potential AArch64 ABI bug wrt handling of 128-bit bit-fields

[Philippe Mathieu-Daudé]

After upgrading to Ubuntu 20.04 LTS, GCC 9.3 displays a lot of notes:

hw/block/pflash_cfi01.c: In function ‘pflash_mem_read_with_attrs’:
hw/block/pflash_cfi01.c:663:20: note: parameter passing for argument of type ‘MemTxAttrs’ {aka ‘struct MemTxAttrs’} changed in GCC 9.1
  663 | static MemTxResult pflash_mem_read_with_attrs(void *opaque, hwaddr addr, uint64_t *value,
      | ^~~~~~~~~~~~~~~~~~~~~~~~~~
hw/block/pflash_cfi01.c: In function ‘pflash_mem_write_with_attrs’:
hw/block/pflash_cfi01.c:677:20: note: parameter passing for argument of type ‘MemTxAttrs’ {aka ‘struct MemTxAttrs’} changed in GCC 9.1
  677 | static MemTxResult pflash_mem_write_with_attrs(void *opaque, hwaddr addr, uint64_t value,
      | ^~~~~~~~~~~~~~~~~~~~~~~~~~~
hw/nvram/fw_cfg.c: In function ‘fw_cfg_dma_mem_valid’:
hw/nvram/fw_cfg.c:475:13: note: parameter passing for argument of type ‘MemTxAttrs’ {aka ‘struct MemTxAttrs’} changed in GCC 9.1
  475 | static bool fw_cfg_dma_mem_valid(void *opaque, hwaddr addr,
      | ^~~~~~~~~~~~~~~~~~~~
hw/nvram/fw_cfg.c: In function ‘fw_cfg_data_mem_valid’:
hw/nvram/fw_cfg.c:483:13: note: parameter passing for argument of type ‘MemTxAttrs’ {aka ‘struct MemTxAttrs’} changed in GCC 9.1
  483 | static bool fw_cfg_data_mem_valid(void *opaque, hwaddr addr,
      | ^~~~~~~~~~~~~~~~~~~~~
hw/nvram/fw_cfg.c: In function ‘fw_cfg_ctl_mem_valid’:
hw/nvram/fw_cfg.c:501:13: note: parameter passing for argument of type ‘MemTxAttrs’ {aka ‘struct MemTxAttrs’} changed in GCC 9.1
  501 | static bool fw_cfg_ctl_mem_valid(void *opaque, hwaddr addr,
      | ^~~~~~~~~~~~~~~~~~~~
hw/nvram/fw_cfg.c: In function ‘fw_cfg_comb_valid’:
hw/nvram/fw_cfg.c:521:13: note: parameter passing for argument of type ‘MemTxAttrs’ {aka ‘struct MemTxAttrs’} changed in GCC 9.1
  521 | static bool fw_cfg_comb_valid(void *opaque, hwaddr addr,
      | ^~~~~~~~~~~~~~~~~
hw/intc/arm_gic.c: In function ‘gic_do_hyp_read’:
hw/intc/arm_gic.c:1996:20: note: parameter passing for argument of type ‘MemTxAttrs’ {aka ‘struct MemTxAttrs’} changed in GCC 9.1
 1996 | static MemTxResult gic_do_hyp_read(void *opaque, hwaddr addr, uint64_t *data,
      | ^~~~~~~~~~~~~~~
hw/intc/arm_gic.c: In function ‘gic_thiscpu_hyp_read’:
hw/intc/arm_gic.c:1979:20: note: parameter passing for argument of type ‘MemTxAttrs’ {aka ‘struct MemTxAttrs’} changed in GCC 9.1
 1979 | static MemTxResult gic_thiscpu_hyp_read(void *opaque, hwaddr addr, uint64_t *data,
      | ^~~~~~~~~~~~~~~~~~~~
hw/intc/arm_gic.c: In function ‘gic_get_current_pending_irq’:
hw/intc/arm_gic.c:419:17: note: parameter passing for argument of type ‘MemTxAttrs’ {aka ‘struct MemTxAttrs’} changed in GCC 9.1
  419 | static uint16_t gic_get_current_pending_irq(GICState *s, int cpu,
      | ^~~~~~~~~~~~~~~~~~~~~~~~~~~

This seems related to:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88469
https://gcc.gnu.org/git/?p=gcc.git&a=commit;h=c590597c45

  This is pretty unlikely in real code, but similar to Arm, the AArch64
  ABI has a bug with the handling of 128-bit bit-fields, where if the
  bit-field dominates the overall alignment the back-end code may end up
  passing the argument correctly. This is a regression that started in
  gcc-6 when the ABI support code was updated to support overaligned
  types. The fix is very similar in concept to the Arm fix. 128-bit
  bit-fields are fortunately extremely rare, so I'd be very surprised if
  anyone has been bitten by this.

[Peter Maydell (pmaydell)]

The warnings aren't a problem for QEMU because we don't expose these functions as public ABI, so the whole compile will be consistently built with the same compiler version. So we added -Wno-psabi in commit bac8d222a19f4a30d to silence the compiler here.