Bug ID 47743
Summary LLDB displays wrong values for packed bitfields
Product lldb
Version unspecified
Hardware All
OS All
Status NEW
Severity normal
Priority P
Component All Bugs
Assignee lldb-dev@lists.llvm.org
Reporter cameron@moodycamel.com
CC jdevlieghere@apple.com, llvm-bugs@lists.llvm.org
When an 'unsigned' bitfield is in a packed structure, its bits may straddle two
aligned dwords in memory.
LLDB calculates the value by reading a single dword from the byte offset of the
first unsigned, then shifting and masking using the bitfield offset/size.
However, when the value straddles two unsigned dwords, the byte offset + bit
offset + size extends past the end of the first dword, and the displayed value
is cut off (missing trailing bits).
To reproduce on a little-endian machine:
struct __attribute__((packed)) foo
{
unsigned : 31;
unsigned u11Sample : 11;
unsigned : 22;
};
__attribute__((noinline))
int quux(struct foo* f)
{
// break here and inspect f->u11Sample
// the value should be 0x50, but it's not displayed as such
return f->u11Sample;
}
int main()
{
struct foo f = { 0 };
f.u11Sample = 0x50;
return quux(&f);
}
To reproduce on a big-endian machine use this struct definition instead:
struct __attribute__((packed)) foo
{
unsigned : 22;
unsigned u11Sample : 11;
unsigned : 31;
};
As a starting point for whoever wants to investigate this, put a breakpoint on
'valobj->GetData(data, error);' (near line 91) in TypeFormat.cpp. Observe the
data read, as well as the
m_byte_offset/m_bitfield_bit_size/m_bitfield_bit_offset in valobj (it's of type
ValueObjectChild).