[[!meta copyright="Copyright © 2013 Free Software Foundation, Inc."]]

[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
id="license" text="Permission is granted to copy, distribute and/or modify this
document under the terms of the GNU Free Documentation License, Version 1.2 or
any later version published by the Free Software Foundation; with no Invariant
Sections, no Front-Cover Texts, and no Back-Cover Texts.  A copy of the license
is included in the section entitled [[GNU Free Documentation
License|/fdl]]."]]"""]]

[[!tag open_issue_gnumach]]


# IRC, freenode, #hurd, 2013-05-10

    <braunr> what code have you used if any (or is it your own implementation)
      ?
    <youpi> I ended up writing my own implementation
    <braunr> eh :)
    <youpi> the libahci/ahci code from linux is full of linux-specific stuff
    <youpi> it would mean working on gluing that
    <youpi> which woudl rather be just done in block-dde
    <youpi> I was told at fosdem that ahci is not actually very difficult
    <youpi> and it isn't indeed
    <braunr> that's why i usually encourage to use netbsd code

    <braunr> any chance using ahci might speed up our virtual machines ?
    <youpi> they are already using DMA, so probably no
    <youpi> (with the driver I've pushed)
    <youpi> adding support for tagged requests would permit to submit several
      requests at a time
    <youpi> _that_ could improve it
    <youpi> (it would make it quite more complex too)
    <youpi> but not so much actually

    <anatoly> What about virtio? will it speed up?

[[open_issues/virtio]].

    <youpi> probably not so much
    <youpi> because in the end it works the same
    <youpi> the guest writes the physical addresse in mapped memory
    <youpi> kvm performs the read into th epointed memory, triggers an irq
    <youpi> the guest takes the irq, marks as done, starts the next request,
      etc.
    <youpi> most enhancements that virtio could bring can already be achieved
      with ahci
    <youpi> one can probably go further with virtio, but doing it with ahci
      will also benefit bare hardware

    <pinotree> http://en.wikipedia.org/wiki/AHCI
    <youpi> anatoly: aka SATA
    <anatoly> some sort of general protocol to work with any SATA drive via
      AHCI-compatible host controller?
    <braunr> yes

    <youpi> braunr: I may be mistaken, but it does seem ahci is faster than ide
    <youpi> possibly because the ide driver is full of hardcoded wait loops
    <braunr> interesting :)
    <youpi> usleeps here and there
    <braunr> oh right
    <braunr> i wonder how they're actually implemented
    <youpi> so it would make sense to use that on shattrath
    <youpi> a nasty buggy busy-loop
    <braunr> yes but ending when ?
    <youpi> when a given number of loops have elapsed
    <youpi> that's where "buggy" applies :)
    <braunr> ok so buggy implies the loop isn't correctly calibrated
    <youpi> it isn't calibrated at all actually
    <braunr> ew
    <youpi> it was probably calibrated on some 486 or pentium hardware :)
    <braunr> yeah that's what i imagined too
    <braunr> we'll need some measurements but if it's actually true, it's even
      better news


## IRC, freenode, #hurd, 2013-05-11

    <youpi> ah, also, worth mentioning: the AHCI driver supports up to 2TiB
      disks
    <youpi> (as opposed to our IDE driver which supports only LBA28, 128GiB)
    <youpi> supporting more than 2TiB would require an RPC change, or using
      bigger sectors
    <youpi> (which wouldn't be a bad idea anyway)
    <braunr> i think we should switch to uint64_t addressable vm_objects
    <braunr> which would allow to support large files too
    <youpi> braunr: yep


## IRC, freenode, #hurd, 2013-05-13

    <braunr> the hurd, running on vbox, with a sata controler :)
    <braunr> hum, problem with an extended partition
    <anatoly_> qemu/kbm doesn't have sata controller, am I right?
    <braunr> anatoly: recent versions might
    <braunr> http://wiki.qemu.org/Features/AHCI
    <braunr> www.linux-kvm.org/wiki/images/7/73/2011-forum-ahci.pdf
    <anatoly> braunr: found first link, too. Thanx for the second one
    <braunr>
      http://git.qemu.org/?p=qemu.git;a=blob;f=hw/ide/ahci.c;h=eab60961bd818c22cf819d85d0bd5485d3a17754;hb=HEAD
    <braunr> looks ok in recent versions
    <braunr> looks useful to have virtio drivers though

[[open_issues/virtio]].

    <anatoly> virtio is shown as fastest way for IO in the presentation
    <anatoly> Hm, failed to run qemu with AHCI enabled
    <anatoly> qemu 1.1 from debian testing
    <anatoly> youpi how do run qemu with AHCI enabled?


## IRC, freenode, #hurd, 2013-05-14

    <anatoly> can somebody ask youpi how he runs qemu with AHCI please?
    <gnu_srs> I think he used vbox?  Did not find any AHCI option for kvm
      (1.1.2-+dfsg-6)
    <anatoly> gnu_srs: http://wiki.qemu.org/ChangeLog/0.14#IDE_.2F_AHCI
    <anatoly> but it doesn't work for me the same version of kvm
    <braunr_> anatoly: have you checked how the debian package builds it ?
    <anatoly> braunr: mach sees AHCI device
    <braunr> oh :)
    <anatoly> the problem is in last option "-device
      ide-drive,drive=disk,bus=ahci.0"
    <anatoly> lvm says 'invalid option'
    <braunr> anatoly: can you give more details please ?
    <braunr> lvm ?
    <anatoly> s/lvm/kvm
    <braunr> i don't understand
    <braunr> how can mach probe an ahci drive if you can't start kvm ?
    <anatoly> I ran it without last option
    <braunr> then why do you want that option ?
    <anatoly> But, actually I entered command with mistake. I retried it and it
      works. But got "start ext2fs: ext2fs: device:hd0s2: No such device or
      address"
    <anatoly> Sorry for confusing
    <braunr> that's normal
    <braunr> it should be sd0s2
    <bddebian2> Right because the device names are different
    <braunr> be aware that gnumach couln't see my extended partitions when i
      tried that yesterday
    <braunr> i don't know what causes the problem
    <anatoly> Yeah, I understand, I just note about it to show that it works
    <braunr> :)
    <anatoly> And I was wring
    <anatoly> s/wring/wrong
    <braunr> is that the version in wheezy ?
    <anatoly> I'm using testing, but it's same
    <braunr> great
    <braunr> the sceen.net VMs will soon use that then
    <anatoly> I don't have extended partions
    <anatoly> Booted with AHCI! :-)
    <anatoly> It freezes while downloading packages for build-essential
      fake-root dependencies with AHCI enabled
    <youpi> anatoly: is the IRQ of the ahci controller the same as for your
      ethernet device? (you can see that in lspci -v)
    <anatoly> youpi: will check
    <anatoly> youpi both uses IRQ 111
    <anatoly> s/111/11
    <braunr> aw
    <youpi> anatoly: ok, that might be why
    <youpi> is this kvm?
    <youpi> if so, you can set up a second ahci controler
    <youpi> and attach devices to it
    <youpi> so the irq is not the same
    <youpi> basically, the issue is about dde disabling the irq
    <youpi> during interrupt handler
    <youpi> which conflicts with ahci driver needs


## IRC, freenode, #hurd, 2013-05-15

    <anatoly> youpi: yes, it's kvm. Will try a second ahci controller

    <Slex> I read recentrly was added ahci driver, is it in userland or
      kernel-land?
    <gnu_srs> kernel-land the change was in gnumach


## IRC, freenode, #hurd, 2013-05-18

    <youpi> about the IRQ conflict, it's simply that both dde and the ahci
      driver need to disable it
    <youpi> it needs to be coherent somehow


## IRC, freenode, #hurd, 2013-05-20

    <anatoly> gnu_srs: kvm -m 1G -drive
      id=disk,file=<path_hurd_disk_img>,if=none,cache=writeback -device
      ahci,id=ahci-1 -device ahci,id=ahci-2 -device
      ide-drive,drive=disk,bus=ahci-2.0
    <anatoly> who knows what does "ich9-ahci.multifunction=on/off" parameter
      for kvm's ahci device mean?
    <anatoly> well, I was a bit incorrect :-) The options is relative to PCI
      miltifunction devices
    <anatoly> s/options is relative/options relates


## IRC, freenode, #hurd, 2013-05-24

    <anatoly> I don't see freezes anymore while downloading packages with AHCI
      enabled
    <youpi> anatoly: by fixing the shared IRQ ?
    <anatoly> youpi: yes, I added second AHCI as you suggested
    <youpi> ok
    <youpi> so it's probably the shared IRQ issue
    <anatoly> NIC and AHCI have similar IRQ when only one AHCI is enabled
    <anatoly> according lspci output
    <youpi> yes


## IRC, freenode, #hurd, 2013-06-18

    <braunr> youpi: is there a simple way from hurd to check interrupts ?
    <youpi> what do you mean by "check interrupts" ?
    <braunr> if they're shared
    <youpi> I still don't understand :)
    <braunr> i'm setting up sata
    <youpi> ah, knowing the number
    <braunr> yes
    <youpi> you can read that from lspci -v
    <braunr> ok
    <braunr> thanks
    <braunr> hum
    <braunr> i get set root='hd-49,msdos1' in grub.cfg when changing the
      device.map file to point to sd0
    <youpi> hum
    <braunr> i wonder if it's necessary
    <braunr> i guess i just have to tell gnumach to look for sd0, not grub
    <braunr> youpi: the trick you mentioned was to add another controler, right
      ?
    <youpi> yes
    <braunr> ok
    <braunr> youpi: looks fine :)
    <braunr> and yes, i left hd0 in grub's device.map
    <braunr> although i have lots of errors on hd0s6 (/home)
    <braunr> youpi: there must be a bug with large sizes
    <braunr> i'll stick with ide for now, but at least setting sata with
      libvirt was quite easy to do
    <braunr> so we can easily switch later