Thank you for the ACKs, commited as r4996.
Here's an updated set of patches which also fixes the udelay issues the previous set brought up (see below)
Am 03.01.2010 23:22, schrieb Stefan Reinauer:
20100103-4-tinybootblock-for-intel-cpus Add tinybootblock handling to Intel's CAR code:
- avoid the normal/fallback decision (we don't have __normal_image, and
we'd use CBFS for loading anyway)
- use CONFIG_XIP_ROM_BASE as external symbol, filled in by the linker.
Necessary to match the XIP region with the place where the romstage ends up in (see next patch)
+#if defined(CONFIG_TINY_BOOTBLOCK) && CONFIG_TINY_BOOTBLOCK +#undef CONFIG_XIP_ROM_BASE
movl $CONFIG_XIP_ROM_BASE, %eaxorl $MTRR_TYPE_WRBACK, %eax+#else movl $(CONFIG_XIP_ROM_BASE | MTRR_TYPE_WRBACK), %eax +#endif
This looks very odd... I'd see why you'd do the undef, given the next patch, but why produce the register value at runtime? Since this is not a CONFIG variable in tinybootblock, maybe it would make sense to call it differently instead of undef'ing it? I.e. AUTO_CALCULATED_XIP_ROM_BASE (or something better ;)
AUTO_XIP_ROM_BASE, and "orl $MTRR_TYPE_WRBACK, %eax" in all code paths. For reference (we discussed this off list): orl (or some other hack) is necessary because AUTO_XIP_ROM_BASE (as it is now called) is included by the linker. The assembler has no idea how to cope with "external-reference | constant", as far as I know.
20100103-5-retarget-xip-base-for-romstage
- Add proper CONFIG_XIP_ROM_BASE to location.ld (used when linking the
romstage)
I guess this is needed for the patch above?
Yes, I merged those two patches for clarity (20100104-3):
20100104-1-fix-udelay-on-kconfig We used to use TSC for udelay on all boards. Now there wasn't a default in case the configuration didn't say anything about timers, which failed. The behaviour matches newconfig's now, using UDELAY_IO in case nothing else is selected
20100104-2-use-TSC-on-via-c7 via c7 knows TSC, so use that.
20100104-3-tinybootblock-and-proper-xip-base-on-intel-cpus Tinybootblock support for intel CPUs, and an updated mechanism to set the XIP area to the right location (otherwise the characters on serial can be counted by sight)
20100104-4-kontron-with-tinybootblock-and-options-cleanup Activate tinybootblock for the kontron/986lcd-m board, and clean up the options: MOVNTI is activated per-cpu, HAVE_ACPI_SLIC is not a kconfig variable.
20100104-4 is the kontron stuff from yesterday's patchset, unchanged but merged.
Signed-off-by: Patrick Georgi patrick.georgi@coresystems.de
On 1/4/10 7:06 PM, Patrick Georgi wrote:
Thank you for the ACKs, commited as r4996.
Here's an updated set of patches which also fixes the udelay issues the previous set brought up (see below)
Am 03.01.2010 23:22, schrieb Stefan Reinauer:
20100103-4-tinybootblock-for-intel-cpus Add tinybootblock handling to Intel's CAR code:
- avoid the normal/fallback decision (we don't have __normal_image, and
we'd use CBFS for loading anyway)
- use CONFIG_XIP_ROM_BASE as external symbol, filled in by the linker.
Necessary to match the XIP region with the place where the romstage ends up in (see next patch)
+#if defined(CONFIG_TINY_BOOTBLOCK) && CONFIG_TINY_BOOTBLOCK +#undef CONFIG_XIP_ROM_BASE
movl $CONFIG_XIP_ROM_BASE, %eaxorl $MTRR_TYPE_WRBACK, %eax+#else movl $(CONFIG_XIP_ROM_BASE | MTRR_TYPE_WRBACK), %eax +#endif
This looks very odd... I'd see why you'd do the undef, given the next patch, but why produce the register value at runtime? Since this is not a CONFIG variable in tinybootblock, maybe it would make sense to call it differently instead of undef'ing it? I.e. AUTO_CALCULATED_XIP_ROM_BASE (or something better ;)
AUTO_XIP_ROM_BASE, and "orl $MTRR_TYPE_WRBACK, %eax" in all code paths. For reference (we discussed this off list): orl (or some other hack) is necessary because AUTO_XIP_ROM_BASE (as it is now called) is included by the linker. The assembler has no idea how to cope with "external-reference | constant", as far as I know.
20100103-5-retarget-xip-base-for-romstage
- Add proper CONFIG_XIP_ROM_BASE to location.ld (used when linking the
romstage)
I guess this is needed for the patch above?
Yes, I merged those two patches for clarity (20100104-3):
20100104-1-fix-udelay-on-kconfig We used to use TSC for udelay on all boards. Now there wasn't a default in case the configuration didn't say anything about timers, which failed. The behaviour matches newconfig's now, using UDELAY_IO in case nothing else is selected
Acked-by: Stefan Reinauer stepan@coresystems.de
Should we drop HAVE_INIT_TIMER and have an empty function instead of another config variable?
20100104-2-use-TSC-on-via-c7 via c7 knows TSC, so use that.
Acked-by: Stefan Reinauer stepan@coresystems.de
20100104-3-tinybootblock-and-proper-xip-base-on-intel-cpus Tinybootblock support for intel CPUs, and an updated mechanism to set the XIP area to the right location (otherwise the characters on serial can be counted by sight)
Acked-by: Stefan Reinauer stepan@coresystems.de
20100104-4-kontron-with-tinybootblock-and-options-cleanup Activate tinybootblock for the kontron/986lcd-m board, and clean up the options: MOVNTI is activated per-cpu, HAVE_ACPI_SLIC is not a kconfig variable.
Acked-by: Stefan Reinauer stepan@coresystems.de
Stefan
Signed-off-by: Patrick Georgi patrick.georgi@coresystems.de
-
Patrick Georgi -
Stefan Reinauer