RE: vrp bitfield miscompilation (WAS: [Regression] snd_emu10k1 doesn't work after GCC 4.2 upgrade)

From: Yuriy Tsibizov <Yuriy.Tsibizov_at_gfk.ru>
Date: Mon, 28 May 2007 11:37:20 +0400
> >>> See http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32100
> >> Use of following variables _can_ be affected by above 
> miscompilation
> >> (these are signed single-bit variables):
> >>
> >> amd64/amd64/mp_machdep.c:	int	cpu_present:1;
> >> amd64/amd64/mp_machdep.c:	int	cpu_bsp:1;
> >> amd64/amd64/mp_machdep.c:	int	cpu_disabled:1;
> >> dev/acpica/acpi_pci_link.c:	int	l_routed:1;
> >> dev/acpica/acpi_pci_link.c:	int	l_isa_irq:1;
> >> dev/sound/isa/ess.c:    	int type, duplex:1, newspeed:1;
> >> dev/sound/pci/emu10k1.c:	int b16:1, stereo:1, busy:1, running:1,
> >> ismaster:1;
> >> dev/sound/pci/solo.c:    	int simplex_dir, type, duplex:1,
> >> newspeed:1, dmasz[2];
> >> dev/puc/puc.c:	int		p_hasintr:1;
> >> dev/puc/puc_bfe.h:	int		sc_fastintr:1;
> >> dev/puc/puc_bfe.h:	int		sc_leaving:1;
> >> dev/puc/puc_bfe.h:	int		sc_polled:1;
> >> dev/rp/rpvar.h:	int			rp_rts_iflow:1;
> >> dev/rp/rpvar.h:	int			rp_disable_writes:1;
> >> dev/rp/rpvar.h:	int			rp_cts:1;
> >> dev/rp/rpvar.h:	int			rp_waiting:1;
> >> dev/rp/rpvar.h:	int			rp_xmit_stopped:1;
> >> dev/scc/scc_bfe.h:	int		m_attached:1;
> >> dev/scc/scc_bfe.h:	int		m_fastintr:1;
> >> dev/scc/scc_bfe.h:	int		m_hasintr:1;
> >> dev/scc/scc_bfe.h:	int		m_probed:1;
> >> dev/scc/scc_bfe.h:	int		m_sysdev:1;
> >> dev/scc/scc_bfe.h:	int		ch_enabled:1;
> >> dev/scc/scc_bfe.h:	int		ch_sysdev:1;
> >> dev/scc/scc_bfe.h:	int		sc_fastintr:1;
> >> dev/scc/scc_bfe.h:	int		sc_leaving:1;
> >> dev/scc/scc_bfe.h:	int		sc_polled:1;
> >> dev/uart/uart_bus.h:	int		sc_callout:1;	
> /* This UART is
> >> opened for callout. */
> >> dev/uart/uart_bus.h:	int		sc_fastintr:1;	
> /* This UART
> >> uses fast interrupts. */
> >> dev/uart/uart_bus.h:	int		sc_hwiflow:1;	
> /* This UART has
> >> HW input flow ctl. */
> >> dev/uart/uart_bus.h:	int		sc_hwoflow:1;	
> /* This UART has
> >> HW output flow ctl. */
> >> dev/uart/uart_bus.h:	int		sc_leaving:1;	
> /* This UART is
> >> going away. */
> >> dev/uart/uart_bus.h:	int		sc_opened:1;	
> /* This UART is
> >> open for business. */
> >> dev/uart/uart_bus.h:	int		sc_polled:1;	
> /* This UART has
> >> no interrupts. */
> >> dev/uart/uart_bus.h:	int		sc_txbusy:1;	
> /* This UART is
> >> transmitting. */
> >> fs/pseudofs/pseudofs_internal.h:	int		 pvd_dead:1;
> >> geom/part/g_part.h:	int		gpe_created:1;	
> /* Entry is
> >> newly created. */
> >> geom/part/g_part.h:	int		gpe_deleted:1;	
> /* Entry has
> >> been deleted. */
> >> geom/part/g_part.h:	int		gpe_modified:1;	
> /* Entry has
> >> been modified. */
> >> geom/part/g_part.h:	int		gpt_isleaf:1;	
> /* Cannot be
> >> sub-partitioned. */
> >> geom/part/g_part.h:	int		gpt_created:1;	/* Newly
> >> created. */
> >> geom/part/g_part.h:	int		gpt_modified:1;	
> /* Table changes
> >> have been made. */
> >> geom/part/g_part.h:	int		gpt_opened:1;	
> /* Permissions
> >> obtained. */
> >> i386/i386/mp_machdep.c:	int	cpu_present:1;
> >> i386/i386/mp_machdep.c:	int	cpu_bsp:1;
> >> i386/i386/mp_machdep.c:	int	cpu_disabled:1;
> >> i386/include/npx.h:	int	fp_sgn:1;	/* 
> mantissa sign */
> >> sparc64/pci/psycho.c:		int apb:1;
> >> sparc64/pci/psycho.c:		int ppb:1;
> >>
> >> I can't say that there _is_ a miscompilation related to 
> all variables in
> >> this list.
> >>
> >> Should most of them (I'm not shure for fp_sign) be changed 
> to unsigned
> >> int? With (signed) int these flags have values of (-1;0), 
> with unsigned
> >> int they will use more obvious (0;1) values.
> >>
> >>
> >> Yuriy.
> > 
> > Interesting find. Another issue:
> > 
> > I've seen that the compile will round the memory size of a 
> bit-fields down to 
> > the nearest byte, if you turn on optimization. I think you should 
> > use "uint8_t".
> 
> My memory is that bitfields are only defined as parts of an 'int'
> in the standard and that using anything else is a gcc 'feature'.
> 
> This is a 14 year old memory (from my SCSI days) so it may not be true
> now.

>From ISO/IEC 9899 (C99): "A bit-field shall have a type that is a
qualified or unqualified version of _Bool, signed
int, unsigned int, or some other implementation-defined type."

Yuriy.
Received on Mon May 28 2007 - 05:34:34 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:11 UTC