Re: [PATCH v9 20/43] arm64: RME: Runtime faulting of memory

[Gavin Shan <gshan@redhat.com>]

Hi Steven,

On 6/11/25 8:48 PM, Steven Price wrote:
> At runtime if the realm guest accesses memory which hasn't yet been
> mapped then KVM needs to either populate the region or fault the guest.
> 
> For memory in the lower (protected) region of IPA a fresh page is
> provided to the RMM which will zero the contents. For memory in the
> upper (shared) region of IPA, the memory from the memslot is mapped
> into the realm VM non secure.
> 
> Signed-off-by: Steven Price <steven.price@arm.com>
> ---
>   arch/arm64/include/asm/kvm_emulate.h |  10 ++
>   arch/arm64/include/asm/kvm_rme.h     |  10 ++
>   arch/arm64/kvm/mmu.c                 | 133 ++++++++++++++++++++-
>   arch/arm64/kvm/rme.c                 | 165 +++++++++++++++++++++++++++
>   4 files changed, 312 insertions(+), 6 deletions(-)
> 

[...]

> @@ -1078,6 +1091,9 @@ void kvm_free_stage2_pgd(struct kvm_s2_mmu *mmu)
>   	if (kvm_is_realm(kvm) &&
>   	    (kvm_realm_state(kvm) != REALM_STATE_DEAD &&
>   	     kvm_realm_state(kvm) != REALM_STATE_NONE)) {
> +		struct realm *realm = &kvm->arch.realm;
> +
> +		kvm_stage2_unmap_range(mmu, 0, BIT(realm->ia_bits - 1), false);
>   		write_unlock(&kvm->mmu_lock);
>   		kvm_realm_destroy_rtts(kvm, pgt->ia_bits);
>   

I'm giving it a try before taking time to review, @may_block needs to be true.
I don't see there is anything why not to do so :)

   kvm_stage2_unmap_range(mmu, 0, BIT(realm->ia_bits - 1), true);

Otherwise, there is RCU stall when the VM is destroyed.

[12730.399825] rcu: INFO: rcu_preempt self-detected stall on CPU
[12730.401922] rcu:     5-....: (5165 ticks this GP) idle=3544/1/0x4000000000000000 softirq=41673/46605 fqs=2625
[12730.404598] rcu:     (t=5251 jiffies g=61757 q=36 ncpus=8)
[12730.406771] CPU: 5 UID: 0 PID: 170 Comm: qemu-system-aar Not tainted 6.16.0-rc1-gavin-gfbc56042a9cf #36 PREEMPT
[12730.407918] Hardware name: QEMU QEMU Virtual Machine, BIOS unknown 02/02/2022
[12730.408796] pstate: 61402009 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
[12730.409515] pc : realm_unmap_private_range+0x1b4/0x310
[12730.410825] lr : realm_unmap_private_range+0x98/0x310
[12730.411377] sp : ffff8000808f3920
[12730.411777] x29: ffff8000808f3920 x28: 0000000104d29000 x27: 000000004229b000
[12730.413410] x26: 0000000000000000 x25: ffffb8c82d23f000 x24: 00007fffffffffff
[12730.414292] x23: 000000004229c000 x22: 0001000000000000 x21: ffff80008019deb8
[12730.415229] x20: 0000000101b3f000 x19: 000000004229b000 x18: ffff8000808f3bd0
[12730.416119] x17: 0000000000000001 x16: ffffffffffffffff x15: 0000ffff91cc5000
[12730.417004] x14: ffffb8c82cfccb48 x13: 0000ffff4b1fffff x12: 0000000000000000
[12730.417876] x11: 0000000038e38e39 x10: 0000000000000004 x9 : ffffb8c82c39a030
[12730.418863] x8 : ffff80008019deb8 x7 : 0000010000000000 x6 : 0000000000000000
[12730.419738] x5 : 0000000038e38e39 x4 : ffff0000c0d80000 x3 : 0000000000000000
[12730.420609] x2 : 000000004229c000 x1 : 0000000104d2a000 x0 : 0000000000000000
[12730.421602] Call trace:
[12730.422209]  realm_unmap_private_range+0x1b4/0x310 (P)
[12730.423096]  kvm_realm_unmap_range+0xbc/0xe0
[12730.423657]  __unmap_stage2_range+0x74/0xa8
[12730.424198]  kvm_free_stage2_pgd+0xc8/0x120
[12730.424746]  kvm_uninit_stage2_mmu+0x24/0x48
[12730.425284]  kvm_arch_flush_shadow_all+0x74/0x98
[12730.425849]  kvm_mmu_notifier_release+0x38/0xa0
[12730.426409]  __mmu_notifier_release+0x80/0x1f0
[12730.427031]  exit_mmap+0x3d8/0x438
[12730.427526]  __mmput+0x38/0x160
[12730.428000]  mmput+0x58/0x78
[12730.428463]  do_exit+0x210/0x9d8
[12730.428945]  do_group_exit+0x3c/0xa0
[12730.429458]  get_signal+0x8d4/0x9c0
[12730.429954]  do_signal+0x98/0x400
[12730.430455]  do_notify_resume+0xec/0x1a0
[12730.431030]  el0_svc+0xe8/0x130
[12730.431536]  el0t_64_sync_handler+0x10c/0x138
[12730.432085]  el0t_64_sync+0x1ac/0x1b0

Thanks,
Gavin