Comments
Patch
Improving BKDG implementation of P-states,
CPU and northbridge frequency and voltage
handling for Fam 10 in SVI mode.
Documentation only.
Adding a checklist to tell which function takes
care of which requirement of BKDG .
Signed-off-by: Xavi Drudis Ferran <xdrudis@tinet.cat>
@@ -16,8 +16,93 @@
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*/
+/*
+ * This file initializes the CPU cores for voltage and frequency settings
+ * in the different power states.
+ */
+/*
+checklist (functions are in this file if no source file named)
+Fam10 Bios and Kernel Development Guide #31116, rev 3.48, April 22, 2010
+
+2.4.2.6 Requirements for p-states
+
+1.- F3x[84:80] According to table 100 : prep_fid_change
+
+2.- COF/VID :
+ 2.4.2.9.1 Steps 1,3-6 and warning for 2,7 if they apply
+ fixPsNbVidBeforeWR(...)
+ 2.4.2.9.1 Step 8 enable_fid_change
+ We do this for all nodes, I don't understand BKDG 100% on
+ whether this is or isn't meant by "on the local
+ processor". Must be OK.
+ 2.4.2.9.1 Steps 9-10 (repeat 1-7 and reset) romstage.c/init_cpus ?
+ 2.4.2.9.1 Steps 11-12 init_fidvid_stage2
+ 2.4.2.9.2 DualPlane PVI : Not supported, don't know how to detect,
+ needs specific circuitry.
+
+3.- 2.4.2.7 dualPlaneOnly(dev)
+
+4.- 2.4.2.8 applyBoostFIDOffset(dev)
+
+5.- enableNbPState1(dev)
+
+6.- 2.4.1.7
+ a) UpdateSinglePlaneNbVid()
+ b) setVSRamp(), called from prep_fid_change
+ c) prep_fid_change
+ d) improperly, for lack of voltage regulator details?,
+ F3xA0[PsiVidEn] in defaults.h
+ F3xA0[PsiVid] in init_cpus.c AMD_SetupPSIVID_d (before prep_fid_change)
+
+7.- TODO (Core Performance Boost is only available in revision E cpus, and we
+ don't seem to support those yet, at least they don't have any
+ constant in amddefs.h )
+
+8.- FIXME ? Transition to min Pstate according to 2.4.2.15.3 is required
+ by 2.4.2.6 after warm reset. But 2.4.2.15 states that it is not required
+ if the warm reset is issued by coreboot to update NbFid. So it is required
+ or not ? How can I tell who issued warm reset ?
+ Coreboot transitions to P0 instead, which is not recommended, and does
+ not follow 2.4.2.15.2 to do so.
+
+9.- TODO Requires information on current delivery capability
+ (depends on mainboard and maybe power supply ?). One might use a config
+ option with the maximum number of Ampers that the board can deliver to CPU.
+
+10.- [Multiprocessor] TODO 2.4.2.12
+ [Uniprocessor] FIXME ? We call setPStateMaxVal() in init_fidvid_stage2,
+ but not sure this is what is meant by "Determine the valid set of
+ P-states based on enabled P-states indicated
+ in MSRC001_00[68:64][PstateEn]" in 2.4.2.6-10
+
+11.- finalPstateChange() from init_fidvid_Stage2 (BKDG says just "may", anyway)
+
+12.- generate ACPI for p-states. FIXME
+ Needs more assesment. There's some kind of fixed support that
+ does not seem to depend on CPU revision or actual MSRC001_00[68:64]
+ as BKDG apparently requires.
+ http://www.coreboot.org/ACPI#CPU_Power_Management
+ At least for Tilapia board:
+ src/mainboard/<vendor>/<model>/acpi_tables.c write_acpi_tables(...) calls
+ acpi_add_ssdt_pstates(...)
+ in /src/northbridge/amd/amdfam10/amdfam10_acpi.c
+ which apparently copies them from static info in
+ src/mainboard/<vendor>/<model>/acpi/cpstate.asl
+
+"must also be completed"
+
+a.- PllLockTime set in ruleset in defaults.h
+ BKDG says set it "If MSRC001_00[68:64][CpuFid] is different between
+ any two enabled P-states", but since it does not say "only if"
+ I guess it is safe to do it always.
+
+b.- prep_fid_change(...)
+
+ */
+
#if CONFIG_SET_FIDVID
+
#include <northbridge/amd/amdht/AsPsDefs.h>
static inline void print_debug_fv(const char *str, u32 val)