xen-devel.lists.xenproject.org archive mirror
 help / color / mirror / Atom feed
From: Corneliu ZUZU <czuzu@bitdefender.com>
To: Julien Grall <julien.grall@arm.com>, xen-devel@lists.xen.org
Cc: Andre Przywara <andre.przywara@arm.com>,
	Stefano Stabellini <sstabellini@kernel.org>,
	Tamas K Lengyel <tamas@tklengyel.com>,
	Razvan Cojocaru <rcojocaru@bitdefender.com>,
	Steve Capper <Steve.Capper@arm.com>
Subject: Re: [PATCH 7/7] vm-event/arm: implement support for control-register write vm-events
Date: Fri, 24 Jun 2016 12:32:48 +0300	[thread overview]
Message-ID: <94545ef5-d527-5f23-83be-ca81ed966575@bitdefender.com> (raw)
In-Reply-To: <576BC3E1.1080707@arm.com>

On 6/23/2016 2:11 PM, Julien Grall wrote:
> Hello,
>
> On 23/06/16 06:49, Corneliu ZUZU wrote:
>> On 6/23/2016 8:31 AM, Corneliu ZUZU wrote:
>>> On 6/22/2016 10:41 PM, Julien Grall wrote:
>>>>
>>>>
>>>> On 22/06/2016 20:37, Corneliu ZUZU wrote:
>>>>> I've also realized that it's a bit complicated to avoid writing HCR
>>>>> from
>>>>> 2 places.
>>>>> That's because:
>>>>> - p2m_restore_state is part of the process of switching to another 
>>>>> vCPU
>>>>> and the HCR write _must be committed_ here because other components
>>>>> depend on that, like address-translation functions
>>>>> - I want vm_event_vcpu_enter to be called _after_ the switch to the
>>>>> vCPU
>>>>> is completed
>>>>> - I want HCR_TVM to be set in vm_event_vcpu_enter because setting
>>>>> necessary traps _for cr vm-events_ to work should be done there
>>>>> (setting
>>>>> HCR_TVM bit makes sense to be there and the purpose is to centralize
>>>>> operations such as this for code comprehensibility; also, on the X86
>>>>> counterpart a similar operation is done for trapping CR3, so it
>>>>> would be
>>>>> nice to keep the symmetry)
>>>>>
>>>>> Would it be such a stretch to have HCR written in 2 places? (the 
>>>>> second
>>>>> time happens rarely anyway: it's unlikely(..) to have to do the
>>>>> write in
>>>>> vm_event_vcpu_enter)
>>>>
>>>> Not really. It was mostly to avoid setting/clearing HCR bits in
>>>> different place in the code. It makes more difficult to know what is
>>>> the final result of the register.
>>>>
>>>> Anyway, let's skip it for now, if it is too difficult.
>>>>
>>>> Regards,
>>>>
>>>
>>> Then perhaps something like the following would be suitable:
>>>
>>> 1. store hcr in arch_domain (register_t hcr)
>>>
>>> 2. add a function in asm-arm/processor.h (or where else?) which only
>>> does:
>>>     static inline void update_hcr(struct domain *d)
>>>     {
>>>         WRITE_SYSREG(d->arch.hcr, HCR_EL2);
>>>         isb();
>>>     }
>>>
>>> 3.  modify p2m_restore_state to do:
>>>     n->domain->arch.hcr &= ~HCR_VM;
>>>     update_hcr(n->domain);
>>>     p2m_load_VTTBR(n->domain);
>>>
>>>     n->domain->arch.hcr |= HCR_VM;
>>>
>>>     if ( is_32bit_domain(n->domain) )
>>>         n->domain->arch.hcr &= ~HCR_RW;
>>>     else
>>>         n->domain->arch.hcr |= HCR_RW;
>
> This is not safe at all, p2m_restore_state is vCPU specific at you 
> modify domain information.
>
> However, if we store the hcr per domain, overriding every context 
> switch is pointless as the domain will always be 32-bit/64-bit.

Oh right, the RW bit needs not be set/unset anymore with this change.

>
>>>
>>>     update_hcr(n->domain);
>>>
>>>     WRITE_SYSREG(n->arch.sctlr, SCTLR_EL1);
>>>     isb();
>>>
>>> 4. and vcpu_enter_adjust_traps to
>>>
>>>     if ( unlikely(0 != v->domain->arch.monitor.write_ctrlreg_enabled) )
>>>     {
>>>          if ( likely(v->domain->arch.hcr & HCR_TVM) )
>>>              return;
>>>          v->domain->arch.hcr |= HCR_TVM;
>>>     }
>>>     else
>>>     {
>>>          if ( likely(!(v->domain->arch.hcr & HCR_TVM)) )
>>>              return;
>>>          v->domain->arch.hcr &= ~HCR_TVM;
>>>     }
>
> This does not need to be done in vcpu_enter_adjust_traps everytime. 
> You can set the bit in arch.hcr in DOMCTL_MONITOR_EVENT_WRITE_CTRLREG.

I wanted to keep X86-ARM symmetry and it seemed more intuitive to have 
these kind of adjustments with the vcpu_enter code motion. But now that 
I think about it, given the fact that we have the guarantee that after 
monitor_domctl and before reentering the vCPU p2m_restore_state gets 
called (due to domain_pause/domain_unpause) - thus actually committing 
the hcr update at the proper time - technically monitor_domctl _is_ the 
optimal place to set arch.hcr in. In conclusion, I'm thinking of 
discarding the entire idea of introducing vm_event_vcpu_enter, it seems 
to me now that this would also render a simpler code.

>
>>>
>>>     update_hcr(v->domain);
>>>
>>> That way at least it's easier to follow where update_hcr is called.
>
> I don't see much reason to store the value in the domain and have 
> multiple update_hcr. If we store the value, then we should only call 
> update_hcr once when returning to the guest.

Yep, that will happen with the above-mentioned discarding of 
vm_event_vcpu_enter idea.

>
>> And also set the initial value of HCR at the moment of creation, i.e. in
>> arch_domain_create as
>>
>> d->arch.hcr = READ_SYSREG(HCR_EL2)
>
> We control the value of HCR_EL2, so it would be better to assign the 
> list of flags here.

Right, that will happen too.

>
> Regards,
>

Thanks for the useful insights,
Corneliu.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

  reply	other threads:[~2016-06-24  9:32 UTC|newest]

Thread overview: 73+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-06-16 14:04 [PATCH 0/7] vm-event: Implement ARM support for control-register writes Corneliu ZUZU
2016-06-16 14:06 ` [PATCH 1/7] minor (formatting) fixes Corneliu ZUZU
2016-06-16 14:24   ` Jan Beulich
2016-06-16 19:19     ` Corneliu ZUZU
2016-06-17  7:06       ` Jan Beulich
2016-06-17 10:46         ` Corneliu ZUZU
2016-06-16 16:02   ` Tamas K Lengyel
2016-06-17  8:33     ` Corneliu ZUZU
2016-06-17  8:36       ` Razvan Cojocaru
2016-06-17  9:29         ` Andrew Cooper
2016-06-17  9:35           ` Jan Beulich
2016-06-17  9:33         ` Jan Beulich
2016-06-17  9:36           ` Razvan Cojocaru
2016-06-17  9:40             ` Jan Beulich
2016-06-17  9:42               ` Razvan Cojocaru
2016-06-17 19:05           ` Tamas K Lengyel
2016-06-16 14:07 ` [PATCH 2/7] vm-event: VM_EVENT_FLAG_DENY requires VM_EVENT_FLAG_VCPU_PAUSED Corneliu ZUZU
2016-06-16 16:11   ` Tamas K Lengyel
2016-06-17  8:43     ` Corneliu ZUZU
2016-06-21 11:26     ` Corneliu ZUZU
2016-06-21 15:09       ` Tamas K Lengyel
2016-06-22  8:34         ` Corneliu ZUZU
2016-06-16 14:08 ` [PATCH 3/7] vm-event: introduce vm_event_vcpu_enter Corneliu ZUZU
2016-06-16 14:51   ` Jan Beulich
2016-06-16 20:10     ` Corneliu ZUZU
2016-06-16 20:33       ` Razvan Cojocaru
2016-06-17 10:41         ` Corneliu ZUZU
2016-06-17  7:17       ` Jan Beulich
2016-06-17 11:13         ` Corneliu ZUZU
2016-06-17 11:27           ` Jan Beulich
2016-06-17 12:13             ` Corneliu ZUZU
2016-06-16 16:17   ` Tamas K Lengyel
2016-06-17  9:19     ` Corneliu ZUZU
2016-06-17  8:55   ` Julien Grall
2016-06-17 11:40     ` Corneliu ZUZU
2016-06-17 13:22       ` Julien Grall
2016-06-16 14:09 ` [PATCH 4/7] vm-event/x86: use vm_event_vcpu_enter properly Corneliu ZUZU
2016-06-16 15:00   ` Jan Beulich
2016-06-16 20:20     ` Corneliu ZUZU
2016-06-17  7:20       ` Jan Beulich
2016-06-17 11:23         ` Corneliu ZUZU
2016-06-16 16:27   ` Tamas K Lengyel
2016-06-17  9:24     ` Corneliu ZUZU
2016-06-16 14:10 ` [PATCH 5/7] x86: replace monitor_write_data.do_write with enum Corneliu ZUZU
2016-06-16 14:12 ` [PATCH 6/7] vm-event/arm: move hvm_event_cr->common vm_event_monitor_cr Corneliu ZUZU
2016-06-16 15:16   ` Jan Beulich
2016-06-17  8:25     ` Corneliu ZUZU
2016-06-17  8:38       ` Jan Beulich
2016-06-17 11:31         ` Corneliu ZUZU
2016-06-21  7:08       ` Corneliu ZUZU
2016-06-21  7:20         ` Jan Beulich
2016-06-21 15:22           ` Tamas K Lengyel
2016-06-22  6:33             ` Jan Beulich
2016-06-16 16:55   ` Tamas K Lengyel
2016-06-17 10:37     ` Corneliu ZUZU
2016-06-16 14:13 ` [PATCH 7/7] vm-event/arm: implement support for control-register write vm-events Corneliu ZUZU
2016-06-16 14:26   ` Julien Grall
2016-06-16 19:24     ` Corneliu ZUZU
2016-06-16 21:28       ` Julien Grall
2016-06-17 11:46         ` Corneliu ZUZU
2016-06-16 16:49   ` Julien Grall
2016-06-17 10:36     ` Corneliu ZUZU
2016-06-17 13:18       ` Julien Grall
2016-06-22 16:35       ` Corneliu ZUZU
2016-06-22 17:17         ` Julien Grall
2016-06-22 18:39           ` Corneliu ZUZU
2016-06-22 19:37             ` Corneliu ZUZU
2016-06-22 19:41               ` Julien Grall
2016-06-23  5:31                 ` Corneliu ZUZU
2016-06-23  5:49                   ` Corneliu ZUZU
2016-06-23 11:11                     ` Julien Grall
2016-06-24  9:32                       ` Corneliu ZUZU [this message]
2016-06-23 11:00           ` Julien Grall

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=94545ef5-d527-5f23-83be-ca81ed966575@bitdefender.com \
    --to=czuzu@bitdefender.com \
    --cc=Steve.Capper@arm.com \
    --cc=andre.przywara@arm.com \
    --cc=julien.grall@arm.com \
    --cc=rcojocaru@bitdefender.com \
    --cc=sstabellini@kernel.org \
    --cc=tamas@tklengyel.com \
    --cc=xen-devel@lists.xen.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).