2024-05-16 03:58:41

by Dev Jain

[permalink] [raw]
Subject: [PATCH] selftests/mm: va_high_addr_switch: Do not skip test and give warning message post FEAT_LPA2

Post FEAT_LPA2, Aarch64 extends the 4KB and 16KB translation granule to
large virtual addresses. Currently, the test is being skipped for said
granule sizes, because the page sizes have been statically defined; to
work around that would mean breaking the nice array of structs used for
adding testcases. Instead, don't skip the test, and encourage the user
to manually change the macros.

Signed-off-by: Dev Jain <[email protected]>
---
.../testing/selftests/mm/va_high_addr_switch.c | 17 ++++++++++++++++-
1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/tools/testing/selftests/mm/va_high_addr_switch.c b/tools/testing/selftests/mm/va_high_addr_switch.c
index cfbc501290d3..ba862f51d395 100644
--- a/tools/testing/selftests/mm/va_high_addr_switch.c
+++ b/tools/testing/selftests/mm/va_high_addr_switch.c
@@ -292,12 +292,24 @@ static int supported_arch(void)
#elif defined(__x86_64__)
return 1;
#elif defined(__aarch64__)
- return getpagesize() == PAGE_SIZE;
+ return 1;
#else
return 0;
#endif
}

+#if defined(__aarch64__)
+void failure_message(void)
+{
+ printf("TEST MAY FAIL: Are you running on a pagesize other than 64K?\n");
+ printf("If yes, please change macros manually. Ensure to change the\n");
+ printf("address macros too if running defconfig on 16K pagesize,\n");
+ printf("since userspace VA = 47 bits post FEAT_LPA2.\n");
+}
+#else
+void failure_message(void) {}
+#endif
+
int main(int argc, char **argv)
{
int ret;
@@ -308,5 +320,8 @@ int main(int argc, char **argv)
ret = run_test(testcases, ARRAY_SIZE(testcases));
if (argc == 2 && !strcmp(argv[1], "--run-hugetlb"))
ret = run_test(hugetlb_testcases, ARRAY_SIZE(hugetlb_testcases));
+
+ if (ret)
+ failure_message();
return ret;
}
--
2.39.2



2024-05-19 23:48:33

by Andrew Morton

[permalink] [raw]
Subject: Re: [PATCH] selftests/mm: va_high_addr_switch: Do not skip test and give warning message post FEAT_LPA2

On Thu, 16 May 2024 09:26:33 +0530 Dev Jain <[email protected]> wrote:

> Post FEAT_LPA2, Aarch64 extends the 4KB and 16KB translation granule to
> large virtual addresses. Currently, the test is being skipped for said
> granule sizes, because the page sizes have been statically defined; to
> work around that would mean breaking the nice array of structs used for
> adding testcases.

Which array is that? testcases[]? If so, we could keep if fairly nice
by doing the array population at runtime. Something like:

static struct testcase *testcases;

static void init_thing()
{
struct testcase t[] = {
...
};

testcases = malloc(sizeof(t));
memcpy(testcases, t, sizeof(t));
}

>
> +#if defined(__aarch64__)
> +void failure_message(void)
> +{
> + printf("TEST MAY FAIL: Are you running on a pagesize other than 64K?\n");
> + printf("If yes, please change macros manually. Ensure to change the\n");
> + printf("address macros too if running defconfig on 16K pagesize,\n");
> + printf("since userspace VA = 47 bits post FEAT_LPA2.\n");
> +}
> +#else
> +void failure_message(void) {}
> +#endif
> +
> int main(int argc, char **argv)
> {
> int ret;
> @@ -308,5 +320,8 @@ int main(int argc, char **argv)
> ret = run_test(testcases, ARRAY_SIZE(testcases));
> if (argc == 2 && !strcmp(argv[1], "--run-hugetlb"))
> ret = run_test(hugetlb_testcases, ARRAY_SIZE(hugetlb_testcases));
> +
> + if (ret)
> + failure_message();
> return ret;
> }

This seems rather lame :(. It would be nice to fix this for once and
for all.


2024-05-21 12:52:35

by Dev Jain

[permalink] [raw]
Subject: Re: [PATCH] selftests/mm: va_high_addr_switch: Do not skip test and give warning message post FEAT_LPA2


On 5/20/24 05:18, Andrew Morton wrote:
> On Thu, 16 May 2024 09:26:33 +0530 Dev Jain <[email protected]> wrote:
>
>> Post FEAT_LPA2, Aarch64 extends the 4KB and 16KB translation granule to
>> large virtual addresses. Currently, the test is being skipped for said
>> granule sizes, because the page sizes have been statically defined; to
>> work around that would mean breaking the nice array of structs used for
>> adding testcases.
> Which array is that? testcases[]? If so, we could keep if fairly nice
> by doing the array population at runtime. Something like:
>
> static struct testcase *testcases;
>
> static void init_thing()
> {
> struct testcase t[] = {
> ...
> };
>
> testcases = malloc(sizeof(t));
> memcpy(testcases, t, sizeof(t));
> }


Great idea! This should work. I shall implement it.