diff mbox

powerpc/pci: Fix IO space breakage after of_pci_range_to_resource() change

Message ID 1413429775-716-1-git-send-email-mpe@ellerman.id.au (mailing list archive)
State Accepted, archived
Headers show

Commit Message

Michael Ellerman Oct. 16, 2014, 3:22 a.m. UTC 
Commit 0b0b0893d49b "of/pci: Fix the conversion of IO ranges into IO
resources" changed the behaviour of of_pci_range_to_resource().

Previously it simply populated the resource based on the arguments. Now
it calls pci_register_io_range() and pci_address_to_pio(). These both
have two implementations depending on whether PCI_IOBASE is defined,
which it is not for powerpc.

Further complicating matters, both routines are weak, and powerpc
implements it's own version of one - pci_address_to_pio(). However
powerpc's implementation depends on other initialisations which are done
later in boot.

The end result is incorrectly initialised IO space. Often we can get
away with that, because we don't make much use of IO space. However
virtio requires it, so we see eg:

  pci_bus 0000:00: root bus resource [io  0xffff] (bus address [0xffffffffffffffff-0xffffffffffffffff])
  PCI: Cannot allocate resource region 0 of device 0000:00:01.0, will remap
  virtio-pci 0000:00:01.0: can't enable device: BAR 0 [io  size 0x0020] not assigned

The simplest fix for now is to just stop using of_pci_range_to_resource(),
and open-code the original implementation, that's all we want it to do.

Fixes: 0b0b0893d49b ("of/pci: Fix the conversion of IO ranges into IO resources")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
---
 arch/powerpc/kernel/pci-common.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

Comments

Arnd Bergmann Oct. 16, 2014, 2:05 p.m. UTC | #1 
(hope my email makes it everywhere, using a webmail client at the moment as I'm
at plumbersconf

Michael Ellerman <mpe@ellerman.id.au> hat am 16. Oktober 2014 um 05:22
geschrieben:
>
>
> Commit 0b0b0893d49b "of/pci: Fix the conversion of IO ranges into IO
> resources" changed the behaviour of of_pci_range_to_resource().

I just looked at this after benh mentioned the problem on IRC, here's a log dump

:26 AM   <benh> argh  
9:27 AM   <benh> the whole ARM OF PCI rework seems to completely break PIO on
powerpc  
9:30 AM   → willy <willy> joined (^willy@62.156.150.204)  
9:35 AM   <benh> and reverting it would mean reverting all of ARM new PCI stuff
 
9:35 AM   <benh> crap  
9:35 AM   <benh> that business with IO space allocation taking over our code
without understanding what it does  
9:35 AM   <benh> yuck  
9:41 AM     → markf <markf> , olaf <olaf> , sarnold <sarnold> , gospo <gospo> ,
cmarinas <cmarinas> , Mahesh1 <Mahesh1> , joern <joern> , clark_ <clark_> and
benhjoined  ⇐ gcl <gcl> and clark <clark> quit  ↔ willy <willy> , jbarnes
<jbarnes> , jbrandeb_ <jbrandeb_> and Mahesh <Mahesh> popped in  ↔ sameo <sameo>
, jbrandeb <jbrandeb> , steved <steved> and jj <jj> nipped out  •  srikar →
srikar_away <srikar_away> , raghu → raghu_away <raghu_away>  
Thursday, October 16th, 2014  
12:06 AM     → fweisbec <fweisbec> , Mahesh <Mahesh> , kamalesh <kamalesh> ,
heiko <heiko> , olaf <olaf> , riel <riel> and willy <willy> joined  ⇐ shaggy
<shaggy> , sammj <sammj> , sameo <sameo> , clark_ <clark_> , lenb <lenb> and jj
<jj> quit  ↔ jbrandeb <jbrandeb> , cdub <cdub> , Mahesh1 <Mahesh1> and gcl <gcl>
popped in  ↔ jbrandeb_ <jbrandeb_> , benh, joern <joern> and BenC <BenC> nipped
out  •  mpe|away → mpe|away <mpe%7Caway> , raghu_away → raghu <raghu> ,
srikar_away → srikar <srikar>  
10:16 AM   <arnd_> benh: is it the of_pci_range_to_resource change?  
10:17 AM   <arnd_> the new pci_ioremap_iospace logic should not get used on
powerpc at all, so I didn't expect any breakage  
10:18 AM   <arnd_> I wasn't too happy with all the details of Liviu's series,
bit in the end it seemed reasonable enough  
10:20 AM   <arnd_> he really wanted to use the pci_address_to_pio code from
powerpc and in the end I stopped complaining  
10:21 AM   <arnd_> the new code can do a few things that simpler versions could
not, e.g. handling multiple host bridges getting registered when they have the
same I/O space window  
10:23 AM   → jj <jj> joined (^jj@static-50-53-60-87.bvtn.or.frontiernet.net)  
10:33 AM   <arnd_> benh: I can see how it breaks your
pci_process_bridge_OF_ranges, we had the same problem in some of the ARM
platforms and Liviu fixed those but apparently didn't realize he had to change
the ppc implementation (and get your ack) too  
10:35 AM   <arnd_> the good news is that it should in fact simplify your code to
fix it, but the fact that this bug got into the kernel in the first place is
extremely annoying  
10:39 AM   <arnd_> benh: the fixup that is done in your
pcibios_reserve_legacy_regions is now already performed in
of_pci_range_to_resource  
10:39 AM   <arnd_> we had duplicated the same thing in each pci host driver (and
they all got it wrong), so the intent was to move it into a common place  
10:40 AM   <arnd_> but of course it's a bug to do it twice  
10:43 AM   <arnd_> pci_register_io_range is trying to do a more generalized
version of how you assign hose->io_base_virt, you should probably override that
to keep the current behavior  
10:45 AM   <arnd_> pcibios_map_phb_io_space I mean, for ppc64  
10:52 AM   → cmarinas <cmarinas> joined (~cmarinas@fw-tnat.cambridge.arm.com)  
10:53 AM   <arnd_> benh: for 3.18, the best approach is likely to #ifdef
<%23ifdef> PCI_IOBASE the changes in of_pci_range_to_resource  
10:54 AM   <arnd_> I suspect you are fine with effectively reverting Liviu's
changes that way, and you can decide whether or not you want to later make the
powerpc code use the common logic  
11:01 AM     → cdub <cdub> and sarnold <sarnold> joined  ⇐ cmarinas <cmarinas>
quit  
11:28 AM   <benh> arnd_: can you shoot the above in an email CCed to mpe ?  
11:28 AM   <benh> arnd_: he did a band aid that works  
11:28 AM   <benh> arnd_: and see the comment I made today about using his stuff
if I can specify where I want the IO ranges  
11:28 AM   <benh> arnd_: I want to keep the way I do the layout on ppc64  

> Previously it simply populated the resource based on the arguments. Now
> it calls pci_register_io_range() and pci_address_to_pio(). These both
> have two implementations depending on whether PCI_IOBASE is defined,
> which it is not for powerpc.
>
> Further complicating matters, both routines are weak, and powerpc
> implements it's own version of one - pci_address_to_pio(). However
> powerpc's implementation depends on other initialisations which are done
> later in boot.

Right, sorry for missing this during the last review of the broken patches.

> The end result is incorrectly initialised IO space. Often we can get
> away with that, because we don't make much use of IO space. However
> virtio requires it, so we see eg:
>
> pci_bus 0000:00: root bus resource [io 0xffff] (bus address
> [0xffffffffffffffff-0xffffffffffffffff])
> PCI: Cannot allocate resource region 0 of device 0000:00:01.0, will remap
> virtio-pci 0000:00:01.0: can't enable device: BAR 0 [io size 0x0020] not
> assigned
>
> The simplest fix for now is to just stop using of_pci_range_to_resource(),
> and open-code the original imp`lementation, that's all we want it to do.

The same bug is likely to be present on microblaze and mips, which may or may
not
care about it. I'll ask Michal about whether microblaze actually has any
I/O space, otherwise we have to fix it too for 3.18.

I believe for 3.19, we should probably migrate microblaze over to use the same
code
as ARM.

> Fixes: 0b0b0893d49b ("of/pci: Fix the conversion of IO ranges into IO
> resources")
> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
> ---
> arch/powerpc/kernel/pci-common.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/arch/powerpc/kernel/pci-common.c
> b/arch/powerpc/kernel/pci-common.c
> index bd70a51d5747..e5dad9a9edc0 100644
> --- a/arch/powerpc/kernel/pci-common.c
> +++ b/arch/powerpc/kernel/pci-common.c
> @@ -747,7 +747,11 @@ void pci_process_bridge_OF_ranges(struct pci_controller
> *hose,
> break;
> }
> if (res != NULL) {
> - of_pci_range_to_resource(&range, dev, res);
> + res->name = dev->full_name;
> + res->flags = range.flags;
> + res->start = range.cpu_addr;
> + res->end = range.cpu_addr + range.size - 1;
> + res->parent = res->child = res->sibling = NULL;
> }
> }

This looks reasonable to me as a hack to work around the breakage. It would be
good
to work together on this for 3.19 to move on to the a common implementation that
works on both ARM and PowerPC. This might be possibly by removing a lot of code
for PowerPC (at least 64-bit) that is now present in common code, but it will
change the structure of the powerpc implementation significantly, since the
returned numbers are now in different memory spaces (logical I/O space rather
than physical). The PowerPC _IO_BASE is the equivalent of the now generic
PCI_IOBASE,
but it's used slightly differently. If you want to use the generic code, 
you should probably change host->io_base_virt to host->io_base in logical
space (i.e. removing the _IO_BASE offset), or using hose->io_resource.start
instead.

       Arnd
diff mbox

Patch

diff --git a/arch/powerpc/kernel/pci-common.c b/arch/powerpc/kernel/pci-common.c
index bd70a51d5747..e5dad9a9edc0 100644
--- a/arch/powerpc/kernel/pci-common.c
+++ b/arch/powerpc/kernel/pci-common.c
@@ -747,7 +747,11 @@  void pci_process_bridge_OF_ranges(struct pci_controller *hose,
 			break;
 		}
 		if (res != NULL) {
-			of_pci_range_to_resource(&range, dev, res);
+			res->name = dev->full_name;
+			res->flags = range.flags;
+			res->start = range.cpu_addr;
+			res->end = range.cpu_addr + range.size - 1;
+			res->parent = res->child = res->sibling = NULL;
 		}
 	}
 }