: Gerd Rausch Signed-off-by: Zhu Yanjun Acked-by: Santosh Shilimkar Signed-off-by: David S. Miller

net/rds: Keep track of and wait for FRWR segments in use upon shutdown

2019-07-17T19:06:52+00:00

Since "rds_ib_free_frmr" and "rds_ib_free_frmr_list" simply put the FRMR memory segments on the "drop_list" or "free_list", and it is the job of "rds_ib_flush_mr_pool" to reap those entries by ultimately issuing a "IB_WR_LOCAL_INV" work-request, we need to trigger and then wait for all those memory segments attached to a particular connection to be fully released before we can move on to release the QP, CQ, etc. So we make "rds_ib_conn_path_shutdown" wait for one more atomic_t called "i_fastreg_inuse_count" that keeps track of how many FRWR memory segments are out there marked "FRMR_IS_INUSE" (and also wake_up rds_ib_ring_empty_wait, as they go away). Signed-off-by: Gerd Rausch Acked-by: Santosh Shilimkar Signed-off-by: David S. Miller