[dpdk-dev] [RFC v2] porting AddressSanitizer feature to DPDK

[Lin, Xueqin]

> -----Original Message-----
> From: David Marchand <david.marchand at redhat.com>
> Sent: Wednesday, June 30, 2021 4:34 PM
> To: Burakov, Anatoly <anatoly.burakov at intel.com>
> Cc: Lin, Xueqin <xueqin.lin at intel.com>; Jerin Jacob <jerinjacobk at gmail.com>;
> Peng, ZhihongX <zhihongx.peng at intel.com>; Ananyev, Konstantin
> <konstantin.ananyev at intel.com>; Stephen Hemminger
> <stephen at networkplumber.org>; dpdk-dev <dev at dpdk.org>
> Subject: Re: [dpdk-dev] [RFC v2] porting AddressSanitizer feature to DPDK
> 
> On Mon, Jun 28, 2021 at 4:22 PM Burakov, Anatoly
> <anatoly.burakov at intel.com> wrote:
> >
> > On 18-Jun-21 10:04 AM, David Marchand wrote:
> > > On Fri, Jun 18, 2021 at 9:49 AM Lin, Xueqin <xueqin.lin at intel.com> wrote:
> > >>>> Suggest listing demo code and tool capture information for user
> > >>>> to try if
> > >>> tool works, also add this part into doc.
> > >>>>
> > >>>>>
> > >>>>> # Also, Please update the release note for this feature.
> > >>>> Sure, we can update the release note if code merge.
> > >>>
> > >>> Probably you can send v1 version next i.e change the RFC status to
> > >>> get merged.
> > >>
> > >> Sure, we will send v1 patch if no obvious objection for that, hope patch
> could receive some ACKs and could success to merge, thanks.
> > >
> > > How did you test this work?
> > >
> > > UNH recently started testing with ASAN and it reveals leaks just in
> > > the unit test.
> > >
> > > Merging these annotations will help catch more issues.
> > > But users will hit the current issues that we must fix first.
> > >
> >
> > As far as i can tell, the regular build is not affected by this patch,
> > so no issues will be hit until someone actually runs the test. IMO
> > it's better to merge it early to catch more issues than to gate the
> > feature on the condition that we fix all bugs unrelated to this feature first.
> 
> - This is affecting more than unit tests.
> 
> $ meson setup build-asan -Db_lundef=false -Db_sanitize=address ...
> 
> $ ninja-build -C build-asan
> ninja: Entering directory `build-asan'
> [2801/2801] Linking target app/test/dpdk-test
> 
> $ ./devtools/test-null.sh build-asan
> EAL: Detected 28 lcore(s)
> EAL: Detected 1 NUMA nodes
> EAL: Detected static linkage of DPDK
> EAL: WARNING! Base virtual address hint (0x100005000 !=
> 0x7fb31c632000) not respected!
> EAL:    This may cause issues with mapping memory into secondary processes
> EAL: Multi-process socket /run/user/1001/dpdk/rte/mp_socket
> EAL: Selected IOVA mode 'VA'
> EAL: VFIO support initialized
> EAL: WARNING! Base virtual address hint (0x10000b000 !=
> 0x7fb31c3b2000) not respected!
> EAL:    This may cause issues with mapping memory into secondary processes
> EAL: WARNING! Base virtual address hint (0x100011000 !=
> 0x7fb31c375000) not respected!
> EAL:    This may cause issues with mapping memory into secondary processes
> EAL: WARNING! Base virtual address hint (0x100017000 !=
> 0x7fb319bfe000) not respected!
> EAL:    This may cause issues with mapping memory into secondary processes
> Interactive-mode selected
> Auto-start selected
> [...]
> Bye...
> EAL: recvmsg failed, Bad file descriptor
> EAL: recvmsg failed, Bad file descriptor
> EAL: recvmsg failed, Bad file descriptor
> EAL: recvmsg failed, Bad file descriptor
> EAL: recvmsg failed, Bad file descriptor
> EAL: recvmsg failed, Bad file descriptor
> 
> Infinite loop of those messages.
> In the thread with Owen, we also noticed what looks like a deadlock with
> multiprocess when ASAN is enabled.
> 
> 
> - Adding a new feature on top of something that does not work yet seems at
> best premature to me.
> This patch does not seem that much tested, since those issues above are
> fairly easy to catch.

Great that you have tried the patch and capture the issue. It is DPDK issue not this patch issue.
We also captured this type issues and selected one, reported in internal system JARA and bugzilla in May. 
Pls check https://bugs.dpdk.org/show_bug.cgi?id=722 for detail, some developers are analysis this issue in JIRA but not resolved so far, Anatoly should also know this.
It is DPDK issue, not this patch issue, it doesn't mean we not do much test. 
We put all of thousands of regression cases as case list for this patch test, some issues still are tracking. 
> 
> Anyway, the memory allocator is your stuff, so your call.
> 
> Prefix for the title of such a patch should be mem:
> 
> 
> --
> David Marchand