On Tue, 18 Jan 2011 17:32:46 +0100, Jean Delvare wrote:
On Tue, 18 Jan 2011 10:03:23 -0600, Nick Hall wrote:
And regarding the sensors-detect script, I understand now. If anyone is
interested, my Sandy Bridge processor would be model "0x2A" according to how
that script does things.
This is an option, yes, but it would be better if we could implement
the same detection logic as the coretemp driver has. For one thing, this
would guarantee that the two are always in sync. For another, it would
lower the maintenance effort from our side, as the new detection logic
is universal and doesn't need to be updated with every new CPU model.
I'm looking into it but this is non-trivial.
I have come up with the following patch:
Index: prog/detect/sensors-detect
===================================================================
--- prog/detect/sensors-detect (révision 5904)
+++ prog/detect/sensors-detect (copie de travail)
@@ -23,7 +23,7 @@
require 5.004;
use strict;
-use Fcntl;
+use Fcntl qw(:DEFAULT :seek);
use File::Basename;
# We will call modprobe, which typically lives in either /sbin,
@@ -2046,14 +2046,10 @@
driver => "k10temp",
detect => \&fam11h_pci_detect,
}, {
- name => "Intel Core family thermal sensor",
+ name => "Intel digital thermal sensor",
driver => "coretemp",
- detect => sub { coretemp_detect(0); },
+ detect => \&coretemp_detect,
}, {
- name => "Intel Atom thermal sensor",
- driver => "coretemp",
- detect => sub { coretemp_detect(1); },
- }, {
name => "Intel AMB FB-DIMM thermal sensor",
driver => "i5k_amb",
detect => \&intel_amb_detect,
@@ -2314,10 +2310,10 @@
while (<INPUTFILE>) {
if (m/^processor\s*:\s*(\d+)/) {
push @cpu, $entry if scalar keys(%{$entry}); # Previous entry
- $entry = {}; # New entry
+ $entry = { nr => $1 }; # New entry
next;
}
- if (m/^(vendor_id|cpu family|model|model name|stepping)\s*:\s*(.+)$/) {
+ if (m/^(vendor_id|cpu family|model|model name|stepping|cpuid level)\s*:\s*(.+)$/) {
my $k = $1;
my $v = $2;
$v =~ s/\s+/ /g; # Merge multiple spaces
@@ -2486,6 +2482,15 @@
$modules_list{$normalized} = 1;
}
+# udev may take some time to create device nodes when loading modules
+sub udev_settle
+{
+ if (!(-x "/sbin/udevadm"&& system("/sbin/udevadm settle") == 0)
+ && !(-x "/sbin/udevsettle"&& system("/sbin/udevsettle") == 0)) {
+ sleep(1);
+ }
+}
+
sub initialize_modules_supported
{
foreach my $chip (@chip_ids) {
@@ -5833,23 +5838,33 @@
return;
}
+sub cpuid
+{
+ my ($cpu_nr, $eax) = @_;
+
+ sysopen(CPUID, "/dev/cpu/$cpu_nr/cpuid", O_RDONLY) or return;
+ binmode CPUID;
+ sysseek(CPUID, $eax, SEEK_SET)
+ or die "Cannot seek /dev/cpu/$cpu_nr/cpuid";
+ sysread(CPUID, my $data, 16)
+ or die "Cannot read /dev/cpu/$cpu_nr/cpuid";
+ close CPUID;
+
+ return unpack("L4", $data);
+}
+
sub coretemp_detect
{
- my $chip = shift;
my $probecpu;
foreach $probecpu (@cpu) {
next unless $probecpu->{vendor_id} eq 'GenuineIntel'&&
- $probecpu->{'cpu family'} == 6;
- return 9 if $chip == 0&&
- ($probecpu->{model} == 14 || # Pentium M DC
- $probecpu->{model} == 15 || # Core 2 DC 65nm
- $probecpu->{model} == 0x16 || # Core 2 SC 65nm
- $probecpu->{model} == 0x17 || # Penryn 45nm
- $probecpu->{model} == 0x1a || # Nehalem
- $probecpu->{model} == 0x1e); # Lynnfield
- return 9 if $chip == 1&&
- ($probecpu->{model} == 0x1c); # Atom
+ $probecpu->{'cpuid level'}>= 6;
+
+ # Now we check for the DTS flag
+ my @regs = cpuid($probecpu->{nr}, 0x06);
+ return unless @regs == 4;
+ return 9 if ($regs[0]& (1<< 0)); # eax, bit 0
}
return;
}
@@ -6203,6 +6218,12 @@
print "Some south bridges, CPUs or memory controllers contain embedded sensors.\n".
"Do you want to scan for them? This is totally safe. (YES/no): ";
unless (<STDIN> =~ /^\s*n/i) {
+ # Load the cpuid driver if needed
+ if (@cpu>= 1&& ! -e "/dev/cpu/$cpu[0]->{nr}/cpuid") {
+ load_module("cpuid");
+ udev_settle();
+ }
+
$| = 1;
foreach my $entry (@cpu_ids) {
scan_cpu($entry);
@@ -6278,12 +6299,7 @@
$by_default = 1 if dmi_match('board_vendor', 'asustek', 'tyan',
'supermicro');
- # udev may take some time to create the device node
- if (!(-x "/sbin/udevadm"&& system("/sbin/udevadm settle") == 0)
- && !(-x "/sbin/udevsettle"&& system("/sbin/udevsettle") == 0)) {
- sleep(1);
- }
-
+ udev_settle();
for (my $dev_nr = 0; $dev_nr< @i2c_adapters; $dev_nr++) {
next unless exists $i2c_adapters[$dev_nr];
scan_i2c_adapter($dev_nr, $by_default);
This seems to do the trick for me. Obviously it assumes that the cpuid
kernel driver is available. All my systems have it available as a
module, but I don't know if we can reasonably assume that it will be
available on all x86 systems where sensors-detect is used.
I would like this change to get as wide a test coverage as possible.
To make testing easier, I've made the full script available for
download at:
http://khali.linux-fr.org/devel/misc/sensors-detect
(It's exactly equivalent to sensors-detect SVN + the patch above.) In
particular, this needs testing on Intel CPUs with cpuid level>= 6 but
without DTS support... if such systems exist (I don't have any here,
for sure.)
If this appears to work for everyone and nobody comes up with a major
objection, then we have our fix.
Thanks,
