From: Avi Kivity <avi@redhat.com>
To: Gleb Natapov <gleb@redhat.com>
Cc: kvm@vger.kernel.org, linux-mm@kvack.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 03/11] Handle asynchronous page fault in a PV guest.
Date: Mon, 02 Nov 2009 14:38:30 +0200 [thread overview]
Message-ID: <4AEED2C6.6030508@redhat.com> (raw)
In-Reply-To: <1257076590-29559-4-git-send-email-gleb@redhat.com>
On 11/01/2009 01:56 PM, Gleb Natapov wrote:
> Asynchronous page fault notifies vcpu that page it is trying to access
> is swapped out by a host. In response guest puts a task that caused the
> fault to sleep until page is swapped in again. When missing page is
> brought back into the memory guest is notified and task resumes execution.
>
> diff --git a/arch/x86/include/asm/kvm_para.h b/arch/x86/include/asm/kvm_para.h
> index 90708b7..61e2aa3 100644
> --- a/arch/x86/include/asm/kvm_para.h
> +++ b/arch/x86/include/asm/kvm_para.h
> @@ -52,6 +52,9 @@ struct kvm_mmu_op_release_pt {
>
> #define KVM_PV_SHM_FEATURES_ASYNC_PF (1<< 0)
>
> +#define KVM_PV_REASON_PAGE_NP 1
> +#define KVM_PV_REASON_PAGE_READY 2
>
_NOT_PRESENT would improve readability.
> +static void apf_task_wait(struct task_struct *tsk, u64 token)
> {
> + u64 key = hash_64(token, KVM_TASK_SLEEP_HASHBITS);
> + struct kvm_task_sleep_head *b =&async_pf_sleepers[key];
> + struct kvm_task_sleep_node n, *e;
> + DEFINE_WAIT(wait);
> +
> + spin_lock(&b->lock);
> + e = _find_apf_task(b, token);
> + if (e) {
> + /* dummy entry exist -> wake up was delivered ahead of PF */
> + hlist_del(&e->link);
> + kfree(e);
> + spin_unlock(&b->lock);
> + return;
> + }
> +
> + n.token = token;
> + init_waitqueue_head(&n.wq);
> + hlist_add_head(&n.link,&b->list);
> + spin_unlock(&b->lock);
> +
> + for (;;) {
> + prepare_to_wait(&n.wq,&wait, TASK_UNINTERRUPTIBLE);
> + if (hlist_unhashed(&n.link))
> + break;
>
Don't you need locking here? At least for the implied memory barriers.
> + schedule();
> + }
> + finish_wait(&n.wq,&wait);
> +
> + return;
> +}
> +
> +static void apf_task_wake(u64 token)
> +{
> + u64 key = hash_64(token, KVM_TASK_SLEEP_HASHBITS);
> + struct kvm_task_sleep_head *b =&async_pf_sleepers[key];
> + struct kvm_task_sleep_node *n;
> +
> + spin_lock(&b->lock);
> + n = _find_apf_task(b, token);
> + if (!n) {
> + /* PF was not yet handled. Add dummy entry for the token */
> + n = kmalloc(sizeof(*n), GFP_ATOMIC);
> + if (!n) {
> + printk(KERN_EMERG"async PF can't allocate memory\n");
>
Worrying. We could have an emergency pool of one node per cpu, and
disable apf if we use it until it's returned. But that's a lot of
complexity for an edge case, so a simpler solution would be welcome.
> +int kvm_handle_pf(struct pt_regs *regs, unsigned long error_code)
> +{
> + u64 reason, token;
> struct kvm_vcpu_pv_shm *pv_shm =
> per_cpu(kvm_vcpu_pv_shm, smp_processor_id());
>
> if (!pv_shm)
> - return;
> + return 0;
> +
> + reason = pv_shm->reason;
> + pv_shm->reason = 0;
> +
> + token = pv_shm->param;
> +
> + switch (reason) {
> + default:
> + return 0;
> + case KVM_PV_REASON_PAGE_NP:
> + /* real page is missing. */
> + apf_task_wait(current, token);
> + break;
> + case KVM_PV_REASON_PAGE_READY:
> + apf_task_wake(token);
> + break;
> + }
>
Ah, reason is not a bitmask but an enumerator. __u32 is more friendly
to i386 in that case.
Much of the code here is arch independent and would work well on non-x86
kvm ports. But we can always lay the burden of moving it on the non-x86
maintainers.
--
error compiling committee.c: too many arguments to function
next prev parent reply other threads:[~2009-11-02 12:38 UTC|newest]
Thread overview: 55+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-11-01 11:56 [PATCH 00/11] KVM: Add asynchronous page fault for PV guest Gleb Natapov
2009-11-01 11:56 ` [PATCH 01/11] Add shared memory hypercall to PV Linux guest Gleb Natapov
2009-11-02 4:27 ` Rik van Riel
2009-11-02 7:07 ` Gleb Natapov
2009-11-02 12:18 ` Avi Kivity
2009-11-02 16:18 ` Gleb Natapov
2009-11-03 5:15 ` Avi Kivity
2009-11-03 7:16 ` Gleb Natapov
2009-11-03 7:40 ` Avi Kivity
2009-11-01 11:56 ` [PATCH 02/11] Add "handle page fault" PV helper Gleb Natapov
2009-11-02 9:22 ` Ingo Molnar
2009-11-02 16:04 ` Gleb Natapov
2009-11-02 16:12 ` Ingo Molnar
2009-11-02 16:22 ` Gleb Natapov
2009-11-02 16:29 ` Ingo Molnar
2009-11-02 16:31 ` Gleb Natapov
2009-11-02 17:42 ` Gleb Natapov
2009-11-08 11:36 ` Ingo Molnar
2009-11-08 12:43 ` Avi Kivity
2009-11-08 12:51 ` Ingo Molnar
2009-11-08 13:01 ` Avi Kivity
2009-11-08 13:05 ` Ingo Molnar
2009-11-08 13:08 ` Avi Kivity
2009-11-08 16:44 ` H. Peter Anvin
2009-11-08 16:47 ` Ingo Molnar
2009-11-02 19:03 ` Rik van Riel
2009-11-02 19:33 ` Avi Kivity
2009-11-02 23:35 ` Rik van Riel
2009-11-03 4:57 ` Avi Kivity
2009-11-05 6:44 ` Tian, Kevin
2009-11-05 8:22 ` Avi Kivity
2009-11-01 11:56 ` [PATCH 03/11] Handle asynchronous page fault in a PV guest Gleb Natapov
2009-11-02 12:38 ` Avi Kivity [this message]
2009-11-02 15:54 ` Gleb Natapov
2009-11-03 14:14 ` Marcelo Tosatti
2009-11-03 14:25 ` Gleb Natapov
2009-11-03 14:32 ` Marcelo Tosatti
2009-11-03 14:38 ` Avi Kivity
2009-11-01 11:56 ` [PATCH 04/11] Export __get_user_pages_fast Gleb Natapov
2009-11-02 9:23 ` Ingo Molnar
2009-11-01 11:56 ` [PATCH 05/11] Add get_user_pages() variant that fails if major fault is required Gleb Natapov
2009-11-02 19:05 ` Rik van Riel
2009-11-01 11:56 ` [PATCH 06/11] Inject asynchronous page fault into a guest if page is swapped out Gleb Natapov
2009-11-02 12:56 ` Avi Kivity
2009-11-02 15:41 ` Gleb Natapov
2009-11-01 11:56 ` [PATCH 07/11] Retry fault before vmentry Gleb Natapov
2009-11-02 13:03 ` Avi Kivity
2009-11-01 11:56 ` [PATCH 08/11] Add "wait for page" hypercall Gleb Natapov
2009-11-02 13:05 ` Avi Kivity
2009-11-02 15:13 ` Gleb Natapov
2009-11-02 15:19 ` Avi Kivity
2009-11-01 11:56 ` [PATCH 09/11] Maintain preemptability count even for !CONFIG_PREEMPT kernels Gleb Natapov
2009-11-02 9:24 ` Ingo Molnar
2009-11-01 11:56 ` [PATCH 10/11] Handle async PF in non preemptable context Gleb Natapov
2009-11-01 11:56 ` [PATCH 11/11] Send async PF when guest is not in userspace too Gleb Natapov
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=4AEED2C6.6030508@redhat.com \
--to=avi@redhat.com \
--cc=gleb@redhat.com \
--cc=kvm@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.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).