Issue created by Wayne Thornton:
https://gitlab.rtems.org/rtems/rtos/rtems/-/work_items/5659
## Background & Motivation
Executing 64-bit e6500 SMP test suites (such as `smpspinwait01.exe`) inside
QEMU (`-M ppce500 -cpu e6500 -m 1024 -smp 4`) using the default bare-metal
`qoriq_e6500_64` BSP results in a silent console hang. This occurs because
QEMU's generic `ppce500` machine model does not emulate physical NXP T-series
CCSR memory maps or the on-chip NS16550 UART registers at their bare-metal
physical addresses.
To successfully run and capture terminal output in QEMU, the BSP must be built
as a **Hypervisor Guest**. Enabling this mode switches the console driver to
use the ePAPR byte channel (routing output cleanly to QEMU's virtual terminal)
and replaces physical MPIC timer dependencies with standard PowerPC decrementer
clocks.
However, enabling Hypervisor Guest mode strips out physical external interrupt
line definitions from `<bsp/irq.h>`, exposing an architectural bug where legacy
VME bus driver code is being compiled against the 64-bit e6500 target.
---
## Environment & Configuration
* **RTEMS Version:** `7.0.not_released` (master branch)
* **Target Architecture:** `powerpc`
* **Target BSP:** `qoriq_e6500_64`
* **Host OS:** Linux x86_64 (WSL2 / GCC 15 toolchain environment)
* **Build System:** Waf
### Custom `config.ini`
```
[powerpc/qoriq_e6500_64]
RTEMS_SMP = True
BUILD_TESTS = True
RTEMS_POSIX_API = True
WARNING_FLAGS = -Wall -Wextra -Werror -Wno-unused-parameter
-Wno-missing-field-initializers
CC_WARNING_FLAGS = -Wmissing-prototypes -Wimplicit-function-declaration
-Wstrict-prototypes -Wnested-externs -Wold-styl>
QORIQ_IS_HYPERVISOR_GUEST = True
QORIQ_UART_0_ENABLE = False
QORIQ_UART_1_ENABLE = False
```
## Steps to Reproduce
1. Apply the 'config.ini' build options shown above
2. Configure the build tree with the following command:
`./waf configure --rtems-bsps=powerpc/qoriq_e6500_64`
3. Execute the build with './waf'
## Actual Behavior
The build aborts during the compilation of
`bsps/powerpc/shared/vme/vme_universe.c`. Because physical external interrupts
are omitted in hypervisor guest mode, the macro `QORIQ_IRQ_EXT_0` is undeclared:
```
In file included from ../../../bsps/powerpc/shared/vme/vme_universe.c:12:
../../../bsps/powerpc/shared/vme/vme_universe.c: In function
'BSP_VMEIrqMgrInstall':
../../../bsps/powerpc/qoriq/include/bsp/VMEConfig.h:82:7: error:
'QORIQ_IRQ_EXT_0' undeclared (first use in this function)
82 | QORIQ_IRQ_EXT_0, \
| ^~~~~~~~~~~~~~~
../../../bsps/powerpc/shared/vme/vme_universe.c:422:3: note: in expansion of
macro 'BSP_VME_INSTALL_IRQ_MGR'
422 | BSP_VME_INSTALL_IRQ_MGR(err);
| ^~~~~~~~~~~~~~~~~~~~~~~
../../../bsps/powerpc/qoriq/include/bsp/VMEConfig.h:82:7: note: each undeclared
identifier is reported only once for each function it appears in
82 | QORIQ_IRQ_EXT_0, \
| ^~~~~~~~~~~~~~~
../../../bsps/powerpc/shared/vme/vme_universe.c:422:3: note: in expansion of
macro 'BSP_VME_INSTALL_IRQ_MGR'
422 | BSP_VME_INSTALL_IRQ_MGR(err);
| ^~~~~~~~~~~~~~~~~~~~~~~
Waf: Leaving directory `/home/wmthornton/src/rtems/build/powerpc/qoriq_e6500_64'
Build failed
-> task in '/bsps/powerpc/objvme' failed with exit status 1
```
## Root Cause Analysis
Unscoped Waf Dependency: In `spec/build/bsps/powerpc/qoriq/`, the VME bus
driver group (`objvme`) is linked to the broader `qoriq` BSP family rather than
being restricted exclusively to the `mvme2500` BSP variant. As a result, the
build system attempts to compile `vme_universe.c` for 64-bit e6500 targets that
do not possess a VME Universe II or Tsi148 bridge.
Hardcoded Legacy Macro: In `bsps/powerpc/qoriq/include/bsp/VMEConfig.h`, the
interrupt handler is hardcoded to `QORIQ_IRQ_EXT_0`. An existing comment in the
source code explicitly acknowledges this technical debt:
```
grep -n -C 10 "QORIQ_IRQ_EXT_0" bsps/powerpc/qoriq/include/bsp/VMEConfig.h
65- */
66-#define _VME_A32_WIN0_ON_PCI (bsp_vme_pcie_base_address + 0x10000000)
67-#define _VME_A24_ON_PCI (bsp_vme_pcie_base_address + 0x03000000)
68-#define _VME_A16_ON_PCI (bsp_vme_pcie_base_address + 0x02000000)
69-#define _VME_CSR_ON_PCI (bsp_vme_pcie_base_address + 0x01000000)
70-
71-/* FIXME: Make this a BSP config option */
72-#define _VME_A32_WIN0_ON_VME 0x20000000
73-
74-/*
75: * FIXME: The fixed QORIQ_IRQ_EXT_0 is valid for the MVME2500 board. In
theory
76- * there should be some possibility to get that information from the device
tree
77- * or from PCI config space. But I didn't find it anywhere.
78- */
79-#define BSP_VME_INSTALL_IRQ_MGR(err) \
80- do { \
81- err = qoriq_pic_set_sense_and_polarity(\
82: QORIQ_IRQ_EXT_0, \
83- QORIQ_EIRQ_TRIGGER_LEVEL_LOW, \
84- NULL \
85- ); \
86- if (err == 0) { \
87: err = vmeTsi148InstallIrqMgrAlt(0, 0, QORIQ_IRQ_EXT_0, -1); \
88- } \
89- } while (0)
90-
91-/* Add prototypes that are in all VMEConfig.h files */
92-extern int BSP_VMEInit(void);
93-extern int BSP_VMEIrqMgrInstall(void);
94-extern unsigned short (*_BSP_clear_vmebridge_errors)(int);
95-
96-#ifdef __cplusplus
97-}
```
Hypervisor IRQ Strip: When `QORIQ_IS_HYPERVISOR_GUEST = True` is defined, RTEMS
alters the interrupt mappings in `<bsp/irq.h>`. Because virtual guest
environments do not map physical external hardware IRQ pins (`QORIQ_IRQ_EXT_0`
through `QORIQ_IRQ_EXT_11`), the symbol evaluates as undeclared during macro
expansion, breaking the build.
## Suggested Solutions
Option A (Waf Specification Fix): Update the YAML build specifications in
`spec/build/bsps/powerpc/qoriq/` so that the `objvme` group (and its associated
drivers like `vme_universe.c`) is compiled only when building the `mvme2500`
target, removing it entirely from `qoriq_e6500_64` build pipelines.
Option B (Header Guard Fix): If the VME files must remain in the shared family
build path, wrap `bsps/powerpc/qoriq/include/bsp/VMEConfig.h` in a preprocessor
guard that checks for VME hardware capability, or provide a safe fallback for
virtualized guest builds:
```
#ifndef QORIQ_IRQ_EXT_0
#define QORIQ_IRQ_EXT_0 0 /* Fallback for hypervisor guest and non-VME
targets */
#endif
```
--
View it on GitLab: https://gitlab.rtems.org/rtems/rtos/rtems/-/work_items/5659
You're receiving this email because of your account on gitlab.rtems.org.
Unsubscribe from this thread:
https://gitlab.rtems.org/-/sent_notifications/4-cgixhl3tyfjhqxienpzzng2w5-1d/unsubscribe
| Manage all notifications: https://gitlab.rtems.org/-/profile/notifications |
Help: https://gitlab.rtems.org/help
_______________________________________________
bugs mailing list
[email protected]
http://lists.rtems.org/mailman/listinfo/bugs
