New: memset hang on pci device memory using kvm

https://bugzilla.kernel.org/show_bug.cgi?id=115451

            Bug ID: 115451
           Summary: memset hang on pci device memory using kvm
           Product: Virtualization
           Version: unspecified
    Kernel Version: RHEL - 2.6.18-308.el5
          Hardware: Intel
                OS: Linux
              Tree: Mainline
            Status: NEW
          Severity: high
          Priority: P1
         Component: kvm
          Assignee: [email protected]
          Reporter: [email protected]
        Regression: No

i have host x86 based rhel system [kernel version - 2.6.18-308.el5]. On this I am trying to boot up a debian amd64 image using qemu with kvm.

On my virtual os i have a pcimem application, through which I R/W any configured memory on pci devices.
Issue is whenever i try to use memset in excess of 156 bytes in my pcimem application to configure any pci device memory, call to memset hangs. However, memcpy work fine in all scenarios.

memset works fine if boot my virtual os without kvm.

my kvm version is :
$/usr/libexec/qemu-kvm --help
QEMU PC emulator version 0.9.1 (kvm-83-maint-snapshot-20090205), Copyright (c) 2003-2008 Fabrice Bellard

You are receiving this mail because:
You are watching the assignee of the bug.

https://bugzilla.kernel.org/show_bug.cgi?id=115451

Huaitong Han <[email protected]> changed:

What |Removed |Added
----------------------------------------------------------------------------
CC| |[email protected]

--- Comment #1 from Huaitong Han <[email protected]> ---
You should try reproduce it on upstream kernel, and list steps if it can be reproduced, so that many people can help to debug it.

You are receiving this mail because:
You are watching the assignee of the bug.

https://bugzilla.kernel.org/show_bug.cgi?id=115451

--- Comment #2 from Gaurav Sharma <[email protected]> --- Following is the system info :
Host machine kernel version : 2.6.18-308.el5
Guest OS : Debian 3.2.0-4-686-pae
KVM vesrion : QEMU PC emulator version 0.9.1

Following are the steps to re-create the issue :
1. Boot the guest OS using QEMU with KVM enabled.
2. lspci will list down the pci devices configured on the guest OS.
3. I have a some PCI device configured with bar0 memory.
4. modify the pcimem.c application to use call memset

// This code is being modified to call memset to do a bulk write transaction case 'w':

         fprintf(stderr, "Before doing memset\n");
         memset((virt_addr), 1, datalen);
         fprintf(stderr, "after doing memset\n");
         break;

5. Call the pcimem application to write on bar0 memory.
6. It results in memset call in hang, and the application needs to be explicitly killed. I do not get the log message after memset call.

You are receiving this mail because:
You are watching the assignee of the bug.

https://bugzilla.kernel.org/show_bug.cgi?id=115451

--- Comment #3 from Gaurav Sharma <[email protected]> --- Created attachment 211901
--> https://bugzilla.kernel.org/attachment.cgi?id=211901&action=edit
Attached is the pcimem application that i use

You are receiving this mail because:
You are watching the assignee of the bug.

https://bugzilla.kernel.org/show_bug.cgi?id=115451

--- Comment #4 from Gaurav Sharma <[email protected]> --- Also, when memset fails, i get following from dmesg on host machine :
emulation failed (pagetable) rip 410bab 66 0f 7f 07 emulation failed (pagetable) rip 410bab 66 0f 7f 07 emulation failed (pagetable) rip 410bab 66 0f 7f 07 emulation failed (pagetable) rip 410bab 66 0f 7f 07

You are receiving this mail because:
You are watching the assignee of the bug.

https://bugzilla.kernel.org/show_bug.cgi?id=115451

Paolo Bonzini <[email protected]> changed:

What |Removed |Added
----------------------------------------------------------------------------
CC| |[email protected]

--- Comment #5 from Paolo Bonzini <[email protected]> --- Please try on KVM and QEMU version that are less than 7 years old...

You are receiving this mail because:
You are watching the assignee of the bug.

https://bugzilla.kernel.org/show_bug.cgi?id=115451

--- Comment #6 from Gaurav Sharma <[email protected]> --- I am using qemu version 2.2.0 and kvm modules are the default with the host kernel version [RHEL 5.8], which doesn't seems to be that old to me. Any data point am i missing ?

You are receiving this mail because:
You are watching the assignee of the bug.

https://bugzilla.kernel.org/show_bug.cgi?id=115451

--- Comment #7 from Paolo Bonzini <[email protected]> ---
First, RHEL5.8 was released in February 2012, which is 4 years old. The latest RHEL release is 5.11 (September 2014). Second, RHEL5's KVM modules were forked from the upstream kernel around 2009. It hasn't had any features added for at least 5 years.

In any case, this bug tracker is only for the upstream kernel. Even if the bug was identified and fixed, you would have to open a support case with Red Hat to get it in the distribution---and you would probably get it only in RHEL7, or at most (quite unlikely) in RHEL6, because RHEL5 only has major security fixes applied at this point in time.

Anyway, from comment 4, it looks like RHEL5 doesn't have support for SSE instructions (movdqa in this case) to and from PCI BARs. This is supported in upstream since April 2012 (generic support for SSE came in March 2011, while movdqa was added one year later).

You are receiving this mail because:
You are watching the assignee of the bug.

https://bugzilla.kernel.org/show_bug.cgi?id=115451

Paolo Bonzini <[email protected]> changed:

What |Removed |Added
----------------------------------------------------------------------------

             Status|NEW                         |RESOLVED
         Resolution|---                         |PATCH_ALREADY_AVAILABLE

You are receiving this mail because:
You are watching the assignee of the bug.