gh0s7/mckernel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

xpmem: fix mapping of attachment and segment

Browse Source 
* Mapping attached part of segment is done at attach time instead of
  make time to work with runtimes (e.g. OpenMPI) xpmem_make-ing the
  entire user-space
* Mapping attached part of segment at attach time can be turned off by
  specifying xpmem_remote_on_demand in kernel argument
* Mapping attachment chooses appropriate page-sizes, i.e., largest
  allowed by memory range and segment page boundary

Fixes: a8696d8 "xpmem: Support large page attachment"
Change-Id: I44663865204036520e5f62fe22b9134ee4629f9b
This commit is contained in:
Masamichi Takagi
2020-05-12 13:17:46 +09:00
parent d370e9241f
commit a5fcc91656
43 changed files with 2876 additions and 285 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
test/issues/1259/large_page.patch
View File

@ -1,8 +1,8 @@
diff --git arch/arm64/kernel/memory.c arch/arm64/kernel/memory.c
index a84bc21..7368ada 100644
index a84bc21..f329fcf 100644
--- arch/arm64/kernel/memory.c
+++ arch/arm64/kernel/memory.c
@@ -2701,6 +2701,13 @@ int set_range_l1(void *args0, pte_t *ptep, uintptr_t base, uintptr_t start,
@@ -2701,6 +2701,16 @@ int set_range_l1(void *args0, pte_t *ptep, uintptr_t base, uintptr_t start,
ptl1_set(ptep, pte);
error = 0;
@ -11,12 +11,15 @@ index a84bc21..7368ada 100644
+ __page_offset(base, PTL1_CONT_SIZE) == 0) {
+ kprintf("%s: large_page_allocation, addr: %016lx, size: 0x%lx , phys: %lx\n",
+ __func__, base, PTL1_CONT_SIZE, phys);
+ if (args->range->private_data)
+ kprintf("%s: xpmem_page_attach, addr: %016lx, size: 0x%lx\n",
+ __func__, base, PTL1_CONT_SIZE);
+ }
+
// call memory_stat_rss_add() here because pgshift is resolved here
if (!(args->attr[0] & PTE_CONT)) {
if (rusage_memory_stat_add(args->range, phys,
@@ -2810,6 +2817,17 @@ retry:
@@ -2810,6 +2820,23 @@ retry:
level);
error = 0;
@ -25,45 +28,57 @@ index a84bc21..7368ada 100644
+ if (__page_offset(base, tbl.cont_pgsize) == 0) {
+ kprintf("%s: large_page_allocation, addr: %016lx, size: 0x%lx , phys: %lx\n",
+ __func__, base, tbl.cont_pgsize, phys);
+ if (args->range->private_data)
+ kprintf("%s: xpmem_page_attach, addr: %016lx, size: 0x%lx\n",
+ __func__, base, tbl.cont_pgsize);
+ }
+ } else {
+ kprintf("%s: large_page_allocation, addr: %016lx, size: 0x%lx , phys: %lx\n",
+ __func__, base, tbl.pgsize, phys);
+ if (args->range->private_data)
+ kprintf("%s: xpmem_page_attach, addr: %016lx, size: 0x%lx\n",
+ __func__, base, tbl.pgsize);
+ }
+
dkprintf("set_range_middle(%lx,%lx,%lx,%d):"
"large page. %d %lx\n",
base, start, end, level, error, *ptep);
diff --git arch/x86_64/kernel/memory.c arch/x86_64/kernel/memory.c
index df545e1..633e390 100644
index 6b34036..4ca3a1a 100644
--- arch/x86_64/kernel/memory.c
+++ arch/x86_64/kernel/memory.c
@@ -1931,6 +1931,10 @@ retry:
@@ -1932,6 +1932,13 @@ retry:
dkprintf("set_range_l2(%lx,%lx,%lx):"
"2MiB page. %d %lx\n",
base, start, end, error, *ptep);
+
+ kprintf("%s: large_page_allocation, addr: %016lx, size: 0x%lx\n",
+ __func__, base, PTL2_SIZE);
+ if (args->range->private_data)
+ kprintf("%s: xpmem_page_attach, addr: %016lx, size: 0x%lx\n",
+ __func__, base, PTL2_SIZE);
+
// Call memory_stat_rss_add() here because pgshift is resolved here
if (rusage_memory_stat_add(args->range, phys, PTL2_SIZE, PTL2_SIZE)) {
dkprintf("%lx+,%s: calling memory_stat_rss_add(),base=%lx,phys=%lx,size=%ld,pgsize=%ld\n", phys, __FUNCTION__, base, phys, PTL2_SIZE, PTL2_SIZE);
@@ -2020,6 +2024,9 @@ retry:
@@ -2021,6 +2028,12 @@ retry:
"1GiB page. %d %lx\n",
base, start, end, error, *ptep);
+ kprintf("%s: large_page_allocation, addr: %016lx, size: 0x%lx\n",
+ __func__, base, PTL3_SIZE);
+ if (args->range->private_data)
+ kprintf("%s: xpmem_page_attach, addr: %016lx, size: 0x%lx\n",
+ __func__, base, PTL3_SIZE);
+
// Call memory_stat_rss_add() here because pgshift is resolved here
if (rusage_memory_stat_add(args->range, phys, PTL3_SIZE, PTL3_SIZE)) {
dkprintf("%lx+,%s: calling memory_stat_rss_add(),base=%lx,phys=%lx,size=%ld,pgsize=%ld\n", phys, __FUNCTION__, base, phys, PTL3_SIZE, PTL3_SIZE);
diff --git kernel/process.c kernel/process.c
index 809f5e0..cba9e5a 100644
index 1624726..d867b4d 100644
--- kernel/process.c
+++ kernel/process.c
@@ -2059,6 +2059,12 @@ retry:
@@ -2138,6 +2138,12 @@ retry:
}
dkprintf("%s: attr=%x\n", __FUNCTION__, attr);
@ -77,10 +92,10 @@ index 809f5e0..cba9e5a 100644
pgsize, phys, attr);
if (error) {
diff --git kernel/xpmem.c kernel/xpmem.c
index e1d0231..c9da711 100644
index 9fedf26..a0a7990 100644
--- kernel/xpmem.c
+++ kernel/xpmem.c
@@ -514,6 +514,7 @@ static int xpmem_make(
@@ -490,6 +490,7 @@ static int xpmem_make(
*segid_p = segid;
XPMEM_DEBUG("return: ret=%d, segid=0x%lx", 0, *segid_p);
@ -88,12 +103,15 @@ index e1d0231..c9da711 100644
return 0;
}
@@ -1994,6 +1995,8 @@ int xpmem_update_process_page_table(
flush_tlb_single(vaddr);
att->flags |= XPMEM_FLAG_VALIDPTEs;
+ kprintf("%s: xpmem_page_attach, addr: %016lx, size: 0x%lx\n",
+ __func__, vaddr, seg_pgsize);
seg_vaddr += seg_pgsize;
vaddr += seg_pgsize;
@@ -1886,6 +1887,11 @@ static int _xpmem_fault_process_memory_range(
goto out;
}
+ if (att_pgsize == PAGE_SIZE) {
+ kprintf("xpmem_page_attach, addr: %016lx, size: 0x%lx\n",
+ vaddr, att_pgsize);
+ }
+
XPMEM_DEBUG("att_pgaddr: %lx, att_pgsize: %lx, "
"seg_vaddr: %lx, seg_pgsize: %lx, seg_phys: %lx\n",
att_pgaddr, att_pgsize, seg_vaddr,