Skip to content

chore: backports for 1.12.8#1552

Merged
talos-bot merged 14 commits into
siderolabs:release-1.12from
shanduur:backports/1.12.8
May 18, 2026
Merged

chore: backports for 1.12.8#1552
talos-bot merged 14 commits into
siderolabs:release-1.12from
shanduur:backports/1.12.8

Conversation

frezbo and others added 10 commits May 18, 2026 12:52
Bump kernel to 6.18.25.

Signed-off-by: Noel Georgi <git@frezbo.dev>
(cherry picked from commit 6a53a93)
Drop the [RFC PATCH net-next 0/3] series 'net: macb: candidate fixes
for silent TX stall on BCM2712/RP1' into kernel/build/patches/.

Patches mirror the netdev submission byte-for-byte:

  0001-net-macb-flush-PCIe-posted-write-after-TSTART-doorbe.patch
        Read-back of NCR after TSTART so the doorbell reaches the
        MAC before the function returns.

  0002-net-macb-re-check-ISR-after-IER-re-enable-in-macb_tx.patch
        Re-reads ISR after IER re-enable in macb_tx_poll() to catch
        TCOMP raised inside the IDR/IER mask window.

  0003-net-macb-add-TX-stall-watchdog-as-defence-in-depth-s.patch
        Per-queue delayed_work safety net that calls macb_tx_restart()
        if tx_tail hasn't advanced for >= 1 s while the ring is
        non-empty.

All three apply cleanly to the v6.18.24 tarball already pulled by
Pkgfile.

Lore thread:
https://lore.kernel.org/netdev/cover.1777064117.git.lukasz@raczylo.com/T/

Issue invitation: siderolabs/sbc-raspberrypi#91

Signed-off-by: Lukasz Raczylo <lukasz@raczylo.com>
Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>
(cherry picked from commit ca3599f)
…DME.md

Cross-reference the three macb patches added in this PR from the
kernel-patches README so the patches table stays in sync with the
files dropped under kernel/build/patches/.

Signed-off-by: Lukasz Raczylo <lukasz@raczylo.com>
Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>
(cherry picked from commit 9a718f6)
Backportable change.

Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>
(cherry picked from commit e192574)
Per siderolabs/talos#13276. Two cherry-picks from Ilpo Järvinen,
both merged to mainline v6.19 and candidates for 6.18.y stable
backport:

  - dc4b4d04e1ca PCI: Prevent shrinking bridge window from required size
  - 1ee4716a5a28 PCI: Fix premature removal from realloc_head

Apply cleanly to v6.18.24 and build successfully. On ASUS X99-E WS
with dual Radeon AI PRO R9700, the platform's PCIe MMIO window is
too small to assign two 32 GB BARs simultaneously regardless of
these patches; this is a board-level address space constraint, not
a regression. The fixes are submitted for the general
large-BAR / eGPU case where MMIO space suffices.

Signed-off-by: YANG JOO WOONG <yjw.cpa@gmail.com>
Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>
(cherry picked from commit 2265fc9)
Using the latest LTS.

Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>
(cherry picked from commit 06ff9dc)
Bump kernel to 6.18.28

Signed-off-by: Noel Georgi <git@frezbo.dev>
(cherry picked from commit dfb09f0)
Bump kernel to 6.18.29

Signed-off-by: Noel Georgi <git@frezbo.dev>
(cherry picked from commit d0c5480)
Update the three net-macb silent-TX-stall patches from RFC v1 to
PATCH net-next v2.  The v2 series is on lore at:

  https://lore.kernel.org/netdev/20260514215459.36109-1-lukasz@raczylo.com/T/

v2 changes from v1 (already merged in siderolabs#1526):

  * 0001 (PCIe posted-write flush after TSTART doorbell) - now gated
    behind a new MACB_CAPS_PCIE_POSTED_WRITES capability, set only on
    raspberrypi_rp1_config.  v1 applied the readback to every macb
    variant; SoC-integrated parts (Atmel, Microchip, SiFive, Xilinx)
    have no fabric posted-write concern and were paying the
    non-posted-read latency for nothing.

  * 0002 (PCIe read barrier before TX completion descriptor check) -
    replaces the v1 form, which was a regression on read-clear ISR
    silicon.  v1 read ISR with a TCOMP mask in macb_tx_poll(); on
    raspberrypi_rp1_config (where MACB_CAPS_ISR_CLEAR_ON_WRITE is
    not set) that read consumes every bit set in ISR, but the
    use-site masks down to TCOMP and discards the rest -- any
    RCOMP / ROVR / TXUBR bit at that instant is silently consumed.
    v2 replaces with (void)queue_readl(queue, IMR), the read-only
    mask mirror -- non-destructive, same PCIe-barrier effect.

  * 0003 (TX stall watchdog) - tracks tail movement via a bool flag
    set by macb_tx_complete() instead of a tx_tail snapshot
    (form suggested by Phil Elwell on raspberrypi/linux#7340).
    Adds a netif_carrier_ok() gate.  Wraps netdev_warn in
    printk_ratelimit() so operators can count occurrences while
    bounding log noise.  (An earlier draft used the macro
    netdev_warn_ratelimited(), which does not exist in this
    kernel -- caught by John Laur's build test on the v2 patches.)

Production runtime so far: 24-node Pi 5 fleet on v2 patch-2
IMR-barrier form since 2026-05-14 14:00 UTC, ~190 cumulative
node-hours, zero mid-runtime TX stalls.  Pre-patch baseline
(~0.5 stall/node-hour) would have predicted ~95 stalls; observed 0.

Related:
  * netdev v1 RFC thread:    https://lore.kernel.org/netdev/cover.1777064117.git.lukasz@raczylo.com/T/
  * netdev v2 series:        https://lore.kernel.org/netdev/20260514215459.36109-1-lukasz@raczylo.com/T/
  * raspberrypi/linux merge: raspberrypi/linux#7340
  * raspberrypi/linux v2 PR: raspberrypi/linux#7369

Signed-off-by: Lukasz Raczylo <lukasz@raczylo.com>
Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>
(cherry picked from commit 02bcfce)
Latest LTS.

Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>
(cherry picked from commit 9fff943)
@github-project-automation github-project-automation Bot moved this to To Do in Planning May 18, 2026
@talos-bot talos-bot moved this from To Do to In Review in Planning May 18, 2026
@shanduur shanduur changed the base branch from main to release-1.12 May 18, 2026 10:57
Run rekres to bump workflows.

Signed-off-by: Mateusz Urbanek <mateusz.urbanek@siderolabs.com>
@shanduur shanduur force-pushed the backports/1.12.8 branch from 7b1ab30 to ea936f1 Compare May 18, 2026 11:03
smira and others added 3 commits May 18, 2026 14:05
Bumping to the latest LTS.

Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>
(cherry picked from commit d616f6c)
Bump kernel 6.18.32

Signed-off-by: Noel Georgi <git@frezbo.dev>
(cherry picked from commit eac5f86)
Updated toolchain to v1.12.0-11-gce759f6
Updated tools to v1.12.0-12-gcbb843d

Signed-off-by: Mateusz Urbanek <mateusz.urbanek@siderolabs.com>
@github-project-automation github-project-automation Bot moved this from In Review to Approved in Planning May 18, 2026
@smira
Copy link
Copy Markdown
Member

smira commented May 18, 2026

bakports

s/backports/g

@shanduur shanduur changed the title chore: bakports for 1.12.8 chore: backports for 1.12.8 May 18, 2026
@shanduur
Copy link
Copy Markdown
Member Author

/m

@talos-bot talos-bot merged commit b00971e into siderolabs:release-1.12 May 18, 2026
13 checks passed
@github-project-automation github-project-automation Bot moved this from Approved to Done in Planning May 18, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: Done

Development

Successfully merging this pull request may close these issues.

6 participants