linux: Add version 4.14.74

[donald]

Add Linux 4.14.74 containing a fix for for a cache invalidation bug in Linux memory management
.

[pmenzel]

Should 4.14.74 be patched like 4.14.68 or remove PCI_IRQ_AFFINITY in the file below?

$ git grep AFFIN ./drivers/scsi/aacraid/
drivers/scsi/aacraid/comminit.c:                                          PCI_IRQ_MSIX | PCI_IRQ_AFFINITY);

[donald]

But linux-4.14.74-225.bee is a copy of linux-4.14.68-222.bee and includes the revert patch ?

buczek@theinternet:~/git/bee-files (rebuild-linux)$ grep PATCH linux-4.14.74-225.bee
PATCHURL+=(/src/mariux/patches/0001-Revert-genirq-affinity-assign-vectors-to-all-possibl.patch)
echo "PATCH $@"

[pmenzel]

Am 05.10.2018 um 19:18 schrieb Donald Buczek:

But linux-4.14.74-225.bee is a copy of linux-4.14.68-222.bee and includes the revert patch ? > ***@***.***:~/git/bee-files (rebuild-linux)$ grep PATCH linux-4.14.74-225.bee > PATCHURL+=(/src/mariux/patches/0001-Revert-genirq-affinity-assign-vectors-to-all-possibl.patch) > echo "PATCH $@"

Sorry, I did not read the versions in your diff command carefully enough, and made the wrong assumptions. Also, I totally forgot, that 4.14.68 had been rebuild with KPROBES support. Anyway, the question is, if only removing `PCI_IRQ_AFFINITY` is the better approach. But, as we are probably going to stay with 4.14.55 for the four “aacraid systems”, it does not matter.

[donald]

related issue: #850

[donald]

I think, 4.14.74 with the Revert "genirq/affinity: assign vectors to all possible CPUs" patch should be valid for all systems, so I'll withdraw 4.14.55 and its nvidia from this PR.

[thomas]

On October 6, 2018 9:23:34 PM GMT+02:00, Donald Buczek ***@***.***> wrote: I think, 4.14.74 with the Revert "genirq/affinity: assign vectors to all possible CPUs" patch should be valid for all systems, so I'll withdraw 4.14.55 and its nvidia from this PR.

I agree that the reversal should be harmless. My only concern is the presence of words like 'think, should, and the like'. So maybe we should run the new kernel on a few selected servers at first.

…

-- Thomas Kreitler - Information Retrieval kreitler@molgen.mpg.de 49/30/8413 1702

[pmenzel]

Alright, I’ll accept the request so that it can be tested easily. After the tests were successful, it can be made the default kernel.

[donald]

My only concern is the presence of words like 'think, should, and the like'.

Whatever anybody says, it is always explicitly or implicitly qualified by "in may opinion which may be wrong". I just formulate cautiously so that I don't look to stupid if I'm really wrong 😎

On a second thought, maybe to remove PCI_IRQ_AFFINITY from the aacraid would have been a better general fix (taking present but offline CPUs into accounts as well), but in our environment both should do.

I've rebooted deadbird to 4.14.74.mx64.225 and the disks are present and no problem in dmesg. claptrap will be next, I love to test on my home server, because any problems will get to me before a normal user suffers. So these two are the few selected servers you asked for.

[donald]

Drat! Left out Add-trace-events-for-all-v4-operations.patch in 4.14.74.