Re: [PATCH v7 00/12] Direct Map Removal Support for guest_memfd

[Nikita Kalyazin <kalyazin@amazon.com>]

On 07/11/2025 18:04, Brendan Jackman wrote:
> On Fri Nov 7, 2025 at 5:23 PM UTC, Nikita Kalyazin wrote:
>>
>>
>> On 07/11/2025 15:54, Brendan Jackman wrote:
>>> On Wed Sep 24, 2025 at 3:10 PM UTC, Patrick Roy wrote:
>>>> From: Patrick Roy <roypat@amazon.co.uk>
>>>>
>>>> [ based on kvm/next ]
>>>>
>>>> Unmapping virtual machine guest memory from the host kernel's direct map is a
>>>> successful mitigation against Spectre-style transient execution issues: If the
>>>> kernel page tables do not contain entries pointing to guest memory, then any
>>>> attempted speculative read through the direct map will necessarily be blocked
>>>> by the MMU before any observable microarchitectural side-effects happen. This
>>>> means that Spectre-gadgets and similar cannot be used to target virtual machine
>>>> memory. Roughly 60% of speculative execution issues fall into this category [1,
>>>> Table 1].
>>>>
>>>> This patch series extends guest_memfd with the ability to remove its memory
>>>> from the host kernel's direct map, to be able to attain the above protection
>>>> for KVM guests running inside guest_memfd.
>>>>
>>>> Additionally, a Firecracker branch with support for these VMs can be found on
>>>> GitHub [2].
>>>>
>>>> For more details, please refer to the v5 cover letter [v5]. No
>>>> substantial changes in design have taken place since.
>>>>
>>>> === Changes Since v6 ===
>>>>
>>>> - Drop patch for passing struct address_space to ->free_folio(), due to
>>>>     possible races with freeing of the address_space. (Hugh)
>>>> - Stop using PG_uptodate / gmem preparedness tracking to keep track of
>>>>     direct map state.  Instead, use the lowest bit of folio->private. (Mike, David)
>>>> - Do direct map removal when establishing mapping of gmem folio instead
>>>>     of at allocation time, due to impossibility of handling direct map
>>>>     removal errors in kvm_gmem_populate(). (Patrick)
>>>> - Do TLB flushes after direct map removal, and provide a module
>>>>     parameter to opt out from them, and a new patch to export
>>>>     flush_tlb_kernel_range() to KVM. (Will)
>>>>
>>>> [1]: https://download.vusec.net/papers/quarantine_raid23.pdf
>>>> [2]: https://github.com/firecracker-microvm/firecracker/tree/feature/secret-hiding
>>>
>>> I just got around to trying this out, I checked out this patchset using
>>> its base-commit and grabbed the Firecracker branch. Things seem OK until
>>> I set the secrets_free flag in the Firecracker config which IIUC makes
>>> it set GUEST_MEMFD_FLAG_NO_DIRECT_MAP.
>>>
>>> If I set it, I find the guest doesn't show anything on the console.
>>> Running it in a VM and attaching GDB suggests that it's entering the
>>> guest repeatedly, it doesn't seem like the vCPU thread is stuck or
>>> anything. I'm a bit clueless about how to debug that (so far, whenever
>>> I've broken KVM, things always exploded very dramatically).
>>>
>>> Anyway, if I then kill the firecracker process, the host sometimes
>>> crashes, I think this is the most suggestive splat I've seen:
>>>
>>> [   99.673420][    T2] BUG: unable to handle page fault for address: ffff888012804000
>>> [   99.676216][    T2] #PF: supervisor write access in kernel mode
>>> [   99.678381][    T2] #PF: error_code(0x0002) - not-present page
>>> [   99.680499][    T2] PGD 2e01067 P4D 2e01067 PUD 2e02067 PMD 12801063 PTE 800fffffed7fb020
>>> [   99.683374][    T2] Oops: Oops: 0002 [#1] SMP
>>> [   99.685004][    T2] CPU: 0 UID: 0 PID: 2 Comm: kthreadd Not tainted 6.17.0-rc7-00366-g473c46a3cb2a #106 NONE
>>> [   99.688514][    T2] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.1 11/11/2019
>>> [   99.691547][    T2] RIP: 0010:clear_page_erms+0x7/0x10
>>> [   99.693440][    T2] Code: 48 89 47 18 48 89 47 20 48 89 47 28 48 89 47 30 48 89 47 38 48 8d 7f 40 75 d9 90 c3 0f 1f 80 00 00 00 00 b9 00 10 00 00 31 c0 <f3> aa c3 66 0f 1f 44 00 00 48 83 f9 40 73 2a 83 f9 08 73 0f 85 c9
>>> [   99.700188][    T2] RSP: 0018:ffff88800318fc10 EFLAGS: 00010246
>>> [   99.702321][    T2] RAX: 0000000000000000 RBX: 0000000000400dc0 RCX: 0000000000001000
>>> [   99.705100][    T2] RDX: ffffea00004a0100 RSI: ffffea00004a0200 RDI: ffff888012804000
>>> [   99.707861][    T2] RBP: 0000000000000801 R08: 0000000000000000 R09: 0000000000000000
>>> [   99.710648][    T2] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000002
>>> [   99.713412][    T2] R13: 0000000000000801 R14: ffffea00004a0100 R15: ffffffff81f4df80
>>> [   99.716191][    T2] FS:  0000000000000000(0000) GS:ffff8880bbf28000(0000) knlGS:0000000000000000
>>> [   99.719316][    T2] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> [   99.721648][    T2] CR2: ffff888012804000 CR3: 0000000007583001 CR4: 0000000000372eb0
>>> [   99.724421][    T2] Call Trace:
>>> [   99.725608][    T2]  <TASK>
>>> [   99.726646][    T2]  get_page_from_freelist+0x6fe/0x14b0
>>> [   99.728583][    T2]  ? fs_reclaim_acquire+0x43/0xe0
>>> [   99.730325][    T2]  ? find_held_lock+0x2b/0x80
>>> [   99.731965][    T2]  __alloc_frozen_pages_noprof+0x147/0x2d0
>>> [   99.734003][    T2]  __alloc_pages_noprof+0x5/0x50
>>> [   99.735766][    T2]  copy_process+0x1b1/0x1b30
>>> [   99.737398][    T2]  ? lock_is_held_type+0x89/0x100
>>> [   99.739157][    T2]  ? kthreadd+0x25/0x190
>>> [   99.740664][    T2]  kernel_clone+0x59/0x390
>>> [   99.742213][    T2]  ? kthreadd+0x25/0x190
>>> [   99.743728][    T2]  kernel_thread+0x55/0x70
>>> [   99.745310][    T2]  ? kthread_complete_and_exit+0x20/0x20
>>> [   99.747265][    T2]  kthreadd+0x117/0x190
>>> [   99.748748][    T2]  ? kthread_is_per_cpu+0x30/0x30
>>> [   99.750509][    T2]  ret_from_fork+0x16b/0x1e0
>>> [   99.752193][    T2]  ? kthread_is_per_cpu+0x30/0x30
>>> [   99.753992][    T2]  ret_from_fork_asm+0x11/0x20
>>> [   99.755717][    T2]  </TASK>
>>> [   99.756861][    T2] CR2: ffff888012804000
>>> [   99.758353][    T2] ---[ end trace 0000000000000000 ]---
>>> [   99.760319][    T2] RIP: 0010:clear_page_erms+0x7/0x10
>>> [   99.762209][    T2] Code: 48 89 47 18 48 89 47 20 48 89 47 28 48 89 47 30 48 89 47 38 48 8d 7f 40 75 d9 90 c3 0f 1f 80 00 00 00 00 b9 00 10 00 00 31 c0 <f3> aa c3 66 0f 1f 44 00 00 48 83 f9 40 73 2a 83 f9 08 73 0f 85 c9
>>> [   99.769129][    T2] RSP: 0018:ffff88800318fc10 EFLAGS: 00010246
>>> [   99.771297][    T2] RAX: 0000000000000000 RBX: 0000000000400dc0 RCX: 0000000000001000
>>> [   99.774126][    T2] RDX: ffffea00004a0100 RSI: ffffea00004a0200 RDI: ffff888012804000
>>> [   99.777013][    T2] RBP: 0000000000000801 R08: 0000000000000000 R09: 0000000000000000
>>> [   99.779827][    T2] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000002
>>> [   99.782641][    T2] R13: 0000000000000801 R14: ffffea00004a0100 R15: ffffffff81f4df80
>>> [   99.785487][    T2] FS:  0000000000000000(0000) GS:ffff8880bbf28000(0000) knlGS:0000000000000000
>>> [   99.788671][    T2] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> [   99.791012][    T2] CR2: ffff888012804000 CR3: 0000000007583001 CR4: 0000000000372eb0
>>> [   99.793863][    T2] Kernel panic - not syncing: Fatal exception
>>> [   99.796760][    T2] Kernel Offset: disabled
>>> [   99.798296][    T2] ---[ end Kernel panic - not syncing: Fatal exception ]---
>>>
>>> This makes me suspect the kvm_gmem_folio_restore_direct_map() path isn't
>>> working or isn't getting called.
>>>
>>> If anyone wants help trying to reproduce this let me know.
>>
>> Hi Brendan,
>>
>> Thanks for trying to run it!
>>
>> Just as a sanity check, the way it is known for us to work is we apply
>> all patches from [1].  For booted VMs (as opposed to restored from
>> snapshot), apart from the v6 of the direct map removal series, the only
>> additional patch is a fix for kvmclock on x86 [2].  Please let me know
>> if you see the same issue with that patch applied too.
>>
>> Nikita
>>
>> [1]
>> https://github.com/firecracker-microvm/firecracker/tree/feature/secret-hiding/resources/hiding_ci/linux_patches
>> [2]
>> https://github.com/firecracker-microvm/firecracker/tree/feature/secret-hiding/resources/hiding_ci/linux_patches/11-kvm-clock
> 
> Ah, thanks! Seems I should have checked my inbox before sending my other
> mail. With the kvmclock fix applied to my host kernel, I start setting
> the other crash immediately when the VM boots. If I comment out the
> actual unmapping of memory, it boots (before, it wouldn't boot even with
> that commented out).
> 
> For the other linux_patches, I couldn't apply them on top of this
> series, do you have a branch I can use as a reference?

Instead of having an explicit branch, we apply all the patches on top of 
[1].  There is a script that performs fetch/build/install end-to-end: [2].

[1] 
https://github.com/firecracker-microvm/firecracker/blob/feature/secret-hiding/resources/hiding_ci/kernel_commit_hash
[2] 
https://github.com/firecracker-microvm/firecracker/blob/feature/secret-hiding/resources/hiding_ci/build_and_install_kernel.sh

> 
> Anyway, the solution I'm hoping to present for your problem gets rid of
> that explicit unmapping code (the allocator will do it for you), so in
> the meantime I have something I can work with.