Re: [PATCH v2 1/1] RDMA/rxe: Bump up default maximum values used via uverbs
From: Shoaib Rao <hidden>
Date: 2021-07-18 19:46:25
Possibly related (same subject, not in this thread)
- 2021-07-16 · Re: [PATCH v2 1/1] RDMA/rxe: Bump up default maximum values used via uverbs · Zhu Yanjun <zyjzyj2000@gmail.com>
- 2021-07-15 · Re: [PATCH v2 1/1] RDMA/rxe: Bump up default maximum values used via uverbs · Shoaib Rao <hidden>
- 2021-07-15 · Re: [PATCH v2 1/1] RDMA/rxe: Bump up default maximum values used via uverbs · Zhu Yanjun <zyjzyj2000@gmail.com>
- 2021-07-14 · [PATCH v2 1/1] RDMA/rxe: Bump up default maximum values used via uverbs · Rao Shoaib <hidden>
Your urls are garbled, so I can not make any sense out of them. [root@ca-dev141 linux]# git remote --v origin https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git (fetch) origin https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git (push) [root@ca-dev141 linux]# git show 8c1b4316c3fa fatal: ambiguous argument '8c1b4316c3fa': unknown revision or path not in the working tree. [root@ca-dev141 linux]# git log --grep='RDMA/efa: Split hardware' [root@ca-dev141 linux]# [root@ca-dev141 linux]# git describe v5.14-rc1-17-gc06676b5953b You need to do a little bit more work on your end and tell me which value rxe is complaining about. Also provide a kernel version that you are using as I have provided above. Shoaib On 7/18/21 12:59 AM, Zhu Yanjun wrote:
On Sat, Jul 17, 2021 at 2:09 AM Shoaib Rao [off-list ref] wrote:quoted
On 7/15/21 9:07 PM, Zhu Yanjun wrote: On Fri, Jul 16, 2021 at 12:44 AM Shoaib Rao [off-list ref] wrote: Following is a link https://urldefense.com/v3/__https://marc.info/?l=linux-rdma&m=162395437604846&w=2__;!!ACWV5N9M2RV99hQ!f9TPWtgUxambtSeQ_L3h-IH7CW3SifyiumB3kjc2v_w6Ec_WYVtjWyusEtgQ60iw$ Or just search for my name in the archive. Do you see any issues with this value? After this commit is applied, I confronted the following problem " [ 639.943561] rdma_rxe: unloaded [ 679.717143] rdma_rxe: loaded [ 691.721055] rdma_rxe: not enough indices for max_elem " Not sure if this problem is introduced by this commit. Please help to check this problem. Thanks a lot. Zhu Yanjun I do not see the issue. [root@ca-dev66 rxe]# lsmod | grep rdma_rxe rdma_rxe 126976 0 ip6_udp_tunnel 16384 1 rdma_rxe udp_tunnel 20480 1 rdma_rxe ib_uverbs 147456 3 rdma_rxe,rdma_ucm,mlx5_ib ib_core 364544 10 rdma_cm,ib_ipoib,rdma_rxe,rds_rdma,iw_cm,ib_umad,rdma_ucm,ib_uverbs,mlx5_ib,ib_cm I am using gdb to dump the values (gdb) ptype RXE_MAX_INLINE_DATA type = enum rxe_device_param {RXE_MAX_MR_SIZE = -1, RXE_PAGE_SIZE_CAP = 4294963200, RXE_MAX_QP_WR = 1048576, RXE_DEVICE_CAP_FLAGS = 137466362998, RXE_MAX_SGE = 32, RXE_MAX_WQE_SIZE = 720, RXE_MAX_INLINE_DATA = 512, RXE_MAX_SGE_RD = 32, RXE_MAX_CQ = 1048576, RXE_MAX_LOG_CQE = 15, RXE_MAX_PD = 1048576, RXE_MAX_QP_RD_ATOM = 128, RXE_MAX_RES_RD_ATOM = 258048, RXE_MAX_QP_INIT_RD_ATOM = 128, RXE_MAX_MCAST_GRP = 8192, RXE_MAX_MCAST_QP_ATTACH = 56, RXE_MAX_TOT_MCAST_QP_ATTACH = 458752, RXE_MAX_AH = 1048576, RXE_MAX_SRQ_WR = 1048576, RXE_MIN_SRQ_WR = 1, RXE_MAX_SRQ_SGE = 27, RXE_MIN_SRQ_SGE = 1, RXE_MAX_FMR_PAGE_LIST_LEN = 512, RXE_MAX_PKEYS = 64, RXE_LOCAL_CA_ACK_DELAY = 15, RXE_MAX_UCONTEXT = 1048576, RXE_NUM_PORT = 1, RXE_MAX_QP = 1048576, RXE_MIN_QP_INDEX = 16, RXE_MAX_QP_INDEX = 262144, RXE_MAX_SRQ = 1048576, RXE_MIN_SRQ_INDEX = 131073, RXE_MAX_SRQ_INDEX = 262144, RXE_MAX_MR = 1048576, RXE_MAX_MW = 4096, RXE_MIN_MR_INDEX = 1, RXE_MAX_MR_INDEX = 262144, RXE_MIN_MW_INDEX = 65537, RXE_MAX_MW_INDEX = 131072, RXE_MAX_PKT_PER_ACK = 64, RXE_MAX_UNACKED_PSNS = 128, RXE_INFLIGHT_SKBS_PER_QP_HIGH = 64, RXE_INFLIGHT_SKBS_PER_QP_LOW = 16, RXE_NSEC_ARB_TIMER_DELAY = 200, RXE_VENDOR_ID = 16777215} It is possible that you are changing some values.I git clone the source code from https://urldefense.com/v3/__https://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma.git__;!!ACWV5N9M2RV99hQ!efPsL9fikzm6Ob1-zsmbsAzPkfqRpxYkp9oJXF9csPOJgoei9njj8RmA5IyuH447$ again. And the first 3 commits are as below. One is your commit. " 6163ad3208c8 (HEAD -> for-next) RDMA/rxe: Bump up default maximum values used via uverbs 8c1b4316c3fa (origin/for-next, origin/HEAD) RDMA/efa: Split hardware stats to device and port stats 916071185b17 MAINTAINERS: Update maintainers of HiSilicon RoCE " And when I tried to add a new rxe0, " error: Invalid argument " And the dmesg logs are as below: # dmesg " [ 70.782302] e1000: enp0s8 NIC Link is Up 1000 Mbps Full Duplex, Flow Control: RX [ 70.782744] IPv6: ADDRCONF(NETDEV_CHANGE): enp0s8: link becomes ready [ 79.467652] rdma_rxe: loaded [ 79.468348] rdma_rxe: not enough indices for max_elem [ 79.468356] rdma_rxe: failed to add enp0s8 " Please follow my steps to reproduce this problem again. Zhu Yanjunquoted
Shoaib Shoaib On 7/14/21 10:02 PM, Zhu Yanjun wrote: On Wed, Jul 14, 2021 at 4:36 PM Rao Shoaib [off-list ref] wrote: From: Rao Shoaib <redacted> In our internal testing we have found that the current maximum are too smalls. Ideally there should be no limits but currently maximum values are reported via ibv_query_device, so we have to keep maximum values but they have been made suffiently large. Resubmitting after fixing an issue reported by test robot. Reported-by: kernel test robot <redacted> Signed-off-by: Rao Shoaib <redacted> --- drivers/infiniband/sw/rxe/rxe_param.h | 26 ++++++++++++++------------ 1 file changed, 14 insertions(+), 12 deletions(-)diff --git a/drivers/infiniband/sw/rxe/rxe_param.h b/drivers/infiniband/sw/rxe/rxe_param.h index 742e6ec93686..092dbff890f2 100644 --- a/drivers/infiniband/sw/rxe/rxe_param.h +++ b/drivers/infiniband/sw/rxe/rxe_param.h@@ -9,6 +9,8 @@ #include <uapi/rdma/rdma_user_rxe.h> +#define DEFAULT_MAX_VALUE (1 << 20)Can you let me know the link in which the above value is discussed? Thanks, Zhu Yanjun + static inline enum ib_mtu rxe_mtu_int_to_enum(int mtu) { if (mtu < 256)@@ -37,7 +39,7 @@ static inline enum ib_mtu eth_mtu_int_to_enum(int mtu) enum rxe_device_param { RXE_MAX_MR_SIZE = -1ull, RXE_PAGE_SIZE_CAP = 0xfffff000, - RXE_MAX_QP_WR = 0x4000, + RXE_MAX_QP_WR = DEFAULT_MAX_VALUE, RXE_DEVICE_CAP_FLAGS = IB_DEVICE_BAD_PKEY_CNTR | IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_AUTO_PATH_MIG@@ -58,40 +60,40 @@ enum rxe_device_param { RXE_MAX_INLINE_DATA = RXE_MAX_WQE_SIZE - sizeof(struct rxe_send_wqe), RXE_MAX_SGE_RD = 32, - RXE_MAX_CQ = 16384, + RXE_MAX_CQ = DEFAULT_MAX_VALUE, RXE_MAX_LOG_CQE = 15, - RXE_MAX_PD = 0x7ffc, + RXE_MAX_PD = DEFAULT_MAX_VALUE, RXE_MAX_QP_RD_ATOM = 128, RXE_MAX_RES_RD_ATOM = 0x3f000, RXE_MAX_QP_INIT_RD_ATOM = 128, RXE_MAX_MCAST_GRP = 8192, RXE_MAX_MCAST_QP_ATTACH = 56, RXE_MAX_TOT_MCAST_QP_ATTACH = 0x70000, - RXE_MAX_AH = 100, - RXE_MAX_SRQ_WR = 0x4000, + RXE_MAX_AH = DEFAULT_MAX_VALUE, + RXE_MAX_SRQ_WR = DEFAULT_MAX_VALUE, RXE_MIN_SRQ_WR = 1, RXE_MAX_SRQ_SGE = 27, RXE_MIN_SRQ_SGE = 1, RXE_MAX_FMR_PAGE_LIST_LEN = 512, - RXE_MAX_PKEYS = 1, + RXE_MAX_PKEYS = 64, RXE_LOCAL_CA_ACK_DELAY = 15, - RXE_MAX_UCONTEXT = 512, + RXE_MAX_UCONTEXT = DEFAULT_MAX_VALUE, RXE_NUM_PORT = 1, - RXE_MAX_QP = 0x10000, + RXE_MAX_QP = DEFAULT_MAX_VALUE, RXE_MIN_QP_INDEX = 16, - RXE_MAX_QP_INDEX = 0x00020000, + RXE_MAX_QP_INDEX = 0x00040000, - RXE_MAX_SRQ = 0x00001000, + RXE_MAX_SRQ = DEFAULT_MAX_VALUE, RXE_MIN_SRQ_INDEX = 0x00020001, RXE_MAX_SRQ_INDEX = 0x00040000, - RXE_MAX_MR = 0x00001000, + RXE_MAX_MR = DEFAULT_MAX_VALUE, RXE_MAX_MW = 0x00001000, RXE_MIN_MR_INDEX = 0x00000001, - RXE_MAX_MR_INDEX = 0x00010000, + RXE_MAX_MR_INDEX = 0x00040000, RXE_MIN_MW_INDEX = 0x00010001, RXE_MAX_MW_INDEX = 0x00020000, --2.27.0