All,
After looking at the problems this "simple" commit caused, I've rolled it back to the revision before my commit (2005->2003). I'll go over these changes with Tom and Eric and try again later. My apologies for any broken builds.
--jason--
My idea about your changes 1. memhole move from raminit.c to northbridge.c, That is goog feature, but need to keep my code in init_ecc, to skip the mem hole. 2. dual core: create entries in cpu_bus_scan or in amd_sibling_init, later one may cause racing, if we disable CPU_INIT_SERILIZE on four way dual core and eight dual core. You know we really need to disable CPU_INIT_SERIALIZE otherwise init_ecc stage will take too long. 8x--->2.1x 3. you need to keep incoherent_ht.c ( K8_ALLOCATE_IO), I'm using that for multiple CK804 early access... 4. ext apic id mode only need for 8 way dual core. 5. dual core, if it is disable the apic id of CPU should be the same as pre-e0, instead of 0, 2, 4.... 6. for CAR purpose, I can not k8_init_and_stop_secondaries, because I need to go to __reset intead of __jmp to it. (somthing related to stack switch). 7. We need to use write_irq_tables intead of copy_irq_tables, I have append write_irq_tables in every MB irq_tables.c to call copy_irq_table. 8. You need to keep CK804_DEVN_BASE=0 in some files, according to NV, for ck804 if you change unit id, you may lose bus master support.
YH
On 8/10/05, Jason Schildt jschildt@lnxi.com wrote:
All,
After looking at the problems this "simple" commit caused, I've rolled it back to the revision before my commit (2005->2003). I'll go over these changes with Tom and Eric and try again later. My apologies for any broken builds.
--jason--
--
Jason W. Schildt LinuxBIOS Software Engineer Linux Networx
LinuxBIOS mailing list LinuxBIOS@openbios.org http://www.openbios.org/mailman/listinfo/linuxbios
Thanks, YH. I'll pass these by Tom and make the necessary changes so as not to break everything. I'll also submit a diff to the mailing list before committing.
--jason--
On Wed, Aug 10, 2005 at 09:17:25AM -0700, yhlu wrote:
My idea about your changes
- memhole move from raminit.c to northbridge.c, That is goog feature,
but need to keep my code in init_ecc, to skip the mem hole. 2. dual core: create entries in cpu_bus_scan or in amd_sibling_init, later one may cause racing, if we disable CPU_INIT_SERILIZE on four way dual core and eight dual core. You know we really need to disable CPU_INIT_SERIALIZE otherwise init_ecc stage will take too long. 8x--->2.1x 3. you need to keep incoherent_ht.c ( K8_ALLOCATE_IO), I'm using that for multiple CK804 early access... 4. ext apic id mode only need for 8 way dual core. 5. dual core, if it is disable the apic id of CPU should be the same as pre-e0, instead of 0, 2, 4.... 6. for CAR purpose, I can not k8_init_and_stop_secondaries, because I need to go to __reset intead of __jmp to it. (somthing related to stack switch). 7. We need to use write_irq_tables intead of copy_irq_tables, I have append write_irq_tables in every MB irq_tables.c to call copy_irq_table. 8. You need to keep CK804_DEVN_BASE=0 in some files, according to NV, for ck804 if you change unit id, you may lose bus master support.
YH
On 8/10/05, Jason Schildt jschildt@lnxi.com wrote:
All,
After looking at the problems this "simple" commit caused, I've rolled it back to the revision before my commit (2005->2003). I'll go over these changes with Tom and Eric and try again later. My apologies for any broken builds.
--jason--
--
Jason W. Schildt LinuxBIOS Software Engineer Linux Networx
LinuxBIOS mailing list LinuxBIOS@openbios.org http://www.openbios.org/mailman/listinfo/linuxbios
Yes, that would be easier if discuss one by one with diff.
YH
On 8/10/05, Jason Schildt jschildt@lnxi.com wrote:
Thanks, YH. I'll pass these by Tom and make the necessary changes so as not to break everything. I'll also submit a diff to the mailing list before committing.
--jason--
On Wed, Aug 10, 2005 at 09:17:25AM -0700, yhlu wrote:
My idea about your changes
- memhole move from raminit.c to northbridge.c, That is goog feature,
but need to keep my code in init_ecc, to skip the mem hole. 2. dual core: create entries in cpu_bus_scan or in amd_sibling_init, later one may cause racing, if we disable CPU_INIT_SERILIZE on four way dual core and eight dual core. You know we really need to disable CPU_INIT_SERIALIZE otherwise init_ecc stage will take too long. 8x--->2.1x 3. you need to keep incoherent_ht.c ( K8_ALLOCATE_IO), I'm using that for multiple CK804 early access... 4. ext apic id mode only need for 8 way dual core. 5. dual core, if it is disable the apic id of CPU should be the same as pre-e0, instead of 0, 2, 4.... 6. for CAR purpose, I can not k8_init_and_stop_secondaries, because I need to go to __reset intead of __jmp to it. (somthing related to stack switch). 7. We need to use write_irq_tables intead of copy_irq_tables, I have append write_irq_tables in every MB irq_tables.c to call copy_irq_table. 8. You need to keep CK804_DEVN_BASE=0 in some files, according to NV, for ck804 if you change unit id, you may lose bus master support.
YH
On 8/10/05, Jason Schildt jschildt@lnxi.com wrote:
All,
After looking at the problems this "simple" commit caused, I've rolled it back to the revision before my commit (2005->2003). I'll go over these changes with Tom and Eric and try again later. My apologies for any broken builds.
--jason--
--
Jason W. Schildt LinuxBIOS Software Engineer Linux Networx
LinuxBIOS mailing list LinuxBIOS@openbios.org http://www.openbios.org/mailman/listinfo/linuxbios
--
Jason W. Schildt LinuxBIOS Software Engineer Linux Networx
-
Jason Schildt -
yhlu