6,534
edits
Changes
created (since 079.4)
== Changes ==
Since {{kernel link|rhel6-testing|042stab079.4}}:
<!-- 042stab080.2 -->
* [fs] limit writeback triggered by fadvise to the specified range should improve the performance
* [ext4] when writeback is done in context, which is followed by sync_fs, skip forced commit in ext4_write_inode, it will be done once. Again the performance should be improved. This feature can be disabled by "echo 0 > /sys/fs/ext4/batch_sync"
<!-- 042stab080.3 -->
* [stat] replaces global counters with per-cpu structures to avoid kstat_glb_lock congestion => performance increases
* [nfs/quota] NFS local quota (vzquota in case /vz over NFS) significantly reworked => no more crashes / stats skews are expected (PCLIN-31747)
* [ploop] kaio must merge bios for aio_kernel_submit => should avoid problem with massive 4k FUSE requests (PSBM-19207)
<!-- 042stab080.4 -->
* [perf] perf events syscall are forbidden now in Containers by default because they are very complicated and potentially vulnerable (use sysctl kernel.perf_event_paranoid = -1 to enable)
<!-- 042stab080.5 -->
<!-- 042stab080.6 -->
* [drv/video/r600] kernel WARNING is substituted with printk() in order not to taint the kernel for nothing (PSBM-21043)
* [mm/compaction] repair should_continue_reclaim logic, this fixes the memory reclaimer under high load (PSBM-20273)
<!-- 042stab080.7 -->
* [vfs] kernel panic when accessing usb storage device has been fixed (PSBM-21015)
* [vfs] kernel warning (and thus taint) by __mark_inode_dirty() function for /dev/zero has been fixed (PSBM-21206)
* [signal] signal pending flag could be lost due to a race between non-atomic bit-reset and set_tsk_thread_flag() (PCLIN-31992)
* [cpt] kernel panic after online migration could happen on destination node due to incorrect packets with SKB_GSO_TCP flag set drop. We can drop only packets with SKB_GSO_UDP, if we cannot dump a packet with SKB_GSO_TCP, need to abort the migration. (PSBM-21181)
* [vzquota] vzquota was mistakenly considered locked after major vzquota rework in previous 080.x kernels (PCLIN-32009)
Since {{kernel link|rhel6-testing|042stab079.4}}:
<!-- 042stab080.2 -->
* [fs] limit writeback triggered by fadvise to the specified range should improve the performance
* [ext4] when writeback is done in context, which is followed by sync_fs, skip forced commit in ext4_write_inode, it will be done once. Again the performance should be improved. This feature can be disabled by "echo 0 > /sys/fs/ext4/batch_sync"
<!-- 042stab080.3 -->
* [stat] replaces global counters with per-cpu structures to avoid kstat_glb_lock congestion => performance increases
* [nfs/quota] NFS local quota (vzquota in case /vz over NFS) significantly reworked => no more crashes / stats skews are expected (PCLIN-31747)
* [ploop] kaio must merge bios for aio_kernel_submit => should avoid problem with massive 4k FUSE requests (PSBM-19207)
<!-- 042stab080.4 -->
* [perf] perf events syscall are forbidden now in Containers by default because they are very complicated and potentially vulnerable (use sysctl kernel.perf_event_paranoid = -1 to enable)
<!-- 042stab080.5 -->
<!-- 042stab080.6 -->
* [drv/video/r600] kernel WARNING is substituted with printk() in order not to taint the kernel for nothing (PSBM-21043)
* [mm/compaction] repair should_continue_reclaim logic, this fixes the memory reclaimer under high load (PSBM-20273)
<!-- 042stab080.7 -->
* [vfs] kernel panic when accessing usb storage device has been fixed (PSBM-21015)
* [vfs] kernel warning (and thus taint) by __mark_inode_dirty() function for /dev/zero has been fixed (PSBM-21206)
* [signal] signal pending flag could be lost due to a race between non-atomic bit-reset and set_tsk_thread_flag() (PCLIN-31992)
* [cpt] kernel panic after online migration could happen on destination node due to incorrect packets with SKB_GSO_TCP flag set drop. We can drop only packets with SKB_GSO_UDP, if we cannot dump a packet with SKB_GSO_TCP, need to abort the migration. (PSBM-21181)
* [vzquota] vzquota was mistakenly considered locked after major vzquota rework in previous 080.x kernels (PCLIN-32009)
