ProHoster > Блог > labaran intanet > Linux 6.3 kernel saki

Linux 6.3 kernel saki

Bayan watanni biyu na haɓakawa, Linus Torvalds ya saki Linux 6.3 kwaya. Daga cikin manyan canje-canje masu mahimmanci: tsaftace tsararren dandamali na ARM da direbobi masu hoto, ci gaba da haɗin kai na tallafin harshen Rust, mai amfani hwnoise, goyan bayan tsarin bishiyar ja-baƙar fata a cikin BPF, BIG TCP yanayin don IPv4, ginanniyar alamar Dhrystone, ikon kashewa. kisa a cikin memfd, tallafi don ƙirƙirar direbobin HID ta amfani da BPF, an yi canje-canje ga Btrfs don rage rarrabuwar rukuni.

Sabuwar sigar ta karɓi gyaran 15637 daga masu haɓaka 2055; girman faci - 76 MB (canji ya shafi fayilolin 14296, 1023183 na lambar da aka ƙara, layukan 883103 sun goge). Don kwatanta, a cikin sigar da ta gabata, an gabatar da gyare-gyaren 16843 daga masu haɓaka 2178; girman faci - 62 MB. Kimanin kashi 39% na duk canje-canjen da aka gabatar a cikin kernel na 6.3 suna da alaƙa da direbobin na'ura, kusan kashi 15% na canje-canjen suna da alaƙa da sabunta lambar musamman ga kayan gine-ginen kayan aiki, 10% suna da alaƙa da tarin hanyar sadarwa, 5% zuwa tsarin fayil, da 3. % zuwa tsarin kernel na ciki.

Mabuɗin sabbin abubuwa a cikin kernel 6.3:

  • Ƙwaƙwalwar ajiya da sabis na tsarin
    • An gudanar da gagarumin tsaftace lambar da ke da alaƙa da tsoffin allon ARM da ba a yi amfani da su ba, wanda ya rage girman tushen kwaya ta layin 150. An cire tsoffin dandamali na ARM sama da 40.
    • An aiwatar da ikon ƙirƙirar direbobi don na'urorin shigarwa tare da haɗin HID (Na'urar Interface Device) wanda aka aiwatar ta hanyar shirye-shiryen BPF.
    • Ci gaba da jigilar kaya daga reshen Rust-for-Linux na ƙarin ayyuka masu alaƙa da amfani da Rust azaman harshe na biyu don haɓaka direbobi da samfuran kwaya. Ba a kunna goyan bayan tsatsa ta tsohuwa, kuma baya haifar da shigar da tsatsa a matsayin dogaron ginin kwaya na tilas. Ayyukan da aka bayar a cikin abubuwan da suka gabata an ƙara su tare da goyan bayan nau'ikan Arc (aiwatar da masu nuni tare da ƙididdigar ƙididdiga), ScopeGuard (ana yin tsaftacewa lokacin da ba a iya yin amfani da shi), da kuma ForeignOwnable (yana ba da motsi mai nuni tsakanin C da Rust code) . Module da aka cire ' aro' daga kunshin 'alloc' (nau'in 'Saniya' da sifa 'ToOwned'). An lura cewa yanayin goyon baya ga Rust a cikin kwaya ya riga ya kusa fara karɓar nau'ikan farko da aka rubuta a cikin Rust a cikin kwaya.
    • Yanayin mai amfani Linux (mai gudanar da kernel azaman tsarin mai amfani) akan tsarin x86-64 yana goyan bayan lambar da aka rubuta a cikin Rust. Ƙara goyon baya don gina Linux-yanayin mai amfani ta amfani da dangi tare da haɓaka haɓaka lokacin haɗin gwiwa (LTO).
    • Ƙara hwnoise mai amfani don bin diddigin jinkirin da ya haifar da halayen kayan masarufi. Ana ƙididdige ɓarna a lokacin aiwatar da ayyuka (jitter) lokacin da aka kashe aikin katsewa, wanda ya wuce microsecond ɗaya a cikin mintuna 10 na lissafin.
    • An ƙara ƙirar kernel tare da aiwatar da ma'auni na Dhrystone wanda za'a iya amfani dashi don kimanta aikin CPU a cikin jeri ba tare da abubuwan da ke amfani da sararin samaniya ba (misali, a matakin jigilar kayayyaki don sabbin SoCs waɗanda ke aiwatar da lodin kwaya kawai).
    • Ƙara "cgroup.memory=nobpf" zaɓin layin umarni na kernel don musaki lissafin yawan ƙwaƙwalwar ajiya don shirye-shiryen BPF, wanda zai iya zama da amfani ga tsarin tare da keɓaɓɓen kwantena.
    • Don shirye-shiryen BPF, an gabatar da aiwatar da tsarin bayanan bishiyar ja-baki, wanda ke amfani da kfunc + kptr (bpf_rbtree_add, bpf_rbtree_remove, bpf_rbtree_first) maimakon ƙara sabon nau'in taswira.
    • A cikin tsarin da za'a iya sake kunnawa (rseq, jerin sake kunnawa) sun kara da ikon wuce masu gano kisa daidai gwargwado (ID ɗin memory-map concurrency) wanda aka gano tare da lambar CPU zuwa matakai. Rseq yana ba da hanyar aiwatar da ayyuka cikin sauri ta atomatik wanda, idan wani zaren ya katse shi, ana tsaftacewa kuma ana sake gwadawa.
    • Masu sarrafawa na ARM suna goyan bayan umarnin SME 2 (Scalable Matrix Extension).
    • Don gine-ginen s390x da RISC-V RV64, ana aiwatar da goyan bayan tsarin "BPF trampoline", wanda ke ba da damar rage girman kai yayin canja wurin kira tsakanin kernel da shirye-shiryen BPF.
    • A kan tsarin da na'urori masu sarrafawa bisa tsarin gine-gine na RISC-V, an aiwatar da amfani da umarnin "ZBB" don hanzarta ayyukan kirtani.
    • Don tsarin da ya danganta da tsarin tsarin koyarwa na LoongArch (amfani da shi a cikin na'urori na Loongson 3 5000 da aiwatar da sabon RISC ISA mai kama da MIPS da RISC-V), goyan bayan bazuwar adireshin kernel (KASLR), ƙaura na kernel a ƙwaƙwalwar ajiya (matsar da wuri). ), hardware maki tsayawa da kprobe inji.
    • Tsarin DAMOS (Data Access Monitoring-based Operation Schemes), wanda ke ba ka damar 'yantar da ƙwaƙwalwar ajiya dangane da yawan damar ƙwaƙwalwar ajiya, tana tallafawa masu tacewa don ware wasu wuraren ƙwaƙwalwar ajiya daga aiki a cikin DAMOS.
    • Ƙananan ma'auni na ɗakin karatu na C Nolibc yana aiwatar da tallafi don gine-ginen s390 da saitin koyarwar Arm Thumb1 (ban da goyan baya ga ARM, AArch64, i386, x86_64, RISC-V da MIPS).
    • An inganta Objtool don haɓaka ginin kwaya da rage yawan amfani da ƙwaƙwalwar ajiya yayin ginawa (lokacin gina kernel a cikin yanayin "allyesconfig", babu matsaloli tare da tilasta dakatarwar matakai akan tsarin tare da 32 GB na RAM).
    • An daina ba da tallafi ga haɗa kernel ta Intel ICC compiler, wanda ya daɗe ba ya aiki kuma babu wanda ya nuna sha'awar gyara shi.
  • Tsarin Disk, I/O da Tsarin Fayil
    • tmpfs yana goyan bayan taswirar ID ɗin mai amfani da tsarin fayil ɗin da aka ɗora, wanda ake amfani da shi don taswirar takamaiman fayilolin mai amfani akan ɓangaren waje da aka ɗora zuwa wani mai amfani akan tsarin na yanzu.
    • A cikin Btrfs, don rage rarrabuwar rukunonin tubalan, ana rarraba madaidaicin ta girman lokacin rarraba tubalan, watau. duk wani rukuni na tubalan yanzu an iyakance shi ga ƙananan (har zuwa 128KB), matsakaici (har zuwa 8MB), da kuma babba. An sake fasalin aiwatar da hari56. Lambar da aka sake tsarawa don duba kuɗin kuɗi. An inganta haɓaka aiki don haɓaka aikin aika har zuwa sau 10 ta hanyar caching utime don kundayen adireshi da aiwatar da umarni kawai lokacin da ake buƙata. 10x ayyukan fiemap cikin sauri ta hanyar tsallake binciken bayanan baya don bayanan da aka raba (snapshots). Ana haɓaka ayyuka tare da metadata da kashi XNUMX% ta haɓaka binciken maɓalli a cikin tsarin itacen b.
    • Inganta aikin ext4 FS ta hanyar barin matakai da yawa don aiwatar da ayyukan I/O kai tsaye a lokaci guda zuwa tubalan da aka riga aka ware ta amfani da makullai na inode da aka raba maimakon keɓaɓɓen makullai.
    • A cikin f2fs, an yi aiki don inganta iya karanta lambar. Kafaffen batutuwa masu mahimmanci da suka shafi rubutun atomic da sabon ma'auni.
    • EROFS (Ingantattun Tsarin Fayil na Karatu-Kawai), wanda aka ƙera don amfani akan ɓangarorin karantawa kawai, yana aiwatar da ikon ɗaure ayyukan lalata fayil ɗin da aka matsa zuwa CPU don rage jinkirin samun damar bayanai.
    • Mai tsara jadawalin I/O na BFQ ya ƙara tallafi don faifan faifai na ci gaba, kamar waɗanda ke amfani da fayafai daban-daban daban (Multi Actuator).
    • Taimako don ɓoye bayanan ta amfani da AES-SHA2 algorithm an ƙara zuwa aiwatar da abokin ciniki na NFS da uwar garken.
    • Ƙarshen tsarin FUSE (Filesystems In User Space) ya ƙara goyan baya ga tsarin haɓaka tambaya, wanda ke ba ka damar saka ƙarin bayani a cikin tambayar. Dangane da wannan fasalin, ana aiwatar da ƙarin abubuwan gano rukuni zuwa buƙatun FS, waɗanda suka wajaba don la'akari da haƙƙin samun dama yayin ƙirƙirar abubuwa a cikin FS (ƙirƙira, mkdir, symlink, mknod).
  • Hankali da Tsaro
    • KVM hypervisor don tsarin x86 yana ƙara goyan baya don tsawaita kiran hyper-V kuma yana ba su da isarwa zuwa mai kulawa da ke gudana a cikin mahalli mai masaukin sararin samaniya. Canjin ya ba da damar aiwatar da goyan baya don ƙaddamar da hypervisor Hyper-V.
    • KVM yana sauƙaƙa don ƙuntata tsarin baƙo zuwa ga PMU (Performance Monitor Unit) abubuwan da suka shafi ma'aunin aiki.
    • Tsarin memfd, wanda ke ba ku damar gano wurin ƙwaƙwalwar ajiya ta hanyar bayanin fayil ɗin da aka wuce tsakanin matakai, an ƙara ikon ƙirƙirar wuraren da aka haramta aiwatar da lambar (wanda ba a aiwatar da memfd) kuma ba shi yiwuwa a saita haƙƙin aiwatarwa a cikin nan gaba.
    • An ƙara sabon aiki na PR_SET_MDWE prctl don toshe ƙoƙarin haɗa haƙƙin samun damar ƙwaƙwalwar ajiya wanda ke ba da damar rubutu da aiwatarwa a lokaci guda.
    • Ƙara da kunna ta tsohuwa kariya daga hare-haren na Specter, aiwatar da tushen IBRS (Ingantattun Hasashen Ƙuntataccen Ƙimar Ƙimar Kai tsaye) Yanayin atomatik wanda aka gabatar a cikin na'urori masu sarrafawa na AMD Zen 4, wanda ke ba ku damar daidaitawa da kuma musaki aiwatar da ƙididdiga na umarni yayin katsewa. sarrafawa, kiran tsarin, da maɓallan mahallin. Kariyar da aka tsara tana haifar da ƙananan sama idan aka kwatanta da kariyar Retpoline.
    • Yana magance raunin da zai iya ƙetare kariyar harin Specter v2 lokacin amfani da fasahar multithreading na lokaci guda (SMT ko Hyper-stringing) kuma wanda ya haifar da kashe STIBP (Masu tsinkaya Tsakanin Tsare-tsare Guda ɗaya) yayin zabar yanayin kariyar IBRS.
    • Don tsarin tushen ARM64, an ƙara sabon maƙasudin gina "virtconfig", wanda, lokacin da aka zaɓa, yana kunna mafi ƙarancin saiti na abubuwan haɗin kernel da ake buƙata don taya a cikin tsarin haɓakawa.
    • Ƙara goyon baya don tace kiran tsarin ta amfani da tsarin seccomp don gine-ginen m68k.
    • Ƙara goyon baya ga na'urorin CRB TPM2 (Command Response Buffer) na'urori masu sarrafawa na AMD Ryzen dangane da fasahar Microsoft Pluton.
  • Tsarin hanyar sadarwa
    • An ƙara haɗin haɗin yanar gizo don saita PLCA (Physical Layer Collision Avoidance) sublayer, wanda aka bayyana a cikin ƙayyadaddun IEEE 802.3cg-2019 kuma an yi amfani da shi a cikin 802.3cg (10Base-T1S) hanyoyin sadarwar Ethernet da aka inganta don haɗa na'urorin IoT da tsarin masana'antu. Amfani da PLCA yana haɓaka aiki akan hanyoyin sadarwar Ethernet da aka raba.
    • An daina goyan bayan API ɗin "marasa waya" don sarrafa hanyoyin sadarwa mara waya ta WiFi 7 (802.11be), tunda wannan API ɗin baya ɗaukar duk saitunan da suka dace. Lokacin ƙoƙarin amfani da API na "marasa waya", wanda ke ci gaba da samun tallafi azaman abin kwaikwaya, yanzu za a nuna gargaɗi don yawancin na'urori na yanzu.
    • An shirya cikakkun takardu akan netlink API (don masu haɓaka kernel da masu haɓaka aikace-aikacen sararin samaniya). An aiwatar da mai amfani na ynl-gen-c don samar da lambar C bisa la'akari da ƙayyadaddun YAML na ka'idar Netlink.
    • An ƙara goyan bayan zaɓin IP_LOCAL_PORT_RANGE zuwa soket ɗin hanyar sadarwa don sauƙaƙa daidaita hanyoyin haɗin da ke fita ta hanyar masu fassarar adireshi ba tare da amfani da SNAT ba. Lokacin amfani da adireshin IP iri ɗaya akan runduna da yawa, IP_LOCAL_PORT_RANGE yana bawa kowane mai watsa shiri damar amfani da nasa kewayon tashoshin sadarwa masu fita, kuma akan ƙofa don tura fakiti bisa lambobi na tashar jiragen ruwa.
    • Don MPTCP (MultiPath TCP), ana aiwatar da ikon sarrafa rafukan gauraye waɗanda ke amfani da ka'idojin IPv4 da IPv6. MPTCP wani tsawo ne na ƙa'idar TCP don tsara aikin haɗin TCP tare da isar da fakiti a lokaci guda tare da hanyoyi da yawa ta hanyoyin mu'amalar cibiyar sadarwa daban-daban waɗanda ke ɗaure zuwa adiresoshin IP daban-daban.
    • Don IPv4, ana aiwatar da yuwuwar yin amfani da tsawo na BIG TCP, wanda ke ba da damar haɓaka matsakaicin girman fakitin TCP har zuwa 4 GB don haɓaka aikin manyan cibiyoyin sadarwa na ciki na cibiyoyin bayanai. Wannan haɓakar girman fakiti tare da filin taken 16-bit yana samuwa ta hanyar aiwatar da fakitin "jumbo" waɗanda ke da girman taken IP da aka saita zuwa 0 da ainihin girman da aka watsa a cikin wani filin 32-bit daban a cikin keɓaɓɓen rubutun da aka haɗe.
    • An ƙara sabon siga na sysctl default_rps_mask, ta inda zaku iya saita tsohowar RPS (Karɓi Jagoran Fakiti), wanda ke da alhakin rarraba sarrafa zirga-zirgar zirga-zirgar da ke shigowa a cikin kwatancen CPU a matakin mai katsewa.
    • An dakatar da goyan bayan layin layi don taƙaita CBQ (jere-jere na aji), ATM (ATM Virtual circuits), dsmark (alamar sabis daban-daban), tcindex (ƙirar sarrafa zirga-zirga), da RSVP (ka'idar ajiyar albarkatu) zirga-zirga. An yi watsi da waɗannan fannonin na dogon lokaci kuma babu wanda ya yarda ya ci gaba da tallafawa.
  • Kayan aiki
    • An cire duk direbobin zane-zane na DRI1: i810 (tsohuwar Intel 8xx hadedde katunan zane), mga (Matrox GPU), r128 (ATI Rage 128 GPU gami da Rage Fury, XPERT 99 da katunan XPERT 128), savage (S3 Savage GPU), sis ( Crusty SiS GPU), tdfx (3dfx Voodoo) da ta (VIA IGP), waɗanda aka yanke a cikin 2016 kuma ba a samun tallafi a Mesa tun 2012.
    • An cire tsoffin direbobin framebuffer (fbdev) omap1, s3c2410, tmiofb da w100fb.
    • An ƙara direban DRM don VPU (Tsarin Gudanar da Mahimmanci) wanda aka haɗa cikin Intel Meteor Lake (ƙarni na 14) CPUs, wanda aka ƙera don haɓaka hangen nesa na kwamfuta da ayyukan koyon injin. Ana aiwatar da direba ta hanyar amfani da tsarin "accel", da nufin samar da tallafi don masu haɓaka lissafi, waɗanda za'a iya ba da su duka a cikin nau'ikan ASIC daban daban kuma azaman toshe IP a cikin SoC da GPU.
    • Direban i915 (Intel) yana faɗaɗa tallafi don katunan zane mai hankali na Intel Arc (DG2/Alchemist), yana ba da tallafi na farko don Meteor Lake GPUs, kuma ya haɗa da tallafi ga Intel Xe HP 4tile GPUs.
    • Direban amdgpu yana ƙara goyan baya don fasahar AdaptiveSync da ikon amfani da Tsararren Nuni tare da nuni da yawa. Tallafin da aka sabunta don DCN 3.2 (Nuna Core Next), SR-IOV RAS, VCN RAS, SMU 13.x da DP 2.1.
    • Ƙara tallafi don SM8350, SM8450 SM8550, SDM845 da dandamali na SC8280XP zuwa direban msm (GPU Qualcomm Adreno).
    • Direban Nouveau ya daina goyan bayan tsohon kiran ioctl.
    • An ƙara goyan bayan gwaji don NPU VerSilicon (VeriSilicon Neural Network Processor) zuwa direban etnaviv.
    • An aiwatar da direban pata_parport don tutocin IDE da aka haɗa ta tashar tashar layi ɗaya. Direban da aka ƙara ya ba da damar cire tsohon direban PARIDE daga kernel da haɓaka tsarin ATA. Ƙayyadaddun sabon direban shine rashin iya haɗa firinta da faifai a lokaci guda ta hanyar tashar jiragen ruwa.
    • Ƙara direban ath12k don katunan mara waya dangane da kwakwalwan Qualcomm tare da goyan bayan Wi-Fi 7. Ƙara goyon baya don katunan mara waya dangane da kwakwalwan kwamfuta na RealTek RTL8188EU.
    • Ƙara goyon baya ga allon 46 dangane da gine-ginen ARM64, gami da Samsung Galaxy tab A (2015), Samsung Galaxy S5, BananaPi R3, Debix Model A, EmbedFire LubanCat 1/2, Facebook Greatlakes, Orange Pi R1 Plus, Tesla FSD, da na'urori dangane da SoC Qualcomm MSM8953 (Snapdragon 610), SM8550 (Snapdragon 8 Gen 2), SDM450 da SDM632, Akwatin TV na Rockchips RK3128, RV1126 Vision, RK3588, RK3568, RK3566, RK3588, RK3328, 3, RK642, da RK654 AM 68/AM69 / AMXNUMX).

A lokaci guda kuma, Gidauniyar Software ta Kyauta ta Latin Amurka ta samar da bambance-bambancen kwaya na 6.3 - Linux-libre 6.3-gnu gaba daya, an share su daga firmware da abubuwan direbobi waɗanda ke ɗauke da abubuwan da ba su da kyauta ko sassan lambobi, wanda iyakar iyakar ta iyakance. masana'anta. A cikin sakin 6.3, an tsabtace ɓangarorin a cikin sabon ath12k, aw88395, da direbobin peb2466, da kuma a cikin sabbin fayilolin kayan aiki don na'urorin qcom na tushen AArch64. Lambobin tsaftacewa da aka sabunta a cikin direbobi da tsarin amdgpu, xhci-rcar, qcom-q6v5-pas, sp8870, av7110, da kuma a cikin direbobi don katunan DVB tare da ƙaddamar da software kuma a cikin fayilolin BPF da aka riga aka tattara. Dakatar da direbobin mga, r128, tm6000, cpia2 da r8188eu yayin da aka cire su daga kwaya. Ingantacciyar tsaftacewa ta i915 direba.

source: budenet.ru

Add a comment