Re: [Xen-users] DomU sees only 512MB RAM with PCI-passthrough

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Wed, May 29, 2013 at 05:23:42PM +0200, Christian Holpert wrote:
> ---- Original-Nachricht ----
>   Von: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
>   Datum: Mittwoch, 29. Mai 2013 17:02:32
> >So with a 2048 guest it boots for me. Here is the diff between
> >your kernel and mine:
> >
> >Can you try the v3.10 or v3.9 and see how it works for you?
> Tried out v3.10-rc3 and can confirm that your patch works.

Great! So something is off with v3.8 kernel. The hint seems
to be:

Populating 100000-160000 pfn range: 393216 pages added		Populating 100000-160000 pfn range: 393216 pages added
..
Xen: [mem 0x0000000100000000-0x000000015fffffff] usable		Xen: [mem 0x0000000100000000-0x000000015fffffff] usable
bootconsole [xenboot0] enabled				      <
NX (Execute Disable) protection: active				NX (Execute Disable) protection: active
DMI not present or invalid.					DMI not present or invalid.
e820: last_pfn = 0x160000 max_arch_pfn = 0x1000000		e820: last_pfn = 0x160000 max_arch_pfn = 0x1000000
init_memory_mapping: [mem 0x00000000-0x2d3fdfff]	      |	Scanning 1 areas for low memory corruption
(XEN) mm.c:864:d6 Non-privileged (6) attempt to map I/O space |	init_memory_mapping: [mem 0x00000000-0x000fffff]
(XEN) mm.c:5317:d6 ptwr_emulate: could not get_page_from_l1e( |	init_memory_mapping: [mem 0x1fe00000-0x1fffffff]
BUG: unable to handle kernel NULL pointer dereference at   (n |	init_memory_mapping: [mem 0x1c000000-0x1fdfffff]
IP: [<c157fe96>] xen_set_pte_init+0x3b/0x43		      |	init_memory_mapping: [mem 0x00100000-0x1bffffff]
*pdpt = 0000000000000000 *pde = c2c2c2c2c2c2c2c2 	      |	RAMDISK: [mem 0x01b51000-0x107b0fff]

That v3.8 is trying to allocate pages up to 2d3fdfff. Which we don't in v3.10.

But we do seem to use said area for HighMem (0x2d3fe000-0x5fffffff) which
works in v3.10.

Oops: 0003 [#1] SMP 					      |	4908MB HIGHMEM available.
Modules linked in:					      |	723MB LOWMEM available.
Pid: 0, comm: swapper Not tainted 3.8.13-gentoo #1  	      |	  mapped low ram: 0 - 20000000
EIP: e019:[<c157fe96>] EFLAGS: 00010046 CPU: 0		      |	  low ram: 0 - 2d3fe000
EIP is at xen_set_pte_init+0x3b/0x43			      |	Zone ranges:
EAX: c2380000 EBX: 00000000 ECX: 20000403 EDX: 20000403	      |	  DMA      [mem 0x00001000-0x00ffffff]
ESI: 00020001 EDI: 00000000 EBP: 20000403 ESP: c1545e18	      |	  Normal   [mem 0x01000000-0x2d3fdfff]
 DS: e021 ES: e021 FS: 00d8 GS: 00e0 SS: e021		      |	  HighMem  [mem 0x2d3fe000-0x5fffffff]
CR0: 80050033 CR2: 00000000 CR3: 015ec000 CR4: 00002660	      |	Movable zone start for each node
DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000	      |	Early memory node ranges
DR6: ffff0ff0 DR7: 00000400				      |	  node   0: [mem 0x00001000-0x0009ffff]
Process swapper (pid: 0, ti=c1544000 task=c1550120 task.ti=c1 |	  node   0: [mem 0x00100000-0x1fffffff]
Stack:							      |	  node   0: [mem 0x00000000-0x5fffffff]
 00000000 20000403 00000000 00020000 e0000000 c158f111 000000 |	Using APIC driver default
 c15ec018 00000003 c15ec018 00000003 20000000 00000000 0002d3 |	smpboot: Allowing 4 CPUs, 0 hotplug CPUs
 00000100 c2380000 00000001 00000000 fffe0000 0016d000 000000 |	Local APIC disabled by BIOS -- you can enable it with "lapic"
Call Trace:						      |	APIC: disable apic facility

Would need to dig in this deeper, but that looks to be
an issue with the kernel, not with the toolstack.