Re: BGRT Pointer in System RAM

[Josh Triplett]

On Mon, Jul 15, 2013 at 04:08:13PM -0700, Andy Lutomirski wrote:
> On Mon, Jul 15, 2013 at 4:04 PM, Josh Triplett <josh@xxxxxxxxxxxxxxxx> wrote:
> > On Mon, Jul 15, 2013 at 01:28:36PM -0700, Andy Lutomirski wrote:
> >> On Sun, Jul 14, 2013 at 12:07 PM, Parag Warudkar <parag.lkml@xxxxxxxxx> wrote:
> >> > Saw this warning running latest git (Ubuntu daily mainline.) It looked
> >> > similar to what Andy saw on MSI hardware -
> >> > http://www.spinics.net/lists/linux-acpi/msg43410.html . The patch for
> >> > it doesn't seem to be merged, although it won't help in my case -
> >> > different hardware with valid status instead of invalid and image
> >> > address falling in system RAM instead of just being wild.
> >> >
> >> > Unsure  how this should be handled - moving the is_ram() check in
> >> > efi_bgrt_init and ignoring the BGRT in case where the check succeeds?
> >> > Doesn't sound completely right to me - since the BGRT is valid and
> >> > exists somewhere, but..
> >> >
> >> > [    0.015141] ------------[ cut here ]------------
> >> > [    0.015147] WARNING: CPU: 0 PID: 0 at
> >> > /home/apw/COD/linux/arch/x86/mm/ioremap.c:102
> >> > __ioremap_caller+0x312/0x390()
> >> > [ snip ]
> >> > [    0.015160] Call Trace:
> >> > [    0.015165]  [<ffffffff8170a704>] dump_stack+0x46/0x58
> >> > [    0.015169]  [<ffffffff8106406c>] warn_slowpath_common+0x8c/0xc0
> >> > [    0.015171]  [<ffffffff810640ba>] warn_slowpath_null+0x1a/0x20
> >> > [    0.015173]  [<ffffffff81054e32>] __ioremap_caller+0x312/0x390
> >> > [    0.015176]  [<ffffffff814013d4>] ? acpi_tb_verify_table+0x54/0x58
> >> > [    0.015179]  [<ffffffff81d35551>] ? efi_bgrt_init+0x8f/0x143
> >> > [    0.015181]  [<ffffffff81054f07>] ioremap_nocache+0x17/0x20
> >> > [    0.015183]  [<ffffffff81d35551>] efi_bgrt_init+0x8f/0x143
> >> > [    0.015186]  [<ffffffff81401d36>] ? acpi_tb_initialize_facs+0x32/0x34
> >> > [    0.015188]  [<ffffffff81d34e7f>] efi_late_init+0x9/0xb
> >> > [    0.015190]  [<ffffffff81d18f17>] start_kernel+0x3fd/0x419
> >> > [    0.015192]  [<ffffffff81d189ac>] ? do_early_param+0x87/0x87
> >> > [    0.015194]  [<ffffffff81d18120>] ? early_idt_handlers+0x120/0x120
> >> > [    0.015196]  [<ffffffff81d185e6>] x86_64_start_reservations+0x2a/0x2c
> >> >
> >> > ioremap.c:102
> >> >     /*
> >> >          * Don't allow anybody to remap normal RAM that we're using..
> >> >          */
> >> >         last_pfn = last_addr >> PAGE_SHIFT;
> >> >         for (pfn = phys_addr >> PAGE_SHIFT; pfn <= last_pfn; pfn++) {
> >> >                 int is_ram = page_is_ram(pfn);
> >> >
> >> >                 if (is_ram && pfn_valid(pfn) && !PageReserved(pfn_to_page(pfn)))
> >> >                         return NULL;
> >> >                 WARN_ON_ONCE(is_ram);
> >> >         }
> >> >
> >> > Looking at the BGRT table from IASL, the status seems to be valid but
> >> > the image address *seems* to me that is falling under system RAM.
> >>
> >> Interesting.  My BGRT says:
> >>
> >> [028h 0040   8]                Image Address : 0D06801800000001
> >>
> >> If I reverse the high and low 32-bit dwords, then I get an address in
> >> system RAM.
> >
> > Does that address in RAM start with a BMP header?
> 
> No idea.  I'd presumably have to modify the driver to find out --
> otherwise something else will overwrite it.

You could boot with a mem= command-line argument that reserves that
memory.

> > Because that would be *special*.  I don't think it's worth trying to
> > cope with that bug; better to just write off the BGRT as invalid if the
> > BIOS can't get endianness right.
> >
> > In theory we could guess at that bug if the unmangled address points to
> > a location in RAM starting with a BMP header.  In practice, let's not; a
> > missing BGRT is a purely cosmetic issue, and BIOS vendors can learn to
> > get that one right if they want to see their logo during Linux boot.
> > This won't break fastboot support, it just breaks fancy crossfades from
> > the BIOS logo to a Linux desktop or splash.
> 
> FWIW, the address that my BIOS gives is non-canonical.  Maybe that's
> good enough.

What do you mean by "non-canonical".

> > So, a "firmware bug" message in dmesg seems sufficent for that case.  We
> > do need to handle the case of a valid pointer into memory that e820
> > calls system RAM, as well as the case of a valid pointer into memory
> > reserved for the BIOS or similar, but not the case of an invalid
> > pointer.
> 
> Is the efi_bgrt code called early enough that data in system RAM will
> still be there?

In theory, it should always point to data in EFI's "reserved until after
boot time" memory.

- Josh Triplett
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/