[ re-send to Darren Hart ]

----- On Aug 23, 2021, at 11:18 AM, Mathieu Desnoyers mathieu.desnoyers@efficios.com wrote:

----- On Aug 20, 2021, at 6:50 PM, Sean Christopherson seanjc@google.com wrote:

...

Add a test to verify an rseq's CPU ID is updated correctly if the task is migrated while the kernel is handling KVM_RUN. This is a regression test for a bug introduced by commit 72c3c0fe54a3 ("x86/kvm: Use generic xfer to guest work function"), where TIF_NOTIFY_RESUME would be cleared by KVM without updating rseq, leading to a stale CPU ID and other badness.

[...]

+#define RSEQ_SIG 0xdeadbeef

Is there any reason for defining a custom signature rather than including tools/testing/selftests/rseq/rseq.h ? This should take care of including the proper architecture header which will define the appropriate signature.

Arguably you don't define rseq critical sections in this test per se, but I'm wondering why the custom signature here.

[...]

...

+static void *migration_worker(void *ign) +{

• cpu_set_t allowed_mask;
• int r, i, nr_cpus, cpu;
• CPU_ZERO(&allowed_mask);
• nr_cpus = CPU_COUNT(&possible_mask);
• for (i = 0; i < 20000; i++) {

• cpu = i % nr_cpus;
  

• if (!CPU_ISSET(cpu, &possible_mask))
  

• 	continue;
  

• CPU_SET(cpu, &allowed_mask);
  

• /*
  

•  * Bump the sequence count twice to allow the reader to detect
  

•  * that a migration may have occurred in between rseq and sched
  

•  * CPU ID reads.  An odd sequence count indicates a migration
  

•  * is in-progress, while a completely different count indicates
  

•  * a migration occurred since the count was last read.
  

•  */
  

• atomic_inc(&seq_cnt);
  

So technically this atomic_inc contains the required barriers because the selftests implementation uses "__sync_add_and_fetch(&addr->val, 1)". But it's rather odd that the semantic differs from the kernel implementation in terms of memory barriers: the kernel implementation of atomic_inc guarantees no memory barriers, but this one happens to provide full barriers pretty much by accident (selftests futex/include/atomic.h documents no such guarantee).

If this full barrier guarantee is indeed provided by the selftests atomic.h header, I would really like a comment stating that in the atomic.h header so the carpet is not pulled from under our feet by a future optimization.

...

• r = sched_setaffinity(0, sizeof(allowed_mask), &allowed_mask);
  

• TEST_ASSERT(!r, "sched_setaffinity failed, errno = %d (%s)",
  

• 	    errno, strerror(errno));
  

• atomic_inc(&seq_cnt);
  

• CPU_CLR(cpu, &allowed_mask);
  

• /*
  

•  * Let the read-side get back into KVM_RUN to improve the odds
  

•  * of task migration coinciding with KVM's run loop.
  

This comment should be about increasing the odds of letting the seqlock read-side complete. Otherwise, the delay between the two back-to-back atomic_inc is so small that the seqlock read-side may never have time to complete the reading the rseq cpu id and the sched_getcpu() call, and can retry forever.

I'm wondering if 1 microsecond is sufficient on other architectures as well. One alternative way to make this depend less on the architecture's implementation of sched_getcpu (whether it's a vDSO, or goes through a syscall) would be to read the rseq cpu id and call sched_getcpu a few times (e.g. 3 times) in the migration thread rather than use usleep, and throw away the value read. This would ensure the delay is appropriate on all architectures.

Thanks!

Mathieu

...

•  */
  

• usleep(1);
  

• }
• done = true;
• return NULL;

+}

+int main(int argc, char *argv[]) +{

• struct kvm_vm *vm;
• u32 cpu, rseq_cpu;
• int r, snapshot;
• /* Tell stdout not to buffer its content */
• setbuf(stdout, NULL);
• r = sched_getaffinity(0, sizeof(possible_mask), &possible_mask);
• TEST_ASSERT(!r, "sched_getaffinity failed, errno = %d (%s)", errno,

•     strerror(errno));
  

• if (CPU_COUNT(&possible_mask) < 2) {

• print_skip("Only one CPU, task migration not possible\n");
  

• exit(KSFT_SKIP);
  

• }
• sys_rseq(0);
• /*

• * Create and run a dummy VM that immediately exits to userspace via
  

• * GUEST_SYNC, while concurrently migrating the process by setting its
  

• * CPU affinity.
  

• */
  

• vm = vm_create_default(VCPU_ID, 0, guest_code);
• pthread_create(&migration_thread, NULL, migration_worker, 0);
• while (!done) {

• vcpu_run(vm, VCPU_ID);
  

• TEST_ASSERT(get_ucall(vm, VCPU_ID, NULL) == UCALL_SYNC,
  

• 	    "Guest failed?");
  

• /*
  

•  * Verify rseq's CPU matches sched's CPU.  Ensure migration
  

•  * doesn't occur between sched_getcpu() and reading the rseq
  

•  * cpu_id by rereading both if the sequence count changes, or
  

•  * if the count is odd (migration in-progress).
  

•  */
  

• do {
  

• 	/*
  

• 	 * Drop bit 0 to force a mismatch if the count is odd,
  

• 	 * i.e. if a migration is in-progress.
  

• 	 */
  

• 	snapshot = atomic_read(&seq_cnt) & ~1;
  

• 	smp_rmb();
  

• 	cpu = sched_getcpu();
  

• 	rseq_cpu = READ_ONCE(__rseq.cpu_id);
  

• 	smp_rmb();
  

• } while (snapshot != atomic_read(&seq_cnt));
  

• TEST_ASSERT(rseq_cpu == cpu,
  

• 	    "rseq CPU = %d, sched CPU = %d\n", rseq_cpu, cpu);
  

• }
• pthread_join(migration_thread, NULL);
• kvm_vm_free(vm);
• sys_rseq(RSEQ_FLAG_UNREGISTER);
• return 0;

+}

2.33.0.rc2.250.ged5fa647cd-goog

-- Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com

----- On Aug 25, 2021, at 8:51 PM, Sean Christopherson seanjc@google.com wrote:

On Mon, Aug 23, 2021, Mathieu Desnoyers wrote:

...

[ re-send to Darren Hart ]

----- On Aug 23, 2021, at 11:18 AM, Mathieu Desnoyers mathieu.desnoyers@efficios.com wrote:

...

----- On Aug 20, 2021, at 6:50 PM, Sean Christopherson seanjc@google.com wrote:

...

Add a test to verify an rseq's CPU ID is updated correctly if the task is migrated while the kernel is handling KVM_RUN. This is a regression test for a bug introduced by commit 72c3c0fe54a3 ("x86/kvm: Use generic xfer to guest work function"), where TIF_NOTIFY_RESUME would be cleared by KVM without updating rseq, leading to a stale CPU ID and other badness.

[...]

+#define RSEQ_SIG 0xdeadbeef

Is there any reason for defining a custom signature rather than including tools/testing/selftests/rseq/rseq.h ? This should take care of including the proper architecture header which will define the appropriate signature.

Arguably you don't define rseq critical sections in this test per se, but I'm wondering why the custom signature here.

Partly to avoid taking a dependency on rseq.h, and partly to try to call out that the test doesn't actually do any rseq critical sections.

It might be good to add a comment near this define stating this then, so nobody attempts to copy this as an example.

...

...

[...]

...

+static void *migration_worker(void *ign) +{

• cpu_set_t allowed_mask;
• int r, i, nr_cpus, cpu;
• CPU_ZERO(&allowed_mask);
• nr_cpus = CPU_COUNT(&possible_mask);
• for (i = 0; i < 20000; i++) {

• cpu = i % nr_cpus;
  

• if (!CPU_ISSET(cpu, &possible_mask))
  

• 	continue;
  

• CPU_SET(cpu, &allowed_mask);
  

• /*
  

•  * Bump the sequence count twice to allow the reader to detect
  

•  * that a migration may have occurred in between rseq and sched
  

•  * CPU ID reads.  An odd sequence count indicates a migration
  

•  * is in-progress, while a completely different count indicates
  

•  * a migration occurred since the count was last read.
  

•  */
  

• atomic_inc(&seq_cnt);
  

So technically this atomic_inc contains the required barriers because the selftests implementation uses "__sync_add_and_fetch(&addr->val, 1)". But it's rather odd that the semantic differs from the kernel implementation in terms of memory barriers: the kernel implementation of atomic_inc guarantees no memory barriers, but this one happens to provide full barriers pretty much by accident (selftests futex/include/atomic.h documents no such guarantee).

Yeah, I got quite lost trying to figure out what atomics the test would actually end up with.

At the very least, until things are clarified in the selftests atomic header, I would recommend adding a comment stating which memory barriers are required alongside each use of atomic_inc here. I would even prefer that we add extra (currently unneeded) write barriers to make extra sure that this stays documented. Performance of the write-side does not matter much here.

...

...

If this full barrier guarantee is indeed provided by the selftests atomic.h header, I would really like a comment stating that in the atomic.h header so the carpet is not pulled from under our feet by a future optimization.

...

• r = sched_setaffinity(0, sizeof(allowed_mask), &allowed_mask);
  

• TEST_ASSERT(!r, "sched_setaffinity failed, errno = %d (%s)",
  

• 	    errno, strerror(errno));
  

• atomic_inc(&seq_cnt);
  

• CPU_CLR(cpu, &allowed_mask);
  

• /*
  

•  * Let the read-side get back into KVM_RUN to improve the odds
  

•  * of task migration coinciding with KVM's run loop.
  

This comment should be about increasing the odds of letting the seqlock read-side complete. Otherwise, the delay between the two back-to-back atomic_inc is so small that the seqlock read-side may never have time to complete the reading the rseq cpu id and the sched_getcpu() call, and can retry forever.

Hmm, but that's not why there's a delay. I'm not arguing that a livelock isn't possible (though that syscall would have to be screaming fast),

I don't think we have the same understanding of the livelock scenario. AFAIU the livelock would be caused by a too-small delay between the two consecutive atomic_inc() from one loop iteration to the next compared to the time it takes to perform a read-side critical section of the seqlock. Back-to-back atomic_inc can be performed very quickly, so I doubt that the sched_getcpu implementation have good odds to be fast enough to complete in that narrow window, leading to lots of read seqlock retry.

but the primary motivation is very much to allow the read-side enough time to get back into KVM proper.

I'm puzzled by your statement. AFAIU, let's say we don't have the delay, then we have:

migration thread KVM_RUN/read-side thread ----------------------------------------------------------------------------------- - ioctl(KVM_RUN) - atomic_inc_seq_cst(&seqcnt) - sched_setaffinity - atomic_inc_seq_cst(&seqcnt) - a = atomic_load(&seqcnt) & ~1 - smp_rmb() - b = LOAD_ONCE(__rseq_abi->cpu_id); - sched_getcpu() - smp_rmb() - re-load seqcnt/compare (succeeds) - Can only succeed if entire read-side happens while the seqcnt is in an even numbered state. - if (a != b) abort() /* no delay. Even counter state is very short. */ - atomic_inc_seq_cst(&seqcnt) /* Let's suppose the lack of delay causes the setaffinity to complete too early compared with KVM_RUN ioctl */ - sched_setaffinity - atomic_inc_seq_cst(&seqcnt)

/* no delay. Even counter state is very short. */ - atomic_inc_seq_cst(&seqcnt) /* Then a setaffinity from a following migration thread loop will run concurrently with KVM_RUN */ - ioctl(KVM_RUN) - sched_setaffinity - atomic_inc_seq_cst(&seqcnt)

As pointed out here, if the first setaffinity runs too early compared with KVM_RUN, a following setaffinity will run concurrently with it. However, the fact that the even counter state is very short may very well hurt progress of the read seqlock.

To encounter the bug, TIF_NOTIFY_RESUME has to be recognized by KVM in its run loop, i.e. sched_setaffinity() must induce task migration after the read-side has invoked ioctl(KVM_RUN).

No argument here.

...

...

I'm wondering if 1 microsecond is sufficient on other architectures as well.

I'm definitely wondering that as well :-)

...

...

One alternative way to make this depend less on the architecture's implementation of sched_getcpu (whether it's a vDSO, or goes through a syscall) would be to read the rseq cpu id and call sched_getcpu a few times (e.g. 3 times) in the migration thread rather than use usleep, and throw away the value read. This would ensure the delay is appropriate on all architectures.

As above, I think an arbitrary delay is required regardless of how fast sched_getcpu() can execute. One thought would be to do sched_getcpu() _and_ usleep() to account for sched_getcpu() overhead and to satisfy the KVM_RUN part, but I don't know that that adds meaningful value.

The real test is if someone could see if the bug repros on non-x86 hardware...

As I explain in the scenario above, I don't think we agree on the reason why the delay is required. One way to confirm this would be to run the code without delay, and count how many seqcnt read-side succeed vs fail. We can then compare that with a run with a 1us delay between the migration thread loops. This data should help us come to a common understanding of the delay's role.

Thanks,

Mathieu

----- On Aug 26, 2021, at 7:54 PM, Sean Christopherson seanjc@google.com wrote:

On Thu, Aug 26, 2021, Mathieu Desnoyers wrote:

...

----- On Aug 25, 2021, at 8:51 PM, Sean Christopherson seanjc@google.com wrote:

...

...

...

...

• r = sched_setaffinity(0, sizeof(allowed_mask), &allowed_mask);
  

• TEST_ASSERT(!r, "sched_setaffinity failed, errno = %d (%s)",
  

• 	    errno, strerror(errno));
  

• atomic_inc(&seq_cnt);
  

• CPU_CLR(cpu, &allowed_mask);
  

• /*
  

•  * Let the read-side get back into KVM_RUN to improve the odds
  

•  * of task migration coinciding with KVM's run loop.
  

This comment should be about increasing the odds of letting the seqlock read-side complete. Otherwise, the delay between the two back-to-back atomic_inc is so small that the seqlock read-side may never have time to complete the reading the rseq cpu id and the sched_getcpu() call, and can retry forever.

Hmm, but that's not why there's a delay. I'm not arguing that a livelock isn't possible (though that syscall would have to be screaming fast),

I don't think we have the same understanding of the livelock scenario. AFAIU the livelock would be caused by a too-small delay between the two consecutive atomic_inc() from one loop iteration to the next compared to the time it takes to perform a read-side critical section of the seqlock. Back-to-back atomic_inc can be performed very quickly, so I doubt that the sched_getcpu implementation have good odds to be fast enough to complete in that narrow window, leading to lots of read seqlock retry.

Ooooh, yeah, brain fart on my side. I was thinking of the two atomic_inc() in the same loop iteration and completely ignoring the next iteration. Yes, I 100% agree that a delay to ensure forward progress is needed. An assertion in main() that the reader complete at least some reasonable number of KVM_RUNs is also probably a good idea, e.g. to rule out a false pass due to the reader never making forward progress.

Agreed.

FWIW, the do-while loop does make forward progress without a delay, but at ~50% throughput, give or take.

I did not expect absolutely no progress, but a significant slow down of the read-side.

...

...

but the primary motivation is very much to allow the read-side enough time to get back into KVM proper.

I'm puzzled by your statement. AFAIU, let's say we don't have the delay, then we have:

migration thread KVM_RUN/read-side thread

                                         - ioctl(KVM_RUN)

• atomic_inc_seq_cst(&seqcnt)

• sched_setaffinity

• atomic_inc_seq_cst(&seqcnt) - a = atomic_load(&seqcnt) & ~1 - smp_rmb() - b = LOAD_ONCE(__rseq_abi->cpu_id); - sched_getcpu() - smp_rmb() - re-load seqcnt/compare (succeeds) - Can only succeed if entire read-side happens while the seqcnt is in an even numbered state. - if (a != b) abort() /* no delay. Even counter state is very short. */

• atomic_inc_seq_cst(&seqcnt) /* Let's suppose the lack of delay causes the setaffinity to complete too early compared with KVM_RUN ioctl */

• sched_setaffinity

• atomic_inc_seq_cst(&seqcnt)

  /* no delay. Even counter state is very short. */

• atomic_inc_seq_cst(&seqcnt) /* Then a setaffinity from a following migration thread loop will run concurrently with KVM_RUN */ - ioctl(KVM_RUN)

• sched_setaffinity

• atomic_inc_seq_cst(&seqcnt)

As pointed out here, if the first setaffinity runs too early compared with KVM_RUN, a following setaffinity will run concurrently with it. However, the fact that the even counter state is very short may very well hurt progress of the read seqlock.

*sigh*

Several hours later, I think I finally have my head wrapped around everything.

Due to the way the test is written and because of how KVM's run loop works, TIF_NOTIFY_RESUME or TIF_NEED_RESCHED effectively has to be set before KVM actually enters the guest, otherwise KVM will exit to userspace without touching the flag, i.e. it will be handled by the normal exit_to_user_mode_loop().

Where I got lost was trying to figure out why I couldn't make the bug reproduce by causing the guest to exit to KVM, but not userspace, in which case KVM should easily trigger the problematic flow as the window for sched_getcpu() to collide with KVM would be enormous. The reason I didn't go down this route for the "official" test is that, unless there's something clever I'm overlooking, it requires arch specific guest code, and ialso I don't know that forcing an exit would even be necessary/sufficient on other architectures.

Anyways, I was trying to confirm that the bug was being hit without a delay, while still retaining the sequence retry in the test. The test doesn't fail because the back-to-back atomic_inc() changes seqcnt too fast. The read-side makes forward progress, but it never observes failure because the do-while loop only ever completes after another sched_setaffinity(), never after the one that collides with KVM because it takes too long to get out of ioctl(KVM_RUN) and back to the test. I.e. the atomic_inc() in the next loop iteration (makes seq_cnt odd) always completes before the check, and so the check ends up spinning until another migration, which correctly updates rseq. This was expected and didn't confuse me.

What confused me is that I was trying to confirm the bug was being hit from within the kernel by confirming KVM observed TIF_NOTIFY_RESUME, but I misunderstood when TIF_NOTIFY_RESUME would get set. KVM can observe TIF_NOTIFY_RESUME directly, but it's rare, and I suspect happens iff sched_setaffinity() hits the small window where it collides with KVM_RUN before KVM enters the guest.

More commonly, the bug occurs when KVM sees TIF_NEED_RESCHED. In that case, KVM calls xfer_to_guest_mode_work(), which does schedule() and _that_ sets TIF_NOTIFY_RESUME. xfer_to_guest_mode_work() then mishandles TIF_NOTIFY_RESUME and the bug is hit, but my confirmation logic in KVM never fired.

So there are effectively three reasons we want a delay:

1. To allow sched_setaffinity() to coincide with ioctl(KVM_RUN) before KVM can enter the guest so that the guest doesn't need an arch-specific VM-Exit source.

2. To let ioctl(KVM_RUN) make its way back to the test before the next round of migration.

3. To ensure the read-side can make forward progress, e.g. if sched_getcpu() involves a syscall.

After looking at KVM for arm64 and s390, #1 is a bit tenuous because x86 is the only arch that currently uses xfer_to_guest_mode_work(), i.e. the test could be tweaked to be overtly x86-specific. But since a delay is needed for #2 and #3, I'd prefer to rely on it for #1 as well in the hopes that this test provides coverage for arm64 and/or s390 if they're ever converted to use the common xfer_to_guest_mode_work().

Now that we have this understanding of why we need the delay, it would be good to write this down in a comment within the test.

Does it reproduce if we randomize the delay to have it picked randomly from 0us to 100us (with 1us step) ? It would remove a lot of the needs for arch-specific magic delay value.

Thanks,

Mathieu

----- On Aug 27, 2021, at 7:23 PM, Sean Christopherson seanjc@google.com wrote:

On Fri, Aug 27, 2021, Mathieu Desnoyers wrote:

[...]

...

Does it reproduce if we randomize the delay to have it picked randomly from 0us to 100us (with 1us step) ? It would remove a lot of the needs for arch-specific magic delay value.

My less-than-scientific testing shows that it can reproduce at delays up to ~500us, but above ~10us the reproducibility starts to drop. The bug still reproduces reliably, it just takes more iterations, and obviously the test runs a bit slower.

Any objection to using a 1-10us delay, e.g. a simple usleep((i % 10) + 1)?

Works for me, thanks!

Mathieu