From: Andrew Jones <andrew.jones@linux.dev>
To: "Bingyu.Xian" <shanbeeyoo@gmail.com>
Cc: anup@brainfault.org, atish.patra@linux.dev, pjw@kernel.org,
palmer@dabbelt.com, aou@eecs.berkeley.edu, alex@ghiti.fr,
kvm@vger.kernel.org, kvm-riscv@lists.infradead.org,
linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
zhouquan@iscas.ac.cn, linux-kselftest@vger.kernel.org,
shuah@kernel.org, pbonzini@redhat.com, seanjc@google.com
Subject: Re: [PATCH] KVM: selftests: riscv: Test the SBI STA shared memory lifecycle
Date: Thu, 19 Nov 2026 14:49:25 +0100 [thread overview]
Message-ID: <av77j8rN593sxIDh@orel> (raw)
In-Reply-To: <20261115085457.889873-1-shanbeeyoo@gmail.com>
On Sun, Nov 15, 2026 at 04:54:57PM +0800, Bingyu.Xian wrote:
> The RISC-V steal-time test checks the UAPI alignment constraints and
> compares stolen time with the host run delay. It does not exercise
> changes to the shared page after accounting has been enabled.
>
> Extend the existing test to check that disabling STA stops shared-page
> updates and that reconfiguring it leaves the old shared page unchanged.
> Place the second STA structure at the end of another page and verify
> that updates mark the appropriate page dirty.
>
> Discard the second backing page with MADV_DONTNEED and check that a
> subsequent KVM_RUN resumes STA updates on the replacement page and
> marks it dirty. Read the replacement page before KVM_RUN so that a
> stale write into the old backing page cannot satisfy the assertions.
>
> Keep these checks in steal_time instead of adding another test binary.
> The test does not depend on a gfn-to-pfn cache implementation.
>
> Assisted-by: YuanSheng: deepseek-v4-pro
> Co-developed-by: Quan Zhou <zhouquan@iscas.ac.cn>
> Signed-off-by: Quan Zhou <zhouquan@iscas.ac.cn>
> Signed-off-by: bingyu.xian <shanbeenyoo@gmail.com>
> ---
> Tested with a freshly built RISC-V kernel and KVM module from the base
> commit below, running as a QEMU TCG guest with H extension enabled:
> - steal_time: 4 lifecycle checks and the existing 4 vCPU cases pass
> - CONFIG_DEBUG_ATOMIC_SLEEP=y and CONFIG_PROVE_LOCKING=y
> - no kernel diagnostics observed
> - x86 steal_time build also passes (not run)
>
> This exercises baseline STA behavior and does not depend on an STA
> caching implementation. The lifecycle checks also apply cleanly to
> riscv_kvm_next at 633b3931a87438282d6454646ff40c669a6896cb.
>
> Test logs (verbatim excerpt from the 2026-11-15 QEMU run):
> STA_LIFECYCLE_KERNEL=7.4.0-rc2-sta-lifecycle+
> Random seed: 0x35d6f94c
> TAP version 13
> 1..8
> ok 1 STA shared page updates and is marked dirty
> ok 2 STA disable stops shared page updates
> ok 3 STA reconfiguration updates only the new page
> ok 4 STA updates recover after backing page invalidation
I'm having trouble understanding what real world scenario these tests are
trying to model. VM migration?
> # VCPU0: total-stolen-time=3269800 test-stolen-time=3269800
> VCPU0:
> sequence: 6
> flags: 0
> steal: 3269800
> preempted: 0
> pad: 00000000000000000000000000000000000000000000000
> ok 5 vcpu0
> # VCPU1: total-stolen-time=898700 test-stolen-time=898700 (BONUS: guest test-stolen-time even exactly matches test-run_delay)
> VCPU1:
> sequence: 4
> flags: 0
> steal: 898700
> preempted: 0
> pad: 00000000000000000000000000000000000000000000000
> ok 6 vcpu1
> # VCPU2: total-stolen-time=654000 test-stolen-time=654000 (BONUS: guest test-stolen-time even exactly matches test-run_delay)
> VCPU2:
> sequence: 4
> flags: 0
> steal: 654000
> preempted: 0
> pad: 00000000000000000000000000000000000000000000000
> ok 7 vcpu2
> # VCPU3: total-stolen-time=800700 test-stolen-time=800700 (BONUS: guest test-stolen-time even exactly matches test-run_delay)
> VCPU3:
> sequence: 4
> flags: 0
> steal: 800700
> preempted: 0
> pad: 00000000000000000000000000000000000000000000000
> ok 8 vcpu3
> # Totals: pass:8 fail:0 xfail:0 xpass:0 skip:0 error:0
> STA_LIFECYCLE_STATUS=0
> STA_LIFECYCLE_PASS
>
> tools/testing/selftests/kvm/steal_time.c | 113 +++++++++++++++++++++++
> 1 file changed, 113 insertions(+)
>
> diff --git a/tools/testing/selftests/kvm/steal_time.c b/tools/testing/selftests/kvm/steal_time.c
> index 82b94a82d..059f7fa60 100644
> --- a/tools/testing/selftests/kvm/steal_time.c
> +++ b/tools/testing/selftests/kvm/steal_time.c
> @@ -8,6 +8,7 @@
> #include <time.h>
> #include <sched.h>
> #include <pthread.h>
> +#include <sys/mman.h>
> #include <linux/kernel.h>
> #include <asm/kvm.h>
> #ifdef __riscv
> @@ -329,6 +330,111 @@ static void steal_time_dump(struct kvm_vm *vm, u32 vcpu_idx)
> pr_info("\n");
> }
>
> +static gpa_t lifecycle_second_gpa;
> +
> +static void sta_lifecycle_guest(void)
> +{
> + sta_set_shmem(ST_GPA_BASE, 0);
> + GUEST_SYNC(0);
> + /* The first update takes place on the next KVM_RUN. */
> + GUEST_SYNC(1);
> +
> + sta_set_shmem(INVALID_GPA, 0);
> + GUEST_SYNC(2);
> + GUEST_SYNC(3);
> +
> + sta_set_shmem(lifecycle_second_gpa, 0);
> + GUEST_SYNC(4);
> + GUEST_SYNC(5);
> +
> + /* Userspace discards the second backing page before resuming us. */
> + GUEST_SYNC(6);
> + sta_set_shmem(INVALID_GPA, 0);
> + GUEST_DONE();
> +}
> +
> +static void sta_lifecycle_run(struct kvm_vcpu *vcpu, unsigned int stage)
> +{
> + struct ucall uc;
> +
> + vcpu_run(vcpu);
> + switch (get_ucall(vcpu, &uc)) {
> + case UCALL_SYNC:
> + TEST_ASSERT_EQ(uc.args[1], stage);
> + break;
> + case UCALL_DONE:
> + TEST_ASSERT_EQ(stage, 7);
> + break;
> + case UCALL_ABORT:
> + REPORT_GUEST_ASSERT(uc);
> + default:
> + TEST_FAIL("Unexpected STA lifecycle exit at stage %u", stage);
> + }
> +}
> +
> +static void check_sta_page(struct sta_struct *st)
> +{
> + TEST_ASSERT(st->sequence && !(st->sequence & 1),
> + "Expected a completed STA update, sequence=%u", st->sequence);
> + TEST_ASSERT_EQ(st->flags, 0);
> + TEST_ASSERT_EQ(st->preempted, 0);
> +}
> +
> +static void check_sta_lifecycle(void)
> +{
> + struct sta_struct snapshot, *first, *second;
> + struct kvm_vcpu *vcpu;
> + struct kvm_vm *vm;
> + unsigned long dirty = 0;
> + void *second_page;
> +
> + vm = vm_create_with_one_vcpu(&vcpu, sta_lifecycle_guest);
> + vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS, ST_GPA_BASE,
> + 1, 2, KVM_MEM_LOG_DIRTY_PAGES);
> + /* Also exercise an STA structure ending exactly at a page boundary. */
> + WRITE_AND_SYNC_TO_GUEST(vm, lifecycle_second_gpa,
> + ST_GPA_BASE + 2 * vm->page_size - sizeof(struct sta_struct));
> + first = addr_gpa2hva(vm, ST_GPA_BASE);
> + second = addr_gpa2hva(vm, lifecycle_second_gpa);
> + second_page = addr_gpa2hva(vm, ST_GPA_BASE + vm->page_size);
> +
> + sta_lifecycle_run(vcpu, 0);
> + sta_lifecycle_run(vcpu, 1);
> + check_sta_page(first);
> + kvm_vm_get_dirty_log(vm, 1, &dirty);
> + TEST_ASSERT(dirty & BIT(0), "STA update did not dirty the first page");
> + ksft_test_result_pass("STA shared page updates and is marked dirty\n");
> +
> + sta_lifecycle_run(vcpu, 2);
> + snapshot = *first;
> + sta_lifecycle_run(vcpu, 3);
> + TEST_ASSERT(!memcmp(first, &snapshot, sizeof(snapshot)),
> + "Disabled STA shared page was modified");
> + ksft_test_result_pass("STA disable stops shared page updates\n");
> +
> + sta_lifecycle_run(vcpu, 4);
> + sta_lifecycle_run(vcpu, 5);
> + check_sta_page(second);
> + TEST_ASSERT(!memcmp(first, &snapshot, sizeof(snapshot)),
> + "Old STA shared page was modified after reconfiguration");
> + dirty = 0;
> + kvm_vm_get_dirty_log(vm, 1, &dirty);
> + TEST_ASSERT(dirty & BIT(1), "STA update did not dirty the new page");
> + ksft_test_result_pass("STA reconfiguration updates only the new page\n");
> +
> + TEST_ASSERT_EQ(madvise(second_page, vm->page_size, MADV_DONTNEED), 0);
> + TEST_ASSERT_EQ(second->sequence, 0);
> + sta_lifecycle_run(vcpu, 6);
> + check_sta_page(second);
> + dirty = 0;
> + kvm_vm_get_dirty_log(vm, 1, &dirty);
> + TEST_ASSERT(dirty & BIT(1), "Recovered STA update did not dirty the page");
> + ksft_test_result_pass("STA updates recover after backing page invalidation\n");
> +
> + sta_lifecycle_run(vcpu, 7);
> + kvm_vm_free(vm);
> +}
> +
> static void check_steal_time_uapi(void)
> {
> struct kvm_vm *vm;
> @@ -535,9 +641,16 @@ int main(int ac, char **av)
>
> ksft_print_header();
> TEST_REQUIRE(is_steal_time_supported(vcpus[0]));
> +#ifdef __riscv
> + ksft_set_plan(NR_VCPUS + 4);
> +#else
> ksft_set_plan(NR_VCPUS);
> +#endif
>
> check_steal_time_uapi();
> +#ifdef __riscv
> + check_sta_lifecycle();
> +#endif
The #ifdefs aren't nice. Again, what are the tests trying to prove? And
why wouldn't the other architectures need the same coverage?
Thanks,
drew
>
> /* Run test on each VCPU */
> for (i = 0; i < NR_VCPUS; ++i) {
>
> base-commit: 23c62304dde29abe92013d85910d03a77dada35d
> --
> 2.34.1
>
>
> --
> kvm-riscv mailing list
> kvm-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/kvm-riscv
prev parent reply other threads:[~2026-11-19 13:50 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-15 8:54 Bingyu.Xian
2026-11-19 13:49 ` Andrew Jones [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=av77j8rN593sxIDh@orel \
--to=andrew.jones@linux.dev \
--cc=alex@ghiti.fr \
--cc=anup@brainfault.org \
--cc=aou@eecs.berkeley.edu \
--cc=atish.patra@linux.dev \
--cc=kvm-riscv@lists.infradead.org \
--cc=kvm@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-kselftest@vger.kernel.org \
--cc=linux-riscv@lists.infradead.org \
--cc=palmer@dabbelt.com \
--cc=pbonzini@redhat.com \
--cc=pjw@kernel.org \
--cc=seanjc@google.com \
--cc=shanbeeyoo@gmail.com \
--cc=shuah@kernel.org \
--cc=zhouquan@iscas.ac.cn \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®