New subject: [PATCH] pci: fix BAR setup

10 May 2012


      Hi Artyom, all,
On 04/19/2012 10:29 PM, Artyom Tarasenko wrote:
...
On Thu, Apr 19, 2012 at 10:04 PM, Blue Swirl blauwirbel@gmail.com wrote:
...
On Thu, Apr 19, 2012 at 19:06, Artyom Tarasenko atar4qemu@gmail.com wrote:
...
On Sun, Mar 11, 2012 at 11:59 AM, Blue Swirl blauwirbel@gmail.com wrote:
...
A change in QEMU on how PCI bridges are setup revealed
a bug in OpenBIOS PCI setup. On Sparc64, the BARs just
happened to get somewhat correct values by accident before
the commit but not after the change.
Don't use arch->io_base for PCI I/O port BARs.
Fix Sparc64 PCI memory base.
Could it be that the ranges property needs to be adjusted to look as
it did before the patch?
This patch breaks HelenOS-0.4.3/sparc boot. But if I modify the ranges
property to the old value, HelenOS boots.
0 > cd /pci@1fe,0  ok
0 > .properties
[...]
ranges                    -- 54 : 00 00 00 00 00 00 00 00 00 00 00 00
00 00 01 fe 01 00 00 00 00 00 00 00 02 00 00 00 01 00 00 00 00 00 00
00 00 00 00 00 00 00 01 fe 02 00 00 00 00 00 00 00 00 01 00 00 02 00
00 00 00 00 00 00 00 10 00 00 00 00 01 ff 00 00 00 00 00 00 00 00 10
00 00 00
Not sure about the best formatting. If I reformat it like this:
00000000 00000000 00000000 000001fe 01000000 00000000 02000000
01000000 00000000 00000000 000001fe 02000000 00000000 00010000
02000000 00000000 00100000 000001ff 00000000 00000000 10000000
                              ^^^^ this one (the third) used to be 0
On an Ultra-5 this should be
00000000.00000000.00000000.000001fe.01000000.00000000.01000000.
01000000.00000000.00000000.000001fe.02000000.00000000.01000000.
02000000.00000000.00000000.000001ff.00000000.00000001.00000000.
03000000.00000000.00000000.000001ff.00000000.00000001.00000000
http://git.kernel.org/?p=linux/kernel/git/davem/prtconfs.git;a=blob;f=ultra5...
I guess the problem is that the properties are added by finding the
host bridge in the PCI bus (which btw isn't so nice), so we use the
pci_mem_base. Using pci_mem_base = 0 would probably fix the BAR, but
then VGA would get in the way. Probably something more complex is
needed.
Hmm. Complicated. What is actually the meaning of this property? Don't
we need the values ourselves?
If these are just BARs, can't we pass the configuration from qemu?
I am now looking into this from the HelenOS point of view. The problem
is that HelenOS miscomputes the framebuffer physical address.
I think I found the culprit in the ranges property of the PCI node. The
mechanism of the bug seems to be as follows.
HelenOS reads the parent-relative address of the VGA node from the reg
property. The address is 0x008000000. It then wants to apply it on the
parent node's (PCI) ranges property. Using a packed C structure, a range
is defined as follows:
typedef struct {
        uint32_t space;
        uint64_t child_base;
        uint64_t parent_base;
        uint64_t size;
};
Note that in our case, the matching range is:
02000000 00000000 00100000 000001ff 00000000 00000000 10000000
child_base is 0x00100000
parent_base is 0x1ff00000000
reg_addr is 0x00800000
HelenOS then computes the resulting physical address using this formula:
pa = parent_base + (reg_addr - child_base);
The result is thus 0x1ff00700000 instead of the correct value 0x1ff00800000.
So it looks like either the child base should be 0, or reg_addr must be
0x00900000, or parent_base must be 0x1ff00100000. From the way this
works e.g. on Sun E6500, the last option seems to be the correct one.
What do you think of it?
Jakub

Re: [OpenBIOS] [PATCH] pci: fix BAR setup