Requests for -current (14.2-->15.0)

volkerdi · 11-13-2019, 08:49 PM

Quote:

Originally Posted by nullptr

Code:

$ dmesg | grep crc32c
[    3.320686] Btrfs loaded, crc32c=crc32c-intel

Maybe it not CONFIG_CRYPTO_CRC32C but CONFIG_LIBCRC32C should be set to m?

Have you tried putting the crc32c-intel.ko kernel module in your initrd?

nullptr · 11-13-2019, 09:02 PM

Quote:

Originally Posted by volkerdi

Have you tried putting the crc32c-intel.ko kernel module in your initrd?

Yes.

volkerdi · 11-13-2019, 10:14 PM

Quote:

Originally Posted by nullptr

Yes.

Well, it _should_ work. Not using huge+initrd, correct?

Anyway, as far as changing crc32c from built-in to a module, I'm inclined to go WONTFIX on that because it complicates things needlessly. Here's a reference to what went on in Fedora:

https://bugzilla.redhat.com/show_bug.cgi?id=208607

Have you actually tested building crc32c as a module and found that it solved the issue? If it's working built-in in Fedora, I'd be surprised if changing crc32c to a module would fix things here. They do build libcrc32c as a module, but the btrfs documentation says that it can also be built into the kernel. If that were not already working, I don't think btrfs would function at all.

I searched for benchmarks on how much hardware accelerated crc32c support would help btrfs, and the closest thing I could find was someone saying any performance gain would be "slight".

Welcome to LQ, BTW :-)

nullptr · 11-14-2019, 01:24 AM

Quote:

Originally Posted by volkerdi

Well, it _should_ work. Not using huge+initrd, correct?

I don't even have huge installed.

Quote:

Originally Posted by volkerdi

Have you actually tested building crc32c as a module and found that it solved the issue?

I am not going to build a kernel with many unnecessary options on, it will takes lots of time, and
as stated before the it works on fedora so the problem may come from CONFIG_LIBCRC32C.

Quote:

Originally Posted by volkerdi

They do build libcrc32c as a module, but the btrfs documentation says that it can also be built into the kernel. If that were not already working, I don't think btrfs would function at all.

But it works on Arch, Debian and Fedora.

Quote:

Originally Posted by volkerdi

I searched for benchmarks on how much hardware accelerated crc32c support would help btrfs, and the closest thing I could find was someone saying any performance gain would be "slight".

https://github.com/htot/crc32c

Quote:

Originally Posted by volkerdi

Welcome to LQ, BTW :-)

Thanks.

EDIT:
I can confirm CONFIG_LIBCRC32C=m (and CONFIG_BLK_DEV_MD=m) fix the problem, would you consider this?

TurboBlaze · 11-14-2019, 02:19 AM

Mesa 19.2.4
https://www.mesa3d.org/relnotes/19.2.4.html
mesa-19.2.4.tar.xz

volkerdi · 11-14-2019, 07:14 AM

Quote:

Originally Posted by nullptr

https://github.com/htot/crc32c

Those are not btrfs benchmarks. Let's see some of those.

Quote:

EDIT:
I can confirm CONFIG_LIBCRC32C=m (and CONFIG_BLK_DEV_MD=m) fix the problem, would you consider this?

Fedora has CONFIG_BLK_DEV_MD=y.

cwizardone · 11-14-2019, 07:40 AM

Quote:

Originally Posted by TurboBlaze

Mesa 19.2.4
https://www.mesa3d.org/relnotes/19.2.4.html
mesa-19.2.4.tar.xz

Thanks for the "heads up" TurboBlaze!

Quote:

Mesa 19.2.4 is an emergency bug fix release to fix on ciritcal bug in 19.2.3.

anon074 · 11-14-2019, 03:02 PM

terminus-font 4.48 provides additional characters. Some install targets have been changed.

Quote:

--- terminus-font.SlackBuild.orig 2019-09-30 01:48:33.000000000 +0200
+++ terminus-font.SlackBuild 2019-11-14 04:50:10.636759092 +0100
@@ -81 +81 @@
-make install install-uni install-ref DESTDIR=$PKG || exit 1
+make install install-pcf-8bit install-psf-ref install-psf-vgaw DESTDIR=$PKG || exit 1

nullptr · 11-14-2019, 08:19 PM

Quote:

Originally Posted by volkerdi

Those are not btrfs benchmarks. Let's see some of those.

Fedora has CONFIG_BLK_DEV_MD=y.

How about CONFIG_MD_RAID456=m?

Code:

--- config-generic-4.19.84.x64	2019-11-13 04:38:11.000000000 +0800
+++ .config	2019-11-15 08:54:15.791606000 +0800
@@ -2562,7 +2562,7 @@
 CONFIG_MD_RAID0=y
 CONFIG_MD_RAID1=y
 CONFIG_MD_RAID10=y
-CONFIG_MD_RAID456=y
+CONFIG_MD_RAID456=m
 CONFIG_MD_MULTIPATH=y
 CONFIG_MD_FAULTY=m
 CONFIG_MD_CLUSTER=m
@@ -7666,12 +7666,12 @@
 # CONFIG_INTEGRITY is not set
 CONFIG_DEFAULT_SECURITY_DAC=y
 CONFIG_DEFAULT_SECURITY=""
-CONFIG_XOR_BLOCKS=y
-CONFIG_ASYNC_CORE=y
-CONFIG_ASYNC_MEMCPY=y
-CONFIG_ASYNC_XOR=y
-CONFIG_ASYNC_PQ=y
-CONFIG_ASYNC_RAID6_RECOV=y
+CONFIG_XOR_BLOCKS=m
+CONFIG_ASYNC_CORE=m
+CONFIG_ASYNC_MEMCPY=m
+CONFIG_ASYNC_XOR=m
+CONFIG_ASYNC_PQ=m
+CONFIG_ASYNC_RAID6_RECOV=m
 CONFIG_CRYPTO=y
 
 #
@@ -7900,7 +7900,7 @@
 #
 # Library routines
 #
-CONFIG_RAID6_PQ=y
+CONFIG_RAID6_PQ=m
 CONFIG_BITREVERSE=y
 CONFIG_RATIONAL=y
 CONFIG_GENERIC_STRNCPY_FROM_USER=y
@@ -7924,7 +7924,7 @@
 CONFIG_CRC64=m
 CONFIG_CRC4=m
 CONFIG_CRC7=m
-CONFIG_LIBCRC32C=y
+CONFIG_LIBCRC32C=m
 CONFIG_CRC8=m
 CONFIG_XXHASH=m
 # CONFIG_RANDOM32_SELFTEST is not set

CONFIG_MD_RAID456=y requires CONFIG_LIBCRC32C=y
CONFIG_BLK_DEV_MD=m requires CONFIG_MD_RAID456=m

Do I have to show you some real btrfs benchmarks to convince you?

--------------------------------------------------------------

Code:

k/kernel-source-4.19.84-noarch-2.txz: Rebuilt.
       CRYPTO_CRC32C_INTEL m -> y

Thank you.

akschu · 11-14-2019, 10:21 PM

Any thoughts on making the GCC libraries their own package? I have some software that links against libatomic which is forcing me to install the entire GCC compiler on a server, which is unwanted.

willysr · 11-14-2019, 10:43 PM

Slackware doesn't split packages like any other Linux distribution. You will get a single package consisting all files, including main program, documentations, and libraries.

akschu · 11-15-2019, 12:43 AM

Quote:

Originally Posted by willysr

Slackware doesn't split packages like any other Linux distribution. You will get a single package consisting all files, including main program, documentations, and libraries.

I totally understand that, I've been using slackware for more than 15 years, but.... Slackware does split off openssl-solibs, glibc-solibs, and seamonkey-solibs. So gcc-solibs isn't unreasonable if software needs those libraries but you don't want the entire gcc suite.

burdi01 · 11-15-2019, 03:50 AM

Quote:

So gcc-solibs isn't unreasonable if software needs those libraries but you don't want the entire gcc suite.

And likewise for qt[45] please!
https://www.linuxquestions.org/quest...ge-4175623664/

allend · 11-15-2019, 07:53 AM

Quote:

So gcc-solibs isn't unreasonable if software needs those libraries but you don't want the entire gcc suite.

Let me correct that for you.
So gcc-solibs isn't unreasonable if software needs those libraries but youI don't want the entire gcc suite.

ehartman · 11-15-2019, 08:25 AM

Quote:

Originally Posted by akschu

Slackware does split off openssl-solibs, glibc-solibs, and seamonkey-solibs.

The last one (seamonkey-solibs) has been removed, wasn't really needed anymore.

Quote:

Mon Sep 9 05:10:25 UTC 2019
l/seamonkey-solibs-2.49.5-i586-1.txz: Removed.
cq
l/seamonkey-solibs-2.49.5-x86_64-1.txz: Removed.

On the other hand openssl-solibs has gotten a companion openssl10-solibs one (as some programs are still built against openssl 1.0).