UbiquitousOS
/
OpenCloudOS-Kernel
mirror of https://gitee.com/OpenCloudOS/OpenCloudOS-Kernel.git
11
0
Fork 0
Code Issues Projects Releases Wiki Activity

IB/hfi1: Improve SDMA engine assignment for user SDMA 

Currently each user context is assigned a single SDMA engine
based on the VL, context id, and subcontext id. That means for
MPI applications, each rank can only use one SDMA engine for
all messages. This may create unwanted backup for independent
messages going to different destinations upon congestion at one
destination.

This patch adds the packet "dlid" to the formula of SDMA engine
selection for user SDMA requests. A simple hash table is used
to maintain even distribution among the available SDMA engines
regardless how the "dlid" values are distributed.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This commit is contained in:
Jianxin Xiong 2016-07-01 16:01:56 -07:00 committed by Doug Ledford
parent e014991d07
commit 14833b8c52
1 changed files with 28 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
drivers/infiniband/hw/hfi1/user_sdma.c
View File

@ -496,6 +496,27 @@ int hfi1_user_sdma_free_queues(struct hfi1_filedata *fd)
return 0; return 0;
} }
static u8 dlid_to_selector(u16 dlid)
{
static u8 mapping[256];
static int initialized;
static u8 next;
int hash;
if (!initialized) {
memset(mapping, 0xFF, 256);
initialized = 1;
}
hash = ((dlid >> 8) ^ dlid) & 0xFF;
if (mapping[hash] == 0xFF) {
mapping[hash] = next;
next = (next + 1) & 0x7F;
}
return mapping[hash];
}
int hfi1_user_sdma_process_request(struct file *fp, struct iovec *iovec, int hfi1_user_sdma_process_request(struct file *fp, struct iovec *iovec,
unsigned long dim, unsigned long *count) unsigned long dim, unsigned long *count)
{ {
@ -511,6 +532,8 @@ int hfi1_user_sdma_process_request(struct file *fp, struct iovec *iovec,
struct user_sdma_request *req; struct user_sdma_request *req;
u8 opcode, sc, vl; u8 opcode, sc, vl;
int req_queued = 0; int req_queued = 0;
u16 dlid;
u8 selector;
if (iovec[idx].iov_len < sizeof(info) + sizeof(req->hdr)) { if (iovec[idx].iov_len < sizeof(info) + sizeof(req->hdr)) {
hfi1_cdbg( hfi1_cdbg(
@ -686,9 +709,13 @@ int hfi1_user_sdma_process_request(struct file *fp, struct iovec *iovec,
idx++; idx++;
} }
dlid = be16_to_cpu(req->hdr.lrh[1]);
selector = dlid_to_selector(dlid);
/* Have to select the engine */ /* Have to select the engine */
req->sde = sdma_select_engine_vl(dd, req->sde = sdma_select_engine_vl(dd,
(u32)(uctxt->ctxt + fd->subctxt), (u32)(uctxt->ctxt + fd->subctxt +
selector),
vl); vl);
if (!req->sde || !sdma_running(req->sde)) { if (!req->sde || !sdma_running(req->sde)) {
ret = -ECOMM; ret = -ECOMM;