* [Patch] small cleanup for powernow-k7.c
@ 2005-01-20 20:38 Hendrik Muhs
2005-01-21 1:57 ` Dave Jones
0 siblings, 1 reply; 4+ messages in thread
From: Hendrik Muhs @ 2005-01-20 20:38 UTC (permalink / raw)
To: davej; +Cc: cpufreq
[-- Attachment #1: Type: text/plain, Size: 998 bytes --]
Hi,
attached is a patch which fixes a wrong entry in the vid table and some
comments.
It is broken down from my powernow-k7 patch to manual set multiplier/voltage
(for Desktop mainboards which does not have PST table in their BIOS
("Another powernow-k7 patch for Desktop XP-M usage")).
A final question:
Is something like this absolutly out of question for inclusion in mainline?
(yes I've read the archives and I have not found a final answer on this)
I see that this feature is not official supported by vendors and although I
have implemented some mechanisms which should prevent serious CPU damage it
can not (by design) completly safe (system may freeze on some setups).
On the other side their is demand for this feature (not only for me). I do not
know what to do with it in the future. If the door for inclusion in mainline
is closed I would consider make some kind of source package and
forking/renaming the driver so that it can be build independent from kernel.
Hendrik
[-- Attachment #2: powernow-k7_smallcleanup.patch --]
[-- Type: text/x-diff, Size: 967 bytes --]
--- linux/arch/i386/kernel/cpu/cpufreq/powernow-k7.c.orig 2005-01-20 19:45:56.000000000 +0100
+++ linux/arch/i386/kernel/cpu/cpufreq/powernow-k7.c 2005-01-20 20:11:04.000000000 +0100
@@ -67,16 +67,20 @@
#endif
#ifdef CONFIG_CPU_FREQ_DEBUG
-/* divide by 1000 to get VID. */
+/* divide by 1000 to get VCore voltage in V. */
static int mobile_vid_table[32] = {
2000, 1950, 1900, 1850, 1800, 1750, 1700, 1650,
1600, 1550, 1500, 1450, 1400, 1350, 1300, 0,
1275, 1250, 1225, 1200, 1175, 1150, 1125, 1100,
- 1075, 1050, 1024, 1000, 975, 950, 925, 0,
+ 1075, 1050, 1025, 1000, 975, 950, 925, 0,
};
#endif
-/* divide by 10 to get FID. */
+/* divide by 10 to get multiplier.
+ * 0x19 should maybe 220 not 225 (see CrystalCPUID), but as their are no CPU's
+ * with this multiplier (and possibly never will be) it does not matter
+ */
+
static int fid_codes[32] = {
110, 115, 120, 125, 50, 55, 60, 65,
70, 75, 80, 85, 90, 95, 100, 105,
[-- Attachment #3: Type: text/plain, Size: 147 bytes --]
_______________________________________________
Cpufreq mailing list
Cpufreq@lists.linux.org.uk
http://lists.linux.org.uk/mailman/listinfo/cpufreq
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [Patch] small cleanup for powernow-k7.c
2005-01-20 20:38 [Patch] small cleanup for powernow-k7.c Hendrik Muhs
@ 2005-01-21 1:57 ` Dave Jones
2005-01-25 20:47 ` [Patch] powernow-k7-manual was " Hendrik Muhs
0 siblings, 1 reply; 4+ messages in thread
From: Dave Jones @ 2005-01-21 1:57 UTC (permalink / raw)
To: Hendrik Muhs; +Cc: cpufreq
On Thu, Jan 20, 2005 at 09:38:35PM +0100, Hendrik Muhs wrote:
> attached is a patch which fixes a wrong entry in the vid table and some
> comments.
Thanks, I've merged this, and will send Linus a request to pull again
soon so this should get into the next -rc release for 2.6.11.
I didn't merge the bit about CrystalCPUID because I'm not convinced
its true at all. I just double-checked the document I got from AMD
which clearly states it's "Reserved for 22.5x"
I don't know where the crystalcpuid folks got their numbers from, but
I'm sceptical. I'll check for an updated version of the powernow documentation,
but afaik, this is the last revision of this doc that was published.
> A final question:
>
> Is something like this absolutly out of question for inclusion in mainline?
> (yes I've read the archives and I have not found a final answer on this)
>
> I see that this feature is not official supported by vendors and although I
> have implemented some mechanisms which should prevent serious CPU damage it
> can not (by design) completly safe (system may freeze on some setups).
I've slowly warmed to the *idea* of having some means for supporting
mobile chips in desktops, as it does seem to be happening more and more
for some reason. I didn't look at your last patch to be honest, so I can't
comment on whether its fit for merging before I take a look.
> On the other side their is demand for this feature (not only for me). I do not
> know what to do with it in the future. If the door for inclusion in mainline
> is closed I would consider make some kind of source package and
> forking/renaming the driver so that it can be build independent from kernel.
Lets see what we can do to get it included. Maintaining out of tree modules
is a real pain in the ass for all concerned, including end-users.
Rather than see endless support requests for 'why wont it build' on cpufreq-list
I'd rather we did what we could to get things like this included.
I'm holding off on 'feature' type things for 2.6.11 from now on, the next
round of cpufreq bits going to Linus should be only bug fixes, so hopefully
by 2.6.12 time, we can close the book on this one. This gives us enough
time to give it some exposure in Andrew's -mm tree too.
Dave
^ permalink raw reply [flat|nested] 4+ messages in thread
* [Patch] powernow-k7-manual was Re: [Patch] small cleanup for powernow-k7.c
2005-01-21 1:57 ` Dave Jones
@ 2005-01-25 20:47 ` Hendrik Muhs
2005-01-26 8:03 ` Jarkko Lavinen
0 siblings, 1 reply; 4+ messages in thread
From: Hendrik Muhs @ 2005-01-25 20:47 UTC (permalink / raw)
To: Dave Jones; +Cc: Adam Jones, cpufreq, Jarkko Lavinen
[-- Attachment #1: Type: text/plain, Size: 2555 bytes --]
Hi,
[...]
> > I see that this feature is not official supported by vendors and
> > although I have implemented some mechanisms which should prevent serious
> > CPU damage it can not (by design) completly safe (system may freeze on
> > some setups).
>
> I've slowly warmed to the *idea* of having some means for supporting
> mobile chips in desktops, as it does seem to be happening more and more
> for some reason. I didn't look at your last patch to be honest, so I can't
> comment on whether its fit for merging before I take a look.
An updated version of the patch is attached to this mail. It is against
2.6.11-rc2, so you should not have problems to appy it.
To be ready for testing and maybe inclusion into one of the next kernel
releases (or mm tree) i have made it optional (see the Kconfig patch). The
txt file explains how to use it (for Documentation/cpu-freq).
(the patch also readds a fix for a comment, this was not included with the
last patch)
>
> > On the other side their is demand for this feature (not only for me). I
> > do not know what to do with it in the future. If the door for inclusion
> > in mainline is closed I would consider make some kind of source package
> > and forking/renaming the driver so that it can be build independent from
> > kernel.
>
> Lets see what we can do to get it included. Maintaining out of tree
> modules is a real pain in the ass for all concerned, including end-users.
> Rather than see endless support requests for 'why wont it build' on
> cpufreq-list I'd rather we did what we could to get things like this
> included.
>
Ack, and making it optional gives distributors the chance to decide if they
want this feature in their precompiled kernels.
> I'm holding off on 'feature' type things for 2.6.11 from now on, the next
> round of cpufreq bits going to Linus should be only bug fixes, so hopefully
> by 2.6.12 time, we can close the book on this one. This gives us enough
> time to give it some exposure in Andrew's -mm tree too.
No, stress.
Just having enough time to discuss it.
To the list: Please complain now!
Hendrik
P.S. I have used this patch nearly 4 months now without any problems, also
people on the list reported success. I use it for a HTPC which has a MATX
mainboard and a XP-M 2400+. With reduced Vcore (BIOS setting), FSB of 100 Mhz
(instead of 133 Mhz) and this patch I am able to run the CPU without a fan.
P.P.S. Kudos also to Adam Jones which inspired the patch, Jarkko Lavinen for
improving it (fsb option) and all others not named here
[-- Attachment #2: cpufreq-powernow-k7-manual.txt --]
[-- Type: text/plain, Size: 1635 bytes --]
The powernow-k7-manual option lets you manual set the PST table
(powerswitching table) with module options.
This is useful if you have a laptop with a broken BIOS or if you use a
mobile CPU on a desktop motherboard, which BIOS does not have PST tables.
The modules has the following options:
overwrite_table: activates the manually settings, set it to >=1
multiplier: specifies an array of multipliers
voltage: specifies an array of VCore voltages (default
BIOS settings)
switch_latency: sets the state transition latency in
microseconds (default 200us)
fsb: sets the frontsidebus (default calculated from
BIOS settings)
Odd multipliers must be multiplied by 10, for example a multiplier of
6.5 becomes 65.
Voltages must be given in mV, the following voltages are valid:
925, 950, 975, 1000, 1025, 1050, 1075, 1100, 1125, 1150, 1175, 1200,
1225, 1250, 1275, 1300, 1350, 1400, 1450, 1500, 1550, 1600, 1650, 1700,
1750, 1800, 1850, 1900, 1950, 2000.
The patch does not allow overclocking: higher multipliers and/or voltages than
the maximum specified on CPU will be ignored.
You do not have to specify voltages, in this case the CPU default will be used
(on some boards voltage switching is not possible anyway).
example usage:
modprobe powernow-k7 overwrite_table=1 multiplier=3,4,5,6,7,8,9,10
Disclaimer:
It should not be possible to burn the CPU, but some combinations could crash
the system. We give no guaranty that it works for your system and/or any
possibly resulting hardware damage.
[-- Attachment #3: powernow-k7-manual.patch --]
[-- Type: text/x-diff, Size: 10534 bytes --]
--- linux/arch/i386/kernel/cpu/cpufreq/powernow-k7.c.orig 2005-01-25 19:54:10.000000000 +0100
+++ linux/arch/i386/kernel/cpu/cpufreq/powernow-k7.c 2005-01-25 20:06:12.000000000 +0100
@@ -66,7 +66,7 @@
};
#endif
-#ifdef CONFIG_CPU_FREQ_DEBUG
+#if defined CONFIG_CPU_FREQ_DEBUG || defined CONFIG_X86_POWERNOW_K7_MANUAL
/* divide by 1000 to get VCore voltage in V. */
static int mobile_vid_table[32] = {
2000, 1950, 1900, 1850, 1800, 1750, 1700, 1650,
@@ -76,7 +76,7 @@
};
#endif
-/* divide by 10 to get FID. */
+/* divide by 10 to get multiplier. */
static int fid_codes[32] = {
110, 115, 120, 125, 50, 55, 60, 65,
70, 75, 80, 85, 90, 95, 100, 105,
@@ -84,6 +84,31 @@
150, 225, 160, 165, 170, 180, -1, -1,
};
+#ifdef CONFIG_X86_POWERNOW_K7_MANUAL
+/* translation table for even multiplier to fid */
+static int even_multiplier[20] = {
+ 16, 18, 4, 6, 8, 10, 12, 14, // 3, 4, 5, 6, 7 ,8 , 9, 10
+ 0, 2, 20, 22, 24, 26, 28, 29, // 11, 12, 13, 14, 15, 16, 17, 18
+ 17, 19, 23, 25, // 19, 20, 21, 22
+};
+
+/* translation table for odd multiplier to fid*/
+static int odd_multiplier[9] = {
+ 5, 7, 9, 11, 13, 15, 1, 3, // 5.5, 6.5, 7.5, 8.5, 9.5, 10.5, 11.5, 12.5
+ 21, // 13.5
+};
+
+/* This parameters can be used to manually overwrite the tables */
+static int overwrite_table = 0;
+#define MAX_PST 10
+static int multiplier_arr_size = MAX_PST;
+static int voltage_arr_size = MAX_PST;
+static int multiplier[MAX_PST] = {[0 ... (MAX_PST - 1)] = 0 };
+static int voltage[MAX_PST] = {[0 ... (MAX_PST - 1)] = 0 };
+static int switch_latency = 0;
+#endif /* CONFIG_X86_POWERNOW_K7_MANUAL */
+
+
/* This parameter is used in order to force ACPI instead of legacy method for
* configuration purpose.
*/
@@ -97,7 +122,7 @@
static unsigned int minimum_speed=-1;
static unsigned int maximum_speed;
static unsigned int number_scales;
-static unsigned int fsb;
+static unsigned int fsb=0;
static unsigned int latency;
static char have_a0;
@@ -413,6 +438,143 @@
}
#endif
+#ifdef CONFIG_X86_POWERNOW_K7_MANUAL
+static int powernow_manual_settings(union msr_fidvidstatus *fidvidstatus)
+{
+ int i, k, validentry;
+ unsigned int max_multiplier, max_voltage;
+ unsigned int speed, cm;
+ u8 vid, fid;
+ static struct cpufreq_frequency_table *powernow_table_tmp;
+
+ if (switch_latency > 0) {
+ if (switch_latency < 100) {
+ printk(KERN_INFO PFX
+ "Settling time passed as %d microseconds."
+ "Should be at least 100. Correcting.\n",
+ switch_latency);
+ switch_latency = 100;
+ }
+ latency = switch_latency;
+ } else {
+ latency = 200;
+ }
+ dprintk("Settling Time: %d microseconds.\n", latency);
+
+ /* get number of specified multipliers */
+ number_scales = multiplier_arr_size;
+ for (i = 0; i < multiplier_arr_size; i++) {
+ if (multiplier[i] == 0) {
+ number_scales = i;
+ break;
+ }
+ }
+
+ /* get maximum values */
+ max_multiplier = fid_codes[fidvidstatus->bits.MFID];
+ max_voltage = mobile_vid_table[fidvidstatus->bits.MVID];
+
+ /* allocate memory */
+ powernow_table =
+ kmalloc((sizeof(struct cpufreq_frequency_table) *
+ (number_scales + 1)), GFP_KERNEL);
+ if (!powernow_table)
+ return -ENOMEM;
+ memset(powernow_table, 0,
+ (sizeof(struct cpufreq_frequency_table) * (number_scales + 1)));
+
+ k = 0;
+ for (i = 0; i < number_scales; i++) {
+ validentry = 0;
+ if (multiplier[i] != 0) {
+ /* fix multiplier */
+ if (multiplier[i] < 30)
+ multiplier[i] = multiplier[i] * 10;
+ if (multiplier[i] < max_multiplier) {
+ cm = (multiplier[i] / 10);
+
+ /* check if odd or even muliplier */
+ if (multiplier[i] % 10) {
+ /* odd multiplier */
+ if (cm == 16) {
+ /* hardcoded because 14.5 and 15.5 fids not possible */
+ fid = 27;
+ validentry = 1;
+ } else if ((cm > 4) && (cm < 14)) {
+ fid = odd_multiplier[cm - 5];
+ validentry = 1;
+ }
+ } else {
+ /* even_multiplier */
+ if ((cm < 23) && (cm > 2)) {
+ fid = even_multiplier[cm - 3];
+ validentry = 1;
+ }
+ }
+ }
+ }
+
+ if (validentry) {
+ /* if no voltage specified use CPU default */
+ if (voltage[i] == 0)
+ voltage[i] = max_voltage;
+ /* we do not allow higher voltages than the CPU's maximum 925 mV is the minimum */
+ if ((voltage[i] <= max_voltage) && (voltage[i] >= 925)) {
+ if (voltage[i] >= 1300) {
+ vid = 40 - (voltage[i] / 50);
+ } else {
+ vid = 67 - (voltage[i] / 25);
+ }
+ /* calculate speed */
+ speed = fsb * fid_codes[fid] / 10;
+ powernow_table[k].frequency = speed;
+ powernow_table[k].index = fid; /*lower 8 bits */
+ powernow_table[k].index |= (vid << 8); /*upper 8 bits */
+
+ if (speed < minimum_speed)
+ minimum_speed = speed;
+ if (speed > maximum_speed)
+ maximum_speed = speed;
+
+ dprintk(" FID: 0x%x (%d.%dx [%dMHz])\t", fid,
+ fid_codes[fid] / 10,
+ fid_codes[fid] % 10, speed / 1000);
+ dprintk("VID: 0x%x (%d.%03dV)\n", vid,
+ mobile_vid_table[vid] / 1000,
+ mobile_vid_table[vid] % 1000);
+ k++;
+ }
+ } else {
+ // invalid entry
+ dprintk("Entry %d is invalid\n", i + 1);
+ }
+ }
+ if (k < number_scales) {
+ /* some entrys were invalid need to realloc table */
+ number_scales = k;
+ powernow_table_tmp =
+ kmalloc((sizeof(struct cpufreq_frequency_table) *
+ (number_scales + 1)), GFP_KERNEL);
+ if (!powernow_table_tmp) {
+ kfree(powernow_table);
+ return -ENOMEM;
+ }
+ memcpy(powernow_table_tmp, powernow_table,
+ (sizeof(struct cpufreq_frequency_table) *
+ (number_scales + 1)));
+ kfree(powernow_table);
+ powernow_table = powernow_table_tmp;
+ }
+
+ /* Terminate frequency list */
+ powernow_table[number_scales].frequency = CPUFREQ_TABLE_END;
+ powernow_table[number_scales].index = 0;
+
+ return 0;
+}
+#endif /* CONFIG_X86_POWERNOW_K7_MANUAL */
+
+
static int powernow_decode_bios (int maxfid, int startvid)
{
struct psb_s *psb;
@@ -585,6 +747,18 @@
return -ENODEV;
rdmsrl (MSR_K7_FID_VID_STATUS, fidvidstatus.val);
+#ifdef CONFIG_X86_POWERNOW_K7_MANUAL
+ if (fsb) {
+ if (fsb < 1000)
+ fsb *= 1000;
+
+ if (fsb < 1000 || fsb > 1000000) {
+ printk(KERN_WARNING PFX "FSB %ukhz out of range\n",
+ fsb);
+ return -EINVAL;
+ }
+ } else
+#endif
/* A K7 with powernow technology is set to max frequency by BIOS */
fsb = (10 * cpu_khz) / fid_codes[fidvidstatus.bits.MFID];
@@ -597,6 +771,12 @@
if (dmi_check_system(powernow_dmi_table) || acpi_force) {
printk (KERN_INFO PFX "PSB/PST known to be broken. Trying ACPI instead\n");
result = powernow_acpi_init();
+#ifdef CONFIG_X86_POWERNOW_K7_MANUAL
+ } else if (overwrite_table) {
+ printk(KERN_INFO PFX
+ "Overwriting PST table with manual settings\n");
+ result = powernow_manual_settings(&fidvidstatus);
+#endif
} else {
result = powernow_decode_bios(fidvidstatus.bits.MFID, fidvidstatus.bits.SVID);
if (result) {
@@ -681,6 +861,22 @@
module_param(acpi_force, int, 0444);
MODULE_PARM_DESC(acpi_force, "Force ACPI to be used.");
+#ifdef CONFIG_X86_POWERNOW_K7_MANUAL
+module_param(overwrite_table, int, 0444);
+MODULE_PARM_DESC(overwrite_table, "overwrite table with manually settings");
+module_param(fsb, uint, 0444);
+MODULE_PARM_DESC(fsb, "(MHz) overrides the calculated FSB");
+module_param_array(multiplier, int, &multiplier_arr_size, 0444);
+MODULE_PARM_DESC(multiplier,
+ "Specifiy up to 10 multipliers, multiply them by 10: 5->50, 5.5->55");
+module_param_array(voltage, int, &voltage_arr_size, 0444);
+MODULE_PARM_DESC(voltage,
+ "Specify voltages in respect to the given multipliers, specify them in mV: 1.275V -> 1275");
+module_param(switch_latency, int, 0444);
+MODULE_PARM_DESC(switch_latency,
+ "Set state transition latency in microseconds (default 200us)");
+#endif
+
MODULE_AUTHOR ("Dave Jones <davej@codemonkey.org.uk>");
MODULE_DESCRIPTION ("Powernow driver for AMD K7 processors.");
MODULE_LICENSE ("GPL");
[-- Attachment #4: powernow-k7-Kconfig.patch --]
[-- Type: text/x-diff, Size: 780 bytes --]
--- linux/arch/i386/kernel/cpu/cpufreq/Kconfig.orig 2005-01-25 19:54:17.000000000 +0100
+++ linux/arch/i386/kernel/cpu/cpufreq/Kconfig 2005-01-25 20:07:52.000000000 +0100
@@ -71,6 +71,17 @@
depends on ((X86_POWERNOW_K7 = "m" && ACPI_PROCESSOR) || (X86_POWERNOW_K7 = "y" && ACPI_PROCESSOR = "y"))
default y
+config X86_POWERNOW_K7_MANUAL
+ bool "Manual setting of PowerNow! table for AMD Mobile Athlon/Duron."
+ depends on X86_POWERNOW_K7
+ default n
+ help
+ This adds the possibility to manual change the PST table.
+
+ For details, take a look at <file:Documentation/cpu-freq/>.
+
+ If in doubt, say N.
+
config X86_POWERNOW_K8
tristate "AMD Opteron/Athlon64 PowerNow!"
depends on CPU_FREQ_TABLE && EXPERIMENTAL
[-- Attachment #5: Type: text/plain, Size: 147 bytes --]
_______________________________________________
Cpufreq mailing list
Cpufreq@lists.linux.org.uk
http://lists.linux.org.uk/mailman/listinfo/cpufreq
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [Patch] powernow-k7-manual was Re: [Patch] small cleanup for powernow-k7.c
2005-01-25 20:47 ` [Patch] powernow-k7-manual was " Hendrik Muhs
@ 2005-01-26 8:03 ` Jarkko Lavinen
0 siblings, 0 replies; 4+ messages in thread
From: Jarkko Lavinen @ 2005-01-26 8:03 UTC (permalink / raw)
To: Hendrik Muhs; +Cc: Adam Jones, Dave Jones, cpufreq
On Tue, Jan 25, 2005 at 09:47:58PM +0100, Hendrik Muhs wrote:
> To the list: Please complain now!
Hi Hendrik
I tried the patch and it works alright. I have few beauty issues.
There are two consequtive dprintks in powernow_manual_settings() which need
to be unified:
- dprintk(" FID: 0x%x (%d.%dx [%dMHz])\t", fid,
- fid_codes[fid] / 10,
- fid_codes[fid] % 10, speed / 1000);
- dprintk("VID: 0x%x (%d.%03dV)\n", vid,
- mobile_vid_table[vid] / 1000,
+ dprintk(" FID: 0x%x (%d.%dx [%dMHz])\t"
+ "VID: 0x%x (%d.%03dV)\n",
+ fid, fid_codes[fid] / 10,
+ fid_codes[fid] % 10, speed / 1000,
+ vid, mobile_vid_table[vid] / 1000,
+ mobile_vid_table[vid] % 1000);
And there is one unneeded check from my fsb code in powernow_cpu_init(). fsb
is unsigned, myltiplied by 1000 and already checked to be nonzero, so checking
for fsb < 1000 is really not needed.
if (fsb < 1000)
fsb *= 1000;
- if (fsb < 1000 || fsb > 1000000) {
+ if (fsb > 1000000) {
printk(KERN_WARNING PFX "FSB %ukhz out of range\n",
fsb);
return -EINVAL;
> I use it for a HTPC which has a MATX mainboard and a XP-M 2400+. With
> reduced Vcore (BIOS setting), FSB of 100 Mhz (instead of 133 Mhz) and
> this patch I am able to run the CPU without a fan.
I am building a VDR/HTPC box with very similar desires.
Jarkko Lavinen
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2005-01-26 8:03 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2005-01-20 20:38 [Patch] small cleanup for powernow-k7.c Hendrik Muhs
2005-01-21 1:57 ` Dave Jones
2005-01-25 20:47 ` [Patch] powernow-k7-manual was " Hendrik Muhs
2005-01-26 8:03 ` Jarkko Lavinen
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.