loopback mount failing on 2.6.0-test7
Looks like some recent changes involving passing more errors up fromget_user_pages is making loopback mounting fail. I've traced this down to mount trying to do an mlockall (it probably shouldn't be doing this for what it's attempting to do). The test that's failing is:
!(flags & vma->vm_flags)
The values in question are:
flags: 0x1 VM_READvm_flags: 0x2070 VM_LOCKED, VM_MAY{READ,WRITE,EXEC}
I tested a 2.4 kernel, and the same test fails, but it's not propagated out to the system call. It appears that the failing region is the "---p" mapping of libc. x86 has no such mapping. Is this a broken use of mlockall, or does the error propagation change make it so mlockall cannot work on ia64? Thanks,
Alex
Signature · 1 line
Alex Williamson HP Linux & Open Source Lab
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Alex Williamson
Alex> Looks like some recent changes involving passing more errors up from Alex> get_user_pages is making loopback mounting fail. I've traced this down Alex> to mount trying to do an mlockall (it probably shouldn't be doing this Alex> for what it's attempting to do). The test that's failing is:
Alex> !(flags & vma->vm_flags)
Alex> The values in question are:
Alex> flags: 0x1 VM_READ
Alex> vm_flags: 0x2070 VM_LOCKED, VM_MAY{READ,WRITE,EXEC}
Alex> I tested a 2.4 kernel, and the same test fails, but it's not propagated Alex> out to the system call. It appears that the failing region is the Alex> "---p" mapping of libc. x86 has no such mapping. Is this a broken use Alex> of mlockall, or does the error propagation change make it so mlockall Alex> cannot work on ia64? Thanks,
Andi Kleen hacked some stuff in this area recently. I didn't follow the discussion too closely, but he complained that mlockall() failed for "no-access" mappings. I thought the fix went into test7 already, but I may be misremembering and perhaps it's only in the current bk tree. Could you check Linus' bk tree and see if it fixes the problem?
--david
Quoted 6 lines · David Mosberger
Andi Kleen hacked some stuff in this area recently. I didn't follow the discussion too closely, but he complained that
mlockall()failed for "no-access" mappings. I thought the fix went into test7 already, but I may be misremembering and perhaps it's only in the current bk tree. Could you check Linus' bk tree and see if it fixes the problem?
Sure enough, seems to be fixed in Linus' bk. It was still broken in
linux-ia64-2.5 as of yesterday though. Thanks,
Alex
Signature · 1 line
Alex Williamson HP Linux & Open Source Lab
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Alex Williamson
Alex> Sure enough, seems to be fixed in Linus' bk. It was still
Alex> broken in linux-ia64-2.5 as of yesterday though.
Yes, my tree is strictly 2.6.0-test7+ia64 fixes, at the moment.
--david