Re: [dpdk-dev] Questions about mmap in Asan APP

[Thomas Monjalon <thomas@monjalon.net>]

06/11/2021 10:27, Min Hu (Connor):
> Hi, Jonas Pfefferle and all,
> 	I met one questions:
> 	When I ran APP on Kasan OS, secondary process occasionally init
> failed.
> 
> BTW:
> testpmd is built with Asan, that is to add "CFLAGS="-fsanitize=address 
> -fno-omit-frame-pointer -pthread"".
> Kasan OS is what opens this:
> CONFIG_SLUB_DEBUG=y
> CONFIG_KASAN=y
> 
> The PRIMARY process APP(testpmd) log:
> estuary:/home$ ./testpmd_kasan_debug    -w 0000:00:01.0 
> --file-prefix=test --proc-type=auto -l 0-1 -n 2 -- -i --rxq=4 --txq=4
> EAL: Detected 16 lcore(s)
> EAL: Detected 1 NUMA nodes
> Option -w, --pci-whitelist is deprecated, use -a, --allow option instead
> EAL: Auto-detected process type: PRIMARY
> EAL: Detected static linkage of DPDK
> EAL: Multi-process socket /var/run/dpdk/test/mp_socket
> EAL: Selected IOVA mode 'PA'
> EAL: No available 32768 kB hugepages reported
> EAL: No available 64 kB hugepages reported
> EAL: No available 1048576 kB hugepages reported
> EAL: VFIO support initialized
> 
> EAL: WARNING! Base virtual address hint (0xd01061000 != 0xfffb89e00000) 
> not respected!
> EAL:    This may cause issues with mapping memory into secondary processes
> 
> EAL:   Invalid NUMA socket, default to 0
> EAL: Using IOMMU type 8 (No-IOMMU)
> EAL: Probe PCI driver: net_hns3_vf (19e5:a22e) device: 0000:00:01.0 
> (socket 0)
> TELEMETRY: No legacy callbacks, legacy socket not created
> Interactive-mode selected
> testpmd: create a new mbuf pool <mb_pool_0>: n=155456, size=2176, socket=0
> testpmd: preferred mempool ops selected: ring_mp_mc
> 
> Warning! port-topology=paired and odd forward ports number, the last 
> port will pair with itself.
> 
> Configuring Port 0 (socket 0)
> 0000:00:01.0 hns3_parse_rss_key(): Default RSS hash key to be set
> 
> Port 0: link state change event
> Port 0: 72:DE:F7:1B:F6:52
> Checking link statuses...
> 
> "EAL: WARNING! Base virtual address hint (0xd01061000 != 0xfffb89e00000) 
> not respected!  EAL:    This may cause issues with mapping memory into
> secondary processes" should be noticed.
> 
> 
> 
> The SECONDARY APP(proc-info) log is:
> estuary:/home$ ./proc-info -w 0000:00:01.0 --file-prefix=test 
> --proc-type=auto -- --xstats-name rx_q0_packets
> Option -w, --pci-whitelist is deprecated, use -a, --allow option instead
> EAL: Cannot get a virtual area at requested address: 0xfffba6000000 (got 
> 0xfffb8ae3b000)
> 
> EAL: Cannot reserve 17179869184 bytes at [0xfffba6000000] - please use 
> '--base-virtaddr' option
> 
> EAL: Cannot preallocate VA space for hugepage memory
> EAL: FATAL: Cannot init memory
> EAL: Cannot init memory
> PANIC in main():
> Cannot init EAL
> 4: [/lib/aarch64-linux-gnu/libc.so.6(__libc_start_main+0xe0) 
> [0xffff8be5a8a0]]
> 3: [./proc-info(main+0x1098) [0x52a3a8]]
> 2: [./proc-info(__rte_panic+0x9c) [0x48cfac]]
> 1: [./proc-info(rte_dump_stack+0x20) [0x80d4b0]]
> Aborted
> 
> "EAL: Cannot get a virtual area at requested address: 0xfffba6000000 
> (got 0xfffb8ae3b000)" should be noticed.
> 
> According to my analysis, the reason is that the system call "mmap"
> return the address which is different with the hint address(first
> parameter). That is to say：
> return_addr = mmap(void* start,size_t length,int prot,int flags,int
> fd,off_t offset);
> return_addr and start is different.
> 
> Why this happened? I guess Asan works based on shadow memory, shadow
> memory will occupy the memory zone and it cannot be one for mmap.
> So, it the shadow memory happens to be in the range of mmap region,
> mmap will return a new address which is free(not occupied by shadow).
> 
> BTW，I think DPDK multiple process is not supported in that situation,
> such as "Asan APP on Kasan OS".
> 
> What is your opinion? any will weclome, thanks.

Secondary process mapping is fragile.
I am not surprised there are more issues when enabling
some kernel features playing with memory ranges.
You may try to find more hints to make it work,
but it is OK stating this scenario is not supported.

Another outcome to this trial could be to document the limitation
in DPDK documentation for Linux.

Thanks