[PATCHv3 10/14] x86/tdx: Convert shared memory back to private on kexec

[kirill.shutemov at linux.intel.com]

On Wed, Nov 22, 2023 at 10:07:44AM +0000, Huang, Kai wrote:
> On Tue, 2023-11-21 at 12:58 +0300, kirill.shutemov at linux.intel.com wrote:
> > On Tue, Nov 21, 2023 at 03:46:41AM +0000, Huang, Kai wrote:
> > > 
> > > >  
> > > > +static atomic_t conversions_in_progress;
> > > > +static bool conversion_allowed = true;
> > > > 
> > > 
> > > [...]
> > > 
> > > >  /* Used while preparing memory map entries for second kernel */
> > > >  struct crash_memmap_data {
> > > > @@ -107,6 +108,9 @@ void native_machine_crash_shutdown(struct pt_regs *regs)
> > > >  
> > > >  	crash_smp_send_stop();
> > > >  
> > > > +	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
> > > > +		x86_platform.guest.enc_kexec_unshare_mem(true);
> > > > +
> > > >  	cpu_emergency_disable_virtualization();
> > > >  
> > > >  	/*
> > > > diff --git a/arch/x86/kernel/reboot.c b/arch/x86/kernel/reboot.c
> > > > index 830425e6d38e..9fb302562bfd 100644
> > > > --- a/arch/x86/kernel/reboot.c
> > > > +++ b/arch/x86/kernel/reboot.c
> > > > @@ -12,6 +12,7 @@
> > > >  #include <linux/delay.h>
> > > >  #include <linux/objtool.h>
> > > >  #include <linux/pgtable.h>
> > > > +#include <linux/kexec.h>
> > > >  #include <acpi/reboot.h>
> > > >  #include <asm/io.h>
> > > >  #include <asm/apic.h>
> > > > @@ -31,6 +32,7 @@
> > > >  #include <asm/realmode.h>
> > > >  #include <asm/x86_init.h>
> > > >  #include <asm/efi.h>
> > > > +#include <asm/tdx.h>
> > > >  
> > > >  /*
> > > >   * Power off function, if any
> > > > @@ -716,6 +718,9 @@ static void native_machine_emergency_restart(void)
> > > >  
> > > >  void native_machine_shutdown(void)
> > > >  {
> > > > +	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT) && kexec_in_progress)
> > > > +		x86_platform.guest.enc_kexec_unshare_mem(false);
> > > > +
> > > >  	/* Stop the cpus and apics */
> > > >  #ifdef CONFIG_X86_IO_APIC
> > > >  	/*
> > > 
> > > In native_machine_crash_shutdown() the "unshare" is called after
> > > crash_smp_send_stop(), but here it's called before the stop_other_cpus().
> > > 
> > > I am wondering if you call "unshare" after stop_other_cpus(), can we guarantee
> > > there's only one cpu running in both normal and crash kexec in which case you
> > > might be able to get rid of the "conversions_in_progress" and
> > > "conversion_allowed" above?
> > 
> > For normal kexec we need to keep other CPUs going so they can finish
> > conversion cleanly and get us to the known state. Note that the conversion
> > is not atomic wrt preemption.
> 
> Yeah makes sense.
> 
> Add a comment to explain this?  That would help to remind us when we look at the
> code like years later.

Okay, will do.

> > In crash scenario we do the best effort: detect if we race with conversion
> > and report. We cannot really wait for conversions to finish: we get into
> > crash path with IRQs disabled and in generally unknown state.
> > 
> 
> Maybe it's good idea to add a TDVMCALL to query page status from the TDX module?
> In that case we can avoid the inaccuracy when looking at the page table.

No. TDX module has no control over what memory is shared. Shared-EPT is in
full control of VMM and therefore we have to track unshared memory on the
guest side to make sure that VMM doesn't do silly things.

-- 
  Kiryl Shutsemau / Kirill A. Shutemov