On 11/28/2012 08:16 PM, JD wrote: > > On 11/28/2012 08:21 PM, Cameron Simpson wrote: >> On 28Nov2012 15:37, JD <jd1008(a)gmail.com&gt; wrote: >> | While building the kernel from source RPM, I run >> | the command ps -ef | grep make in another terminal, >> | and I do see that 4 make processes are running, with >> | each one building objects for a different kernel source >> | directory. >> | >> | However, something else seems to be single threaded >> | which is causing the huge delay in finishing kernel build. >> | >> | I found that when I run top to see the upper-most >> | cpu consumers, I see only one cc1 process at the top. >> | >> | So I ran ps again to see how many cc1 processes there >> | are. >> | >> | Always - just one cc1 process. >> | >> | I would expect to see anywhere from 1 to 4 cc1 processes >> | at a time. But that is not the case. >> | >> | What is forcing only a single instance of cc1 at a time? >> >> Is make running with the "-j 4" (or whatever value) option? > I had already said that after running rpmbuild, > ps -ef shows 4 make processes running. > So, it must be picking up -j4 from somewhere, > perhaps from one of the rpmbuild macros. > > But as I also stated, when files are being compiled, one of the stages > of compilation is cc1. Since there are 4 make processes running at a > time, > with each make process running for one of the dirs containing kernel c > source files, then one woould expect to see anywhere from 1 to cc1 > processes running during the build. > I only see a single cc1 process running during the entire kernel > build duration, > by running ps -ef | grep cc1 > every 5 seconds. > > I never see more than a single instance of cc1. > I can't help with this question, but I build a new kernel (from kernel.org) whenever they appear. This never takes more than 5 or 6 minutes on a slow 2-cpu AMD box with only 1 GB memory. I usually don't specify "-j", so I only have one process running. I think you're seeing something seriously broken if your builds are taking a huge amount of time. Regards, John

Setting -j to be equal to the number of cores/CPUs is just playing it safe.

This is especially true when building a rpm which involves not just compiles and link but also packaging, dependency resolution and documentation tasks which are even more disk I/O intensive than compiles. Chances are that several of these operations that not dependent on successfully compiling the code are fired off in parallel rather than doing them sequentially and that old workhorse make gets the task of managing each of these tasks.

Thus, you end up with multiple copies of make running at the same time even on a single core/CPU system to perform these tasks.

My best results were with -j 12.

-----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.11 (GNU/Linux) iQIcBAEBAgAGBQJQt/nwAAoJENMrftdElS8VxKgP+wcnGdrorvYioNl8m/QSfROa ZXU5e3rkbIfRPBTMNusnwTu5ngiIs9N03jTZKdrgvRftyYWrDAI+QB0JSGZfO7X4 waeI80sgDnr4Km4zluw67RLpP7h3UNQomMgYX9CSL0gsXlou5MpvxV5lO/sDM4HH hLAa2+cL2TCf/sL6jKbSrvmCgxQhAquMrfLb3JigGUDDDW8LXqOz0yq5rl0Yu37a ZnVJ003dGkgIxs3g27ZFmuEWQpV7nro6JGZDhYemuPGRMLK2psbCfJwpM/tG+KPZ W9h7ZA1+J1XUmZ9T/Ejv5AFO2CiF9QG5GedjSYV2DTc0DMKwX/70iugoSpy3qa/l y40v6+lP1UkLYjL1HpUvhcxZjIJJlt8qG/VlZwjby011S7xOZZpyOiRIy+52Gq9p w3kEE8MciHc8rpEV11d+gKj38rTA3kLlNgb69PlGk3nvk7Ce9nD0Kpj7ZM7LE+5r cmRn/8U5is28lq+FnSjihDKOEXCX0raFB85i/sh6nUnSyRzUPHUu8AJPnjlckJqB WlUn5wxmaeERtzLrDOIzu3w3wstmpx/wb/rKShRBhftTXclYUCfkHF4BR2aUvGk/ 5BGJ95/2qnQZryk1yaCoz8c8jsz/0iCHkeN1cjKK5DolXMxkCs4xEeZ7/cmHKOlr jVUsu1Mp0Ri3nxGKcLf0 =NwWn -----END PGP SIGNATURE-----

On Thu, 29 Nov 2012, JD wrote: > That does not make sense. > I have a unicore!! An old Athlon64 3700+, In this case %{?_smp_mflags} is expanded to "-j1" or "" and you will have exactly only one "cc1".

> and yet ps -ef shows up to 4 makes running at a time. Read the previous mail again. Gabriel

That does not make sense.

I have a unicore!! An old Athlon64 3700+,

and yet ps -ef shows up to 4 makes running at a time.

-----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.11 (GNU/Linux) iQIcBAEBAgAGBQJQt5dTAAoJENMrftdElS8VrskP/1TMPmSx6nVDRJf76dgpT7pW nu6uwFfkifTAT2RcBwAuCYzYpRF0emQvPXuQdMTUukeP4cnNk5GKvSaK7ZfJbSly nsiSwXIxChVx+un9odEQ6eUumaygW2L172hFo1VsQS6oP1wV3YRA+itXoogCJu/m pzkfvBA7NDXjxBz/bQNtO2Zy/oAnv5PhxGaqKBPegyqUspgrUT2FtzWg8DD26+n3 D6ge+i8Oo3fgNvq9qf/itcLUFjmWCP9THqHkuTFRf6NvAsdMtoQwjN1iXZASmg2y HDboYFNun4hyp4yg92v4JV3H56/GvIQjuH6ScexKzSQdorjcPB73JrleUUoqbO19 eRN+XggnPZ2cElzOsPLsD6e7DDIrlvRaTSEnrcJgEzskS3npKAEKO85iR9L6vjcr JCoX5ZO7gBnq5SLctVrQY0/sDKwFCETPYCKvVx/OAc8T+Q1AIV7bVQU3Uct3lkGV dxhDTiuSzPRnmlUxw4AQYqmrgVS+ebhRStPXnd6y4a96fZfY0mhnFv45S85x83c/ TTNOutgv1JjZmN/oQJm0tCNXZlAMPv56v/2aOqVFza86CE7fN5c3tTJ8iHeuIcHV PT77qNolWRVdtcHQgPvWu8qCTEyZrVOHht34kYqMUC00kxazM5V2z94zzGjfPk/b a4C+fAbZIxMp1V+ipw0w =koAE -----END PGP SIGNATURE-----

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Wed, 28 Nov 2012, JD wrote: > Date: Wed, 28 Nov 2012 15:37:54 -0700 > From: JD <jd1008(a)gmail.com&gt; > To: Community support for Fedora users <users(a)lists.fedoraproject.org&gt; > User-Agent: Mozilla/5.0 (X11; Linux i686; > rv:15.0) Gecko/20120911 Thunderbird/15.0.1 > X-Spam-Status: No, score=-4.7 required=5.0 tests=AWL,BAYES_00, > DKIM_ADSP_CUSTOM_MED,DKIM_SIGNED,FREEMAIL_FROM,RCVD_IN_DNSWL_MED, > RP_MATCHES_RCVD,SPF_PASS,T_DKIM_INVALID autolearn=ham version=3.3.1 > Subject: Multi-Threaded make > > While building the kernel from source RPM, I run > the command ps -ef | grep make in another terminal, > and I do see that 4 make processes are running, with > each one building objects for a different kernel source > directory. > > However, something else seems to be single threaded > which is causing the huge delay in finishing kernel build. > > I found that when I run top to see the upper-most > cpu consumers, I see only one cc1 process at the top. > > So I ran ps again to see how many cc1 processes there > are. > > Always - just one cc1 process. > > I would expect to see anywhere from 1 to 4 cc1 processes > at a time. But that is not the case. > > What is forcing only a single instance of cc1 at a time? > > I hope the devs can have a look at either the compiler > tools or the make util. make is invoked from kernel.spec file like this: make -s ARCH=$Arch V=1 %{?_smp_mflags} $MakeTarget %{?sparse_mflags} make -s ARCH=$Arch V=1 %{?_smp_mflags} modules %{?sparse_mflags} || exit 1 where %{?_smp_mflags} is: 1 core : -j 1 2 cores: -j 2 4 cores: -j 4 ... You can see multiple make proccesses. But only 2 (in my case) do the real work: user1 22073 12810 0 13:13 pts/0 00:00:00 make -f scripts/Makefile.build obj=fs user1 24474 22073 0 13:15 pts/0 00:00:00 make -f scripts/Makefile.build obj=fs/cifs user1 24530 22073 0 13:15 pts/0 00:00:00 make -f scripts/Makefile.build obj=fs/coda Make proccess with pid 22073 only invoke two addititonal make proccesses which call cc1: user1 3617 3616 0 13:21 pts/0 00:00:00 /usr/libexec/gcc/x86_64-redhat-linux/4.7.2/cc1 -quiet -nostdinc .... user1 3627 3626 0 13:21 pts/0 00:00:00 /usr/libexec/gcc/x86_64-redhat-linux/4.7.2/cc1 -quiet -nostdinc ... One cc1 -> one core or you called make with -j 1. Gabriel

As a side note to Bruno, who wanted me to build and try this kernel: I will reboot immediately after sending this message with this kernel I just built.

On 11/28/2012 08:21 PM, Cameron Simpson wrote: > On 28Nov2012 15:37, JD <jd1008(a)gmail.com&gt; wrote: > | While building the kernel from source RPM, I run > | the command ps -ef | grep make in another terminal, > | and I do see that 4 make processes are running, with > | each one building objects for a different kernel source > | directory. > | > | However, something else seems to be single threaded > | which is causing the huge delay in finishing kernel build. > | > | I found that when I run top to see the upper-most > | cpu consumers, I see only one cc1 process at the top. > | > | So I ran ps again to see how many cc1 processes there > | are. > | > | Always - just one cc1 process. > | > | I would expect to see anywhere from 1 to 4 cc1 processes > | at a time. But that is not the case. > | > | What is forcing only a single instance of cc1 at a time? > > Is make running with the "-j 4" (or whatever value) option? I had already said that after running rpmbuild, ps -ef shows 4 make processes running. So, it must be picking up -j4 from somewhere, perhaps from one of the rpmbuild macros. But as I also stated, when files are being compiled, one of the stages of compilation is cc1. Since there are 4 make processes running at a time, with each make process running for one of the dirs containing kernel c source files, then one woould expect to see anywhere from 1 to cc1 processes running during the build. I only see a single cc1 process running during the entire kernel build duration, by running ps -ef | grep cc1 every 5 seconds. I never see more than a single instance of cc1.