From: "Jürgen Groß" <jgross@suse.com>
To: "Roger Pau Monné" <roger.pau@citrix.com>
Cc: xen-devel@lists.xenproject.org, Jan Beulich <jbeulich@suse.com>,
Andrew Cooper <andrew.cooper3@citrix.com>, Wei Liu <wl@xen.org>,
George Dunlap <george.dunlap@citrix.com>,
Ian Jackson <iwj@xenproject.org>, Julien Grall <julien@xen.org>,
Stefano Stabellini <sstabellini@kernel.org>
Subject: Re: [PATCH v2 1/2] xen/x86: add nmi continuation framework
Date: Thu, 8 Oct 2020 10:58:40 +0200 [thread overview]
Message-ID: <e774053e-1f84-fd84-bafc-a2f254d70286@suse.com> (raw)
In-Reply-To: <20201008084306.GJ19254@Air-de-Roger>
On 08.10.20 10:43, Roger Pau Monné wrote:
> On Wed, Oct 07, 2020 at 03:30:10PM +0200, Juergen Gross wrote:
>> Actions in NMI context are rather limited as e.g. locking is rather
>> fragile.
>>
>> Add a generic framework to continue processing in softirq context after
>> leaving NMI processing. This is working for NMIs happening in guest
>> context as NMI exit handling will issue an IPI to itself in case a
>> softirq is pending, resulting in the continuation running before the
>> guest gets control again.
>
> There's already kind of a nmi callback framework using nmi_callback.
> I assume that moving this existing callback code to be executed in
> softirq context is not suitable because that would be past the
> execution of an iret instruction?
>
> Might be worth mentioning in the commit message.
>
>> Signed-off-by: Juergen Gross <jgross@suse.com>
>> ---
>> xen/arch/x86/traps.c | 37 +++++++++++++++++++++++++++++++++++++
>> xen/include/asm-x86/nmi.h | 8 +++++++-
>> xen/include/xen/softirq.h | 5 ++++-
>> 3 files changed, 48 insertions(+), 2 deletions(-)
>>
>> diff --git a/xen/arch/x86/traps.c b/xen/arch/x86/traps.c
>> index bc5b8f8ea3..f433fe5acb 100644
>> --- a/xen/arch/x86/traps.c
>> +++ b/xen/arch/x86/traps.c
>> @@ -1799,6 +1799,42 @@ void unset_nmi_callback(void)
>> nmi_callback = dummy_nmi_callback;
>> }
>>
>> +static DEFINE_PER_CPU(void (*)(void *), nmi_cont_func);
>> +static DEFINE_PER_CPU(void *, nmi_cont_par);
>> +
>> +static void nmi_cont_softirq(void)
>> +{
>> + unsigned int cpu = smp_processor_id();
>> + void (*func)(void *par) = per_cpu(nmi_cont_func, cpu);
>> + void *par = per_cpu(nmi_cont_par, cpu);
>
> I think you can use this_cpu here and below, and avoid having to worry
> about the processor id at all? Also less likely to mess up and assign
> a NMI callback to a wrong CPU.
this_cpu() and smp_processor_id() are both based on get_cpu_info().
So multiple uses of this_cpu() are less efficient than per_cpu() with
cpu in a local variable (regarding code-size and speed, at least I think
so).
>
>> +
>> + /* Reads must be done before following write (local cpu ordering only). */
>> + barrier();
>
> Is this added because of the usage of RELOC_HIDE by per_cpu?
This is added because reordering of loads by the compiler must be
avoided as a NMI using those fields again might happen anytime.
>
>> +
>> + per_cpu(nmi_cont_func, cpu) = NULL;
>
> Since we are using RELOC_HIDE, doesn't it imply that the compiler
> cannot reorder this, since it has no idea what variable we are
> accessing?
I'd rather be safe here than relying on per_cpu() internals.
>
>> +
>> + if ( func )
>> + func(par);
>> +}
>> +
>> +int set_nmi_continuation(void (*func)(void *par), void *par)
>> +{
>> + unsigned int cpu = smp_processor_id();
>> +
>> + if ( per_cpu(nmi_cont_func, cpu) )
>> + {
>> + printk("Trying to set NMI continuation while still one active!\n");
>> + return -EBUSY;
>> + }
>> +
>> + per_cpu(nmi_cont_func, cpu) = func;
>> + per_cpu(nmi_cont_par, cpu) = par;
>> +
>> + raise_softirq(NMI_CONT_SOFTIRQ);
>> +
>> + return 0;
>> +}
>> +
>> void do_device_not_available(struct cpu_user_regs *regs)
>> {
>> #ifdef CONFIG_PV
>> @@ -2132,6 +2168,7 @@ void __init trap_init(void)
>>
>> cpu_init();
>>
>> + open_softirq(NMI_CONT_SOFTIRQ, nmi_cont_softirq);
>> open_softirq(PCI_SERR_SOFTIRQ, pci_serr_softirq);
>> }
>>
>> diff --git a/xen/include/asm-x86/nmi.h b/xen/include/asm-x86/nmi.h
>> index a288f02a50..da40fb6599 100644
>> --- a/xen/include/asm-x86/nmi.h
>> +++ b/xen/include/asm-x86/nmi.h
>> @@ -33,5 +33,11 @@ nmi_callback_t *set_nmi_callback(nmi_callback_t *callback);
>> void unset_nmi_callback(void);
>>
>> DECLARE_PER_CPU(unsigned int, nmi_count);
>> -
>> +
>> +/**
>> + * set_nmi_continuation
>> + *
>> + * Schedule a function to be started in softirq context after NMI handling.
>> + */
>> +int set_nmi_continuation(void (*func)(void *par), void *par);
>
> I would introduce a type for the nmi callback, as I think it's easier
> than having to retype the prototype everywhere:
>
> typedef void nmi_continuation_t(void *);
Fine with me.
Juergen
next prev parent reply other threads:[~2020-10-08 8:59 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-10-07 13:30 [PATCH v2 0/2] xen/x86: implement NMI continuation as softirq Juergen Gross
2020-10-07 13:30 ` [PATCH v2 1/2] xen/x86: add nmi continuation framework Juergen Gross
2020-10-08 8:43 ` Roger Pau Monné
2020-10-08 8:58 ` Jürgen Groß [this message]
2020-10-07 13:30 ` [PATCH v2 2/2] xen/oprofile: use set_nmi_continuation() for sending virq to guest Juergen Gross
2020-10-15 10:49 ` [PATCH v2 0/2] xen/x86: implement NMI continuation as softirq Roger Pau Monné
2020-10-15 10:52 ` Jürgen Groß
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=e774053e-1f84-fd84-bafc-a2f254d70286@suse.com \
--to=jgross@suse.com \
--cc=andrew.cooper3@citrix.com \
--cc=george.dunlap@citrix.com \
--cc=iwj@xenproject.org \
--cc=jbeulich@suse.com \
--cc=julien@xen.org \
--cc=roger.pau@citrix.com \
--cc=sstabellini@kernel.org \
--cc=wl@xen.org \
--cc=xen-devel@lists.xenproject.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).