>>>>> (gdb) p nr_swapper_spaces >>>>> $4 = <optimized out> >>>> Thank you folks, for the various information. >>>> >>>> Apparently the data of nr_swapper_spaces exists in the kernel, but its >>>> symbol does not exist. I think this means that we cannot calculate the >>>> number of pages from swapper_spaces[] with such a debuginfo, because we >>>> cannot get the array size of a swapper_spaces entry. >>> That's correct, Kazu. >>> >>> >>>> So I'd like to go with Johan's patch. >>>> >>>> cc -c -g -DX86_64 -DLZO -DSNAPPY -DGDB_10_2 memory.c -Wall -O2 >>>> -Wstrict-prototypes -Wmissing-prototypes -fstack-protector >>>> -Wformat-security >>>> memory.c: In function dump_kmeminfo: >>>> memory.c:8612:50: warning: pointer targets in passing argument 2 of >>>> dump_vm_stat differ in signedness [-Wpointer-sign] >>>> 8612 | if (dump_vm_stat("NR_SWAPCACHE", >>>> &swapper_space_nrpages, 0)) { >>>> | >>>> ^~~~~~~~~~~~~~~~~~~~~~ >>>> | | >>>> | ulong * {aka >>>> long unsigned int *} >>>> memory.c:298:33: note: expected long int * but argument is of type ulong >>>> * {aka long unsigned int *} >>>> 298 | static int dump_vm_stat(char *, long *, ulong); >>>> | ^~~~~~ >>>> >>>> So with the following, >>>> >>>> - ulong swapper_space_nrpages; >>>> + long swapper_space_nrpages; >>> Looks good. Maybe the patch log also needs to be improved a little bit. >>> >>> Although the checking of kernel version looks not very good, seems no >>> better way. >> hmm, this sounds strange. My ack is for Johan's patch (attached), which > >Ah, I did not see the attached patch. Thank you for pointing out this >issue, Kazu. > >> does not have kernel version check, is this ok? sorry for confusing. > >Sounds good idea if there is no kernel version check. > >I still have one question: the following two commits were introduced >separately in kernel 5.12 and 4.11. > >b6038942480e ("mm: memcg: add swapcache stat for memcg v2") > >4b3ef9daa4fc ("mm/swap: split swap cache into 64MB trunks") > >It indicates that the attached patch may not work well between kernel >4.11 and 5.12, is it possible to encounter the similar issue(the >nr_swapper_spaces may be optimized away)? I have not noticed any problems with the attached patch on earlier kernel versions. For instance on GKI kernels based on 5.10 [1] the 'nr_swapper_spaces' variable is not optimized out.

- [1] https://source.android.com/docs/core/architecture/kernel/gki-android13-5_... Regards Johan