[Bug 668872] New: kdump does not work - ioremap: invalid physical address
https://bugzilla.novell.com/show_bug.cgi?id=668872 https://bugzilla.novell.com/show_bug.cgi?id=668872#c0 Summary: kdump does not work - ioremap: invalid physical address Classification: openSUSE Product: openSUSE 11.4 Version: Milestone 5 of 6 Platform: Other OS/Version: Other Status: NEW Severity: Minor Priority: P5 - None Component: Kernel AssignedTo: joerg.roedel@amd.com ReportedBy: trenn@novell.com QAContact: qa@suse.de CC: hare@novell.com, agraf@novell.com, boris.ostrovsky@amd.com Found By: Development Blocker: --- On a Dinar AMD system kdump does not work with latest kernel (openSUSE 11.4, 2.6.37.20). I run into this when finding something else causing a kernel crash. Might be related to iommu? -> already assigning to Joerg. I doubt someone else here would otherwise look at this issue for 11.4. iommu=soft does not help. Hmm, it might behave a bit different with iommu=soft. I reserved 128M for the crash kernel, with iommu=soft it run into an out of memory condition, but when increasing to 256M also iommu=soft shows the same __ioremap WARN_ON and backtrace. I'll attach the full serial console output of the productive kernel and the kdump kernel booting. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=668872 https://bugzilla.novell.com/show_bug.cgi?id=668872#c1 --- Comment #1 from Thomas Renninger <trenn@novell.com> 2011-02-02 12:41:08 UTC --- Created an attachment (id=411744) --> (http://bugzilla.novell.com/attachment.cgi?id=411744) Serial console output - __ioremap segfault somewhere in the end Causing /proc/vmcore not getting exported in the crash kernel and the dump fails. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=668872 https://bugzilla.novell.com/show_bug.cgi?id=668872#c2 --- Comment #2 from Joerg Roedel <joerg.roedel@amd.com> 2011-02-02 14:12:08 UTC --- The key seems to be this line:
[ 2.462838] ioremap: invalid physical address db74000000000000
This address is larger than the biggest physical address supported by AMD cpus (support up to 48 bits) and even bigger than the architectural physical address limit of 52 bits. According to the backtrace the address is read from the elf-note section. Is it possible that the elf-image is broken and contains such an address for some reason? Is this address always the same when starting the kdump kernel? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=668872 https://bugzilla.novell.com/show_bug.cgi?id=668872#c3 --- Comment #3 from Thomas Renninger <trenn@novell.com> 2011-02-02 14:42:33 UTC ---
Is it possible that the elf-image is broken and contains such an address for some reason? afaik this code creates an elf header (or a helper first) per continuous memory chunks it likes to map and export later in /proc/vmcore. Looks like it needs some sanity check there?
Is this address always the same when starting the kdump kernel? Yes, doing a grep (grep ioremap screenlog.krommer) on my serial console log for this machine I get 8 times exactly the same matches: ioremap: invalid physical address db74000000000000 WARNING: at /usr/src/packages/BUILD/kernel-desktop-2.6.37/linux-2.6.37/arch/x86/mm/ioremap.c:83 __ioremap_caller+0x304/0x380() [<ffffffff810324a4>] __ioremap_caller+0x304/0x380
-- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=668872 https://bugzilla.novell.com/show_bug.cgi?id=668872#c4 --- Comment #4 from Joerg Roedel <joerg.roedel@amd.com> 2011-02-02 15:28:37 UTC --- To me it looks like the wrong physical address is read from the old kernels memory. In the parse_crash_elf64_headers() function the elfcorebuf is filled with read_from_old_mem(). This buffer is passed later to the merge_note_headers_elf64() function which iterates over the buffer and reads an offset from it (offset = phdr_ptr->p_offset;). This offset is later passed to read_from_oldmem again which causes the warning at some point. So my guess is that the old kernel contains the invalid physical address in one of its PT_NOTE type headers. The question is why this address is there. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=668872 https://bugzilla.novell.com/show_bug.cgi?id=668872#c Joerg Roedel <joerg.roedel@amd.com> changed: What |Removed |Added ---------------------------------------------------------------------------- AssignedTo|joerg.roedel@amd.com |trenn@novell.com -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=668872 https://bugzilla.novell.com/show_bug.cgi?id=668872#c5 Joerg Roedel <joerg.roedel@amd.com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |joerg.roedel@amd.com --- Comment #5 from Joerg Roedel <joerg.roedel@amd.com> 2011-06-20 09:05:08 UTC --- Re-Assigned since this does not look like an IOMMU problem. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=668872 https://bugzilla.novell.com/show_bug.cgi?id=668872#c6 Thomas Renninger <trenn@novell.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Priority|P5 - None |P4 - Low Status|NEW |NEEDINFO InfoProvider| |trenn@novell.com --- Comment #6 from Thomas Renninger <trenn@novell.com> 2011-06-20 14:49:35 UTC --- I'll try to reproduce with latest mainline kernel. This may take a while because of other deadlines and kdump has not high prio for 11.4. -> Setting needinfo to myself. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
participants (1)
-
bugzilla_noreply＠novell.com