Bug 6367
| Summary: | acpi-cpufreq: fancy CPUfreq values - can't load p4-clockmod anymore | ||
|---|---|---|---|
| Product: | Power Management | Reporter: | paolo (oopla) |
| Component: | cpufreq | Assignee: | cpufreq (cpufreq) |
| Status: | REJECTED UNREPRODUCIBLE | ||
| Severity: | normal | CC: | akpm, bunk |
| Priority: | P2 | ||
| Hardware: | i386 | ||
| OS: | Linux | ||
| Kernel Version: | 2.6.16.1 + suspend2-2.2.4 | Subsystem: | |
| Regression: | --- | Bisected commit-id: | |
| Attachments: | oops on modprobe -r cpufreq_stats | ||
|
Description
paolo
2006-04-10 06:06:56 UTC
Created attachment 7875 [details]
oops on modprobe -r cpufreq_stats
tested 2.6.16.5: better, p4-clockmod would load, but avail_scaling_freq shows just:
2025000 2362500 2700000
while I know this CPU offers a larger f-table - I'd say the old one reported is
correct/closer to what ToshibaHCI offers.
But get an oops on modprobe -r cpufreq_stats - see attachment #7875 [details].
changing severity to 'normal', as p4-clockmod now loads and cpufreq works albeit with very limited (nearly useless wrt pw.saving) cpu-freq. Also, cpufreq_stats seems not essentials, so temp w.around is to not load it at all. Does this reproduceable with latest RC kernel? 2.6.17-rc1 won't even boot - pls see #6368 k2.6.16.7 (some printk() added - note: PREEMPT in on)
1- p4-clockmod as shipped :
in cpufreq_p4_cpi_init()
p4-clockmod: P4/Xeon(TM) CPU On-Demand Clock Modulation available
has_N44_O17_errata[policy->cpu]=0, has_N60_errata[policy->cpu]=1
p4clockmod_table[1].frequency = -1
p4clockmod_table[2].frequency = -1
p4clockmod_table[3].frequency = -1
p4clockmod_table[4].frequency = -1
p4clockmod_table[5].frequency = -1
p4clockmod_table[6].frequency = 2025000
p4clockmod_table[7].frequency = 2362500
p4clockmod_table[8].frequency = 2700000
- removing 'else if (has_N60...)':
p4-clockmod: P4/Xeon(TM) CPU On-Demand Clock Modulation available
has errata -- disabling frequencies lower than 2ghz
p4clockmod_table[1].frequency = 337500
p4clockmod_table[2].frequency = 675000
p4clockmod_table[3].frequency = 1012500
p4clockmod_table[4].frequency = 1350000
p4clockmod_table[5].frequency = 1687500
p4clockmod_table[6].frequency = 2025000
p4clockmod_table[7].frequency = 2362500
p4clockmod_table[8].frequency = 2700000
and it works fine as before, in 2.6.15. I can change F, bogomips/actual
performance are consistent with F selected. I'm almost always using
'ondemand', other governors seem to get stuck to either L/H limits.
So, looks like such 'has_N60_errata' test is wrong here.
2- oops on 'rmmod cpufreq_stats': same as original report, still b0rked:
Unable to handle kernel paging request at virtual address dfa6e000
printing eip:
dfa6e000
*pde = 01465067
*pte = 00000000
Oops: 0000 [#1]
PREEMPT
Modules linked in: cpufreq_stats cpufreq_conservative cpufreq_powersave cpufreq_
userspace cpufreq_ondemand p4_clockmod freq_table usb_storage ipt_ULOG ipv6 twof
ish serpent des sha256 crypto_null capi capifs kernelcapi pcmcia firmware_class
iptable_filter xfrm_user ipcomp esp4 ah4 iptable_mangle ip_tables x_tables nls_i
so8859_15 nls_cp850 vfat fat dm_mod sg sr_mod sd_mod scsi_mod lp speedstep_lib s
oftdog i915 drm hangcheck_timer toshiba_acpi thermal processor fan button batter
y ac video non_fatal capability commoncap cpuid usbhid evdev 8250_pci e100 mii y
enta_socket rsrc_nonstatic pcmcia_core 8250 serial_core snd_intel8x0 snd_intel8x
0m snd_ac97_codec snd_ac97_bus snd_pcm_oss snd_mixer_oss snd_pcm intel_agp agpga
rt ide_cd psmouse ehci_hcd i2c_i801 uhci_hcd snd_timer snd soundcore snd_page_al
loc parport_pc parport cdrom serio_raw rtc i2c_core pcspkr
CPU: 0
EIP: 0060:[<dfa6e000>] Not tainted VLI
EFLAGS: 00010282 (2.6.16.7-ss2.2.4-p4.1 #1)
EIP is at 0xdfa6e000
eax: 00000000 ebx: dfba2500 ecx: 00000000 edx: ffff0001
esi: 00000000 edi: 00000880 ebp: d625c000 esp: d625df4c
ds: 007b es: 007b ss: 0068
Process rmmod (pid: 8558, threadinfo=d625c000 task=d33a3a90)
Stack: <0>dfba16e7 dfba247c 00000007 00000000 c01361a2 dfba2500 00000880 d625df6
c
00000000 66757063 5f716572 74617473 c2fa0073 d143d494 b7f0e000 b7f0f000
d143d494 cef4d200 cef4d234 00000000 d625c000 c015b2fe cef4d200 00f0e000
Call Trace:
[<dfba16e7>] cpufreq_stats_exit+0x47/0x4b [cpufreq_stats]
[<c01361a2>] sys_delete_module+0x133/0x161
[<c015b2fe>] sys_munmap+0x44/0x64
[<c0102bd9>] syscall_call+0x7/0xb
Code: Bad EIP value.
Is this crash still happening in 2.6.20-rc7? can't tell, have no more that P4 notebook handy, sorry. Can only report that, on a "Intel(R) Celeron(R) M processor 1400MHz" which looks like a PM but isn't - ie speedstep won't work while p4_clockmod does - I had similar problems, but 2.6.18.6 works fine. rmmod cpufreq_stats won't oops here, either. Also, 2.6.20-rc* broke too many things on this pc, running Debian Etch, that I refrained from going on 2.6.19 and above. Not 100% sure, but it might be reasonable to close this bugrep. thx this happens again also in 2.6.19.7 acpi_cpufreq sets these scaling freqs, as in original report: # cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 2700000 65535000 65535000 65535000 65535000 65535000 65535000 65535000 65535000 65535000 65535000 65535000 65535000 65535000 65535000 once loaded, acpi_cpufreq can't be removed and p4-clockmod can't be loaded. same for p4_clockmod as in original report, scaling freqs are bound > 2GHz, which turns out to be wrong on this CPU. Seems that p4-clockmod can be unloaded. Seems that loading/unloading cpufreq-stats doesn't oopses anymore, though, so there seems to be an improvement. 2.6.19.7 is prehistoric. Please try again with a newer kernel, as some similar bugs were definitely fixed after that version. Please reopen this bug if it's still present with kernel 2.6.22. I will, provided that I'll find a way to get 2.6.22.* to work: 2.6.22.* doesn't even boot on any of 4 brands/flavours of notebooks I tried: P3, P4, PM; it hangs on ACPI stuff :( |