From: Jan Kiszka <jan.kiszka@web.de>
To: Blue Swirl <blauwirbel@gmail.com>
Cc: qemu-devel@nongnu.org, Gleb Natapov <gleb@redhat.com>,
Juan Quintela <quintela@redhat.com>
Subject: Re: [Qemu-devel] Re: [RFT][PATCH 07/15] qemu_irq: Add IRQ handlers with delivery feedback
Date: Sat, 29 May 2010 11:36:45 +0200 [thread overview]
Message-ID: <4C00E02D.9050804@web.de> (raw)
In-Reply-To: <AANLkTimvdjU1zunypgO1VikK3Tw47TDZyrtR7m1YC2o_@mail.gmail.com>
[-- Attachment #1: Type: text/plain, Size: 5786 bytes --]
Blue Swirl wrote:
>>> On the contrary, APIC is actually the only source of the IRQ ack
>>> information. RTC hack would not work without APIC (or the
>>> bidirectional IRQ) passing this info to RTC.
>>>
>>> What APIC doesn't have now is the timer frequency or period info. This
>>> is known by RTC and also higher levels managing the clocks.
>>>
>> So APIC has one bit of information and RTC everything else.
>
> The information known by RTC (timer period) is also known by higher levels.
Curious to see where you'll find this.
>
>> The current
>> approach (and proposed patch) brings this one bit of information to RTC,
>> you are arguing that RTC should be able to communicate all its info to
>> APIC. Sorry I don't see that your way has any advantage. Just more
>> complex interface and it is much easier to get it wrong for other time
>> sources.
>
> I don't think anymore that APIC should be handling this but the
> generic stuff, like vl.c or exec.c. Then there would be only
> information passing from APIC to higher levels.
You neglect the the information required to associate a periodic source
(e.g. RTC) with an IRQ sink (e.g. APIC). Without that, you will have a
hard time figuring out if a reported IRQ coalescing requires any
activities or should simply be welcomed (for I/O IRQs).
>
>>> I keep ignoring the idea that the current model, where both RTC and
>>> APIC must somehow work together to make coalescing work, is the only
>>> possible just because it is committed and it happens to work in some
>>> cases. It would be much better to concentrate this to one place, APIC
>>> or preferably higher level where it may benefit other timers too.
>>> Provided of course that the other models can be made to work.
>>>
>> So write the code and show us. You haven't show any evidence that RTC is
>> the wrong place. RTC knows when interrupt was acknowledge to RTC, it
>> know when clock frequency changes, it know when device reset happened.
>> APIC knows only that interrupt was coalesced. It doesn't even know that
>> it may be masked by a guest in IOAPIC (interrupts delivered while they
>> are masked not considered coalesced).
>
> Oh, I thought interrupt masking was the reason for coalescing! What
> exactly is the reason then?
Missing acks, ie. the IRQ is still pending when the next one arrives.
You want to filter out masked/suppressed IRQs to avoid running the
de-coalescing logic on sources that are actually cut off (like the RTC
IRQ when the HPET took over).
>
>> Time source knows only when
>> frequency changes and may be when device reset happens if timer is
>> stopped by device on reset. So RTC is actually a sweet spot if you want
>> to minimize amount of info you need to pass between various layers.
>>
>>>>> Maybe that version would not bend backwards as much as the current to
>>>>> cater for buggy hosts.
>>>>>
>>>> You mean "buggy guests"?
>>> Yes, sorry.
>>>
>>>> What guests are not buggy in your opinion?
>>>> Linux tries hard to be smart and as a result the only way to have stable
>>>> clock with it is to go paravirt.
>>> I'm not an OS designer, but I think an OS should never crash, even if
>>> a burst of IRQs is received. Reprogramming the timer should consider
>>> the pending IRQ situation (0 or 1 with real HW). Those bugs are one
>>> cause of the problem.
>> OS should never crash in the absence of HW bugs? I doubt you can design
>> an OS that can run in a face of any HW failure. Anyway here we are
>> trying to solve guests time keeping problem not crashes. Do you think
>> you can design OS that can keep time accurately no matter how crazy all
>> HW clock behaves?
>
> I think my OS design skills are not relevant in this discussion, but
> IIRC there are fault tolerant operating systems for extreme conditions
> so it can be done.
No one can influence the design of released OS versions anymore.
>
>>>>>> The fact is that timer device is not "just like any
>>>>>> other device" in virtual world. Any other device is easy: you just
>>>>>> implement spec as close as possible and everything works. For time
>>>>>> source device this is not enough. You can implement RTC+HPET to the
>>>>>> letter and your guest will drift like crazy.
>>>>> It's doable: a cycle accurate emulator will not cause any drift,
>>>>> without any voodoo. The interrupts would come after executing the same
>>>>> instruction as the real HW. For emulating any sufficiently buggy
>>>>> guests in any sufficiently desperate low resource conditions, this may
>>>>> be the only option that will always work.
>>>>>
>>>> Yes, but qemu and kvm are not cycle accurate emulators and don't strive
>>>> to be one. On the contrary KVM runs at native host CPU speed most of the
>>>> time, so any emulation done between two instruction is theoretically
>>>> noticeable for a guest. TSC is bypassed directly to a guest too, so
>>>> keeping all time source in perfect sync is also impossible.
>>> That is actually another cause of the problem. KVM gives the guest an
>>> illusion that the VCPU speed is equal to host speed. When they don't
>>> match, especially in critical code, there can be problems. It would be
>>> better to tell the guest a lower speed, which also can be guaranteed.
>>>
>> Not possible. It's that simple. You should take it into account in your
>> architecture design stage. In case of KVM real physical CPU executes guest
>> instruction and it does this as fast as it can. The only way we can hide
>> that from a guest is by intercepting each access to TSC and at that
>> point we can use bochs instead.
>
> Well, as Paul pointed out, there's also icount option.
Which is not available in virtualization mode.
Jan
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]
next prev parent reply other threads:[~2010-05-29 15:48 UTC|newest]
Thread overview: 122+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-05-24 20:13 [Qemu-devel] [RFT][PATCH 00/15] HPET cleanups, fixes, enhancements Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 01/15] hpet: Catch out-of-bounds timer access Jan Kiszka
2010-05-24 20:34 ` [Qemu-devel] " Juan Quintela
2010-05-24 20:36 ` Jan Kiszka
2010-05-24 20:50 ` Juan Quintela
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 02/15] hpet: Coding style cleanups and some refactorings Jan Kiszka
2010-05-24 20:37 ` [Qemu-devel] " Juan Quintela
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 03/15] hpet: Silence warning on write to running main counter Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 04/15] hpet: Move static timer field initialization Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 05/15] hpet: Convert to qdev Jan Kiszka
2010-05-25 9:37 ` Paul Brook
2010-05-25 10:14 ` Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 06/15] hpet: Start/stop timer when HPET_TN_ENABLE is modified Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 07/15] qemu_irq: Add IRQ handlers with delivery feedback Jan Kiszka
2010-05-25 6:07 ` Gleb Natapov
2010-05-25 6:31 ` Jan Kiszka
2010-05-25 6:40 ` Gleb Natapov
2010-05-25 6:54 ` Jan Kiszka
2010-05-25 19:09 ` [Qemu-devel] " Blue Swirl
2010-05-25 20:16 ` Anthony Liguori
2010-05-25 21:44 ` Jan Kiszka
2010-05-26 8:08 ` Gleb Natapov
2010-05-26 20:14 ` Blue Swirl
2010-05-27 5:42 ` Gleb Natapov
2010-05-26 19:55 ` Blue Swirl
2010-05-26 20:09 ` Jan Kiszka
2010-05-26 20:35 ` Blue Swirl
2010-05-26 22:35 ` Jan Kiszka
2010-05-26 23:26 ` Paul Brook
2010-05-27 17:56 ` Blue Swirl
2010-05-27 18:31 ` Jan Kiszka
2010-05-27 18:53 ` Blue Swirl
2010-05-27 19:08 ` Jan Kiszka
2010-05-27 19:19 ` Blue Swirl
2010-05-27 22:19 ` Jan Kiszka
2010-05-28 19:00 ` Blue Swirl
2010-05-30 12:00 ` Avi Kivity
2010-05-27 22:21 ` Paul Brook
2010-05-28 19:10 ` Blue Swirl
2010-05-27 22:21 ` Paul Brook
2010-05-27 6:13 ` Gleb Natapov
2010-05-27 18:37 ` Blue Swirl
2010-05-28 7:31 ` Gleb Natapov
2010-05-28 20:06 ` Blue Swirl
2010-05-28 20:47 ` Gleb Natapov
2010-05-29 7:58 ` Jan Kiszka
2010-05-29 9:35 ` Blue Swirl
2010-05-29 9:45 ` Jan Kiszka
2010-05-29 10:04 ` Blue Swirl
2010-05-29 10:16 ` Jan Kiszka
2010-05-29 10:26 ` Blue Swirl
2010-05-29 10:38 ` Jan Kiszka
2010-05-29 14:46 ` Gleb Natapov
2010-05-29 16:13 ` Blue Swirl
2010-05-29 16:37 ` Gleb Natapov
2010-05-29 21:21 ` Blue Swirl
2010-05-30 6:02 ` Gleb Natapov
2010-05-30 12:10 ` Blue Swirl
2010-05-30 12:24 ` Jan Kiszka
2010-05-30 12:58 ` Blue Swirl
2010-05-31 7:46 ` Jan Kiszka
2010-05-30 12:33 ` Gleb Natapov
2010-05-30 12:56 ` Blue Swirl
2010-05-30 13:49 ` Gleb Natapov
2010-05-30 16:54 ` Blue Swirl
2010-05-30 19:37 ` Blue Swirl
2010-05-30 20:07 ` Gleb Natapov
2010-05-30 20:21 ` Blue Swirl
2010-05-31 5:19 ` Gleb Natapov
2010-06-01 18:00 ` Blue Swirl
2010-06-01 18:30 ` Gleb Natapov
2010-06-02 19:05 ` Blue Swirl
2010-06-03 6:23 ` Jan Kiszka
2010-06-03 6:34 ` Gleb Natapov
2010-06-03 6:59 ` Jan Kiszka
2010-06-03 7:03 ` Gleb Natapov
2010-06-03 7:06 ` Gleb Natapov
2010-06-04 19:05 ` Blue Swirl
2010-06-05 0:04 ` Jan Kiszka
2010-06-05 7:20 ` Blue Swirl
2010-06-05 8:27 ` Jan Kiszka
2010-06-05 9:23 ` Blue Swirl
2010-06-05 12:14 ` Jan Kiszka
2010-06-06 7:15 ` Gleb Natapov
2010-06-06 7:39 ` Jan Kiszka
2010-06-06 7:49 ` Gleb Natapov
2010-06-06 8:07 ` Jan Kiszka
2010-06-06 9:23 ` Gleb Natapov
2010-06-06 10:10 ` Jan Kiszka
2010-06-06 10:27 ` Gleb Natapov
2010-06-06 7:39 ` Blue Swirl
2010-06-06 8:07 ` Gleb Natapov
2010-05-30 13:22 ` Blue Swirl
2010-05-29 9:15 ` Blue Swirl
2010-05-29 9:36 ` Jan Kiszka [this message]
2010-05-29 14:38 ` Gleb Natapov
2010-05-29 16:03 ` Blue Swirl
2010-05-29 16:32 ` Gleb Natapov
2010-05-29 20:52 ` Blue Swirl
2010-05-30 5:41 ` Gleb Natapov
2010-05-30 11:41 ` Blue Swirl
2010-05-30 11:52 ` Gleb Natapov
2010-05-30 12:05 ` Avi Kivity
2010-05-27 5:58 ` Gleb Natapov
2010-05-26 19:49 ` Blue Swirl
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 08/15] x86: Refactor RTC IRQ coalescing workaround Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 09/15] hpet/rtc: Rework RTC IRQ replacement by HPET Jan Kiszka
2010-05-25 9:29 ` Paul Brook
2010-05-25 10:23 ` Jan Kiszka
2010-05-25 11:05 ` Paul Brook
2010-05-25 11:19 ` Jan Kiszka
2010-05-25 11:23 ` Paul Brook
2010-05-25 11:26 ` Jan Kiszka
2010-05-25 12:03 ` Paul Brook
2010-05-25 12:39 ` Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 10/15] hpet: Drop static state Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 11/15] hpet: Add support for level-triggered interrupts Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 12/15] vmstate: Add VMSTATE_STRUCT_VARRAY_UINT8 Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 13/15] hpet: Make number of timers configurable Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 14/15] hpet: Add MSI support Jan Kiszka
2010-05-24 20:13 ` [Qemu-devel] [RFT][PATCH 15/15] monitor/QMP: Drop info hpet / query-hpet Jan Kiszka
2010-05-24 22:16 ` [Qemu-devel] [RFT][PATCH 00/15] HPET cleanups, fixes, enhancements Anthony Liguori
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=4C00E02D.9050804@web.de \
--to=jan.kiszka@web.de \
--cc=blauwirbel@gmail.com \
--cc=gleb@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=quintela@redhat.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).