Skip to contentSkip to conversation

s390x: edat: test 2G large page spanning end of memory

PATCH In review Claudio Imbrenda 2 participants 5 messages Oct 1, 2024 – Oct 2, 2024

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 s390x: edat: test 2G large page spanning end of memory

Create a region 3 table with fc=1 (i.e. a 2G large page) mapping across the end of memory.

Check that the part of the large page before the end of memory is accessible, and the part that is after the end of memory is not.

Also fix a typo in the existing edat2 test.

---
 s390x/edat.c | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)
s390x/edat.c +17 -1
index 16138397..1f582efc 100644
@@ -196,6 +196,8 @@ static void test_edat1(void)
 
 static void test_edat2(void)
 {
+	uint64_t mem_end, i;
+
 	report_prefix_push("edat2");
 	p[0] = 42;
 
@@ -206,7 +208,21 @@ static void test_edat2(void)
 	/* Prefixing should not work with huge pages, just like large pages */
 	report(!memcmp(0, VIRT(prefix_buf), LC_SIZE) &&
 		!memcmp(prefix_buf, VIRT(0), LC_SIZE),
-		"pmd, large, prefixing");
+		"pud, large, prefixing");
+
+	mem_end = get_ram_size();
+	if (mem_end >= BIT_ULL(REGION3_SHIFT)) {
+		report_skip("pud spanning end of memory");
+	} else {
+		for (i = 0; i < mem_end; i += PAGE_SIZE)
+			READ_ONCE(*(uint64_t *)VIRT(i));
+		for (i = mem_end; i < BIT_ULL(REGION3_SHIFT); i += PAGE_SIZE) {
+			expect_pgm_int();
+			READ_ONCE(*(uint64_t *)VIRT(i));
+			assert(clear_pgm_int() == PGM_INT_CODE_ADDRESSING);
+		}
+		report_pass("pud spanning end of memory");
+	}
 
 	report_prefix_pop();
 }
Signature · 1 line
2.46.2
#1 lore raw

Quoting Claudio Imbrenda (2024-10-01 13:36:40)
[...]

Quoted 8 lines
s390x/edat.c +0 -0
index 16138397..1f582efc 100644
@@ -196,6 +196,8 @@ static void test_edat1(void)
 
 static void test_edat2(void)
 {

[...]

s390x/edat.c static void test_edat2(void) · line 215

Quoted 10 lines
s390x/edat.c +5 -1
@@ -206,7 +208,21 @@ static void test_edat2(void)
        /* Prefixing should not work with huge pages, just like large pages */
        report(!memcmp(0, VIRT(prefix_buf), LC_SIZE) &&
                !memcmp(prefix_buf, VIRT(0), LC_SIZE),
-               "pmd, large, prefixing");
+               "pud, large, prefixing");
+
+       mem_end = get_ram_size();
+       if (mem_end >= BIT_ULL(REGION3_SHIFT)) {
+               report_skip("pud spanning end of memory");

Does it make sense to explicitly add a mem parameter in unittests.cfg so this will never be the case?

#2 lore raw
Quoted 11 lines · On Wed, 02 Oct 2024 16:25:39 +0200 Nico Boehr <[email protected]> wrote:

Quoting Claudio Imbrenda (2024-10-01 13:36:40)
[...]

Quoted 8 lines
s390x/edat.c +0 -0
index 16138397..1f582efc 100644
@@ -196,6 +196,8 @@ static void test_edat1(void)
 
 static void test_edat2(void)
 {  

[...]

Quoted 13 lines

s390x/edat.c static void test_edat2(void) · line 215

Quoted 10 lines
s390x/edat.c +5 -1
@@ -206,7 +208,21 @@ static void test_edat2(void)
        /* Prefixing should not work with huge pages, just like large pages */
        report(!memcmp(0, VIRT(prefix_buf), LC_SIZE) &&
                !memcmp(prefix_buf, VIRT(0), LC_SIZE),
-               "pmd, large, prefixing");
+               "pud, large, prefixing");
+
+       mem_end = get_ram_size();
+       if (mem_end >= BIT_ULL(REGION3_SHIFT)) {
+               report_skip("pud spanning end of memory");  

Does it make sense to explicitly add a mem parameter in unittests.cfg so this will never be the case?

hmmm, I did not consider this case; I kinda assumed we would never increase the default guest size

I do not have any strong opinions

#3 lore raw

Quoting Claudio Imbrenda (2024-10-02 16:31:33)

Quoted 28 lines
Quoted 11 lines · On Wed, 02 Oct 2024 16:25:39 +0200 Nico Boehr <[email protected]> wrote:

Quoting Claudio Imbrenda (2024-10-01 13:36:40)
[...]

Quoted 8 lines
s390x/edat.c +0 -0
index 16138397..1f582efc 100644
@@ -196,6 +196,8 @@ static void test_edat1(void)
 
 static void test_edat2(void)
 {  

[...]

Quoted 13 lines

s390x/edat.c static void test_edat2(void) · line 215

Quoted 10 lines
s390x/edat.c +5 -1
@@ -206,7 +208,21 @@ static void test_edat2(void)
        /* Prefixing should not work with huge pages, just like large pages */
        report(!memcmp(0, VIRT(prefix_buf), LC_SIZE) &&
                !memcmp(prefix_buf, VIRT(0), LC_SIZE),
-               "pmd, large, prefixing");
+               "pud, large, prefixing");
+
+       mem_end = get_ram_size();
+       if (mem_end >= BIT_ULL(REGION3_SHIFT)) {
+               report_skip("pud spanning end of memory");  

Does it make sense to explicitly add a mem parameter in unittests.cfg so this will never be the case?

Quoted 4 lines

hmmm, I did not consider this case; I kinda assumed we would never increase the default guest size

I do not have any strong opinions

As long as the default mem size is OK, I think it's fine to leave as-is.

Reviewed-by on v1 1/1
#4 lore raw
In reply to Claudio Imbrenda

Quoting Claudio Imbrenda (2024-10-01 13:36:40)
[...]

Quoted 4 lines
s390x/edat.c +0 -0
index 16138397..1f582efc 100644

[...]

s390x/edat.c static void test_edat2(void) · line 214

Quoted 9 lines
s390x/edat.c +4 -1
@@ -206,7 +208,21 @@ static void test_edat2(void)
        /* Prefixing should not work with huge pages, just like large pages */
        report(!memcmp(0, VIRT(prefix_buf), LC_SIZE) &&
                !memcmp(prefix_buf, VIRT(0), LC_SIZE),
-               "pmd, large, prefixing");
+               "pud, large, prefixing");
+
+       mem_end = get_ram_size();
+       if (mem_end >= BIT_ULL(REGION3_SHIFT)) {

Do you mind introducting REGION3_SIZE like the kernel has?

s390x/edat.c static void test_edat2(void) · line 221

Quoted 7 lines
s390x/edat.c +7 -0
+               report_skip("pud spanning end of memory");
+       } else {
+               for (i = 0; i < mem_end; i += PAGE_SIZE)
+                       READ_ONCE(*(uint64_t *)VIRT(i));
+               for (i = mem_end; i < BIT_ULL(REGION3_SHIFT); i += PAGE_SIZE) {
+                       expect_pgm_int();
+                       READ_ONCE(*(uint64_t *)VIRT(i));

Would a write behave any different here?

With or without the suggestions above: