-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 06/14/2012 04:53 AM, Olaf Hering wrote:

On Tue, Jun 12, Jan Beulich wrote:

...

While I questioned whether doing this (for a slim boot time benefit this might provide) is really the right thing already in the past, bug 766284 points out another reason why this is a bad idea. Can this be reverted (in master and 12.2 at least)?

The change below for ata_piix (and other related changes in April 2009) was made without a bugzilla or fate reference. Whats the practical benefit of having the drivers built into the kernel? At least ata_piix should be a module because it also drives emulated hardware (not only in a Xen HVM guest).

The changes came out of the faster booting effort at Intel for Meego (or Moblin, or whatever it was called then), since the initialization can happen in parallel rather than waiting on each module to load in the initrd as happens with a mostly-modular kernel. Greg KH added the changes and presented numbers as to why it was faster at the time. The fact that ata_piix is also used by emulated hardware is not a reason to make it modular. It's a reason to fix the drivers to work together, or to pull the functionality into the original driver. BTW, patches.suse/suse-hv-pata_piix-ignore-disks.patch pretty clearly violates the one-commit-per-patch rule. - -Jeff - -- Jeff Mahoney SUSE Labs -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.19 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iQIcBAEBAgAGBQJP2ekWAAoJEB57S2MheeWyP1AQALplGP2/2nqajxbBF2CHEEdZ ATB1iOWwPWUc0hH4bbxXmrjeOYBMxdldGE9W5/M1FlPfSoPQ99eLKkbJARQP7lsQ 2JMqhpfJS3tqnqVz/W1SjV4GHzWNWtM7O4PIMKOhVlYnxDcvbAVxLwQdNeywoA9W p8+2zMGbTWRDwg27l0ugK5voxv9AM4UQAPkXmdXziIntP//UtRQeX79cO6aQT1ON piJqV8HQ4rD+MHMH5Z2lqc/oo8bISQFAVot5ySripN18RSnEjjiywsuQkroCwn4B f6kHndqsEzT3hKbObxSW7lD5v8o2LPCZUNRWKlp89h+zCmEPQ5NzmQ0YBkIFoT+/ l8U9aWi0amZMmwwItwmapAx9FLrrEKNclh/ESdLT/XpKoueMMv+yzcVWYo7MJSxQ TAcFJkcNhtpG0pgWvV5LpTiUVsfkV9/lPDh1ViBIRYLH6vkTEGWlrEpWa0K7jcgQ PnJpRKBPJ9lThkERrXVaYvSAnjnFwdAZQTfz6Y557djGNYQyTqPDIYAXjNIGpAtC XrwMX44qp9vC3qnBuwVqVNwql8vqIaCAk94bEK4+p0nKzIYsDk/1uJcaZX74F1Nb GwMX2zs5DulHikUtx20iu2BOBCuRvbMav3I7f9buwKR//WmdNw9ebkqR4Ec5Eld1 mEBNoras0f2SFJ3xBiNF =+DwQ -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 06/14/2012 10:10 AM, Jan Beulich wrote:

...

...

...

On 14.06.12 at 15:37, Jeff Mahoney wrote: -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

On 06/14/2012 04:53 AM, Olaf Hering wrote:

...

On Tue, Jun 12, Jan Beulich wrote:

...

While I questioned whether doing this (for a slim boot time benefit this might provide) is really the right thing already in the past, bug 766284 points out another reason why this is a bad idea. Can this be reverted (in master and 12.2 at least)?

The change below for ata_piix (and other related changes in April 2009) was made without a bugzilla or fate reference. Whats the practical benefit of having the drivers built into the kernel? At least ata_piix should be a module because it also drives emulated hardware (not only in a Xen HVM guest).

The changes came out of the faster booting effort at Intel for Meego (or Moblin, or whatever it was called then), since the initialization can happen in parallel rather than waiting on each module to load in the initrd as happens with a mostly-modular kernel. Greg KH added the changes and presented numbers as to why it was faster at the time.

The fact that ata_piix is also used by emulated hardware is not a reason to make it modular. It's a reason to fix the drivers to work together, or to pull the functionality into the original driver.

If the intended driver is modular, then that driver has no way to tell a built-in one to not control certain devices. So the only option then is to use Olaf's patch (which isn't really desirable imo).

From my perspective, if there are two or more drivers for one piece of hardware (ignore the fact that it's virtual/emulated here), they _all_ should be modular, so that the rest of the system (e.g. modprobe rules) can figure out the right one. The situation really is no different than for other cases where a KMP override is wanted/necessary.

- From my perspective, there shouldn't be more than one driver for a given piece of hardware unless one is planned to phase out the other. That doesn't seem to be the case. The emulated one is a special case of the the canonical one and should be integrated. - -Jeff - -- Jeff Mahoney SUSE Labs -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.19 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iQIcBAEBAgAGBQJP2fHBAAoJEB57S2MheeWyhB0P/3GorW9ia7mC0QBM85udbzyr nuZ8ArjgrAlJAPgx+825cvgNeVXnKA/jQa/wih38szr93r4RUqIyEWfe84bXVIuv bi49/rYDup6jkOuOsPd2Igilj+TWVRIPgcthD0Fsy2UFTz64NrVaNgtgt2rwAdOk +WgneGbaKIRwy8bB1xszkXF6phNrmmys9XRWbjXiUeRoZ6IwhD68AlRcJ8qbEkoj Xcvpqg/H/kjXKdUllIUFRxd38lAKOlNJ/xMei+7nKLJV9UP8dsjUFGPpvbgudYwA ScM9DlEs/PY7ca58V2b0RRGlRoalgsYKxYPNThry0biIZOklgDSIfouIWCaQyRsN 1eHl3+rdty1wmIRWpp0jWBWHS/csqhJVUa5UQU2UGCcCAmNcCqQcuhpDW3h3jPU/ SCUZO8bg9v8il70DchIyFYV6ddFesek2z/FTR+4QD3rKpMF3W87iOeTnQYyaWgQT 40xrNLUbZk83uRPXlhIw+xcWcdTpbCvBP2FyvF2rAV0GbSu25c/iM3T7eyWw1vsh KrIjxE2LNNIovLwyti22YzMFS0KCt9ftUYaE2EY7LidZTlXl/B6xiv9TO0XHVe6f JHmf61ILo60S2JKfmsM4sJUeF2GaSKbaPl5NbM4/Z5isAJvGMy79kRxtFh+ZhMfX LpeOWD8kyGODQKtpmGCj =4Etr -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 06/14/2012 10:29 AM, Jan Beulich wrote:

...

...

...

On 14.06.12 at 16:14, Jeff Mahoney wrote: -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

On 06/14/2012 10:10 AM, Jan Beulich wrote:

...

...

...

...

On 14.06.12 at 15:37, Jeff Mahoney wrote: -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

On 06/14/2012 04:53 AM, Olaf Hering wrote:

...

On Tue, Jun 12, Jan Beulich wrote:

...

While I questioned whether doing this (for a slim boot time benefit this might provide) is really the right thing already in the past, bug 766284 points out another reason why this is a bad idea. Can this be reverted (in master and 12.2 at least)?

The change below for ata_piix (and other related changes in April 2009) was made without a bugzilla or fate reference. Whats the practical benefit of having the drivers built into the kernel? At least ata_piix should be a module because it also drives emulated hardware (not only in a Xen HVM guest).

The changes came out of the faster booting effort at Intel for Meego (or Moblin, or whatever it was called then), since the initialization can happen in parallel rather than waiting on each module to load in the initrd as happens with a mostly-modular kernel. Greg KH added the changes and presented numbers as to why it was faster at the time.

The fact that ata_piix is also used by emulated hardware is not a reason to make it modular. It's a reason to fix the drivers to work together, or to pull the functionality into the original driver.

If the intended driver is modular, then that driver has no way to tell a built-in one to not control certain devices. So the only option then is to use Olaf's patch (which isn't really desirable imo).

From my perspective, if there are two or more drivers for one piece of hardware (ignore the fact that it's virtual/emulated here), they _all_ should be modular, so that the rest of the system (e.g. modprobe rules) can figure out the right one. The situation really is no different than for other cases where a KMP override is wanted/necessary.

- From my perspective, there shouldn't be more than one driver for a given piece of hardware unless one is planned to phase out the other. That doesn't seem to be the case. The emulated one is a special case of the the canonical one and should be integrated.

Integrated into what? You're aware that we're talking about the PV Hyper-V and Xen drivers, aren't you?

Furthermore, there are already multiple drivers for ATA devices - many specialized ones, and ata_generic. I doubt you could force the use of ata_generic on a kernel with built-in ata_piix on devices matched by the latter's MODULE_DEVICE_TABLE() without tedious (almost) manual unbinding (which isn't an option for the problem at hand, since afaict there's no direct correlation between the PV and emulated devices).

Ok. I see your point. All of that only works if they're all modular or all static, not a mix of both. Go ahead and modularize ata_piix. - -Jeff - -- Jeff Mahoney SUSE Labs -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.19 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iQIcBAEBAgAGBQJP2fYVAAoJEB57S2MheeWyREcQAJeGMxbblT0SMSE3n1i9BPG0 KDs/8ikqusH0emBbc2a0YgD+O/XbzH2asXnM3nKhzeE5HYaKzO1KjvRqhE63LLjJ qHw2jFTCDXLYT4bCYuH0UyHUubMwjJ12LfpmzqKcjUWEHjE5W7+0wV7jrv9kCWaV /aIybAJ2VyTbEaawfENuyR4YvHOBxnt4EYDAEu7sHKiz6mS8ShwWSaW5sF4Z0QYi 0AA4uLQoXmEVR8vahfoZZ4UotpyQuK7m8jlg803f/0MNY778Owu8zdJ7CtIoH4dN 79hm69q8fCXbaIzetp8bx3g/jZqMMypK0HNWOw7PRZNt3oCV7rTBlSsL5rSxbQOo duRYTtAe0UcUOeBrKHKnJnzJMvxd7kvTE/5QZMRc3FbSgtTEB09qBRSIRoEomSqx BQagJn2a+jxlQ7TzprcxkZqvz9RlIiMo4c0pIEEQUEmVpxGpg4a48Q4CHJyZSJpT HIAhDR236QwG4GMFY4Kn6Dxv9YUY1ugud3hJRnmVVS/vC7Ny+8BnFddTLzxpCzcp E+fc+/1OL2KZ9AZ3clzNCqU7UR1Uoewzugy3wnEwM5othAod3sqECr1usiEguRTj YqCSEVqMCf5uWJQFZtpCQiYAYyvx0Es8+ru8+4d1IljMILRBJX12VxIM7X7Oe1hV kNVTOaSjQCdp6auBYX1A =lV6t -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 06/14/2012 11:07 AM, Jan Beulich wrote:

...

...

...

On 14.06.12 at 16:32, Jeff Mahoney wrote: Ok. I see your point. All of that only works if they're all modular or all static, not a mix of both. Go ahead and modularize ata_piix.

Done so, but indeed only for ATA_PIIX (despite my thinking that this should be done to AHCI and probably also ATA and SCSI as well).

What's the justification for doing all of them? The benefits for faster booting are still there. I wouldn't have modularized PIIX if there was wasn't a conflict. - -Jeff - -- Jeff Mahoney SUSE Labs -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.19 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iQIcBAEBAgAGBQJP2f5MAAoJEB57S2MheeWyeRMQAJ7Zq8GI+eOhAzDm3Fo45Dx4 4zp72EN6vSJyzU7agAFZWE/14lyXNFBVoeN5Xyw7+PFAfyjzU3griMXzpCw5paww KPywOFrxL/1GEgsHvzQY6Z7sVNeZ+ZFuyUG2IPUaNmfUuzbwyPcK6zNoX2yI1zZ6 hmI75YVhlinw7VgPLxfZ72K+nHtSOLt7ouCNQnoXx2GnA82970/HhhggxbXXv41N gW0lT42Hh0jEZX7NyM/Eg8SMlN8bqgPS8A5RI3NnBjf/CER4S0gydOgjdyFx3l+L ag+qCC1tD1BmDeLTylQBAeq7+BRDP6T/TBVQimi+aRV7oyRjc75I64f1LgYUzxzx ZEI1J7WwA1bhRAjfi8lJSlX1qTFNId97IxX0E6JeyqJ7lm+Z9tUiP7PJqkqWB+a1 035myWJP/+7e5MkjZDzSh8qjdhz4wHRJBsf5UX8aJJoxuAJdrR7SPZHEfOnfZETS wo8ceIf3bl/fKx67q25lwzYnQspeTOo2z649qmc06eSbOquORS6o2l/aKqbF5m8s G2b4pklC4mtQ3IZr06md4XG1YsB1IZE7ExyN9F/b5z6tA5HicbOm3PjapQj5LDT2 EeQY1FMU03xxFS2iodYrTjQvLv30oUhtTJ2uc93SvlsHGz9rJjONji6t8zsRKNII BPMdTjdkMEtdN/t5mzHC =jMIq -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org

On 06/14/2012 05:07 PM, Jeff Mahoney wrote:

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

On 06/14/2012 11:07 AM, Jan Beulich wrote:

...

...

...

...

On 14.06.12 at 16:32, Jeff Mahoney wrote: Ok. I see your point. All of that only works if they're all modular or all static, not a mix of both. Go ahead and modularize ata_piix.

Done so, but indeed only for ATA_PIIX (despite my thinking that this should be done to AHCI and probably also ATA and SCSI as well).

What's the justification for doing all of them? The benefits for faster booting are still there. I wouldn't have modularized PIIX if there was wasn't a conflict.

This has become largely obsolete now with systemd. The overall gain we'll have from systemd far offsets any benefit we'll incur from having ata_piix built-in. I still think this was a bad idea. We had the very same issue during the IDE/libata transition, and I thought we had some lessons learned from them. But no, of _course_ this time it would be different. You wish. Cheers, Hannes -- Dr. Hannes Reinecke zSeries & Storage hare@suse.de +49 911 74053 688 SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg GF: Markus Rex, HRB 16746 (AG Nürnberg) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org

...

...

On 15.06.12 at 14:35, Andreas Jaeger wrote: On 06/14/2012 05:07 PM, Jan Beulich wrote:

...

...

On 14.06.12 at 16:32, Jeff Mahoney wrote: Ok. I see your point. All of that only works if they're all modular or all static, not a mix of both. Go ahead and modularize ata_piix.

Done so, but indeed only for ATA_PIIX (despite my thinking that this should be done to AHCI and probably also ATA and SCSI as well).

This was only done for the master branch but not for 12.2. Should it get done there as well?

I'm sure I did it there too (commit 8677f43b2bc1915d826c7a476a7b3200dfc89907). Jan -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org

On 06/15/2012 02:54 PM, Jan Beulich wrote:

...

...

...

On 15.06.12 at 14:35, Andreas Jaeger wrote: On 06/14/2012 05:07 PM, Jan Beulich wrote:

...

...

On 14.06.12 at 16:32, Jeff Mahoney wrote: Ok. I see your point. All of that only works if they're all modular or all static, not a mix of both. Go ahead and modularize ata_piix.

Done so, but indeed only for ATA_PIIX (despite my thinking that this should be done to AHCI and probably also ATA and SCSI as well).

This was only done for the master branch but not for 12.2. Should it get done there as well?

I'm sure I did it there too (commit 8677f43b2bc1915d826c7a476a7b3200dfc89907).

Yes, you did indeed - I forgot to pull ;) thanks, Andreas -- Andreas Jaeger aj@{suse.com,opensuse.org} Twitter/Identica: jaegerandi SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Jeff Hawn,Jennifer Guild,Felix Imendörffer,HRB16746 (AG Nürnberg) GPG fingerprint = 93A3 365E CE47 B889 DF7F FED1 389A 563C C272 A126 -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org