Comment 0 for bug 1023365

Impact: When running PXE installations of 32bit HVM Xen guest there is a high chance of the installation crashing with one or more messages about bad page table entries (the error code states that reserved page table flags were detected as being set). This affects Precise. Testing with Quantal (after merging 4.1.3) has not shown this issue.

<1>[ 563.253594] frontend: Corrupted page table at address 91ead10
<4>[ 563.253611] *pdpt = 00000000377e1001 *pde = 000000003db5b067
<0>[ 563.253624] Bad pagetable: 000c [#1] SMP
<4>[ 563.253634] Modules linked in: xfs reiserfs jfs btrfs zlib_deflate libcrc32c ntfs vfat fat ext2 usb_storage vga16fb vgastate floppy
<4>[ 563.253849]
<4>[ 563.253858] Pid: 9699, comm: frontend Not tainted 3.2.0-23-generic-pae #36-Ubuntu Xen HVM domU
<4>[ 563.253871] EIP: 0073:[<b75ceb65>] EFLAGS: 00010246 CPU: 0
<4>[ 563.253880] EIP is at 0xb75ceb65
<4>[ 563.253887] EAX: 0931d737 EBX: 090070f8 ECX: 0000000a EDX: 091ead10
<4>[ 563.253894] ESI: 00000000 EDI: 0931d738 EBP: 0923f4f0 ESP: bf930208
<4>[ 563.253901] DS: 007b ES: 007b FS: 0000 GS: 0033 SS: 007b
<0>[ 563.253909] Process frontend (pid: 9699, ti=ed3a0000 task=ed2d0ca0 task.ti=ed3a0000)
<4>[ 563.253914]
<0>[ 563.253918] EIP: [<b75ceb65>] 0xb75ceb65 SS:ESP 007b:bf930208
<4>[ 563.253976] ---[ end trace e78469a17274d536 ]---

Fix: There is a patch in the upstream Xen repositories which came between Xen version 4.1.2 and 4.1.3. That change modifies the lookup of guest pages so it will not check for reserved flags when the page is not set to be present.

Testcase: Installing a Xen HVM 32bit guest (2 VCPUs, 1MB memory) only rarely succeeded without the patch. Testing with the modified Xen hypervisor source succeeded 3 times in a row.