Iqela le Abaphandi abavela kwiDyunivesithi yaseVirginia naseCalifornia bazise uhlobo olutsha lokuhlaselwa kulwakhiwo lobuchwephesha beeprosesa Intel kunye ne-AMD.
Indlela yokuhlasela ecetywayo inxulunyaniswa nokusetyenziswa kwendawo yokugcina efihlakeleyo esecaleni yemisebenzi emincinci (i-micro-op cache) kwiiprosesa, ezinokusetyenziselwa ukukhupha ulwazi oluzinzileyo ekuqhubekeni kwengqikelelo yomyalelo.
Kuyaqapheleka ukuba indlela entsha iphumeze uhlaselo lweSpecter v1 ngokwentsebenzo, yenza kubenzima ukubona uhlaselo kwaye aluthintelwanga ziindlela ezikhoyo zokuzikhusela kuhlaselo ngamajelo asecaleni ayilelwe ukuthintela ubungozi obubangelwa kukwenziwa kwemiyalelo yokucinga.
Umzekelo, ukusetyenziswa kwesiteyitimenti se-LFENCE kuthintela ukuvuza kumanqanaba okugqibela okuphunyezwa kwentelekelelo, kodwa ayikhuseli ekuvuzeni ngenxa yolwakhiwo oluncinci.
Indlela ichaphazela iimodeli zeprosesa ye-Intel kunye ne-AMD ekhutshwe ukusukela ngo-2011 kubandakanya uluhlu lwe-Intel Skylake kunye ne-AMD Zen. ezigcinwe kwindawo efihlakeleyo eyahlukileyo.
Le cache yahluke ngokusisiseko kwizikhephe ezikumgangatho ophezulu, ayifikeleleki ngokuthe ngqo kwaye isebenza njenge-buffer yomjelo ukufikelela ngokukhawuleza kwiziphumo zokucacisa imiyalelo ye-CISC kwi-RISC microinstruction.
Nangona kunjalo, abaphandi Ndifumene indlela yokwenza iimeko ezivelayo ngexesha lokungqubana kwe-cache kunye nokuvumela ukugweba umxholo we-cache yemisebenzi emincinci ngokuhlalutya umahluko kwixesha lokuphunyezwa kwezenzo ezithile.
I-cache ye-micro-op kwiiprosesa ze-Intel yahlulwe ngokuhambelana nemisonto yeCPU (I-Hyper-Threading), ngelixa iiprosesa I-AMD Zen isebenzisa i-cache ekwabelwana ngayo, edala iimeko zokuvuza kwedatha kungekuphela kwintambo enye yokuphunyezwa, kodwa naphakathi kwemisonto eyahlukeneyo kwi-SMT (ukuvuza kwedatha kunokwenzeka phakathi kwekhowudi eqhuba ngeendlela ezahlukeneyo ze-CPU).
Abaphandi bacebise indlela esisiseko ukufumanisa utshintsho kwindawo yokugcina ii-micro-ops kunye neemeko ezahlukeneyo zokuhlaselwa ezivumela ukuba zenze iindlela zokuhambisa idatha ezifihlakeleyo kunye nokusebenzisa ikhowudi esesichengeni ukucoca idatha eyimfihlo, kokubini kwinkqubo enye (umzekelo, ukuhlela inkqubo yokuvuza kwedatha xa uqhuba okwesithathu ikhowudi yeqela kwiinjini zeJIT kunye noomatshini ababonakalayo) naphakathi kwekernel kunye neenkqubo kwindawo yomsebenzisi.
Ngokubeka into eyahlukileyo kuhlaselo lweSpecter kusetyenziswa i-cache encinci ye-op, abaphandi bakwazile ukufezekisa ukuphuma kwe-965.59 Kbps ngezinga lempazamo le-0.22% kunye ne-785.56 Kbps xa usebenzisa ukulungiswa kwempazamo, kwimeko yokuvuza ngaphakathi kwimemori efanayo isithuba. iidilesi. kunye nenqanaba lelungelo.
Ngokuvuza kwamanqanaba ngamalungelo awodwa (phakathi kwekernel kunye nendawo yomsebenzisi), ukuphuma kwaba yi-85,2 Kbps ngokulungiswa kwempazamo eyongeziweyo kunye ne-110,96 Kbps enepesenti ye-4%.
Xa uhlasela iprosesa ye-AMD Zen, edala ukuvuza phakathi kwe-cores ezahlukeneyo ze-CPU, ukuphuma kwaba yi-250 Kbps ngezinga lempazamo le-5,59% kunye ne-168,58 Kbps ngokulungiswa kwempazamo. Xa kuthelekiswa nendlela yakudala yeSpecter v1, uhlaselo olutsha luye lwangamaxesha ama-2,6 ngokukhawuleza.
Ukunciphisa uhlaselo lwecache ye-micro-op kulindeleke ukuba lifune utshintsho olungaphezulu kokuthoba intsebenzo kunangexesha lokukhuselwa kweSpecter.
Njengolungelelwaniso olululo, kuyacetyiswa ukuba kuthintelwe olo hlaselo hayi ngokukhubaza i-caching, kodwa kwinqanaba lokujonga okungalunganga kunye nokumisela imeko ye-cache yohlaselo.
Njengasekuhlaselweni kweSpecter, Ukulungiselela ukuvuza kwe-kernel okanye ezinye iinkqubo kufuna ukwenziwa kweskripthi esithile (iijethi) kwicala lexhoba kwiinkqubo, zikhokelela ekwenziweni kwengqikelelo yemiyalelo.
Zijikeleze i-100 yezixhobo zifunyenwe kwi-kernel ye-Linux kwaye ziya kususwa, kodwa izisombululo ziyafumaneka rhoqo ukuzenza, umzekelo ezo zinxulunyaniswa nokwazisa iinkqubo ze-BPF eziyilwe ngokukodwa kwi-kernel.
Gqibela ukuba unomdla wokwazi okungakumbi ngayo, ungajonga iinkcukacha Kule khonkco ilandelayo.