net/smc: Introduce SMC-D-based OS internal communication acceleration
Subsystems (5)
Wen Gu
reviewerv3 cover net/smc: Introduce SMC-D-based OS internal communication acceleration
Hi, all
# Background
The background and previous discussion can be referred from [1].
We found SMC-D can be used to accelerate OS internal communication, such as loopback or between two containers within the same OS instance. So this patch set provides a kind of SMC-D dummy device (we call it the SMC-D loopback device) to emulate an ISM device, so that SMC-D can also be used on architectures other than s390. The SMC-D loopback device are designed as a system global device, visible to all containers.
This version is implemented based on the generalized interface provided by [2]. And there is an open issue of this version, which will be mentioned later.
# Design
This patch set basically follows the design of the previous version.
Patch #1/9 ~ #3/9 attempt to decouple ISM-related structures from the SMC-D generalized code and extract some helpers to make SMC-D protocol compatible with devices other than s390 ISM device.
Patch #4/9 introduces a kind of loopback device, which is defined as SMC-D v2 device and designed to provide communication between SMC sockets in the same OS instance.
+-------------------------------------------+ | +--------------+ +--------------+ | | | SMC socket A | | SMC socket B | | | +--------------+ +--------------+ | | ^ ^ | | | +----------------+ | | | | | SMC stack | | | | +--->| +------------+ |<--| | | | | dummy | | | | | | device | | | | +-+------------+-+ | | OS | +-------------------------------------------+
Patch #5/9 ~ #8/9 expand SMC-D protocol interface (smcd_ops) for scenarios where SMC-D is used to communicate within VM (loopback here) or between VMs on the same host (based on virtio-ism device, see [3]). What these scenarios have in common is that the local sndbuf and peer RMB can be mapped to same physical memory region, so the data copy between the local sndbuf and peer RMB can be omitted. Performance improvement brought by this extension can be found in # Benchmark Test.
+----------+ +----------+
| socket A | | socket B |
+----------+ +----------+
| ^
| +---------+ |
regard as | | ----------|
local sndbuf | B's | regard as
| | RMB | local RMB
|-------> | |
+---------+Patch #9/9 realizes the support of loopback device for the above-mentioned expanded SMC-D protocol interface.
# Benchmark Test
* Test environments:
- VM with Intel Xeon Platinum 8 core 2.50GHz, 16 GiB mem.
- SMC sndbuf/RMB size 1MB.
* Test object:
- TCP lo: run on TCP loopback.
- domain: run on UNIX domain.
- SMC lo: run on SMC loopback device with patch #1/9 ~ #4/9.
- SMC lo-nocpy: run on SMC loopback device with patch #1/9 ~ #9/9.
1. ipc-benchmark (see [4])
- ./<foo> -c 1000000 -s 100
TCP-lo domain SMC-lo SMC-lo-nocpy
Message
rate (msg/s) 79025 115736(+46.45%) 146760(+85.71%) 149800(+89.56%)
2. sockperf
- serv: <smc_run> taskset -c <cpu> sockperf sr --tcp
- clnt: <smc_run> taskset -c <cpu> sockperf { tp | pp } --tcp --msg-size={ 64000 for tp | 14 for pp } -i 127.0.0.1 -t 30
TCP-lo SMC-lo SMC-lo-nocpy
Bandwidth(MBps) 4822.388 4940.918(+2.56%) 8086.67(+67.69%)
Latency(us) 6.298 3.352(-46.78%) 3.35(-46.81%)
3. iperf3
- serv: <smc_run> taskset -c <cpu> iperf3 -s
- clnt: <smc_run> taskset -c <cpu> iperf3 -c 127.0.0.1 -t 15
TCP-lo SMC-lo SMC-lo-nocpy
Bitrate(Gb/s) 40.7 40.5(-0.49%) 72.4(+77.89%)
4. nginx/wrk
- serv: <smc_run> nginx
- clnt: <smc_run> wrk -t 8 -c 500 -d 30 http://127.0.0.1:80
TCP-lo SMC-lo SMC-lo-nocpy
Requests/s 155994.57 214544.79(+37.53%) 215538.55(+38.17%)
# Open issue
The open issue has not been resolved now is about how to detect that the source and target of CLC proposal are within the same OS instance and can communicate through the SMC-D loopback device. Similar issue also exists when using virtio-ism devices (the background and details of virtio-ism device can be referred from [3]). In previous discussions, multiple options were proposed (see [5]). Thanks again for the help of the community. cc Alexandra Winter :)
But as we discussed, these solutions have some imperfection. So this version of RFC continues to use previous workaround, that is, a 64-bit random GID is generated for SMC-D loopback device. If the GIDs of the devices found by two peers are the same, then they are considered to be in the same OS instance and can communicate with each other by the loopback device.
This approach has very small risk. Assume the following situations:
(1) Assume that the SMC-D loopback devices of the two OS instances happen to generate the same 64-bit GID.
For the convenience of description, we refer to the sockets on these two different OS instance as server A and client B.
A will misjudge that the two are on the same OS instance because the same GID in CLC proposal message. Then A creates its RMB and sends 64-bit token-A to B in CLC accept message.
B receives the CLC accept message. And according to patch #7/9, B tries to attach its sndbuf to A's RMB by token-A.
(2) Assume that the OS instance where B is located happens to have an unattached RMB whose 64-bit token is same as token-A.
Then B successfully attaches its sndbuf to the wrong RMB, and creates its RMB, sends token-B to A in CLC confirm message.
Similarly, A receives the message and tries to attach its sndbuf to B's RMB by token-B.
(3) Similar to (2), assume that the OS instance where A is located happens to have an unattached RMB whose 64-bit token is same as token-B.
Then A successfully attach its sndbuf to the wrong RMB. Both sides mistakenly believe that an SMC-D connection based on the loopback device is established between them.
If the above 3 coincidences all happen, that is, 64-bit random number conflicts occur 3 times, then an unreachable SMC-D connection will be established, which is nasty. If one of above is not satisfied, it will safely fallback to TCP.
Since the chances of these happening are very small, I wonder if this risk of 1/2^(64*3) probability can be tolerated ?
Another way to solve this open issue is using a 128-bit UUID to identify SMC-D loopback device or virtio-ism device, because the probability of a 128-bit UUID collision is considered negligible. But it may need to extend the CLC message to carry a longer GID, which is the last option.
v3->v2
1. Adapt new generalized interface provided by [2];
2. Select loopback device through SMC-D v2 protocol;
3. Split the loopback-related implementation and generic implementation into different patches more reasonably.
v1->v2
1. Fix some build WARNINGs complained by kernel test rebot
2. Add iperf3 test data.
[1] https://lore.kernel.org/netdev/[email protected]/
[2] https://lore.kernel.org/netdev/[email protected]/
[3] https://lists.oasis-open.org/archives/virtio-comment/202302/msg00148.html
[4] https://github.com/goldsborough/ipc-bench
[5] https://lore.kernel.org/netdev/[email protected]/
Wen Gu (9):
net/smc: Decouple ism_dev from SMC-D device dump net/smc: Decouple ism_dev from SMC-D DMB registration net/smc: Extract v2 check helper from SMC-D device registration net/smc: Introduce SMC-D loopback device
net/smc: Introduce an interface for getting DMB attribute net/smc: Introudce interfaces for DMB attach and detach net/smc: Avoid data copy from sndbuf to peer RMB in SMC-D net/smc: Modify cursor update logic when using mappable DMB net/smc: Add interface implementation of loopback device
drivers/s390/net/ism_drv.c | 5 +- include/net/smc.h | 18 +- net/smc/Makefile | 2 +- net/smc/af_smc.c | 26 ++- net/smc/smc_cdc.c | 59 ++++-- net/smc/smc_cdc.h | 1 + net/smc/smc_core.c | 70 ++++++- net/smc/smc_core.h | 1 + net/smc/smc_ism.c | 79 ++++++-- net/smc/smc_ism.h | 4 + net/smc/smc_loopback.c | 442 +++++++++++++++++++++++++++++++++++++++++++++ net/smc/smc_loopback.h | 55 ++++++ 12 files changed, 725 insertions(+), 37 deletions(-) create mode 100644 net/smc/smc_loopback.c create mode 100644 net/smc/smc_loopback.h
Signature · 1 line
1.8.3.1
Wen Gu
reviewerRFC PATCH v3 1/9 net/smc: Decouple ism_dev from SMC-D device dump
This patch helps to decouple SMC-D device and ISM device, allowing different underlying device forms, such as non-PCI devices.
--- include/net/smc.h | 1 + net/smc/smc_ism.c | 6 +++--- 2 files changed, 4 insertions(+), 3 deletions(-)
include/net/smc.h +1 -0
@@ -76,6 +76,7 @@ struct smcd_ops {struct smcd_dev {const struct smcd_ops *ops;void *priv; +struct device *parent_pci_dev;struct list_head list;spinlock_t lock;struct smc_connection **conn;
net/smc/smc_ism.c +3 -3
@@ -231,11 +231,9 @@ static int smc_nl_handle_smcd_dev(struct smcd_dev *smcd,struct smc_pci_dev smc_pci_dev;struct nlattr *port_attrs;struct nlattr *attrs; -struct ism_dev *ism;int use_cnt =0 ;void *nlh; - ism = smcd->priv; nlh =genlmsg_put (skb,NETLINK_CB (cb->skb).portid, cb->nlh->nlmsg_seq, &smc_gen_nl_family,NLM_F_MULTI ,SMC_NETLINK_GET_DEV_SMCD );
@@ -250,7 +248,8 @@ static int smc_nl_handle_smcd_dev(struct smcd_dev *smcd,if (nla_put_u8 (skb,SMC_NLA_DEV_IS_CRIT , use_cnt >0 )) goto errattr;memset (&smc_pci_dev,0 ,sizeof (smc_pci_dev)); -smc_set_pci_values (to_pci_dev (ism->dev.parent), &smc_pci_dev); +if (smcd->parent_pci_dev) +smc_set_pci_values (to_pci_dev (smcd->parent_pci_dev), &smc_pci_dev);if (nla_put_u32 (skb,SMC_NLA_DEV_PCI_FID , smc_pci_dev.pci_fid)) goto errattr;if (nla_put_u16 (skb,SMC_NLA_DEV_PCI_CHID , smc_pci_dev.pci_pchid))
@@ -420,6 +419,7 @@ static void smcd_register_dev(struct ism_dev *ism)if (!smcd)return ; smcd->priv = ism; + smcd->parent_pci_dev = ism->dev.parent;ism_set_priv (ism, &smc_ism_client, smcd);if (smc_pnetid_by_dev_port (&ism->pdev->dev,0 , smcd->pnetid))smc_pnetid_by_table_smcd (smcd);
Signature · 1 line
1.8.3.1
Wen Gu
reviewerRFC PATCH v3 2/9 net/smc: Decouple ism_dev from SMC-D DMB registration
In reply to Wen GuThis patch tries to decouple ISM device from SMC-D DMB registration, So that the register_dmb option is not restricted to be used by ISM device.
--- drivers/s390/net/ism_drv.c | 5 +++-- include/net/smc.h | 4 ++-- net/smc/smc_ism.c | 7 ++----- 3 files changed, 7 insertions(+), 9 deletions(-)
drivers/s390/net/ism_drv.c +3 -2
@@ -799,9 +799,10 @@ static int smcd_query_rgid(struct smcd_dev *smcd, u64 rgid, u32 vid_valid, }static int smcd_register_dmb (struct smcd_dev *smcd,struct smcd_dmb *dmb, -struct ism_client *client) +void *client_priv) { -return ism_register_dmb (smcd->priv, (struct ism_dmb *)dmb, client); +return ism_register_dmb (smcd->priv, (struct ism_dmb *)dmb, + (struct ism_client *)client_priv); }static int smcd_unregister_dmb (struct smcd_dev *smcd,struct smcd_dmb *dmb)
include/net/smc.h +2 -2
@@ -50,13 +50,12 @@ struct smcd_dmb {#define ISM_ERROR 0xFFFFstruct smcd_dev ; -struct ism_client ;struct smcd_ops {int (*query_remote_gid)(struct smcd_dev *dev,u64 rgid,u32 vid_valid,u32 vid);int (*register_dmb)(struct smcd_dev *dev,struct smcd_dmb *dmb, -struct ism_client *client); +void *client_priv);int (*unregister_dmb)(struct smcd_dev *dev,struct smcd_dmb *dmb);int (*add_vlan_id)(struct smcd_dev *dev,u64 vlan_id);int (*del_vlan_id)(struct smcd_dev *dev,u64 vlan_id);
@@ -76,6 +75,7 @@ struct smcd_ops {struct smcd_dev {const struct smcd_ops *ops;void *priv; +void *client_priv;struct device *parent_pci_dev;struct list_head list;spinlock_t lock;
net/smc/smc_ism.c +2 -5
@@ -200,7 +200,6 @@ int smc_ism_unregister_dmb(struct smcd_dev *smcd, struct smc_buf_desc *dmb_desc)int smc_ism_register_dmb (struct smc_link_group *lgr,int dmb_len,struct smc_buf_desc *dmb_desc) { -#if IS_ENABLED (CONFIG_ISM )struct smcd_dmb dmb;int rc;
@@ -209,7 +208,7 @@ int smc_ism_register_dmb(struct smc_link_group *lgr, int dmb_len, dmb.sba_idx = dmb_desc->sba_idx; dmb.vlan_id = lgr->vlan_id; dmb.rgid = lgr->peer_gid; - rc = lgr->smcd->ops->register_dmb (lgr->smcd, &dmb, &smc_ism_client); + rc = lgr->smcd->ops->register_dmb (lgr->smcd, &dmb, lgr->smcd->client_priv);if (!rc) { dmb_desc->sba_idx = dmb.sba_idx; dmb_desc->token = dmb.dmb_tok;
@@ -218,9 +217,6 @@ int smc_ism_register_dmb(struct smc_link_group *lgr, int dmb_len, dmb_desc->len = dmb.dmb_len; }return rc; -#else -return 0 ; -#endif } static int smc_nl_handle_smcd_dev(struct smcd_dev *smcd,
@@ -419,6 +415,7 @@ static void smcd_register_dev(struct ism_dev *ism)if (!smcd)return ; smcd->priv = ism; + smcd->client_priv = &smc_ism_client; smcd->parent_pci_dev = ism->dev.parent;ism_set_priv (ism, &smc_ism_client, smcd);if (smc_pnetid_by_dev_port (&ism->pdev->dev,0 , smcd->pnetid))
Signature · 1 line
1.8.3.1
Wen Gu
reviewerRFC PATCH v3 3/9 net/smc: Extract v2 check helper from SMC-D device registration
In reply to Wen GuThis patch extracts v2-capable logic from the process of registering the ISM device as an SMC-D device, so that the registration process of other underlying devices can also use it.
--- net/smc/smc_ism.c | 27 +++++++++++++++++---------- net/smc/smc_ism.h | 1 + 2 files changed, 18 insertions(+), 10 deletions(-)
net/smc/smc_ism.c +17 -10
@@ -69,6 +69,22 @@ bool smc_ism_is_v2_capable(void)return smc_ism_v2_capable; } +/* must be called under smcd_dev_list.mutex lock */ +void smc_ism_check_v2_capable (struct smcd_dev *smcd) +{ +u8 *system_eid =NULL ; + +if (!list_empty (&smcd_dev_list.list)) +return ; + + system_eid = smcd->ops->get_system_eid (); +if (system_eid[24 ] !='0' || system_eid[28 ] !='0' ) { + smc_ism_v2_capable = true; +memcpy (smc_ism_v2_system_eid, system_eid, +SMC_MAX_EID_LEN ); + } +} +/* Set a connection using this DMBE. */ void smc_ism_set_conn (struct smc_connection *conn) {
@@ -422,16 +438,7 @@ static void smcd_register_dev(struct ism_dev *ism)smc_pnetid_by_table_smcd (smcd);mutex_lock (&smcd_dev_list.mutex); -if (list_empty (&smcd_dev_list.list)) { -u8 *system_eid =NULL ; - - system_eid = smcd->ops->get_system_eid (); -if (system_eid[24 ] !='0' || system_eid[28 ] !='0' ) { - smc_ism_v2_capable = true; -memcpy (smc_ism_v2_system_eid, system_eid, -SMC_MAX_EID_LEN ); - } - } +smc_ism_check_v2_capable (smcd);/* sort list: devices without pnetid before devices with pnetid */ if (smcd->pnetid[0 ])list_add_tail (&smcd->list, &smcd_dev_list.list);
net/smc/smc_ism.h +1 -0
@@ -42,6 +42,7 @@ int smc_ism_register_dmb(struct smc_link_group *lgr, int buf_size,void smc_ism_get_system_eid (u8 **eid);u16 smc_ism_get_chid (struct smcd_dev *dev);bool smc_ism_is_v2_capable (void ); +void smc_ism_check_v2_capable (struct smcd_dev *dev);int smc_ism_init (void );void smc_ism_exit (void );int smcd_nl_get_device (struct sk_buff *skb,struct netlink_callback *cb);
Signature · 1 line
1.8.3.1
Wen Gu
reviewerRFC PATCH v3 4/9 net/smc: Introduce SMC-D loopback device
In reply to Wen GuThis patch introduces a kind of loopback device for SMC-D, thus enabling the SMC communication between two local sockets within one OS instance.
The loopback device supports basic capabilities defined by SMC-D options, and exposed as an SMC-D v2 device.
--- include/net/smc.h | 6 + net/smc/Makefile | 2 +- net/smc/af_smc.c | 12 +- net/smc/smc_cdc.c | 9 +- net/smc/smc_cdc.h | 1 + net/smc/smc_loopback.c | 363 +++++++++++++++++++++++++++++++++++++++++++++++++ net/smc/smc_loopback.h | 51 +++++++ 7 files changed, 441 insertions(+), 3 deletions(-) create mode 100644 net/smc/smc_loopback.c create mode 100644 net/smc/smc_loopback.h
include/net/smc.h +6 -0
@@ -41,6 +41,12 @@ struct smcd_dmb {dma_addr_t dma_addr; }; +struct smcd_seid { +u8 seid_string[24 ]; +u8 serial_number[4 ]; +u8 type[4 ]; +}; +#define ISM_EVENT_DMB 0#define ISM_EVENT_GID 1#define ISM_EVENT_SWR 2
net/smc/Makefile +1 -1
@@ -4,5 +4,5 @@ obj-$(CONFIG_SMC) += smc.o obj-$(CONFIG_SMC_DIAG) += smc_diag.o smc-y := af_smc.o smc_pnet.o smc_ib.o smc_clc.o smc_core.o smc_wr.o smc_llc.o smc-y += smc_cdc.o smc_tx.o smc_rx.o smc_close.o smc_ism.o smc_netlink.o smc_stats.o -smc-y += smc_tracepoint.o +smc-y += smc_tracepoint.o smc_loopback.o smc-$(CONFIG_SYSCTL) += smc_sysctl.o
net/smc/af_smc.c +11 -1
@@ -53,6 +53,7 @@#include "smc_stats.h" #include "smc_tracepoint.h" #include "smc_sysctl.h" +#include "smc_loopback.h" static DEFINE_MUTEX (smc_server_lgr_pending);/* serialize link group * creation on server
@@ -3454,15 +3455,23 @@ static int __init smc_init(void) goto out_sock; } + rc =smc_loopback_init (); +if (rc) { +pr_err ("%s: smc_loopback_init fails with %d\n" , __func__, rc); + goto out_ib; + } + rc =tcp_register_ulp (&smc_ulp_ops);if (rc) {pr_err ("%s: tcp_ulp_register fails with %d\n" , __func__, rc); - goto out_ib; + goto out_lo; }static_branch_enable (&tcp_have_smc);return 0 ; +out_lo: +smc_loopback_exit (); out_ib:smc_ib_unregister_client (); out_sock:
@@ -3499,6 +3508,7 @@ static void __exit smc_exit(void)tcp_unregister_ulp (&smc_ulp_ops);sock_unregister (PF_SMC );smc_core_exit (); +smc_loopback_exit ();smc_ib_unregister_client ();smc_ism_exit ();destroy_workqueue (smc_close_wq);
net/smc/smc_cdc.c +8 -1
@@ -407,7 +407,14 @@ static void smc_cdc_msg_recv(struct smc_sock *smc, struct smc_cdc_msg *cdc)*/ static void smcd_cdc_rx_tsklet (struct tasklet_struct *t) { -struct smc_connection *conn =from_tasklet (conn, t, rx_tsklet); +struct smc_connection *conn = +from_tasklet (conn, t, rx_tsklet); + +smcd_cdc_rx_handler (conn); +} + +void smcd_cdc_rx_handler (struct smc_connection *conn) +{struct smcd_cdc_msg *data_cdc;struct smcd_cdc_msg cdc;struct smc_sock *smc;
net/smc/smc_cdc.h +1 -0
@@ -301,5 +301,6 @@ int smcr_cdc_msg_send_validation(struct smc_connection *conn, struct smc_wr_buf *wr_buf);int smc_cdc_init (void )__init ;void smcd_cdc_rx_init (struct smc_connection *conn); +void smcd_cdc_rx_handler (struct smc_connection *conn);#endif /* SMC_CDC_H */
net/smc/smc_loopback.c new file +363 -0
363 lines not on this page · raw message
net/smc/smc_loopback.h new file +51 -0
51 lines not on this page · raw message
Signature · 1 line
1.8.3.1
Wen Gu
reviewerRFC PATCH v3 5/9 net/smc: Introduce an interface for getting DMB attribute
In reply to Wen GuOn s390, since all OSs run on a kind of machine level hypervisor which is a partitioning hypervisor without paging, the sndbufs and DMBs in such case are unable to be mapped to the same physical memory.
However, in other scene, such as communication within the same OS instance (loopback) or between guests of a paging hypervisor, eg. KVM, the sndbufs and DMBs can be mapped to the same physical memory to avoid memory copy from sndbufs to DMBs.
So this patch introduces an interface to smcd_ops for users to judge whether DMB-map is available. And for reuse, the interface is designed to return DMB attribute, not only mappability.
--- include/net/smc.h | 5 +++++ net/smc/smc_ism.c | 8 ++++++++ net/smc/smc_ism.h | 1 + 3 files changed, 14 insertions(+)
include/net/smc.h +5 -0
@@ -55,6 +55,10 @@ struct smcd_seid {#define ISM_ERROR 0xFFFF +enum { +ISM_DMB_MAPPABLE =0 , +}; +struct smcd_dev ;struct smcd_ops {
@@ -76,6 +80,7 @@ struct smcd_ops {u64 (*get_local_gid)(struct smcd_dev *dev);u16 (*get_chid)(struct smcd_dev *dev);struct device * (*get_dev)(struct smcd_dev *dev); +int (*get_dev_dmb_attr)(struct smcd_dev *dev); };struct smcd_dev {
net/smc/smc_ism.c +8 -0
@@ -213,6 +213,14 @@ int smc_ism_unregister_dmb(struct smcd_dev *smcd, struct smc_buf_desc *dmb_desc)return rc; } +bool smc_ism_dmb_mappable (struct smcd_dev *smcd) +{ +if (smcd->ops->get_dev_dmb_attr && + (smcd->ops->get_dev_dmb_attr (smcd) & (1 <<ISM_DMB_MAPPABLE ))) +return true; +return false; +} +int smc_ism_register_dmb (struct smc_link_group *lgr,int dmb_len,struct smc_buf_desc *dmb_desc) {
net/smc/smc_ism.h +1 -0
@@ -38,6 +38,7 @@ struct smc_ism_vlanid { /* VLAN id set on ISM device */int smc_ism_register_dmb (struct smc_link_group *lgr,int buf_size,struct smc_buf_desc *dmb_desc);int smc_ism_unregister_dmb (struct smcd_dev *dev,struct smc_buf_desc *dmb_desc); +bool smc_ism_dmb_mappable (struct smcd_dev *smcd);int smc_ism_signal_shutdown (struct smc_link_group *lgr);void smc_ism_get_system_eid (u8 **eid);u16 smc_ism_get_chid (struct smcd_dev *dev);
Signature · 1 line
1.8.3.1
Wen Gu
reviewerRFC PATCH v3 6/9 net/smc: Introudce interfaces for DMB attach and detach
In reply to Wen GuThis patch extends smcd_ops, adding two more semantic for SMC-D DMB:
- attach_dmb:
Attach an already registered DMB to a specific buf_desc, so that we can refer to the DMB through this buf_desc.
- detach_dmb:
Reverse operation of attach_dmb. detach the DMB from the buf_desc.
This interface extension is to prepare for the avoidance of memory copy from sndbuf to DMB with SMC-D device whose DMBs has ISM_DMB_MAPPABLE attribute.
--- include/net/smc.h | 2 ++ net/smc/smc_ism.c | 31 +++++++++++++++++++++++++++++++ net/smc/smc_ism.h | 2 ++ 3 files changed, 35 insertions(+)
include/net/smc.h +2 -0
@@ -67,6 +67,8 @@ struct smcd_ops {int (*register_dmb)(struct smcd_dev *dev,struct smcd_dmb *dmb,void *client_priv);int (*unregister_dmb)(struct smcd_dev *dev,struct smcd_dmb *dmb); +int (*attach_dmb)(struct smcd_dev *dev,struct smcd_dmb *dmb); +int (*detach_dmb)(struct smcd_dev *dev,u64 token);int (*add_vlan_id)(struct smcd_dev *dev,u64 vlan_id);int (*del_vlan_id)(struct smcd_dev *dev,u64 vlan_id);int (*set_vlan_required)(struct smcd_dev *dev);
net/smc/smc_ism.c +31 -0
@@ -243,6 +243,37 @@ int smc_ism_register_dmb(struct smc_link_group *lgr, int dmb_len,return rc; } +int smc_ism_attach_dmb (struct smcd_dev *dev,u64 token, +struct smc_buf_desc *dmb_desc) +{ +struct smcd_dmb dmb; +int rc =0 ; + +memset (&dmb,0 ,sizeof (dmb)); + dmb.dmb_tok = token; + +if (!dev->ops->attach_dmb) +return -EINVAL ; + + rc = dev->ops->attach_dmb (dev, &dmb); +if (!rc) { + dmb_desc->sba_idx = dmb.sba_idx; + dmb_desc->token = dmb.dmb_tok; + dmb_desc->cpu_addr = dmb.cpu_addr; + dmb_desc->dma_addr = dmb.dma_addr; + dmb_desc->len = dmb.dmb_len; + } +return rc; +} + +int smc_ism_detach_dmb (struct smcd_dev *dev,u64 token) +{ +if (!dev->ops->detach_dmb) +return -EINVAL ; + +return dev->ops->detach_dmb (dev, token); +} +static int smc_nl_handle_smcd_dev (struct smcd_dev *smcd,struct sk_buff *skb,struct netlink_callback *cb)
net/smc/smc_ism.h +2 -0
@@ -39,6 +39,8 @@ int smc_ism_register_dmb(struct smc_link_group *lgr, int buf_size, struct smc_buf_desc *dmb_desc);int smc_ism_unregister_dmb (struct smcd_dev *dev,struct smc_buf_desc *dmb_desc);bool smc_ism_dmb_mappable (struct smcd_dev *smcd); +int smc_ism_attach_dmb (struct smcd_dev *dev,u64 token,struct smc_buf_desc *dmb_desc); +int smc_ism_detach_dmb (struct smcd_dev *dev,u64 token);int smc_ism_signal_shutdown (struct smc_link_group *lgr);void smc_ism_get_system_eid (u8 **eid);u16 smc_ism_get_chid (struct smcd_dev *dev);
Signature · 1 line
1.8.3.1
Wen Gu
reviewerRFC PATCH v3 7/9 net/smc: Avoid data copy from sndbuf to peer RMB in SMC-D
In reply to Wen GuThis patch aims to avoid data copy from local sndbuf to peer RMB by attaching local sndbuf to peer RMB when DMBs have ISM_DMB_MAPPABLE attribute.
After this, local sndbuf and peer RMB share the same physical memory.
+----------+ +----------+
| socket A | | socket B |
+----------+ +----------+
| ^
| +---------+ |
regard as | | ----------|
local sndbuf | B's | regard as
| | RMB | local RMB
|-------> | |
+---------+1. Actions on local RMB.
a. Create or reuse RMB when connection is created;
b. Unuse RMB when connection is freed;
c. Free RMB when link group is freed;2. Actions on local sndbuf.Attach local sndbuf to peer RMB.
a. Attach local sndbuf to peer RMB by the rtoken exchanged through
CLC message. Since then, accessing local sndbuf is equivalent to accessing peer RMB
b. sndbuf_desc is exclusive to specific connection and won't be
added to lgr buffer pool for reuse.
c. Local sndbuf is detached from peer RMB when connection is freed,
and sndbuf_desc will be freed.
Therefore, the data written to local sndbuf will directly reach peer RMB.
--- net/smc/af_smc.c | 14 +++++++++++ net/smc/smc_core.c | 70 +++++++++++++++++++++++++++++++++++++++++++++++++++++- net/smc/smc_core.h | 1 + 3 files changed, 84 insertions(+), 1 deletion(-)
net/smc/af_smc.c +14 -0
@@ -1366,6 +1366,12 @@ static int smc_connect_ism(struct smc_sock *smc, }smc_conn_save_peer_info (smc, aclc); + +if (smc_ism_dmb_mappable (smc->conn.lgr->smcd)) { + rc =smcd_buf_attach (smc); +if (rc) + goto connect_abort; + }smc_close_init (smc);smc_rx_init (smc);smc_tx_init (smc);
@@ -2422,6 +2428,14 @@ static void smc_listen_work(struct work_struct *work)mutex_unlock (&smc_server_lgr_pending); }smc_conn_save_peer_info (new_smc, cclc); + +if (ini->is_smcd && +smc_ism_dmb_mappable (new_smc->conn.lgr->smcd)) { + rc =smcd_buf_attach (new_smc); +if (rc) + goto out_decl; + } +smc_listen_out_connected (new_smc);SMC_STAT_SERV_SUCC_INC (sock_net (newclcsock->sk), ini); goto out_free;
net/smc/smc_core.c +69 -1
102 lines not on this page · raw message
net/smc/smc_core.h +1 -0
@@ -518,6 +518,7 @@ void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid,void smc_smcd_terminate_all (struct smcd_dev *dev);void smc_smcr_terminate_all (struct smc_ib_device *smcibdev);int smc_buf_create (struct smc_sock *smc,bool is_smcd); +int smcd_buf_attach (struct smc_sock *smc);int smc_uncompress_bufsize (u8 compressed);int smc_rmb_rtoken_handling (struct smc_connection *conn,struct smc_link *link,struct smc_clc_msg_accept_confirm *clc);
Signature · 1 line
1.8.3.1
Wen Gu
reviewerRFC PATCH v3 8/9 net/smc: Modify cursor update logic when using mappable DMB
In reply to Wen GuSince local sndbuf shares the same physical memory region with peer RMB when using mappable DMBs, the cursor update logic needs to be adapted.
The main concern is to ensure that the data written by local to this memory region won't overwrite the data that has not been consumed by the peer.
So in this scene, the fin_curs and sndbuf_space that were originally updated when sending out CDC message are not updated until the cons_curs update from the peer is received.
--- net/smc/smc_cdc.c | 50 +++++++++++++++++++++++++++++++++++++++----------- 1 file changed, 39 insertions(+), 11 deletions(-)
net/smc/smc_cdc.c +39 -11
@@ -18,6 +18,7 @@#include "smc_tx.h" #include "smc_rx.h" #include "smc_close.h" +#include "smc_ism.h" /********************************** send *************************************/
@@ -253,17 +254,24 @@ int smcd_cdc_msg_send(struct smc_connection *conn)return rc;smc_curs_copy (&conn->rx_curs_confirmed, &curs, conn); conn->local_rx_ctrl.prod_flags.cons_curs_upd_req =0 ; -/* Calculate transmitted data and increment free send buffer space */ - diff =smc_curs_diff (conn->sndbuf_desc->len, &conn->tx_curs_fin, - &conn->tx_curs_sent); -/* increased by confirmed number of bytes */ -smp_mb__before_atomic (); -atomic_add (diff, &conn->sndbuf_space); -/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */ -smp_mb__after_atomic (); -smc_curs_copy (&conn->tx_curs_fin, &conn->tx_curs_sent, conn); +if (!smc_ism_dmb_mappable (conn->lgr->smcd)) { +/* If local sndbuf has been mapped to peer RMB, then +* don't update the tx_curs_fin and sndbuf_space until +* peer has consumed the data in RMB. +*/ -smc_tx_sndbuf_nonfull (smc); +/* Calculate transmitted data and increment free send buffer space */ + diff =smc_curs_diff (conn->sndbuf_desc->len, &conn->tx_curs_fin, + &conn->tx_curs_sent); +/* increased by confirmed number of bytes */ +smp_mb__before_atomic (); +atomic_add (diff, &conn->sndbuf_space); +/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */ +smp_mb__after_atomic (); +smc_curs_copy (&conn->tx_curs_fin, &conn->tx_curs_sent, conn); + +smc_tx_sndbuf_nonfull (smc); + }return rc; }
@@ -321,7 +329,7 @@ static void smc_cdc_msg_recv_action(struct smc_sock *smc, {union smc_host_cursor cons_old, prod_old;struct smc_connection *conn = &smc->conn; -int diff_cons, diff_prod; +int diff_cons, diff_prod, diff_tx;smc_curs_copy (&prod_old, &conn->local_rx_ctrl.prod, conn);smc_curs_copy (&cons_old, &conn->local_rx_ctrl.cons, conn);
@@ -337,6 +345,26 @@ static void smc_cdc_msg_recv_action(struct smc_sock *smc,atomic_add (diff_cons, &conn->peer_rmbe_space);/* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */ smp_mb__after_atomic (); + +if (conn->lgr->is_smcd && +smc_ism_dmb_mappable (conn->lgr->smcd)) { +/* If local sndbuf has been mapped to peer RMB, then +* update tx_curs_fin and sndbuf_space when peer has +* consumed the data in it's RMB. +*/ + +/* calculate peer rmb consumed data */ + diff_tx =smc_curs_diff (conn->sndbuf_desc->len, &conn->tx_curs_fin, + &conn->local_rx_ctrl.cons); +/* increase local sndbuf space and fin_curs */ +smp_mb__before_atomic (); +atomic_add (diff_tx, &conn->sndbuf_space); +/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */ +smp_mb__after_atomic (); +smc_curs_copy (&conn->tx_curs_fin, &conn->local_rx_ctrl.cons, conn); + +smc_tx_sndbuf_nonfull (smc); + } } diff_prod = smc_curs_diff(conn->rmb_desc->len, &prod_old,
Signature · 1 line
1.8.3.1
Wen Gu
reviewerRFC PATCH v3 9/9 net/smc: Add interface implementation of loopback device
In reply to Wen GuThis patch completes the specific implementation of loopback device for the newly added SMC-D DMB-related interface.
The loopback device always provides mappable DMB because the device users are in the same OS instance.
--- net/smc/smc_loopback.c | 93 ++++++++++++++++++++++++++++++++++++++++++++++---- net/smc/smc_loopback.h | 4 +++ 2 files changed, 90 insertions(+), 7 deletions(-)
net/smc/smc_loopback.c +86 -7
@@ -77,6 +77,7 @@ static int smc_lo_register_dmb(struct smcd_dev *smcd, struct smcd_dmb *dmb, goto err_node; } dmb_node->len = dmb->dmb_len; +refcount_set (&dmb_node->refcnt,1 );/* TODO: token is random but not exclusive ! * suppose to find token in dmb hask table, if has this token
@@ -87,6 +88,7 @@ static int smc_lo_register_dmb(struct smcd_dev *smcd, struct smcd_dmb *dmb,write_lock (&ldev->dmb_ht_lock);hash_add (ldev->dmb_ht, &dmb_node->list, dmb_node->token);write_unlock (&ldev->dmb_ht_lock); +atomic_inc (&ldev->dmb_cnt); dmb->sba_idx = dmb_node->sba_idx; dmb->dmb_tok = dmb_node->token;
@@ -124,12 +126,76 @@ static int smc_lo_unregister_dmb(struct smcd_dev *smcd, struct smcd_dmb *dmb)write_unlock (&ldev->dmb_ht_lock);clear_bit (dmb_node->sba_idx, ldev->sba_idx_mask); + +/* wait for dmb refcnt to be 0 */ +if (!refcount_dec_and_test (&dmb_node->refcnt)) +wait_event (ldev->dmbs_release, !refcount_read (&dmb_node->refcnt));kfree (dmb_node->cpu_addr);kfree (dmb_node); +if (atomic_dec_and_test (&ldev->dmb_cnt)) +wake_up (&ldev->ldev_release); +return 0 ; +} + +static int smc_lo_attach_dmb (struct smcd_dev *smcd,struct smcd_dmb *dmb) +{ +struct smc_lo_dmb_node *dmb_node =NULL , *tmp_node; +struct smc_lo_dev *ldev = smcd->priv; + +/* find dmb_node according to dmb->dmb_tok */ +read_lock (&ldev->dmb_ht_lock); +hash_for_each_possible (ldev->dmb_ht, tmp_node, list, dmb->dmb_tok) { +if (tmp_node->token == dmb->dmb_tok) { + dmb_node = tmp_node; +break ; + } + } +if (!dmb_node) { +read_unlock (&ldev->dmb_ht_lock); +return -EINVAL ; + } +read_unlock (&ldev->dmb_ht_lock); +refcount_inc (&dmb_node->refcnt); + +/* provide dmb information */ + dmb->sba_idx = dmb_node->sba_idx; + dmb->dmb_tok = dmb_node->token; + dmb->cpu_addr = dmb_node->cpu_addr; + dmb->dma_addr = dmb_node->dma_addr; + dmb->dmb_len = dmb_node->len; +return 0 ; +} + +static int smc_lo_detach_dmb (struct smcd_dev *smcd,u64 token) +{ +struct smc_lo_dmb_node *dmb_node =NULL , *tmp_node; +struct smc_lo_dev *ldev = smcd->priv; + +/* find dmb_node according to dmb->dmb_tok */ +read_lock (&ldev->dmb_ht_lock); +hash_for_each_possible (ldev->dmb_ht, tmp_node, list, token) { +if (tmp_node->token == token) { + dmb_node = tmp_node; +break ; + } + } +if (!dmb_node) { +read_unlock (&ldev->dmb_ht_lock); +return -EINVAL ; + } +read_unlock (&ldev->dmb_ht_lock); + +if (refcount_dec_and_test (&dmb_node->refcnt)) +wake_up_all (&ldev->dmbs_release);return 0 ; } +static int smc_lo_get_dev_dmb_attr (struct smcd_dev *smcd) +{ +return (1 <<ISM_DMB_MAPPABLE ); +} +static int smc_lo_add_vlan_id (struct smcd_dev *smcd,u64 vlan_id) {return -EOPNOTSUPP ;
@@ -156,7 +222,15 @@ static int smc_lo_move_data(struct smcd_dev *smcd, u64 dmb_tok, unsigned int idx {struct smc_lo_dmb_node *rmb_node =NULL , *tmp_node;struct smc_lo_dev *ldev = smcd->priv; - +struct smc_connection *conn; + +if (!sf) { +/* local sndbuf shares the same physical memory with +* peer RMB, so no need to copy data from local sndbuf +* to peer RMB. +*/ +return 0 ; + }read_lock (&ldev->dmb_ht_lock);hash_for_each_possible (ldev->dmb_ht, tmp_node, list, dmb_tok) {if (tmp_node->token == dmb_tok) {
@@ -172,13 +246,10 @@ static int smc_lo_move_data(struct smcd_dev *smcd, u64 dmb_tok, unsigned int idxmemcpy ((char *)rmb_node->cpu_addr + offset, data, size); -if (sf) { -struct smc_connection *conn = - smcd->conn[rmb_node->sba_idx]; + conn = smcd->conn[rmb_node->sba_idx]; +if (conn && !conn->killed) +smcd_cdc_rx_handler (conn); -if (conn && !conn->killed) -smcd_cdc_rx_handler (conn); - }return 0 ; }
@@ -206,6 +277,8 @@ static struct device *smc_lo_get_dev(struct smcd_dev *smcd) .query_remote_gid = smc_lo_query_rgid, .register_dmb = smc_lo_register_dmb, .unregister_dmb = smc_lo_unregister_dmb, + .attach_dmb = smc_lo_attach_dmb, + .detach_dmb = smc_lo_detach_dmb, .add_vlan_id = smc_lo_add_vlan_id, .del_vlan_id = smc_lo_del_vlan_id, .set_vlan_required = smc_lo_set_vlan_required,
@@ -216,6 +289,7 @@ static struct device *smc_lo_get_dev(struct smcd_dev *smcd) .get_local_gid = smc_lo_get_local_gid, .get_chid = smc_lo_get_chid, .get_dev = smc_lo_get_dev, + .get_dev_dmb_attr = smc_lo_get_dev_dmb_attr, }; static struct smcd_dev *smcd_lo_alloc_dev(const struct smcd_ops *ops,
@@ -296,6 +370,9 @@ static int smc_lo_dev_init(struct smc_lo_dev *ldev)smc_lo_gen_id (ldev);rwlock_init (&ldev->dmb_ht_lock);hash_init (ldev->dmb_ht); +atomic_set (&ldev->dmb_cnt,0 ); +init_waitqueue_head (&ldev->dmbs_release); +init_waitqueue_head (&ldev->ldev_release);return smcd_lo_register_dev (ldev); }
@@ -334,6 +411,8 @@ static int smc_lo_dev_probe(void)static void smc_lo_dev_exit (struct smc_lo_dev *ldev) {smcd_lo_unregister_dev (ldev); +if (atomic_read (&ldev->dmb_cnt)) +wait_event (ldev->ldev_release, !atomic_read (&ldev->dmb_cnt)); }static void smc_lo_dev_remove (void )
net/smc/smc_loopback.h +4 -0
@@ -33,6 +33,7 @@ struct smc_lo_dmb_node {u32 sba_idx;void *cpu_addr;dma_addr_t dma_addr; +refcount_t refcnt; };struct smc_lo_dev {
@@ -43,6 +44,9 @@ struct smc_lo_dev {DECLARE_BITMAP (sba_idx_mask,SMC_LODEV_MAX_DMBS );rwlock_t dmb_ht_lock;DECLARE_HASHTABLE (dmb_ht,SMC_LODEV_MAX_DMBS_BUCKETS ); +atomic_t dmb_cnt; +wait_queue_head_t dmbs_release; +wait_queue_head_t ldev_release; };int smc_loopback_init (void );
Signature · 1 line
1.8.3.1
Wen Gu
reviewerQuoted 155 lines · Wen Gu
Hi, all
# Background
The background and previous discussion can be referred from [1].
We found SMC-D can be used to accelerate OS internal communication, such as loopback or between two containers within the same OS instance. So this patch set provides a kind of SMC-D dummy device (we call it the SMC-D loopback device) to emulate an ISM device, so that SMC-D can also be used on architectures other than s390. The SMC-D loopback device are designed as a system global device, visible to all containers.
This version is implemented based on the generalized interface provided by [2]. And there is an open issue of this version, which will be mentioned later.
# Design
This patch set basically follows the design of the previous version.
Patch #1/9 ~ #3/9 attempt to decouple ISM-related structures from the SMC-D generalized code and extract some helpers to make SMC-D protocol compatible with devices other than s390 ISM device.
Patch #4/9 introduces a kind of loopback device, which is defined as SMC-D v2 device and designed to provide communication between SMC sockets in the same OS instance.
+-------------------------------------------+ | +--------------+ +--------------+ | | | SMC socket A | | SMC socket B | | | +--------------+ +--------------+ | | ^ ^ | | | +----------------+ | | | | | SMC stack | | | | +--->| +------------+ |<--| | | | | dummy | | | | | | device | | | | +-+------------+-+ | | OS | +-------------------------------------------+Patch #5/9 ~ #8/9 expand SMC-D protocol interface (
smcd_ops) for scenarios where SMC-D is used to communicate within VM (loopback here) or between VMs on the same host (based on virtio-ism device, see [3]). What these scenarios have in common is that the local sndbuf and peer RMB can be mapped to same physical memory region, so the data copy between the local sndbuf and peer RMB can be omitted. Performance improvement brought by this extension can be found in # Benchmark Test.+----------+ +----------+ | socket A | | socket B | +----------+ +----------+ | ^ | +---------+ | regard as | | ----------| local sndbuf | B's | regard as | | RMB | local RMB |-------> | | +---------+Patch #9/9 realizes the support of loopback device for the above-mentioned expanded SMC-D protocol interface.
# Benchmark Test
* Test environments:
- VM with Intel Xeon Platinum 8 core 2.50GHz, 16 GiB mem.
- SMC sndbuf/RMB size 1MB.* Test object:
- TCP lo: run on TCP loopback.
- domain: run on UNIX domain.
- SMC lo: run on SMC loopback device with patch #1/9 ~ #4/9.
- SMC lo-nocpy: run on SMC loopback device with patch #1/9 ~ #9/9.1. ipc-benchmark (see [4])
- ./<foo> -c 1000000 -s 100
TCP-lo domain SMC-lo SMC-lo-nocpy
Message
rate (msg/s) 79025 115736(+46.45%) 146760(+85.71%) 149800(+89.56%)2. sockperf
- serv: <
smc_run> taskset -c <cpu> sockperf sr --tcp
- clnt: <smc_run> taskset -c <cpu> sockperf { tp | pp } --tcp --msg-size={ 64000 for tp | 14 for pp } -i 127.0.0.1 -t 30TCP-lo SMC-lo SMC-lo-nocpy
Bandwidth(MBps) 4822.388 4940.918(+2.56%) 8086.67(+67.69%)
Latency(us) 6.298 3.352(-46.78%) 3.35(-46.81%)3. iperf3
- serv: <
smc_run> taskset -c <cpu> iperf3 -s
- clnt: <smc_run> taskset -c <cpu> iperf3 -c 127.0.0.1 -t 15TCP-lo SMC-lo SMC-lo-nocpy
Bitrate(Gb/s) 40.7 40.5(-0.49%) 72.4(+77.89%)4. nginx/wrk
- serv: <
smc_run> nginx
- clnt: <smc_run> wrk -t 8 -c 500 -d 30 http://127.0.0.1:80TCP-lo SMC-lo SMC-lo-nocpy
Requests/s 155994.57 214544.79(+37.53%) 215538.55(+38.17%)# Open issue
The open issue has not been resolved now is about how to detect that the source and target of CLC proposal are within the same OS instance and can communicate through the SMC-D loopback device. Similar issue also exists when using virtio-ism devices (the background and details of virtio-ism device can be referred from [3]). In previous discussions, multiple options were proposed (see [5]). Thanks again for the help of the community. cc Alexandra Winter :)
But as we discussed, these solutions have some imperfection. So this version of RFC continues to use previous workaround, that is, a 64-bit random GID is generated for SMC-D loopback device. If the GIDs of the devices found by two peers are the same, then they are considered to be in the same OS instance and can communicate with each other by the loopback device.
This approach has very small risk. Assume the following situations:
(1) Assume that the SMC-D loopback devices of the two OS instances happen to generate the same 64-bit GID.
For the convenience of description, we refer to the sockets on these two different OS instance as server A and client B.
A will misjudge that the two are on the same OS instance because the same GID in CLC proposal message. Then A creates its RMB and sends 64-bit token-A to B in CLC accept message.
B receives the CLC accept message. And according to patch #7/9, B tries to attach its sndbuf to A's RMB by token-A.
(2) Assume that the OS instance where B is located happens to have an unattached RMB whose 64-bit token is same as token-A.
Then B successfully attaches its sndbuf to the wrong RMB, and creates its RMB, sends token-B to A in CLC confirm message.
Similarly, A receives the message and tries to attach its sndbuf to B's RMB by token-B.
(3) Similar to (2), assume that the OS instance where A is located happens to have an unattached RMB whose 64-bit token is same as token-B.
Then A successfully attach its sndbuf to the wrong RMB. Both sides mistakenly believe that an SMC-D connection based on the loopback device is established between them.
Quoted 6 lines
If the above 3 coincidences all happen, that is, 64-bit random number conflicts occur 3 times, then an unreachable SMC-D connection will be established, which is nasty. If one of above is not satisfied, it will safely fallback to TCP.
Since the chances of these happening are very small, I wonder if this risk of 1/2^(64*3) probability can be tolerated ?
Hi,
Any comments about this open issue or other parts of this RFC patch set? :)
Thanks,
Wen Gu
Quoted 50 lines
Another way to solve this open issue is using a 128-bit UUID to identify SMC-D loopback device or virtio-ism device, because the probability of a 128-bit UUID collision is considered negligible. But it may need to extend the CLC message to carry a longer GID, which is the last option.
v3->v2
1. Adapt new generalized interface provided by [2];
2. Select loopback device through SMC-D v2 protocol;
3. Split the loopback-related implementation and generic implementation into different patches more reasonably.
v1->v2
1. Fix some build WARNINGs complained by kernel test rebotReported-by kernel test robot <[email protected]>2. Add iperf3 test data.
[1] https://lore.kernel.org/netdev/[email protected]/
[2] https://lore.kernel.org/netdev/[email protected]/
[3] https://lists.oasis-open.org/archives/virtio-comment/202302/msg00148.html
[4] https://github.com/goldsborough/ipc-bench
[5] https://lore.kernel.org/netdev/[email protected]/Wen Gu (9):
net/smc: Decoupleism_devfrom SMC-D device dump net/smc: Decoupleism_devfrom SMC-D DMB registration net/smc: Extract v2 check helper from SMC-D device registration net/smc: Introduce SMC-D loopback device
net/smc: Introduce an interface for getting DMB attribute net/smc: Introudce interfaces for DMB attach and detach net/smc: Avoid data copy from sndbuf to peer RMB in SMC-D net/smc: Modify cursor update logic when using mappable DMB net/smc: Add interface implementation of loopback devicedrivers/s390/net/ism_drv.c | 5 +- include/net/smc.h | 18 +- net/smc/Makefile | 2 +- net/smc/af_smc.c | 26 ++- net/smc/smc_cdc.c | 59 ++++-- net/smc/smc_cdc.h | 1 + net/smc/smc_core.c | 70 ++++++- net/smc/smc_core.h | 1 + net/smc/smc_ism.c | 79 ++++++-- net/smc/smc_ism.h | 4 + net/smc/smc_loopback.c | 442 +++++++++++++++++++++++++++++++++++++++++++++ net/smc/smc_loopback.h | 55 ++++++ 12 files changed, 725 insertions(+), 37 deletions(-)create mode 100644
net/smc/smc_loopback.c
create mode 100644net/smc/smc_loopback.h
Quoted 167 lines · Wen Gu
Quoted 156 lines · Wen Gu
Hi, all
# Background
The background and previous discussion can be referred from [1].
We found SMC-D can be used to accelerate OS internal communication, such as loopback or between two containers within the same OS instance. So this patch set provides a kind of SMC-D dummy device (we call it the SMC-D loopback device) to emulate an ISM device, so that SMC-D can also be used on architectures other than s390. The SMC-D loopback device are designed as a system global device, visible to all containers.
This version is implemented based on the generalized interface provided by [2]. And there is an open issue of this version, which will be mentioned later.
# Design
This patch set basically follows the design of the previous version.
Patch #1/9 ~ #3/9 attempt to decouple ISM-related structures from the SMC-D generalized code and extract some helpers to make SMC-D protocol compatible with devices other than s390 ISM device.
Patch #4/9 introduces a kind of loopback device, which is defined as SMC-D v2 device and designed to provide communication between SMC sockets in the same OS instance.
+-------------------------------------------+ | +--------------+ +--------------+ | | | SMC socket A | | SMC socket B | | | +--------------+ +--------------+ | | ^ ^ | | | +----------------+ | | | | | SMC stack | | | | +--->| +------------+ |<--| | | | | dummy | | | | | | device | | | | +-+------------+-+ | | OS | +-------------------------------------------+Patch #5/9 ~ #8/9 expand SMC-D protocol interface (
smcd_ops) for scenarios where SMC-D is used to communicate within VM (loopback here) or between VMs on the same host (based on virtio-ism device, see [3]). What these scenarios have in common is that the local sndbuf and peer RMB can be mapped to same physical memory region, so the data copy between the local sndbuf and peer RMB can be omitted. Performance improvement brought by this extension can be found in # Benchmark Test.+----------+ +----------+ | socket A | | socket B | +----------+ +----------+ | ^ | +---------+ | regard as | | ----------| local sndbuf | B's | regard as | | RMB | local RMB |-------> | | +---------+Patch #9/9 realizes the support of loopback device for the above-mentioned expanded SMC-D protocol interface.
# Benchmark Test
* Test environments:
- VM with Intel Xeon Platinum 8 core 2.50GHz, 16 GiB mem.
- SMC sndbuf/RMB size 1MB.* Test object:
- TCP lo: run on TCP loopback.
- domain: run on UNIX domain.
- SMC lo: run on SMC loopback device with patch #1/9 ~ #4/9.
- SMC lo-nocpy: run on SMC loopback device with patch #1/9 ~ #9/9.1. ipc-benchmark (see [4])
- ./<foo> -c 1000000 -s 100
TCP-lo domain SMC-lo SMC-lo-nocpy
Message
rate (msg/s) 79025 115736(+46.45%) 146760(+85.71%) 149800(+89.56%)2. sockperf
- serv: <
smc_run> taskset -c <cpu> sockperf sr --tcp
- clnt: <smc_run> taskset -c <cpu> sockperf { tp | pp } --tcp --msg-size={ 64000 for tp | 14 for pp } -i 127.0.0.1 -t 30TCP-lo SMC-lo SMC-lo-nocpy
Bandwidth(MBps) 4822.388 4940.918(+2.56%) 8086.67(+67.69%)
Latency(us) 6.298 3.352(-46.78%) 3.35(-46.81%)3. iperf3
- serv: <
smc_run> taskset -c <cpu> iperf3 -s
- clnt: <smc_run> taskset -c <cpu> iperf3 -c 127.0.0.1 -t 15TCP-lo SMC-lo SMC-lo-nocpy
Bitrate(Gb/s) 40.7 40.5(-0.49%) 72.4(+77.89%)4. nginx/wrk
- serv: <
smc_run> nginx
- clnt: <smc_run> wrk -t 8 -c 500 -d 30 http://127.0.0.1:80TCP-lo SMC-lo SMC-lo-nocpy
Requests/s 155994.57 214544.79(+37.53%) 215538.55(+38.17%)# Open issue
The open issue has not been resolved now is about how to detect that the source and target of CLC proposal are within the same OS instance and can communicate through the SMC-D loopback device. Similar issue also exists when using virtio-ism devices (the background and details of virtio-ism device can be referred from [3]). In previous discussions, multiple options were proposed (see [5]). Thanks again for the help of the community. cc Alexandra Winter :)
But as we discussed, these solutions have some imperfection. So this version of RFC continues to use previous workaround, that is, a 64-bit random GID is generated for SMC-D loopback device. If the GIDs of the devices found by two peers are the same, then they are considered to be in the same OS instance and can communicate with each other by the loopback device.
This approach has very small risk. Assume the following situations:
(1) Assume that the SMC-D loopback devices of the two OS instances happen to generate the same 64-bit GID.
For the convenience of description, we refer to the sockets on these two different OS instance as server A and client B.
A will misjudge that the two are on the same OS instance because the same GID in CLC proposal message. Then A creates its RMB and sends 64-bit token-A to B in CLC accept message.
B receives the CLC accept message. And according to patch #7/9, B tries to attach its sndbuf to A's RMB by token-A.
(2) Assume that the OS instance where B is located happens to have an unattached RMB whose 64-bit token is same as token-A.
Then B successfully attaches its sndbuf to the wrong RMB, and creates its RMB, sends token-B to A in CLC confirm message.
Similarly, A receives the message and tries to attach its sndbuf to B's RMB by token-B.
(3) Similar to (2), assume that the OS instance where A is located happens to have an unattached RMB whose 64-bit token is same as token-B.
Then A successfully attach its sndbuf to the wrong RMB. Both sides mistakenly believe that an SMC-D connection based on the loopback device is established between them.
Quoted 6 lines
If the above 3 coincidences all happen, that is, 64-bit random number conflicts occur 3 times, then an unreachable SMC-D connection will be established, which is nasty. If one of above is not satisfied, it will safely fallback to TCP.
Since the chances of these happening are very small, I wonder if this risk of 1/2^(64*3) probability can be tolerated ?
Quoted 7 lines
Hi,
Any comments about this open issue or other parts of this RFC patch set? :)
Thanks,
Wen Gu
Hi Wen,
I don't forget it ;) I'm trying to run it by myself. Please give us more time for the trying and review.
Thanks
Wenjia
Quoted 50 lines
Quoted 50 lines
Another way to solve this open issue is using a 128-bit UUID to identify SMC-D loopback device or virtio-ism device, because the probability of a 128-bit UUID collision is considered negligible. But it may need to extend the CLC message to carry a longer GID, which is the last option.
v3->v2
1. Adapt new generalized interface provided by [2];
2. Select loopback device through SMC-D v2 protocol;
3. Split the loopback-related implementation and generic implementation into different patches more reasonably.
v1->v2
1. Fix some build WARNINGs complained by kernel test rebotReported-by kernel test robot <[email protected]>2. Add iperf3 test data.
[1] https://lore.kernel.org/netdev/[email protected]/
[2] https://lore.kernel.org/netdev/[email protected]/
[3] https://lists.oasis-open.org/archives/virtio-comment/202302/msg00148.html
[4] https://github.com/goldsborough/ipc-bench
[5] https://lore.kernel.org/netdev/[email protected]/Wen Gu (9):
net/smc: Decoupleism_devfrom SMC-D device dump net/smc: Decoupleism_devfrom SMC-D DMB registration net/smc: Extract v2 check helper from SMC-D device registration net/smc: Introduce SMC-D loopback device
net/smc: Introduce an interface for getting DMB attribute net/smc: Introudce interfaces for DMB attach and detach net/smc: Avoid data copy from sndbuf to peer RMB in SMC-D net/smc: Modify cursor update logic when using mappable DMB net/smc: Add interface implementation of loopback device
drivers/s390/net/ism_drv.c| 5 +-
include/net/smc.h | 18 +-
net/smc/Makefile | 2 +-
net/smc/af_smc.c | 26 ++-
net/smc/smc_cdc.c | 59 ++++--
net/smc/smc_cdc.h | 1 +
net/smc/smc_core.c | 70 ++++++-
net/smc/smc_core.h | 1 +
net/smc/smc_ism.c | 79 ++++++--
net/smc/smc_ism.h | 4 +
net/smc/smc_loopback.c | 442
+++++++++++++++++++++++++++++++++++++++++++++
net/smc/smc_loopback.h | 55 ++++++12 files changed, 725 insertions(+), 37 deletions(-)create mode 100644
net/smc/smc_loopback.c
create mode 100644net/smc/smc_loopback.h
Wen Gu
reviewerQuoted 177 lines · Wenjia Zhang
Quoted 167 lines · Wen Gu
Quoted 156 lines · Wen Gu
Hi, all
# Background
The background and previous discussion can be referred from [1].
We found SMC-D can be used to accelerate OS internal communication, such as loopback or between two containers within the same OS instance. So this patch set provides a kind of SMC-D dummy device (we call it the SMC-D loopback device) to emulate an ISM device, so that SMC-D can also be used on architectures other than s390. The SMC-D loopback device are designed as a system global device, visible to all containers.
This version is implemented based on the generalized interface provided by [2]. And there is an open issue of this version, which will be mentioned later.
# Design
This patch set basically follows the design of the previous version.
Patch #1/9 ~ #3/9 attempt to decouple ISM-related structures from the SMC-D generalized code and extract some helpers to make SMC-D protocol compatible with devices other than s390 ISM device.
Patch #4/9 introduces a kind of loopback device, which is defined as SMC-D v2 device and designed to provide communication between SMC sockets in the same OS instance.
+-------------------------------------------+ | +--------------+ +--------------+ | | | SMC socket A | | SMC socket B | | | +--------------+ +--------------+ | | ^ ^ | | | +----------------+ | | | | | SMC stack | | | | +--->| +------------+ |<--| | | | | dummy | | | | | | device | | | | +-+------------+-+ | | OS | +-------------------------------------------+Patch #5/9 ~ #8/9 expand SMC-D protocol interface (
smcd_ops) for scenarios where SMC-D is used to communicate within VM (loopback here) or between VMs on the same host (based on virtio-ism device, see [3]). What these scenarios have in common is that the local sndbuf and peer RMB can be mapped to same physical memory region, so the data copy between the local sndbuf and peer RMB can be omitted. Performance improvement brought by this extension can be found in # Benchmark Test.+----------+ +----------+ | socket A | | socket B | +----------+ +----------+ | ^ | +---------+ | regard as | | ----------| local sndbuf | B's | regard as | | RMB | local RMB |-------> | | +---------+Patch #9/9 realizes the support of loopback device for the above-mentioned expanded SMC-D protocol interface.
# Benchmark Test
* Test environments:
- VM with Intel Xeon Platinum 8 core 2.50GHz, 16 GiB mem.
- SMC sndbuf/RMB size 1MB.* Test object:
- TCP lo: run on TCP loopback.
- domain: run on UNIX domain.
- SMC lo: run on SMC loopback device with patch #1/9 ~ #4/9.
- SMC lo-nocpy: run on SMC loopback device with patch #1/9 ~ #9/9.1. ipc-benchmark (see [4])
- ./<foo> -c 1000000 -s 100
TCP-lo domain SMC-lo SMC-lo-nocpy Message
rate (msg/s) 79025 115736(+46.45%) 146760(+85.71%) 149800(+89.56%)2. sockperf
- serv: <
smc_run> taskset -c <cpu> sockperf sr --tcp
- clnt: <smc_run> taskset -c <cpu> sockperf { tp | pp } --tcp --msg-size={ 64000 for tp | 14 for pp } -i 127.0.0.1 -t 30TCP-lo SMC-lo SMC-lo-nocpy Bandwidth(MBps) 4822.388 4940.918(+2.56%) 8086.67(+67.69%) Latency(us) 6.298 3.352(-46.78%) 3.35(-46.81%)
3. iperf3
- serv: <
smc_run> taskset -c <cpu> iperf3 -s
- clnt: <smc_run> taskset -c <cpu> iperf3 -c 127.0.0.1 -t 15TCP-lo SMC-lo SMC-lo-nocpy Bitrate(Gb/s) 40.7 40.5(-0.49%) 72.4(+77.89%)
4. nginx/wrk
- serv: <
smc_run> nginx
- clnt: <smc_run> wrk -t 8 -c 500 -d 30 http://127.0.0.1:80TCP-lo SMC-lo SMC-lo-nocpy Requests/s 155994.57 214544.79(+37.53%) 215538.55(+38.17%)
# Open issue
The open issue has not been resolved now is about how to detect that the source and target of CLC proposal are within the same OS instance and can communicate through the SMC-D loopback device. Similar issue also exists when using virtio-ism devices (the background and details of virtio-ism device can be referred from [3]). In previous discussions, multiple options were proposed (see [5]). Thanks again for the help of the community. cc Alexandra Winter :)
But as we discussed, these solutions have some imperfection. So this version of RFC continues to use previous workaround, that is, a 64-bit random GID is generated for SMC-D loopback device. If the GIDs of the devices found by two peers are the same, then they are considered to be in the same OS instance and can communicate with each other by the loopback device.
This approach has very small risk. Assume the following situations:
(1) Assume that the SMC-D loopback devices of the two OS instances happen to generate the same 64-bit GID.
For the convenience of description, we refer to the sockets on these two different OS instance as server A and client B.
A will misjudge that the two are on the same OS instance because the same GID in CLC proposal message. Then A creates its RMB and sends 64-bit token-A to B in CLC accept message.
B receives the CLC accept message. And according to patch #7/9, B tries to attach its sndbuf to A's RMB by token-A.
(2) Assume that the OS instance where B is located happens to have an unattached RMB whose 64-bit token is same as token-A.
Then B successfully attaches its sndbuf to the wrong RMB, and creates its RMB, sends token-B to A in CLC confirm message.
Similarly, A receives the message and tries to attach its sndbuf to B's RMB by token-B.
(3) Similar to (2), assume that the OS instance where A is located happens to have an unattached RMB whose 64-bit token is same as token-B.
Then A successfully attach its sndbuf to the wrong RMB. Both sides mistakenly believe that an SMC-D connection based on the loopback device is established between them.
Quoted 6 lines
If the above 3 coincidences all happen, that is, 64-bit random number conflicts occur 3 times, then an unreachable SMC-D connection will be established, which is nasty. If one of above is not satisfied, it will safely fallback to TCP.
Since the chances of these happening are very small, I wonder if this risk of 1/2^(64*3) probability can be tolerated ?
Quoted 7 lines
Hi,
Any comments about this open issue or other parts of this RFC patch set? :)
Thanks,
Wen Gu
Quoted 7 lines
Hi Wen,
I don't forget it ;) I'm trying to run it by myself. Please give us more time for the trying and review.
Thanks
Wenjia
Sure, Wenjia. Thank you!
Please feel free to add comments. I will wait for you to complete the review before deciding what to do next.
Regards,
Wen Gu
Quoted 49 lines
Quoted 49 lines
Quoted 49 lines
Another way to solve this open issue is using a 128-bit UUID to identify SMC-D loopback device or virtio-ism device, because the probability of a 128-bit UUID collision is considered negligible. But it may need to extend the CLC message to carry a longer GID, which is the last option.
v3->v2
1. Adapt new generalized interface provided by [2];
2. Select loopback device through SMC-D v2 protocol;
3. Split the loopback-related implementation and generic implementation into different patches more reasonably.
v1->v2
1. Fix some build WARNINGs complained by kernel test rebotReported-by kernel test robot <[email protected]>2. Add iperf3 test data.
[1] https://lore.kernel.org/netdev/[email protected]/
[2] https://lore.kernel.org/netdev/[email protected]/
[3] https://lists.oasis-open.org/archives/virtio-comment/202302/msg00148.html
[4] https://github.com/goldsborough/ipc-bench
[5] https://lore.kernel.org/netdev/[email protected]/Wen Gu (9):
net/smc: Decoupleism_devfrom SMC-D device dump net/smc: Decoupleism_devfrom SMC-D DMB registration net/smc: Extract v2 check helper from SMC-D device registration net/smc: Introduce SMC-D loopback device
net/smc: Introduce an interface for getting DMB attribute net/smc: Introudce interfaces for DMB attach and detach net/smc: Avoid data copy from sndbuf to peer RMB in SMC-D net/smc: Modify cursor update logic when using mappable DMB net/smc: Add interface implementation of loopback device
drivers/s390/net/ism_drv.c| 5 +-
include/net/smc.h | 18 +-
net/smc/Makefile | 2 +-
net/smc/af_smc.c | 26 ++-
net/smc/smc_cdc.c | 59 ++++--
net/smc/smc_cdc.h | 1 +
net/smc/smc_core.c | 70 ++++++-
net/smc/smc_core.h | 1 +
net/smc/smc_ism.c | 79 ++++++--
net/smc/smc_ism.h | 4 +
net/smc/smc_loopback.c | 442 +++++++++++++++++++++++++++++++++++++++++++++ net/smc/smc_loopback.h | 55 ++++++12 files changed, 725 insertions(+), 37 deletions(-)create mode 100644
net/smc/smc_loopback.c
create mode 100644net/smc/smc_loopback.h