Re: [patch V9 30/39] x86/entry: Convert various hypervisor vectors to IDTENTRY_SYSVEC

[Boqun Feng]

On Wed, May 27, 2020 at 12:09:18PM +0000, Wei Liu wrote:
> On Wed, May 27, 2020 at 09:38:03AM +0100, Wei Liu wrote:
> > On Wed, May 27, 2020 at 09:46:16AM +0800, Boqun Feng wrote:
> > > Hi Thomas,
> > > 
> > > On Thu, May 21, 2020 at 10:05:43PM +0200, Thomas Gleixner wrote:
> > > > From: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > > > 
> > > > Convert various hypervisor vectors to IDTENTRY_SYSVEC
> > > >   - Implement the C entry point with DEFINE_IDTENTRY_SYSVEC
> > > >   - Emit the ASM stub with DECLARE_IDTENTRY_SYSVEC
> > > >   - Remove the ASM idtentries in 64bit
> > > >   - Remove the BUILD_INTERRUPT entries in 32bit
> > > >   - Remove the old prototypes
> > > > 
> > > > No functional change.
> > > > 
> > > > Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > > > Acked-by: Andy Lutomirski <luto@xxxxxxxxxx>
> > > 
> > > I hit the following while trying to test the whole patchset on a Hyper-V
> > > guest, and git bisect told me this patch introduced the problem, the
> > > config file is in the attachment.
> > > 
> > > Regards,
> > > Boqun
> > > 
> > 
> > Hi Boqun,
> > 
> > Thanks for testing and reporting.
> > 
> > This patch modifies Hyper-V's entry point so no wonder you pin down this
> > patch.
> > 
> > > [    3.366637] BUG: kernel NULL pointer dereference, address: 0000000000000010
> > > [    3.369959] #PF: supervisor instruction fetch in kernel mode
> > > [    3.369959] #PF: error_code(0x0010) - not-present page
> > > [    3.369959] PGD 0 P4D 0
> > > [    3.369959] Oops: 0010 [#1] PREEMPT SMP PTI
> > > [    3.369959] CPU: 27 PID: 0 Comm: swapper/27 Not tainted 5.7.0-rc5-00374-ge2d215d23d72 #23
> > > [    3.369959] Hardware name: Microsoft Corporation Virtual Machine/Virtual Machine, BIOS Hyper-V UEFI Release v4.0 05/16/2019
> > > [    3.369959] RIP: 0010:0x10
> > 
> > Hmmm... NULL pointer here.
> > 
> > > [    3.369959] Code: Bad RIP value.
> > > [    3.369959] RSP: 0000:ffffbf444018beb8 EFLAGS: 00010086
> > > [    3.369959] RAX: 000000000000001b RBX: ffffa04620900000 RCX: 00000000c8aab6dc
> > > [    3.369959] RDX: 0000000000000001 RSI: ffffffffb390feaf RDI: ffffffffb3919e60
> > > [    3.369959] RBP: 000000000000001b R08: 0000000000000000 R09: 0000000000000101
> > > [    3.369959] R10: 0000000000dda899 R11: 0000000001d417ad R12: 0000000000000000
> > > [    3.369959] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
> > > [    3.369959] FS:  0000000000000000(0000) GS:ffffa04627ac0000(0000) knlGS:0000000000000000
> > > [    3.369959] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > [    3.369959] CR2: ffffffffffffffe6 CR3: 000000101bcba002 CR4: 00000000003606e0
> > > [    3.369959] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > > [    3.369959] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> > > [    3.369959] Call Trace:
> > > [    3.369959]  ? default_idle+0x1c/0x190
> > > [    3.369959]  ? do_idle+0x1cd/0x230
> > > [    3.369959]  ? cpu_startup_entry+0x19/0x20
> > > [    3.369959]  ? secondary_startup_64+0xa4/0xb0
> > 
> > And I certainly didn't expect this.
> > 
> > > [    3.369959] Modules linked in: crc32c_intel hv_vmbus(+)
> > 
> > VMBus is being loaded. That gives me some clue.
> > 
> > I notice a hunk in the patch:
> > 
> > @@ -331,17 +327,19 @@ static void __init ms_hyperv_init_platfo
> >         x86_platform.apic_post_init = hyperv_init;
> >         hyperv_setup_mmu_ops();
> >         /* Setup the IDT for hypervisor callback */
> > -       alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, hyperv_callback_vector);
> > +       alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
> > 
> > It is not using the asm variant.
> > 
> > Could this be the culprit? Thomas?
> > 
> 
> I'm able to reproduce the issue locally.  And after applying the
> following diff I'm able to get a VM booting again on Hyper-V.
> 
> diff --git a/arch/x86/kernel/cpu/mshyperv.c b/arch/x86/kernel/cpu/mshyperv.c
> index a103e1c0b90e..af94f05a5c66 100644
> --- a/arch/x86/kernel/cpu/mshyperv.c
> +++ b/arch/x86/kernel/cpu/mshyperv.c
> @@ -327,7 +327,7 @@ static void __init ms_hyperv_init_platform(void)
>         x86_platform.apic_post_init = hyperv_init;
>         hyperv_setup_mmu_ops();
>         /* Setup the IDT for hypervisor callback */
> -       alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, sysvec_hyperv_callback);
> +       alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_hyperv_callback);
> 
>         /* Setup the IDT for reenlightenment notifications */
>         if (ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT) {
> 
> Boqun, can you give this a try?
> 

Sure, I just build a kernel with your modification, and the guest kernel
booted without any problem, thanks!

> Thomas, feel free to fold this in.
> 

FWIW, feel free to add

Tested-by: Boqun Feng <boqun.feng@xxxxxxxxx>

, for the version with this change folded.

Regards,
Boqun

> Wei.