On 2018-Nov-1, at 5:41 PM, Konstantin Belousov <kostikbel at gmail.com> wrote:
> On Tue, Oct 30, 2018 at 12:45:13PM -0700, Mark Millard wrote: >> Konstantin Belousov kostikbel at gmail.com wrote on >> Tue Oct 30 18:04:04 UTC 2018 : >> >>> On Tue, Oct 30, 2018 at 03:32:40PM +0000, Alexander Richardson wrote: >>>> On Tue, 30 Oct 2018 at 10:17, Michael Tuexen >>>> <Michael.Tuexen at macmic.franken.de> wrote: >>>>> >>>>>> On 29. Oct 2018, at 22:08, Alex Richardson <arichardson at FreeBSD.org> >>>>>> wrote: >>>>>> >>>>>> Author: arichardson >>>>>> Date: Mon Oct 29 21:08:02 2018 >>>>>> New Revision: 339876 >>>>>> URL: https://svnweb.freebsd.org/changeset/base/339876 >>>>>> >>>>>> Log: >>>>>> rtld: set obj->textsize correctly >>>>>> >>>>>> With lld-generated binaries the first PT_LOAD will usually be a read-only >>>>>> segment unless you pass --no-rosegment. For those binaries the textsize >>>>>> is >>>>>> determined by the next PT_LOAD. To allow both LLD and bfd 2.17 binaries >>>>>> to >>>>>> be parsed correctly use the end of the last PT_LOAD that is marked as >>>>>> executable instead. >>>>>> >>>>>> I noticed that the value was wrong while adding some debug prints for >>>>>> some rtld >>>>>> changes for CHERI binaries. `obj->textsize` only seems to be used by PPC >>>>>> so the >>>>>> effect is untested. However, the value before was definitely wrong and >>>>>> the new >>>>>> result matches the phdrs. >>>>> I build kernel and world with a revision later than this on a PPC. >>>>> Buildword >>>>> ends up with a world where almost all binaries are segfaulting.... >>>>> Especially gdb >>>>> (but svn, ls or so all segfault). >>>>> >>>>> Best regards >>>>> Michael >>>> >>>> This is rather surprising since if anything the range of the icache >>>> flush should increase rather than decrease after this change. >>>> >>>> I can only see this causing a behaviour change if we actually need to >>>> flush more than just the executable segments. >>>> Is it possible that some binary/library contains a non-executable >>>> segment as the first PT_LOAD? >>>> Or is there some linker script that adds custom PHDRS? >>>> >>> Could it be that there is a hole between start of the object mapping and >>> the last PT_LOADable segment eligible for execution ? >> >> [This note may be easier to deal with than the first >> note that I sent out.] >> >> [My examples are from devel/powerpc64-xtoolchain-gcc used >> to buildworld buildkernel targeting a head -r339076 based >> powerpc64 environment. I do that on powerpc64 as well.] >> >> powerpc64 loads the readonly data and the readonly code in one PT_LOAD, >> the first. The 2nd PT_LOAD is for sections without the readonly status, >> that includes .got and .plt being spanned. See below from >> objdump -ph for /bin/ls : >> >> Program Header: >> PHDR off 0x0000000000000040 vaddr 0x0000000010000040 paddr >> 0x0000000010000040 align 2**3 >> filesz 0x0000000000000188 memsz 0x0000000000000188 flags r-- >> INTERP off 0x00000000000001c8 vaddr 0x00000000100001c8 paddr >> 0x00000000100001c8 align 2**0 >> filesz 0x0000000000000015 memsz 0x0000000000000015 flags r-- >> LOAD off 0x0000000000000000 vaddr 0x0000000010000000 paddr >> 0x0000000010000000 align 2**16 >> filesz 0x000000000000910c memsz 0x000000000000910c flags r-x >> LOAD off 0x0000000000009110 vaddr 0x0000000010019110 paddr >> 0x0000000010019110 align 2**16 >> filesz 0x0000000000000ee0 memsz 0x00000000000010e8 flags rw- >> DYNAMIC off 0x0000000000009138 vaddr 0x0000000010019138 paddr >> 0x0000000010019138 align 2**3 >> filesz 0x00000000000001c0 memsz 0x00000000000001c0 flags rw- >> NOTE off 0x00000000000001e0 vaddr 0x00000000100001e0 paddr >> 0x00000000100001e0 align 2**2 >> filesz 0x0000000000000030 memsz 0x0000000000000030 flags r-- >> STACK off 0x0000000000000000 vaddr 0x0000000000000000 paddr >> 0x0000000000000000 align 2**4 >> filesz 0x0000000000000000 memsz 0x0000000000000000 flags rw- > > We only need program headers, and we only need them from the object > which load causes the fault. It might be not the binary but some library > that triggers the fault. So the backtrace and some information from the > state of the image is needed. > > You can build only rtld and use it as the standalone program to initiate > the image activation: > <path>/ld-elf.so.1 /bin/ls > or similar. It will be a while before I can get an environment set up which allows me to get a backtrace from a 32-bit powerpc system. I have demonstrated that I get the crash with what I tried to verify such. I'll work on getting an environment in place that allows what you are asking for. Problem reproduction details (if you care) . . . I've been trying to provide evidence but my 32-bit powerpc environment is not as modern as has the problem. Also: the buildworld it was installed from was built via clang, which means that the libgcc_s.so that was built does not correctly support C++ exceptions. (clang's generated output is the source of that problem.) I did not build /libexec/gdb and the /usr/local/gdb uses C++ exceptions heavily and will not work. All I can report for now is how I reproduced the problem. . . # mkdir /339870 # cd /339870/ then I used wget to get: https://artifact.ci.freebsd.org/snapshot/head/r339870/powerpc/powerpc/base*.txz and expanded them in /339870/ and did: # ./libexec/ld-elf.so.1 /bin/ls Segmentation fault (core dumped) Looking at where my environment is putting cores: -rw------- 1 root wheel 2752512 Nov 1 18:12:40 2018 /var/crash/ld-elf.so.1.9906.core So at least I know the problem reproduces via such a procedure. For reference for the above: # uname -apKU FreeBSD FBSDG4S 12.0-CURRENT FreeBSD 12.0-CURRENT r327364M powerpc powerpc 1200054 1200054 === Mark Millard marklmi at yahoo.com ( dsl-only.net went away in early 2018-Mar) _______________________________________________ svn-src-head@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/svn-src-head To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
