From: Yu Zhang <yu.c.zhang@linux.intel.com>
To: Jan Beulich <JBeulich@suse.com>
Cc: Kevin Tian <kevin.tian@intel.com>,
George Dunlap <george.dunlap@eu.citrix.com>,
Andrew Cooper <andrew.cooper3@citrix.com>,
xen-devel@lists.xen.org, Paul Durrant <paul.durrant@citrix.com>,
zhiyuan.lv@intel.com, Jun Nakajima <jun.nakajima@intel.com>
Subject: Re: [PATCH v5 4/4] x86/ioreq server: Reset outstanding p2m_ioreq_server entries when an ioreq server unmaps.
Date: Tue, 9 Aug 2016 18:21:09 +0800 [thread overview]
Message-ID: <57A9AE95.9050300@linux.intel.com> (raw)
In-Reply-To: <57A9C2710200007800104216@prv-mh.provo.novell.com>
On 8/9/2016 5:45 PM, Jan Beulich wrote:
>>>> On 09.08.16 at 11:25, <yu.c.zhang@linux.intel.com> wrote:
>> On 8/9/2016 4:13 PM, Jan Beulich wrote:
>>>>>> On 09.08.16 at 09:39, <yu.c.zhang@linux.intel.com> wrote:
>>>> On 8/9/2016 12:29 AM, Jan Beulich wrote:
>>>>>>>> On 12.07.16 at 11:02, <yu.c.zhang@linux.intel.com> wrote:
>>>>>> @@ -5512,6 +5513,12 @@ static int hvmop_set_mem_type(
>>>>>> if ( rc )
>>>>>> goto out;
>>>>>>
>>>>>> + if ( t == p2m_ram_rw && memtype[a.hvmmem_type] == p2m_ioreq_server )
>>>>>> + p2m->ioreq.entry_count++;
>>>>>> +
>>>>>> + if ( t == p2m_ioreq_server && memtype[a.hvmmem_type] == p2m_ram_rw )
>>>>>> + p2m->ioreq.entry_count--;
>>>>>> +
>>>>> These (and others below) happen, afaict, outside of any lock, which
>>>>> can't be right.
>>>> How about we make this entry_count as atomic_t and use atomic_inc/dec
>>>> instead?
>>> That's certainly one of the options, but beware of overflow.
>> Oh, thanks for your remind, Jan. I just found that atomic_t is defined
>> as "typedef struct { int counter; } atomic_t;"
>> which do have overflow issues if entry_count is supposed to be a uint32
>> or uint64.
>>
>> Now I have another thought: the entry_count is referenced in 3
>> different scenarios:
>> 1> the hvmop handlers - hvmop_set_mem_type() and
>> hvmop_map_mem_type_to_ioreq_server(),
>> which are triggered by device model, and will not be concurrent. And
>> hvmop_set_mem_type() will
>> permit the mem type change only when an ioreq server has already been
>> mapped to this type.
> You shouldn't rely on a well behaved dm, and that's already
> leaving aside the question whether there couldn't even be well
> behaved use cases of parallel invocations of this op.
Oh. This is a good point. Thanks. :-)
>> 2> the misconfiguration handlers - resolve_misconfig() or do_recalc(),
>> which are triggered by HVM's
>> vm exit, yet this will only happen after the ioreq server has already
>> been unmapped. This means the
>> accesses to the entry_count will not be concurrent with the above hvmop
>> handlers.
> This case may be fine, but not for (just) the named reason:
> Multiple misconfig invocations can happen at the same time, but
> presumably your addition is inside the p2m-locked region (you'd
> have to double check that).
Yes, both are wrapped in p2m-locked region. So this give me another thought:
if we put the increment/decrement of entry_count inside
p2m_change_type_one(),
which has the p2m_lock/p2m_unlock, we could avoid introducing another
spinlock
and can also avoid the overflow possibility by using atomic_t.
[snip]
Thanks
Yu
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel
next prev parent reply other threads:[~2016-08-09 10:21 UTC|newest]
Thread overview: 32+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-07-12 9:02 [PATCH v5 0/4] x86/ioreq server: Introduce HVMMEM_ioreq_server mem type Yu Zhang
2016-07-12 9:02 ` [PATCH v5 1/4] x86/ioreq server: Rename p2m_mmio_write_dm to p2m_ioreq_server Yu Zhang
2016-07-12 9:02 ` [PATCH v5 2/4] x86/ioreq server: Add new functions to get/set memory types Yu Zhang
2016-07-12 9:02 ` [PATCH v5 3/4] x86/ioreq server: Add HVMOP to map guest ram with p2m_ioreq_server to an ioreq server Yu Zhang
2016-08-08 15:40 ` Jan Beulich
2016-08-09 7:39 ` Yu Zhang
2016-08-09 8:11 ` Jan Beulich
2016-08-09 8:20 ` Paul Durrant
2016-08-09 8:51 ` Yu Zhang
2016-08-09 9:07 ` Paul Durrant
2016-08-10 8:09 ` Yu Zhang
2016-08-10 10:33 ` Jan Beulich
2016-08-10 10:43 ` Paul Durrant
2016-08-10 12:32 ` Yu Zhang
2016-08-10 10:43 ` Yu Zhang
2016-08-11 8:47 ` Yu Zhang
2016-08-11 8:58 ` Jan Beulich
2016-08-11 9:19 ` Yu Zhang
2016-07-12 9:02 ` [PATCH v5 4/4] x86/ioreq server: Reset outstanding p2m_ioreq_server entries when an ioreq server unmaps Yu Zhang
2016-08-08 16:29 ` Jan Beulich
2016-08-09 7:39 ` Yu Zhang
2016-08-09 8:13 ` Jan Beulich
2016-08-09 9:25 ` Yu Zhang
2016-08-09 9:45 ` Jan Beulich
2016-08-09 10:21 ` Yu Zhang [this message]
2016-08-16 13:35 ` George Dunlap
2016-08-16 13:54 ` Jan Beulich
2016-08-30 12:17 ` Yu Zhang
2016-09-02 11:00 ` Yu Zhang
2016-08-05 2:44 ` [PATCH v5 0/4] x86/ioreq server: Introduce HVMMEM_ioreq_server mem type Yu Zhang
2016-08-05 6:16 ` Jan Beulich
2016-08-05 12:46 ` George Dunlap
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=57A9AE95.9050300@linux.intel.com \
--to=yu.c.zhang@linux.intel.com \
--cc=JBeulich@suse.com \
--cc=andrew.cooper3@citrix.com \
--cc=george.dunlap@eu.citrix.com \
--cc=jun.nakajima@intel.com \
--cc=kevin.tian@intel.com \
--cc=paul.durrant@citrix.com \
--cc=xen-devel@lists.xen.org \
--cc=zhiyuan.lv@intel.com \
/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).