On 08/29/2016 08:10 PM, Benoît Canet wrote:
On Mon, Aug 29, 2016 at 7:08 PM, Avi Kivity <a...@scylladb.com <mailto:a...@scylladb.com>> wrote:On 08/29/2016 07:49 PM, Nadav Har'El wrote: GDB uses DWARF CFI meta-instructions to know how to backtrace through function call frames, and which registers get saved where. This patch fixes the CFI instructions in the syscall_entry() assembly code. After this patch, if we are in system call code we can backtrack through the syscall_entry() - for example: 5 0x000000000059c10b in syscall_wrapper (number=186) at linux.cc:332 6 0x0000000000480245 in syscall_entry () at arch/x64/entry.S:238 7 0x0000100000c00bce in main (argc=1, argv=<optimized out>) at /home/nyh/osv/tests/tst-syscall.cc:49 8 0x000000000063526e in osv::application::run_main (this=0xffffa00003079910, path="tests/tst-syscall.so", argc=1, argv=0xffffa00001f88170) at core/app.cc:338 Note how syscall_entry() looks like a normal function - it no longer pretends to be a "signal frame". In particular, we no longer need to obey a specific layout of the registers saved on the stack, so I took this opportunity to remove some of the useless things we saved or saved multiple times. Signed-off-by: Nadav Har'El <n...@scylladb.com <mailto:n...@scylladb.com>> --- arch/x64/entry.S | 104 +++++++++++++++++++++---------------------------------- 1 file changed, 40 insertions(+), 64 deletions(-) diff --git a/arch/x64/entry.S b/arch/x64/entry.S index e265a30..e3be0bc 100644 --- a/arch/x64/entry.S +++ b/arch/x64/entry.S @@ -164,66 +164,43 @@ call_signal_handler_thunk: syscall_entry: .type syscall_entry, @function .cfi_startproc simple Probably, we need to remove simple here, likely it sets up incorrect references. + .cfi_undefined rcx # was overwritten with rip by the syscall instruction Not .cfi_register1 %rip, %rcx? So if we trap here, we know who the caller is. From what I understand Nadav does it by hand later when pushing.
He does, but when you debug between this point and the push, you will have a broken stack trace.
+ .cfi_undefined r11 # was overwritten with rflags by the syscall instruction # There is no ring transition and rflags are left unchanged. # Skip the "red zone" allowed by the AMD64 ABI (the caller used a # SYSCALL instruction and doesn't know he called a function): subq $128, %rsp + .cfi_def_cfa %rsp, 0 # We need to save and restore the caller's %rbp anyway, so let's also # set it up properly for old-style frame-pointer backtracing to work # (e.g., backtrace_safe()). Also need to push the return address before # the rbp to get a normal frame. Our return address is in rcx. - pushq %rcx - pushq %rbp + pushq_cfi %rcx + .cfi_rel_offset %rip, 0 + pushq_cfi %rbp movq %rsp, %rbp + .cfi_rel_offset %rsp, 0 # # From http://stackoverflow.com/questions/2535989/what-are-the-calling-conventions-for-unix-linux-system-calls-on-x86-64 <http://stackoverflow.com/questions/2535989/what-are-the-calling-conventions-for-unix-linux-system-calls-on-x86-64>: # "User-level applications use as integer registers for passing the sequence %rdi, %rsi, %rdx, %rcx, %r8 and %r9. The kernel interface uses %rdi, %rsi, %rdx, %r10, %r8 and %r9" # FIXME: fpu - # build the stack frame by hand - pushq %rsp - subq $8, %rsp # rip was saved in rcx by the syscall instruction - pushq %rax - pushq %rbx - pushq %rcx # contains rip before syscall instruction - pushq %rdx - pushq %rsi - pushq %rdi - pushq %r8 - pushq %r9 - pushq %r10 - pushq %r11 # contains rflags before syscall instruction - pushq %r12 - pushq %r13 - pushq %r14 - pushq %r15 - - # stack contains a signal_frame - .cfi_signal_frame - .cfi_def_cfa %rsp, 0 - .cfi_register rip,rcx # rcx took previous rip value - .cfi_register rflags,r11 # r11 took previous rflags value - .cfi_undefined rcx # was overwritten with rip by the syscall instruction - .cfi_undefined r11 # was overwritten with rflags by the syscall instruction - .cfi_offset %r15, 0x00 - .cfi_offset %r14, 0x08 - .cfi_offset %r13, 0x10 - .cfi_offset %r12, 0x18 - .cfi_offset %r11, 0x20 - .cfi_offset %r10, 0x28 - .cfi_offset %r9, 0x30 - .cfi_offset %r8, 0x38 - .cfi_offset %rbp, 0x40 - .cfi_offset %rdi, 0x48 - .cfi_offset %rsi, 0x50 - .cfi_offset %rdx, 0x58 - .cfi_offset %rcx, 0x60 - .cfi_offset %rbx, 0x68 - .cfi_offset %rax, 0x70 - .cfi_offset %rip, 0x80 - .cfi_offset %rsp, 0x98 + pushq_cfi %rax + pushq_cfi %rbx + pushq_cfi %rdx + pushq_cfi %rsi + pushq_cfi %rdi + pushq_cfi %r8 + pushq_cfi %r9 + pushq_cfi %r10 + pushq_cfi %r11 # contains rflags before syscall instruction + .cfi_rel_offset %rflags, 0 + pushq_cfi %r12 + pushq_cfi %r13 + pushq_cfi %r14 + pushq_cfi %r15 # The kernel interface use r10 as fourth argument while the user interface use rcx # so overwrite rcx with r10 @@ -231,7 +208,7 @@ syscall_entry: # prepare function call parameter: r9 is on the stack since it's the seventh param # because we shift existing params by one to make room for syscall number - pushq %r9 + pushq_cfi %r9 movq %r8, %r9 movq %rcx, %r8 movq %rdx, %rcx @@ -254,34 +231,33 @@ syscall_entry: # restore it from 8(%rsp). pushq %rsp pushq (%rsp) + .cfi_adjust_cfa_offset 16 andq $-0x10, %rsp + .cfi_rel_offset %rsp, 8 callq syscall_wrapper movq 8(%rsp), %rsp + .cfi_adjust_cfa_offset -16 - popq %r9 + popq_cfi %r9 # in Linux user and kernel return value are in rax so we have nothing to do for return values - popq %r15 - popq %r14 - popq %r13 - popq %r12 - popq %r11 - popq %r10 - popq %r9 - popq %r8 - popq %rdi - popq %rsi - popq %rdx - popq %rcx - popq %rbx - addq $8, %rsp # skip rax emplacement (return value is in rax) - addq $8, %rsp # rip emplacement (rip cannot be popped) - popq %rsp + popq_cfi %r15 + popq_cfi %r14 + popq_cfi %r13 + popq_cfi %r12 + popq_cfi %r11 + popq_cfi %r10 + popq_cfi %r9 + popq_cfi %r8 + popq_cfi %rdi + popq_cfi %rsi + popq_cfi %rdx + popq_cfi %rbx - popq %rbp - popq %rcx + popq_cfi %rbp + popq_cfi %rcx addq $128, %rsp # undo red-zone skip-- You received this message because you are subscribed to the GoogleGroups "OSv Development" group. To unsubscribe from this group and stop receiving emails from it, send an email to osv-dev+unsubscr...@googlegroups.com <mailto:osv-dev%2bunsubscr...@googlegroups.com>. For more options, visit https://groups.google.com/d/optout <https://groups.google.com/d/optout>.
-- You received this message because you are subscribed to the Google Groups "OSv Development" group. To unsubscribe from this group and stop receiving emails from it, send an email to osv-dev+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
