| commit | 410b3bf09e76fd2b6d68b424a26d407a0bc4bc11 | [log] [tgz] |
|---|---|---|
| author | Alexandru Elisei <alexandru.elisei@arm.com> | Fri Jan 31 16:37:28 2020 +0000 |
| committer | Andrew Jones <drjones@redhat.com> | Fri Apr 03 09:40:33 2020 +0200 |
| tree | 81e10f2a72c2d2c4aaba027b3d5ceecec838b82c | |
| parent | 24dd2adab559390ebbd7043cbe50bca5f631b29b [diff] |
arm/arm64: Perform dcache clean + invalidate after turning MMU off
When the MMU is off, data accesses are to Device nGnRnE memory on arm64 [1]
or to Strongly-Ordered memory on arm [2]. This means that the accesses are
non-cacheable.
Perform a dcache clean to PoC so we can read the newer values from the
cache after we turn the MMU off, instead of the stale values from memory.
Perform an invalidation so we can access the data written to memory after
we turn the MMU back on. This prevents reading back the stale values we
cleaned from the cache when we turned the MMU off.
Data caches are PIPT and the VAs are translated using the current
translation tables, or an identity mapping (what Arm calls a "flat
mapping") when the MMU is off [1, 2]. Do the clean + invalidate when the
MMU is off so we don't depend on the current translation tables and we can
make sure that the operation applies to the entire physical memory.
The patch was tested by hacking arm/selftest.c:
+#include <alloc_page.h>
+#include <asm/mmu.h>
int main(int argc, char **argv)
{
+ int *x = alloc_page();
+
report_prefix_push("selftest");
+ *x = 0x42;
+ mmu_disable();
+ report(*x == 0x42, "read back value written with MMU on");
+ *x = 0x50;
+ mmu_enable(current_thread_info()->pgtable);
+ report(*x == 0x50, "read back value written with MMU off");
+
if (argc < 2)
report_abort("no test specified");
Without the fix, the first report fails, and the test usually hangs before
the second report. This is because mmu_enable pushes the LR register on the
stack when the MMU is off, which means that the value will be written to
memory. However, after asm_mmu_enable, the MMU is enabled, and we read it
back from the dcache, thus getting garbage.
With the fix, the two reports pass.
[1] ARM DDI 0487E.a, section D5.2.9
[2] ARM DDI 0406C.d, section B3.2.1
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
Signed-off-by: Andrew Jones <drjones@redhat.com>
See http://www.linux-kvm.org/page/KVM-unit-tests for a high-level description of this project, as well as running tests and adding tests HOWTOs.
This directory contains sources for a KVM test suite.
To create the test images do:
./configure make
in this directory. Test images are created in ./ARCH/*.flat
The tests can be built as standalone. To create and use standalone tests do:
./configure make standalone (send tests/some-test somewhere) (go to somewhere) ./some-test
make install will install all tests in PREFIX/share/kvm-unit-tests/tests, each as a standalone test.
Then use the runner script to detect the correct invocation and invoke the test:
./x86-run ./x86/msr.flat
or:
./run_tests.sh
to run them all.
By default the runner script searches for a suitable QEMU binary in the system. To select a specific QEMU binary though, specify the QEMU=path/to/binary environment variable:
QEMU=/tmp/qemu/x86_64-softmmu/qemu-system-x86_64 ./x86-run ./x86/msr.flat
To select an accelerator, for example “kvm” or “tcg”, specify the ACCEL=name environment variable:
ACCEL=kvm ./x86-run ./x86/msr.flat
The test case may need specific runtime configurations, for example, extra QEMU parameters and time to execute limited, the runner script reads those information from a configuration file found at ./ARCH/unittests.cfg.
The configuration file also contain the groups (if any) each test belong to. So that a given group can be executed by specifying its name in the runner's -g option.
Unit tests use QEMU's ‘-append args...’ parameter for command line inputs, i.e. all args will be available as argv strings in main(). Additionally a file of the form
KEY=VAL KEY2=VAL ...
may be passed with ‘-initrd file’ to become the unit test's environ, which can then be accessed in the usual ways, e.g. VAL = getenv(“KEY”). Any key=val strings can be passed, but some have reserved meanings in the framework. The list of reserved environment variables is below
QEMU_ACCEL either kvm or tcg QEMU_VERSION_STRING string of the form `qemu -h | head -1` KERNEL_VERSION_STRING string of the form `uname -r`
Additionally these self-explanatory variables are reserved
QEMU_MAJOR, QEMU_MINOR, QEMU_MICRO, KERNEL_VERSION, KERNEL_PATCHLEVEL, KERNEL_SUBLEVEL, KERNEL_EXTRAVERSION
Some tests are not safe to run by default, as they may crash the host. kvm-unit-tests provides two ways to handle tests like those.
Adding ‘nodefault’ to the groups field for the unit test in the unittests.cfg file. When a unit test is in the nodefault group it is only run when invoked
a) independently, ARCH-run ARCH/test
b) by specifying any other non-nodefault group it is in, groups = nodefault,mygroup : ./run_tests.sh -g mygroup
c) by specifying all tests should be run, ./run_tests.sh -a
Making the test conditional on errata in the code,
if (ERRATA(abcdef012345)) {
do_unsafe_test();
}
With the errata condition the unsafe unit test is only run when
a) the ERRATA_abcdef012345 environment variable is provided and ‘y’
b) the ERRATA_FORCE environment variable is provided and ‘y’
c) by specifying all tests should be run, ./run_tests.sh -a (The -a switch ensures the ERRATA_FORCE is provided and set to ‘y’.)
The ./errata.txt file provides a mapping of the commits needed by errata conditionals to their respective minimum kernel versions. By default, when the user does not provide an environ, then an environ generated from the ./errata.txt file and the host's kernel version is provided to all unit tests.
.: configure script, top-level Makefile, and run_tests.sh ./scripts: helper scripts for building and running tests ./lib: general architecture neutral services for the tests ./lib/<ARCH>: architecture dependent services for the tests ./<ARCH>: the sources of the tests and the created objects/images
See ./ARCH/README for architecture specific documentation.
Currently there is a mix of indentation styles so any changes to existing files should be consistent with the existing style. For new files:
Exceptions:
Patches are welcome at the KVM mailing list kvm@vger.kernel.org.
Please prefix messages with: [kvm-unit-tests PATCH]
You can add the following to .git/config to do this automatically for you:
[format]
subjectprefix = kvm-unit-tests PATCH
Additionally it‘s helpful to have a common order of file types in patches. Our chosen order attempts to place the more declarative files before the code files. We also start with common code and finish with unit test code. git-diff’s orderFile feature allows us to specify the order in a file. The orderFile we use is scripts/git.difforder; adding the config with git config diff.orderFile scripts/git.difforder enables it.