[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <6efb18e38a7812b62652ce5c403455811336f063.camel@intel.com>
Date: Thu, 26 May 2022 04:00:49 +1200
From: Kai Huang <kai.huang@...el.com>
To: Dave Hansen <dave.hansen@...el.com>, linux-kernel@...r.kernel.org,
kvm@...r.kernel.org
Cc: seanjc@...gle.com, pbonzini@...hat.com, len.brown@...el.com,
tony.luck@...el.com, rafael.j.wysocki@...el.com,
reinette.chatre@...el.com, dan.j.williams@...el.com,
peterz@...radead.org, ak@...ux.intel.com,
kirill.shutemov@...ux.intel.com,
sathyanarayanan.kuppuswamy@...ux.intel.com,
isaku.yamahata@...el.com
Subject: Re: [PATCH v3 09/21] x86/virt/tdx: Get information about TDX module
and convertible memory
On Wed, 2022-05-25 at 16:57 +1200, Kai Huang wrote:
> On Wed, 2022-05-25 at 16:47 +1200, Kai Huang wrote:
> > On Fri, 2022-04-29 at 10:47 -0700, Dave Hansen wrote:
> > > On 4/28/22 16:14, Kai Huang wrote:
> > > > On Thu, 2022-04-28 at 07:06 -0700, Dave Hansen wrote:
> > > > > On 4/27/22 17:15, Kai Huang wrote:
> > > > > > > Couldn't we get rid of that comment if you did something like:
> > > > > > >
> > > > > > > ret = tdx_get_sysinfo(&tdx_cmr_array, &tdx_sysinfo);
> > > > > >
> > > > > > Yes will do.
> > > > > >
> > > > > > > and preferably make the variables function-local.
> > > > > >
> > > > > > 'tdx_sysinfo' will be used by KVM too.
> > > > >
> > > > > In other words, it's not a part of this series so I can't review whether
> > > > > this statement is correct or whether there's a better way to hand this
> > > > > information over to KVM.
> > > > >
> > > > > This (minor) nugget influencing the design also isn't even commented or
> > > > > addressed in the changelog.
> > > >
> > > > TDSYSINFO_STRUCT is 1024B and CMR array is 512B, so I don't think it should be
> > > > in the stack. I can change to use dynamic allocation at the beginning and free
> > > > it at the end of the function. KVM support patches can change it to static
> > > > variable in the file.
> > >
> > > 2k of stack is big, but it isn't a deal breaker for something that's not
> > > nested anywhere and that's only called once in a pretty controlled
> > > setting and not in interrupt context. I wouldn't cry about it.
> >
> > Hi Dave,
> >
> > I got below complaining when I use local variable for TDSYSINFO_STRUCT and CMR
> > array:
> >
> > arch/x86/virt/vmx/tdx/tdx.c:383:1: warning: the frame size of 3072 bytes is
> > larger than 1024 bytes [-Wframe-larger-than=]
> > 383 | }
> >
> > So I don't think we can use local variable for them. I'll still use static
> > variables to avoid dynamic allocation. In the commit message, I'll explain they
> > are too big to put into the stack, and KVM will need to use TDSYSINFO_STRUCT
> > reported by TDX module anyway.
> >
> > Let me know if you don't agree?
>
> Btw, CMR array alone can be put into the stack. It will never be used by KVM,
> so I'll put CMR array as local variable, but keep tdx_sysinfo as static
> variable.
>
Sorry for multiple emails about this. If I put CMR array to the stack, I still
sometimes get the build warning. So will use static variables.
Also, constructing TDMRs internally needs to use tdx_sysinfo (max_tdmrs,
pamt_entry_size, max_rsvd_per_tdmr), so with static variable they don't need to
be passed around as function arguments.
--
Thanks,
-Kai
Powered by blists - more mailing lists