AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Set guard page as disabled as default

Browse Source 
Due to max_mmap_count issue, set default of guard page as disabled.


Change-Id: Ic9dfe69b621733e9fac86831b008a122994a67e7
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
This commit is contained in:
Amber Lin
2017-08-23 17:17:09 -04:00
parent 395ecaa985
commit 65d680c035
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/fmm.c
+3 -3
View File
@@ -1423,7 +1423,7 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
uint32_t num_of_nodes;
HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;
char *disableCache, *pagedUserptr, *checkUserptr, *guardPagesStr;
unsigned int guardPages = 1;
unsigned int guardPages = 0;
struct pci_access *pacc;
/* If HSA_DISABLE_CACHE is set to a non-0 value, disable caching */
@@ -1443,10 +1443,10 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
checkUserptr = getenv("HSA_CHECK_USERPTR");
svm.check_userptr = (checkUserptr && strcmp(checkUserptr, "0"));
/* Specify number of guard pages for SVM apertures, default is 1 */
/* Specify number of guard pages for SVM apertures, default is 0 */
guardPagesStr = getenv("HSA_SVM_GUARD_PAGES");
if (!guardPagesStr || sscanf(guardPagesStr, "%u", &guardPages) != 1)
guardPages = 1;
guardPages = 0;
/* Trade off - NumNodes includes GPU nodes + CPU Node. So in
* systems with CPU node, slightly more memory is allocated than