提交 dfc1ed1c 编写于 作者: J Justin Ernst 提交者: Ingo Molnar

x86/platform/uv/BAU: Fix congested_response_us not taking effect

Bug fix for the BAU tunable congested_cycles not being set to the user
defined value. Instead of referencing a global variable when deciding
on BAU shutdown, a node will reference its own tunable set
value ( cong_response_us). This results in the user set
tunable value congested_response_us taking effect correctly.
Signed-off-by: NJustin Ernst <justin.ernst@hpe.com>
Acked-by: NAndrew Banman <abanman@hpe.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: mike.travis@hpe.com
Cc: sivanich@hpe.com
Link: http://lkml.kernel.org/r/1499970803-282432-1-git-send-email-justin.ernst@hpe.comSigned-off-by: NIngo Molnar <mingo@kernel.org>
上级 5f8a1615
......@@ -40,7 +40,6 @@ static int timeout_base_ns[] = {
static int timeout_us;
static bool nobau = true;
static int nobau_perm;
static cycles_t congested_cycles;
/* tunables: */
static int max_concurr = MAX_BAU_CONCURRENT;
......@@ -829,10 +828,10 @@ static void record_send_stats(cycles_t time1, cycles_t time2,
if ((completion_status == FLUSH_COMPLETE) && (try == 1)) {
bcp->period_requests++;
bcp->period_time += elapsed;
if ((elapsed > congested_cycles) &&
if ((elapsed > usec_2_cycles(bcp->cong_response_us)) &&
(bcp->period_requests > bcp->cong_reps) &&
((bcp->period_time / bcp->period_requests) >
congested_cycles)) {
usec_2_cycles(bcp->cong_response_us))) {
stat->s_congested++;
disable_for_period(bcp, stat);
}
......@@ -2228,7 +2227,6 @@ static int __init uv_bau_init(void)
}
nuvhubs = uv_num_possible_blades();
congested_cycles = usec_2_cycles(congested_respns_us);
uv_base_pnode = 0x7fffffff;
for (uvhub = 0; uvhub < nuvhubs; uvhub++) {
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册