On Mon, Apr 15, 2024 at 04:17:35PM +0800, Xiaoyao Li <xiaoyao.li@xxxxxxxxx> wrote: > On 2/26/2024 4:25 PM, isaku.yamahata@xxxxxxxxx wrote: > > ... > > > + > > + kvm_tdx->tdcs_pa = tdcs_pa; > > + for (i = 0; i < tdx_info->nr_tdcs_pages; i++) { > > + err = tdh_mng_addcx(kvm_tdx->tdr_pa, tdcs_pa[i]); > > + if (err == TDX_RND_NO_ENTROPY) { > > + /* Here it's hard to allow userspace to retry. */ > > + ret = -EBUSY; > > So userspace is expected to stop creating TD and quit on this? > > If so, it exposes an DOS attack surface that malicious users in another can > drain the entropy with busy-loop on RDSEED. > > Can you clarify why it's hard to allow userspace to retry? To me, it's OK to > retry that "teardown" cleans everything up, and userspace and issue the > KVM_TDX_INIT_VM again. The current patch has complicated error recovery path. After simplifying the code, it would be possible to return -EAGAIN in this patch. For the retry case, we need to avoid TDH.MNG.CREATE() and TDH.MNG.KEY.CONFIG(). -- Isaku Yamahata <isaku.yamahata@xxxxxxxxx>