Re: [PATCH RFC v1 06/18] x86/hyperv: allocate output arg pages if required

From: Vitaly Kuznetsov <vkuznets@redhat.com>
To: Wei Liu <wei.liu@kernel.org>
Cc: Wei Liu <wei.liu@kernel.org>,
	Linux on Hyper-V List <linux-hyperv@vger.kernel.org>,
	virtualization@lists.linux-foundation.org,
	Linux Kernel List <linux-kernel@vger.kernel.org>,
	Michael Kelley <mikelley@microsoft.com>,
	Vineeth Pillai <viremana@linux.microsoft.com>,
	Sunil Muthuswamy <sunilmut@microsoft.com>,
	Nuno Das Neves <nudasnev@microsoft.com>,
	Lillian Grassin-Drake <ligrassi@microsoft.com>,
	"K. Y. Srinivasan" <kys@microsoft.com>,
	Haiyang Zhang <haiyangz@microsoft.com>,
	Stephen Hemminger <sthemmin@microsoft.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
	"maintainer\:X86 ARCHITECTURE \(32-BIT AND 64-BIT\)"
	<x86@kernel.org>, "H. Peter Anvin" <hpa@zytor.com>
Subject: Re: [PATCH RFC v1 06/18] x86/hyperv: allocate output arg pages if required
Date: Wed, 16 Sep 2020 18:10:26 +0200	[thread overview]
Message-ID: <87zh5phfd9.fsf@vitty.brq.redhat.com> (raw)
In-Reply-To: <20200916154200.7nf74vjmqu3f6sfq@liuwe-devbox-debian-v2>

Wei Liu <wei.liu@kernel.org> writes:

> On Tue, Sep 15, 2020 at 12:43:18PM +0000, Wei Liu wrote:
>> On Tue, Sep 15, 2020 at 12:16:58PM +0200, Vitaly Kuznetsov wrote:
>> > Wei Liu <wei.liu@kernel.org> writes:
>> > 
>> > > When Linux runs as the root partition, it will need to make hypercalls
>> > > which return data from the hypervisor.
>> > >
>> > > Allocate pages for storing results when Linux runs as the root
>> > > partition.
>> > >
>> > > Signed-off-by: Lillian Grassin-Drake <ligrassi@microsoft.com>
>> > > Co-Developed-by: Lillian Grassin-Drake <ligrassi@microsoft.com>
>> > > Signed-off-by: Wei Liu <wei.liu@kernel.org>
>> > > ---
>> > >  arch/x86/hyperv/hv_init.c       | 45 +++++++++++++++++++++++++++++----
>> > >  arch/x86/include/asm/mshyperv.h |  1 +
>> > >  2 files changed, 41 insertions(+), 5 deletions(-)
>> > >
>> > > diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c
>> > > index cac8e4c56261..ebba4be4185d 100644
>> > > --- a/arch/x86/hyperv/hv_init.c
>> > > +++ b/arch/x86/hyperv/hv_init.c
>> > > @@ -45,6 +45,9 @@ EXPORT_SYMBOL_GPL(hv_vp_assist_page);
>> > >  void  __percpu **hyperv_pcpu_input_arg;
>> > >  EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
>> > >  
>> > > +void  __percpu **hyperv_pcpu_output_arg;
>> > > +EXPORT_SYMBOL_GPL(hyperv_pcpu_output_arg);
>> > > +
>> > >  u32 hv_max_vp_index;
>> > >  EXPORT_SYMBOL_GPL(hv_max_vp_index);
>> > >  
>> > > @@ -75,14 +78,29 @@ static int hv_cpu_init(unsigned int cpu)
>> > >  	u64 msr_vp_index;
>> > >  	struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
>> > >  	void **input_arg;
>> > > -	struct page *pg;
>> > > +	struct page *input_pg;
>> > >  
>> > >  	input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
>> > >  	/* hv_cpu_init() can be called with IRQs disabled from hv_resume() */
>> > > -	pg = alloc_page(irqs_disabled() ? GFP_ATOMIC : GFP_KERNEL);
>> > > -	if (unlikely(!pg))
>> > > +	input_pg = alloc_page(irqs_disabled() ? GFP_ATOMIC : GFP_KERNEL);
>> > > +	if (unlikely(!input_pg))
>> > >  		return -ENOMEM;
>> > > -	*input_arg = page_address(pg);
>> > > +	*input_arg = page_address(input_pg);
>> > > +
>> > > +	if (hv_root_partition) {
>> > > +		struct page *output_pg;
>> > > +		void **output_arg;
>> > > +
>> > > +		output_pg = alloc_page(irqs_disabled() ? GFP_ATOMIC :
>> > >  	GFP_KERNEL);
>> > 
>> > To simplify the code, can we just rename 'input_arg' to 'hypercall_args'
>> > and do alloc_pages(rqs_disabled() ? GFP_ATOMIC : GFP_KERNEL, 1) to
>> > allocate two pages above?
>> 
>> It should be doable, but I need to code it up and test it to be 100%
>> sure.
>> 
>
> I switch to alloc_pages to allocate an order of 2 page if necessary, but
> I keep input_arg and output_arg separate because I want to avoid code
> churn in other places.
>

My idea was that we're free to choose how to use these pages, e.g. with
two pages allocated we can now do a hypercall which takes two pages of
input and produces no output other than the return value. This doesn't
contradict your suggestion to keep input_arg/output_arg as these are
just pointers to the continuous space, we can still do the trick.

I don't feel strong about this and you probably have more patches in
your stash, no need for massive re-work I believe. 

-- 
Vitaly