[dpdk-dev] mem: fix how to calculate space left in a hugetlbfs
Commit Message
This patch enables calculating space left in a hugetlbfs.
There are three sources to get the information: 1. from
sysfs; 2. from option size specified when mount; 3. use
statfs. We should use the minimum one of these three sizes.
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
---
lib/librte_eal/linuxapp/eal/eal_hugepage_info.c | 85 ++++++++++++++++++++++++-
1 file changed, 84 insertions(+), 1 deletion(-)
Comments
Hi Jianfeng,
> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jianfeng Tan
> Sent: Thursday, November 12, 2015 12:18 AM
> To: dev@dpdk.org
> Subject: [dpdk-dev] [PATCH] mem: fix how to calculate space left in a
> hugetlbfs
>
> This patch enables calculating space left in a hugetlbfs.
> There are three sources to get the information: 1. from
> sysfs; 2. from option size specified when mount; 3. use
> statfs. We should use the minimum one of these three sizes.
>
> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
You should reword the title of the patch, as this does not look like a fix.
On Thu, 12 Nov 2015 08:17:57 +0800
Jianfeng Tan <jianfeng.tan@intel.com> wrote:
> This patch enables calculating space left in a hugetlbfs.
> There are three sources to get the information: 1. from
> sysfs; 2. from option size specified when mount; 3. use
> statfs. We should use the minimum one of these three sizes.
>
> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Thanks, the hugetlbfs usage up until now has been rather brute force.
I wonder if long term it might be better to defer all this stuff
to another library like libhugetlbfs.
https://github.com/libhugetlbfs/libhugetlbfs
Especially wen dealing with other architectures it might provide
some nice abstraction.
2015-11-12 09:38, Stephen Hemminger:
> On Thu, 12 Nov 2015 08:17:57 +0800
> Jianfeng Tan <jianfeng.tan@intel.com> wrote:
>
> > This patch enables calculating space left in a hugetlbfs.
> > There are three sources to get the information: 1. from
> > sysfs; 2. from option size specified when mount; 3. use
> > statfs. We should use the minimum one of these three sizes.
> >
> > Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
>
> Thanks, the hugetlbfs usage up until now has been rather brute force.
> I wonder if long term it might be better to defer all this stuff
> to another library like libhugetlbfs.
> https://github.com/libhugetlbfs/libhugetlbfs
>
> Especially wen dealing with other architectures it might provide
> some nice abstraction.
Maybe, maybe not :)
Sergio arleady looked at it:
http://dpdk.org/ml/archives/dev/2015-July/022080.html
@@ -44,6 +44,8 @@
#include <unistd.h>
#include <errno.h>
#include <sys/queue.h>
+#include <sys/vfs.h>
+#include <mntent.h>
#include <rte_memory.h>
#include <rte_memzone.h>
@@ -189,6 +191,70 @@ get_hugepage_dir(uint64_t hugepage_sz)
return retval;
}
+/* Caller to make sure this mnt_dir exist
+ */
+static uint64_t
+get_hugetlbfs_mount_size(const char *mnt_dir)
+{
+ char *start, *end, *opt_size;
+ struct mntent *ent;
+ uint64_t size;
+ FILE *f;
+ int len;
+
+ f = setmntent("/proc/mounts", "r");
+ if (f == NULL) {
+ RTE_LOG(ERR, EAL, "setmntent() error: %s\n",
+ strerror(errno));
+ return 0;
+ }
+ while (NULL != (ent = getmntent(f))) {
+ if (!strcmp(ent->mnt_dir, mnt_dir))
+ break;
+ }
+
+ start = hasmntopt(ent, "size");
+ if (start == NULL) {
+ RTE_LOG(DEBUG, EAL, "option size not specified for %s\n",
+ mnt_dir);
+ size = 0;
+ goto end;
+ }
+ start += strlen("size=");
+ end = strstr(start, ",");
+ if (end != NULL)
+ len = end - start;
+ else
+ len = strlen(start);
+ opt_size = strndup(start, len);
+ size = rte_str_to_size(opt_size);
+ free(opt_size);
+
+end:
+ endmntent(f);
+ return size;
+}
+
+/* Caller to make sure this mount has option size
+ * so that statfs is not zero.
+ */
+static uint64_t
+get_hugetlbfs_free_size(const char *mnt_dir)
+{
+ int r;
+ struct statfs stats;
+
+ r = statfs(mnt_dir, &stats);
+ if (r != 0) {
+ RTE_LOG(ERR, EAL, "statfs() error: %s\n",
+ strerror(errno));
+ return 0;
+ }
+
+ return stats.f_bfree * stats.f_bsize;
+}
+
+
/*
* Clear the hugepage directory of whatever hugepage files
* there are. Checks if the file is locked (i.e.
@@ -329,9 +395,26 @@ eal_hugepage_info_init(void)
if (clear_hugedir(hpi->hugedir) == -1)
break;
+ /* there are three souces of how much space left in a
+ * hugetlbfs dir.
+ */
+ uint64_t sz_left, sz_sysfs, sz_option, sz_statfs;
+
+ sz_sysfs = get_num_hugepages(dirent->d_name) *
+ hpi->hugepage_sz;
+ sz_left = sz_sysfs;
+ sz_option = get_hugetlbfs_mount_size(hpi->hugedir);
+ if (sz_option) {
+ sz_statfs = get_hugetlbfs_free_size(hpi->hugedir);
+ sz_left = RTE_MIN(sz_sysfs, sz_statfs);
+ RTE_LOG(INFO, "sz_sysfs: %"PRIu64", sz_option: "
+ "%"PRIu64", sz_statfs: %"PRIu64"\n",
+ sz_sysfs, sz_option, sz_statfs);
+ }
+
/* for now, put all pages into socket 0,
* later they will be sorted */
- hpi->num_pages[0] = get_num_hugepages(dirent->d_name);
+ hpi->num_pages[0] = sz_left / hpi->hugepage_sz;
#ifndef RTE_ARCH_64
/* for 32-bit systems, limit number of hugepages to