summaryrefslogtreecommitdiff
path: root/kernel/dma/map_benchmark.c
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/dma/map_benchmark.c')
-rw-r--r--kernel/dma/map_benchmark.c92
1 files changed, 54 insertions, 38 deletions
diff --git a/kernel/dma/map_benchmark.c b/kernel/dma/map_benchmark.c
index da95df381483..cc19a3efea89 100644
--- a/kernel/dma/map_benchmark.c
+++ b/kernel/dma/map_benchmark.c
@@ -1,6 +1,6 @@
// SPDX-License-Identifier: GPL-2.0-only
/*
- * Copyright (C) 2020 Hisilicon Limited.
+ * Copyright (C) 2020 HiSilicon Limited.
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
@@ -11,6 +11,7 @@
#include <linux/dma-mapping.h>
#include <linux/kernel.h>
#include <linux/kthread.h>
+#include <linux/map_benchmark.h>
#include <linux/math64.h>
#include <linux/module.h>
#include <linux/pci.h>
@@ -18,27 +19,6 @@
#include <linux/slab.h>
#include <linux/timekeeping.h>
-#define DMA_MAP_BENCHMARK _IOWR('d', 1, struct map_benchmark)
-#define DMA_MAP_MAX_THREADS 1024
-#define DMA_MAP_MAX_SECONDS 300
-
-#define DMA_MAP_BIDIRECTIONAL 0
-#define DMA_MAP_TO_DEVICE 1
-#define DMA_MAP_FROM_DEVICE 2
-
-struct map_benchmark {
- __u64 avg_map_100ns; /* average map latency in 100ns */
- __u64 map_stddev; /* standard deviation of map latency */
- __u64 avg_unmap_100ns; /* as above */
- __u64 unmap_stddev;
- __u32 threads; /* how many threads will do map/unmap in parallel */
- __u32 seconds; /* how long the test will last */
- __s32 node; /* which numa node this benchmark will run on */
- __u32 dma_bits; /* DMA addressing capability */
- __u32 dma_dir; /* DMA data direction */
- __u8 expansion[84]; /* For future use */
-};
-
struct map_benchmark_data {
struct map_benchmark bparam;
struct device *dev;
@@ -56,9 +36,11 @@ static int map_benchmark_thread(void *data)
void *buf;
dma_addr_t dma_addr;
struct map_benchmark_data *map = data;
+ int npages = map->bparam.granule;
+ u64 size = npages * PAGE_SIZE;
int ret = 0;
- buf = (void *)__get_free_page(GFP_KERNEL);
+ buf = alloc_pages_exact(size, GFP_KERNEL);
if (!buf)
return -ENOMEM;
@@ -74,10 +56,10 @@ static int map_benchmark_thread(void *data)
* 66 means evertything goes well! 66 is lucky.
*/
if (map->dir != DMA_FROM_DEVICE)
- memset(buf, 0x66, PAGE_SIZE);
+ memset(buf, 0x66, size);
map_stime = ktime_get();
- dma_addr = dma_map_single(map->dev, buf, PAGE_SIZE, map->dir);
+ dma_addr = dma_map_single(map->dev, buf, size, map->dir);
if (unlikely(dma_mapping_error(map->dev, dma_addr))) {
pr_err("dma_map_single failed on %s\n",
dev_name(map->dev));
@@ -87,8 +69,11 @@ static int map_benchmark_thread(void *data)
map_etime = ktime_get();
map_delta = ktime_sub(map_etime, map_stime);
+ /* Pretend DMA is transmitting */
+ ndelay(map->bparam.dma_trans_ns);
+
unmap_stime = ktime_get();
- dma_unmap_single(map->dev, dma_addr, PAGE_SIZE, map->dir);
+ dma_unmap_single(map->dev, dma_addr, size, map->dir);
unmap_etime = ktime_get();
unmap_delta = ktime_sub(unmap_etime, unmap_stime);
@@ -104,10 +89,26 @@ static int map_benchmark_thread(void *data)
atomic64_add(map_sq, &map->sum_sq_map);
atomic64_add(unmap_sq, &map->sum_sq_unmap);
atomic64_inc(&map->loops);
+
+ /*
+ * We may test for a long time so periodically check whether
+ * we need to schedule to avoid starving the others. Otherwise
+ * we may hangup the kernel in a non-preemptible kernel when
+ * the test kthreads number >= CPU number, the test kthreads
+ * will run endless on every CPU since the thread resposible
+ * for notifying the kthread stop (in do_map_benchmark())
+ * could not be scheduled.
+ *
+ * Note this may degrade the test concurrency since the test
+ * threads may need to share the CPU time with other load
+ * in the system. So it's recommended to run this benchmark
+ * on an idle system.
+ */
+ cond_resched();
}
out:
- free_page((unsigned long)buf);
+ free_pages_exact(buf, size);
return ret;
}
@@ -116,7 +117,6 @@ static int do_map_benchmark(struct map_benchmark_data *map)
struct task_struct **tsk;
int threads = map->bparam.threads;
int node = map->bparam.node;
- const cpumask_t *cpu_mask = cpumask_of_node(node);
u64 loops;
int ret = 0;
int i;
@@ -133,11 +133,13 @@ static int do_map_benchmark(struct map_benchmark_data *map)
if (IS_ERR(tsk[i])) {
pr_err("create dma_map thread failed\n");
ret = PTR_ERR(tsk[i]);
+ while (--i >= 0)
+ kthread_stop(tsk[i]);
goto out;
}
if (node != NUMA_NO_NODE)
- kthread_bind_mask(tsk[i], cpu_mask);
+ kthread_bind_mask(tsk[i], cpumask_of_node(node));
}
/* clear the old value in the previous benchmark */
@@ -154,13 +156,17 @@ static int do_map_benchmark(struct map_benchmark_data *map)
msleep_interruptible(map->bparam.seconds * 1000);
- /* wait for the completion of benchmark threads */
+ /* wait for the completion of all started benchmark threads */
for (i = 0; i < threads; i++) {
- ret = kthread_stop(tsk[i]);
- if (ret)
- goto out;
+ int kthread_ret = kthread_stop_put(tsk[i]);
+
+ if (kthread_ret)
+ ret = kthread_ret;
}
+ if (ret)
+ goto out;
+
loops = atomic64_read(&map->loops);
if (likely(loops > 0)) {
u64 map_variance, unmap_variance;
@@ -185,8 +191,6 @@ static int do_map_benchmark(struct map_benchmark_data *map)
}
out:
- for (i = 0; i < threads; i++)
- put_task_struct(tsk[i]);
put_device(map->dev);
kfree(tsk);
return ret;
@@ -198,7 +202,6 @@ static long map_benchmark_ioctl(struct file *file, unsigned int cmd,
struct map_benchmark_data *map = file->private_data;
void __user *argp = (void __user *)arg;
u64 old_dma_mask;
-
int ret;
if (copy_from_user(&map->bparam, argp, sizeof(map->bparam)))
@@ -218,12 +221,23 @@ static long map_benchmark_ioctl(struct file *file, unsigned int cmd,
return -EINVAL;
}
+ if (map->bparam.dma_trans_ns > DMA_MAP_MAX_TRANS_DELAY) {
+ pr_err("invalid transmission delay\n");
+ return -EINVAL;
+ }
+
if (map->bparam.node != NUMA_NO_NODE &&
- !node_possible(map->bparam.node)) {
+ (map->bparam.node < 0 || map->bparam.node >= MAX_NUMNODES ||
+ !node_possible(map->bparam.node))) {
pr_err("invalid numa node\n");
return -EINVAL;
}
+ if (map->bparam.granule < 1 || map->bparam.granule > 1024) {
+ pr_err("invalid granule size\n");
+ return -EINVAL;
+ }
+
switch (map->bparam.dma_dir) {
case DMA_MAP_BIDIRECTIONAL:
map->dir = DMA_BIDIRECTIONAL;
@@ -258,6 +272,9 @@ static long map_benchmark_ioctl(struct file *file, unsigned int cmd,
* dma_mask changed by benchmark
*/
dma_set_mask(map->dev, old_dma_mask);
+
+ if (ret)
+ return ret;
break;
default:
return -EINVAL;
@@ -362,4 +379,3 @@ module_exit(map_benchmark_cleanup);
MODULE_AUTHOR("Barry Song <song.bao.hua@hisilicon.com>");
MODULE_DESCRIPTION("dma_map benchmark driver");
-MODULE_LICENSE("GPL");
generated by cgit (git 2.34.1) at 2025-07-24 08:11:21 +0000