[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5f8aaa652cc112fc61b9b13b6d77b998a2461172.camel@intel.com>
Date: Fri, 22 Mar 2024 04:33:21 +0000
From: "Huang, Kai" <kai.huang@...el.com>
To: "Yamahata, Isaku" <isaku.yamahata@...el.com>
CC: "Zhang, Tina" <tina.zhang@...el.com>, "isaku.yamahata@...ux.intel.com"
<isaku.yamahata@...ux.intel.com>, "seanjc@...gle.com" <seanjc@...gle.com>,
"Yuan, Hang" <hang.yuan@...el.com>, "sean.j.christopherson@...el.com"
<sean.j.christopherson@...el.com>, "Chen, Bo2" <chen.bo@...el.com>,
"sagis@...gle.com" <sagis@...gle.com>, "isaku.yamahata@...il.com"
<isaku.yamahata@...il.com>, "Aktas, Erdem" <erdemaktas@...gle.com>,
"kvm@...r.kernel.org" <kvm@...r.kernel.org>, "pbonzini@...hat.com"
<pbonzini@...hat.com>, "Yao, Yuan" <yuan.yao@...el.com>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"binbin.wu@...ux.intel.com" <binbin.wu@...ux.intel.com>
Subject: Re: [PATCH v19 029/130] KVM: TDX: Add C wrapper functions for
SEAMCALLs to the TDX module
> >
> > So how about we have some macros:
> >
> > static inline bool is_seamcall_err_kernel_defined(u64 err)
> > {
> > return err & TDX_SW_ERROR;
> > }
> >
> > #define TDX_KVM_SEAMCALL(_kvm, _seamcall_func, _fn, _args) \
> > ({ \
> > u64 _ret = _seamcall_func(_fn, _args);
> > KVM_BUG_ON(_kvm, is_seamcall_err_kernel_defined(_ret));
> > _ret;
> > })
>
> As we can move out KVM_BUG_ON() to the call site, we can simply have
> seamcall() or seamcall_ret().
> The call site has to check error. whether it is TDX_SW_ERROR or not.
> And if it hit the unexpected error, it will mark the guest bugged.
How many call sites are we talking about?
I think handling KVM_BUG_ON() in macro should be able to eliminate bunch of
individual KVM_BUG_ON()s in these call sites?
>
>
> > #define tdx_kvm_seamcall(_kvm, _fn, _args) \
> > TDX_KVM_SEAMCALL(_kvm, seamcall, _fn, _args)
> >
> > #define tdx_kvm_seamcall_ret(_kvm, _fn, _args) \
> > TDX_KVM_SEAMCALL(_kvm, seamcall_ret, _fn, _args)
> >
> > #define tdx_kvm_seamcall_saved_ret(_kvm, _fn, _args) \
> > TDX_KVM_SEAMCALL(_kvm, seamcall_saved_ret, _fn, _args)
> >
> > This is consistent with what we have in TDX host code, and this handles
> > NO_ENTROPY error internally.
> >
> >
[...]
> >
> > > Because only TDH.MNG.CREATE() and TDH.MNG.ADDCX() can return TDX_RND_NO_ENTROPY, > we can use __seamcall(). The TDX spec doesn't guarantee such error code
> > > convention. It's very unlikely, though.
> >
> > I don't quite follow the "convention" part. Can you elaborate?
> >
> > NO_ENTROPY is already handled in seamcall() variants. Can we just use them
> > directly?
>
> I intended for bad code generation. If the loop on NO_ENTRY error harms the
> code generation, we might be able to use __seamcall() or __seamcall_ret()
> instead of seamcall(), seamcall_ret().
This doesn't make sense to me.
Firstly, you have to *prove* the loop generates worse code.
Secondly, if it does generate worse code, and we care about it, we should fix it
in the host seamcall() code. No?
Powered by blists - more mailing lists