01:32 <Guest23417> Hi Can I access local variable by name ?

01:37 <Guest23417> e.g. in vfs_open() there is local variable struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags, 0);, Can want access this local variable by name, instead of using $$locals

02:31 <fche> $dentry

20:05 <irker474> systemtap: smakarov systemtap.git:master * release-4.2-64-g4f9715d3b / buildrun.cxx: PR25485: also disable -Wno-{pointer<->int}-cast for kernel modules :p

22:43 <agentzh> fche: what do you think of this patch? https://gist.github.com/agentzh/a465564e2c41f1353de46690eeb4f4cf

22:43 <agentzh> the kernel func snprintf would output hashed pointer values which obscure the userland pointers.

22:43 <agentzh> and recent kernels even obscure NULL pointers as well, even more confusing.

22:44 <agentzh> i already spent several hours on this...

22:44 <agentzh> wasted actually :)

22:50 <fche> ugh

22:50 <fche> yeah you know that makes sense

22:50 <fche> we can divorce from the kernel's %p hiding games in the few places we use the kernel's snprintf

22:51 <agentzh> okay, cool, will commit.

22:53 <agentzh> another quick question, will vmalloc() evict pages from the page cache/buffer when the free memory is not enough?

22:53 <fche> I believe it depends on the flags we pass

22:53 <agentzh> just wondering if we should reserve enough truly free pages for stap modules.

22:53 <fche> a little bit of resource freeing on our behalf should be fine

22:54 <agentzh> how about stap's own _stp_vzalloc()?

22:54 <fche> git grep :)

22:56 <agentzh> seems like vmalloc() does not take any flags. under the hood, it seems like GFP_KERNEL being passed.

22:57 <agentzh> return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);

22:57 <fche> ok

22:58 <agentzh> __vmalloc in turn calls kmalloc().

22:58 <agentzh> with some special flags.

23:19 <Guest97310> HI I using symname to get symbol name from address, but it desn't work, if the address is from a module, how can I let systemtap also look at the module symbol?

23:20 <fche> Guest97310, stap -d modulename or stap --all-modules

23:21 <Guest97310> Thanks!

23:27 <Guest97310> some header file is not in kernel header, such as fs/proc/internal.h,

23:27 <Guest97310> struct proc_inode { +0 struct pid *pid; +8 unsigned int fd; +16 union proc_op op; +24 struct proc_dir_entry *pde; +32 struct ctl_table_header *sysctl; +40 struct ctl_table *sysctl_entry; +48 struct hlist_node sysctl_inodes; +64 struct proc_ns_operations *ns_ops; +72 struct inode

23:27 <Guest97310> vfs_inode;};

23:27 <Guest97310> I need to manually adjust the offset, like this

23:28 <Guest97310> function get_proc_dir_entry:long (file:long) %{ /* pure */ long addr = (long)file_inode((void*)STAP_ARG_file); long pde = kread((unsigned long *)(addr - 72 + 24)); long ops = kread((unsigned long *)(pde + 40)); STAP_RETURN(ops); CATCH_DEREF_FAULT();%}

23:29 <Guest97310> I manually calcuate the offset by using dwarfdump or gdb.

23:29 <fche> eww

23:30 <fche> couldn't make @cast() work ?

23:30 <agentzh> stap can do dwarf magic for you :)

23:31 <Guest97310> OH, I haven't tried cast yet, I thought cast only works for those header file in linux header, let me try it now

23:31 <fche> if the data type can be fished out of the dwarf debuginfo, no header file needed

23:31 <Guest97310> That's great

23:34 <Guest97310> how can I use @cast do some work like C code container_of does?

23:35 <Guest97310> convert a pointer from member to it container, which need subtract some offset from original pointer?

23:35 <fche> we have a macro that computes the equivalent pointer arithmetic

23:35 <fche> called @container_of :-)

23:36 <Guest97310> Thanks!!!