Skip to contentSkip to conversation

Linux 2.6.30-rc8

PATCH In review Linus Torvalds 12 participants 84 messages Jun 3, 2009 – Jun 15, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

This is almost certainly the last -rc, and I debated even doing it. But rather than just do 2.6.30, I decided that I'd be better off doing a last -rc8 and then a real release probably this weekend.

This has mostly driver and arch updates, with perhaps the intel drm/kms and network driver changes standing out, but there's powerpc, blackfin and arm updates too. Admittedly, the two biggest parts of the powerpc update are a revert and a defconfig update.

A lot of small stuff, fixing a few regressions (and at least one bugzilla entry going back to 2.6.24). The small stuff does matter. Please test.

Linus

---
Alan Cox (3):
8250: Fix oops from setserial
pata_netcell: LBA48 force identify bits correct
parport: quickfix the proc registration bug

Alan Stern (1):
usb-serial: fix crash when sub-driver updates firmware

Alex Chiang (2):
atlx: move modinfo data from atlx.h to atl1.c
PCI Hotplug: acpiphp: don't store a pci_dev in acpiphp_func

Alex Riesen (1):
Use a format for linux_banner

Alexander Beregalov (2):
parport_gsc: fix printk format error
serial: 8250_gsc: fix printk format error

Alexey Dobriyan (1):
cred: #include init.h in cred.h

Andreas Herrmann (1):
[CPUFREQ] powernow-k8: determine exact CPU frequency for HW Pstates

Andrew Morton (1):
sysfs: file.c: use create_singlethread_workqueue()

Avi Kivity (2):
KVM: Make paravirt tlb flush also reload the PAE PDPTRs
KVM: Fix PDPTR reloading on CR4 writes

Bartlomiej Zolnierkiewicz (1):
ide_pci_generic: add quirk for Netcell ATA RAID

Benjamin Herrenschmidt (5):
Revert "powerpc: Rework dma-noncoherent to use generic vmalloc layer"
powerpc: Move dma-noncoherent.c from arch/powerpc/lib to arch/powerpc/mm
powerpc: Minor cleanups of kernel virt address space definitions
powerpc: Fix up dma_alloc_coherent() on platforms without cache coherency. powerpc/pmac: Update PowerMac 32-bit defconfig

Christian Engelmayer (1):
hwmon: Update documentation on fan_max

Clemens Ladisch (1):
sound: usb-audio: make the MotU Fastlane work again

Coly Li (1):
[ARM] pxa: add parameter to clksrc_read() for pxa168/910

Daisuke Nishimura (1):
memcg: fix deadlock between lock_page_cgroup and mapping tree_lock

Dan Carpenter (1):
RxRPC: Error handling for rxrpc_alloc_connection()

Daniel Ribeiro (3):
[ARM] pxa: save/restore PGSR on suspend/resume. [ARM] pxa: allow gpio_reset drive high during normal work [ARM] pxa/ezx: fix pin configuration for low power mode

Dave Jones (1):
[CPUFREQ] powernow-k7 build fix when ACPI=n

Dave Young (1):
Bluetooth: Remove useless flush_work() causing lockdep warnings

David Dillow (1):
r8169: avoid losing MSI interrupts

David Howells (2):
FS-Cache: Fixup renamed filenames in comments in internal.h
CacheFiles: Fixup renamed filenames in comments in internal.h

David Rientjes (1):
oom: fix possible oom_dump_tasks NULL pointer

David S. Miller (2):
sparc64: Fix SET_PERSONALITY to not clip bits outside of PER_MASK.
sparc64: Fix section attribute warnings.

Divy Le Ray (2):
cxgb3: fix dma mapping regression
cxgb3: link fault fixes

Dmitry Eremin-Solenikov (1):
[ARM] pxa/spitz: provide spitz_ohci_exit() that unregisters USB_HOST GPIO

Dmitry Torokhov (1):
Input: libps2 - better handle bad scheduler decisions

Doug Leith (1):
tcp: tcp_vegas ssthresh bugfix

Ed Swierk (1):
forcedeth: add phy_power_down parameter, leave phy powered up by default (v2)

Eric Anholt (2):
drm/i915: Fix tiling pitch handling on 8xx.
drm/i915: Apply a big hammer to 865 GEM object CPU cache flushing.

Eric Dumazet (2):
net: fix length computation in rt_check_expire()
net: fix rtable leak in net/ipv4/route.c

Eric Sandeen (1):
xfs: fix overflow in xfs_growfs_data_private

Fabio Rossi (1):
ath5k: fix interpolation with equal power levels

Felix Blyakher (2):
xfs: fix double unlock in xfs_swap_extents()
xfs: prevent deadlock in xfs_qm_shake()

Finn Thain (2):
mac8390: fix regression caused during net_device_ops conversion
mac8390: fix build with NET_POLL_CONTROLLER

Florian Fainelli (1):
MAINTAINERS: take maintainership of the cpmac Ethernet driver

Florian Westphal (1):

      pktgen: do not access flows[] beyond its length

Forrest Zhang (1):
ath5k: fix exp off-by-one when computing OFDM delta slope

Frans Pop (1):
ACPI processor: remove spurious newline from warning message

Greg Kroah-Hartman (1):
ath1e: add new device id for asus hardware

H. Peter Anvin (1):
x86, setup: revert ACPI 3 E820 extended attributes support

Haavard Skinnemoen (1):
USB: atmel_usb_udc: Use kzalloc() to allocate ep structures

Harry Ciao (2):

      edac: AMD8111 & AMD8131 use dev_name()
      edac: AMD8111 & AMD8131 Kconfig fixup

Henrik Rydberg (2):
Input: multitouch - add tracking ID to the protocol
Input: multitouch - augment event semantics documentation

Herbert Xu (1):
crypto: hash - Fix handling of sg entry that crosses page boundary

Herton Ronaldo Krzesinski (1):
tomoyo: add missing call to cap_bprm_set_creds

Hideo Saito (1):
powerpc/mm: Fix broken MMU PID stealing on !SMP

Inaky Perez-Gonzalez (1):
wimax/i2400m: usb: fix device reset on autosuspend while not yet idle

Ira Snyder (4):
fsldma: fix "DMA halt timeout!" errors
fsldma: fix infinite loop on multi-descriptor DMA chain completion
fsldma: snooping is not enabled for last entry in descriptor chain
fsldma: fix memory leak on error path in fsl_dma_prep_memcpy()

J. Bruce Fields (1):
nfsd: Revert "svcrpc: take advantage of tcp autotuning"

James Bottomley (1):
async: make sure independent async domains can't accidentally entangle

Jarod Wilson (1):
[CPUFREQ] add atom family to p4-clockmod

Jaswinder Singh Rajput (3):
drm/i915: acpi/video.c fix section mismatch warning headers_check fix: linux/auto_fs.h headers_check fix: linux/net_dropmon.h

Jay Sternberg (1):
iwlwifi: update 5000 ucode support to version 2 of API

Jean-Mickael Guerin (1):
IPv6: set RTPROT_KERNEL to initial route

Jesper Dangaard Brouer (1):
netfilter: xt_hashlimit does a wrong SEQ_SKIP

Jesse Barnes (2):
drm/i915: allocate large pointer arrays with vmalloc
i915: support 8xx desktop cursors

Joakim Tjernlund (1):
jffs2: Fix corruption when flash erase/write failure

Joe Perches (2):
MAINTAINERS: pair EDAC-E752X P: and M: entries
acpi-cpufreq: fix printk typo and indentation

Johannes Berg (1):
wext: verify buffer size for SIOCSIWENCODEEXT

John W. Linville (3):
airo: fix airo_get_encode{,ext} buffer overflow like I mean it...
at76c50x-usb: avoid mutex deadlock in at76_dwork_hw_scan
rtl8187: add USB ID for Linksys WUSB54GC-EU v2 USB wifi dongle

Jonas Bonn (1):
drm/i915: Determine type before initialising connector

Jozsef Kadlecsik (1):
netfilter: nf_ct_tcp: fix accepting invalid RST segments

KOSAKI Motohiro (1):
procfs: make errno values consistent when open pident vs exit(2) race occurs

Kay Sievers (1):
Driver Core: do not oops when driver_unregister() is called for unregistered drivers

Kenji Kaneshige (1):
PCI/ACPI: fix wrong ref count handling in acpi_pci_bind()

Kristian Høgsberg (1):
i915: Set object to gtt domain when faulting it back in

Kumar Gala (1):
fsldma: Fix compile warnings

Len Brown (3):
i7300_idle: allow testing on i5000-series hardware w/o re-compile
ACPI: sanity check _PSS frequency to prevent cpufreq crash ACPI, i915: build fix (v2)

Lennert Buytenhek (1):
gianfar: fix BUG under load after introduction of skb recycling

Li Yang (1):
fsldma: update mailling list address in MAINTAINERS

Linus Torvalds (1):

      Linux 2.6.30-rc8

Luis R. Rodriguez (1):
cfg80211: fix race between core hint and driver's custom apply

Ma Ling (4):
drm/i915: Fetch SDVO LVDS mode lines from VBT, then reserve them drm/i915: Return SDVO LVDS VBT mode if no EDID modes are detected. drm/i915: Use an I2C algo to do the flip to SDVO DDC bus. drm/i915: Add support for VGA load detection (pre-945).

Maciej W. Rozycki (1):
3c509: Add missing EISA IDs

Manuel Traut (1):
Input: usb1400_ts - fix access to "device data" in resume function

Marek Szyprowski (1):
S3C-fb: PM fix

Marek Vasut (1):
[ARM] pxa/palm: fix PalmLD/T5/TX AC97 MFP

Martin Fuzzey (1):
USB: atmel-usba-udc : fix control out requests.

Martin Michlmayr (1):
[ARM] Orion: Remove explicit name for platform device resources

Mathieu Desnoyers (4):
[CPUFREQ] remove rwsem lock from CPUFREQ_GOV_STOP call [CPUFREQ] fix timer teardown in conservative governor [CPUFREQ] fix timer teardown in ondemand governor [ARM] Add cmpxchg support for ARMv6+ systems (v5)

Matt Kraai (1):
net/firmare: Ignore .cis files

Mel Gorman (2):
x86: ignore VM_LOCKED when determining if hugetlb-backed page tables can be shared or not
mm: account for MAP_SHARED mappings using VM_MAYSHARE and not VM_SHARED in hugetlbfs

Mike Frysinger (9):
bfin_mac: fix build error due to net_device_ops convert
Blackfin: hook up preadv/pwritev syscalls
MAINTAINERS: update Blackfin items
MAINTAINERS: drop (subscribers-only) markings on Blackfin lists
Blackfin: ignore generated vmlinux.lds
Blackfin: drop unneeded asm/.gitignore
Blackfin: fix strncmp.o build error
Revert "USB: Correct Makefile to make isp1760 buildable"
hwmon: (lm78) Add missing __devexit_p()

Mingwei Wang (1):

      [ARM] pxa: fix the incorrectly defined drive strength macros for pxa{168,910}

Minoru Usui (1):
net_cls: fix unconfigured struct tcf_proto keeps chaining and avoid kernel panic when we use cls_cgroup

Neil Horman (1):
e1000: add missing length check to e1000 receive routine

NeilBrown (8):
md: always update level / chunk_size / layout when writing v1.x metadata.
md: improve errno return when setting array_size
md: bitmap: improve bitmap maintenance code.
md: export 'frozen' resync state through sysfs
md: raid5: avoid sector values going negative when testing reshape progress.
md: don't update curr_resync_completed without also updating reshape_position.
md: don't use locked_ioctl.
md: raid5: change incorrect usage of 'min' macro to 'min_t'

Nicolas Ferre (1):
atmel_lcdfb: correct fifo size for some products

Nicolas Pitre (1):
[ARM] add coherent DMA mask for mv643xx_eth

Nikanth Karthikesan (1):

      memcg: fix build warning and avoid checking for mem != null again and again

Oskar Schirmer (1):
flat: fix data sections alignment

Ozan Çağlayan (1):
ALSA: hda - Add forced codec-slots for ASUS W5Fm

Pablo Neira Ayuso (2):
netfilter: nf_ct_dccp: add missing DCCP protocol changes in event cache
netfilter: nfnetlink_log: fix wrong skbuff size calculation

Pallipadi, Venkatesh (1):
x86: avoid back to back on_each_cpu in cpa_flush_array

Paul Menage (1):
cls_cgroup: read classid atomically in classifier

Paulius Zaleckas (2):
MAINTAINER: Add F: entries for Gemini and FA526
Gemini: Fix SRAM/ROM location after memory swap

Pavel Roskin (1):
ath5k: fix scanning in AR2424

Rafael J. Wysocki (1):
PM: Do not hold dpm_list_mtx while disabling/enabling nonboot CPUs

Reinette Chatre (1):
iwlwifi: do not cancel delayed work inside spin_lock_irqsave

Robert Olsson (1):
ipv4: Fix oops with FIB_TRIE

Robert Richter (1):
oprofile: fix cpu buffer size

Roel Kluin (4):

      wireless: beyond ARRAY_SIZE of intf->crypto_stats
      gigaset: beyond ARRAY_SIZE of iwb->data
      fsldma: fix check on potential fdev->chan[] overflow
      drivers/serial/mpc52xx_uart.c: fix array overindexing check

Russell King (3):
[ARM] disable NX support for OABI-supporting kernels [ARM] barriers: improve xchg, bitops and atomic SMP barriers [ARM] update mach-types

Rusty Russell (1):
lguest: fix on Intel when KVM loaded (unhandled trap 13)

Ryusuke Konishi (1):
nilfs2: fix bh leak in nilfs_cpfile_delete_checkpoints function

Sam Ravnborg (1):
nfs: fix build error in nfsroot with initconst

Shaohua Li (2):
cpuidle: makes AMD C1E work in acpi_idle
cpuidle: fix AMD C1E suspend hang

Steve Wise (1):
svcrdma: dma unmap the correct length for the RPCRDMA header page.

Suresh Siddha (1):
x86: introduce noxsave boot parameter

Takashi Iwai (4):
ALSA: hda - Add 5stack-no-fp model for STAC927x
ALSA: hda - Add missing check of pin vref 50 and others in Realtek codecs
ALSA: Fix invalid jiffies check after pause
ALSA: Enable PCM hw_ptr_jiffies check only in xrun_debug mode

Tejun Heo (2):
x86: Remove remap percpu allocator for the time being x86, relocs: ignore R_386_NONE in kernel relocation entries

Tetsuo Handa (1):
kmod: Release sub_info on cred allocation failure.

Thomas Dahlmann (1):
MAINTAINERS: change email address for Thomas Dahlmann

Thomas Reitmayr (1):
[ARM] Kirkwood: Correct MPP for SATA activity/presence LEDs of QNAP TS-119/TS-219.

Thomas Renninger (1):
[CPUFREQ] powernow-k8 cleanup msg if BIOS does not export ACPI _PSS cpufreq data

Timothy Clacy (1):
[ARM] pxa: enable GPIO receivers after configuring pins

Tony Vroon (1):
ALSA: hda - Compaq Presario CQ60 patching for Conexant

Trond Myklebust (1):
NFSv4: Fix the case where NFSv4 renewal fails

Vladimir Barinov (1):
mtd: MXC NAND driver fixes (v5)

Vu Pham (1):
XPRTRDMA: fix client rpcrdma FRMR registration on mlx4 devices

Warren Free (1):
USB: isp1760: urb_dequeue doesn't always find the urbs

Wei Yongjun (1):
nfsd: fix hung up of nfs client while sync write data to nfs server

Xiao Kaijian (1):
USB: Yet another Conexant Clone to add to cdc-acm.c

Xiaotian Feng (1):
gianfar: fix babbling rx error event bug

Yevgeny Petrilin (1):
mlx4_en: Fix a kernel panic when waking tx queue

Zhang Rui (3):
x86: DMI match for the Sony VGN-Z540N as it needs BIOS reboot
ACPI: video: DMI workaround broken eMachines E510 BIOS enabling display brightness
ACPI: video: DMI workaround broken Acer 5315 BIOS enabling display brightness

[email protected] (2):
x86: bugfix wbinvd() model check instead of family check
x86: cpa_flush_array wbinvd should be done on all CPUs

#1 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 10 lines · Linus Torvalds

This is almost certainly the last -rc, and I debated even doing it. But rather than just do 2.6.30, I decided that I'd be better off doing a last -rc8 and then a real release probably this weekend.

This has mostly driver and arch updates, with perhaps the intel drm/kms and network driver changes standing out, but there's powerpc, blackfin and arm updates too. Admittedly, the two biggest parts of the powerpc update are a revert and a defconfig update.

Quoted 5 lines

A lot of small stuff, fixing a few regressions (and at least one bugzilla entry going back to 2.6.24). The small stuff does matter. Please test.

Linus

Today's first 2.6.30-rc8 build for VIA NetBooks is posted. Details at:
http://forum.netbookuser.com/viewtopic.php?pid=7034#p7034

- - summary - -

This is the kernel.org repository tag: 2.6.30-rc8 codebase ***without*** local patches; Using the same configuration as the record holding (-09149), variable speed kernel.

The -09149 announcement is at:
http://forum.netbookuser.com/viewtopic.php?pid=7002#p7002
(Note: that build had local patches.)

I have also re-posted the record holding (-09143lk), fixed speed build, see above links.

#2 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 17 lines · Michael S. Zick
Quoted 10 lines · Linus Torvalds

This is almost certainly the last -rc, and I debated even doing it. But rather than just do 2.6.30, I decided that I'd be better off doing a last -rc8 and then a real release probably this weekend.

This has mostly driver and arch updates, with perhaps the intel drm/kms and network driver changes standing out, but there's powerpc, blackfin and arm updates too. Admittedly, the two biggest parts of the powerpc update are a revert and a defconfig update.

Quoted 5 lines

A lot of small stuff, fixing a few regressions (and at least one bugzilla entry going back to 2.6.24). The small stuff does matter. Please test.

Linus

Quoted 9 lines

Today's first 2.6.30-rc8 build for VIA NetBooks is posted. Details at:
http://forum.netbookuser.com/viewtopic.php?pid=7034#p7034

- - summary - -

This is the kernel.org repository tag: 2.6.30-rc8 codebase ***without*** local patches; Using the same configuration as the record holding (-09149), variable speed kernel.

Well, that one isn't anything to brag about"

top - 09:45:44 up 1:44, 2 users, load average: 1.46, 1.48, 1.47

Tasks: 116 total,   1 running, 115 sleeping,   0 stopped,   0 zombie
Cpu(s): 30.3%us,  8.3%sy,  0.0%ni, 60.6%id,  0.0%wa,  0.9%hi,  0.0%si,  0.0%st
Mem:    446720k total,   419160k used,    27560k free,    26352k buffers
Swap:  2199324k total,        0k used,  2199324k free,   214908k cached
- - - -
 3208 mszick    20   0  156m 8752 7424 S 19.4  2.0  16:21.69 pulseaudio
 3392 mszick    20   0  311m  44m  23m S 10.2 10.1  11:38.47 vlc
 3937 root      20   0  2448 1188  920 R  3.7  0.3   3:18.88 top
   49 root      15  -5     0    0    0 S  2.8  0.0   2:55.46 hd-audio0
 3009 root      20   0  222m  24m 7372 S  0.9  5.5   1:14.42 Xorg
 3798 root      20   0  8500 3048 2456 S  0.9  0.7   1:08.91 sshd
    1 root      20   0  3084 1888  564 S  0.0  0.4   0:01.58 init
    2 root      15  -5     0    0    0 S  0.0  0.0   0:00.00 kthreadd

Testing the -09155 now (-09154 + local patches).

Mike

Quoted 7 lines

The -09149 announcement is at:
http://forum.netbookuser.com/viewtopic.php?pid=7002#p7002
(Note: that build had local patches.)

I have also re-posted the record holding (-09143lk), fixed speed build, see above links.

mike

#3 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 2 lines · Michael S. Zick

Well, that one isn't anything to brag about"

So what's the problem on that machine? Is the VIA C7-M (or some random device in it) just buggy and it eventually hangs, or what? Has _any_ kernel ever worked on that machine for longer than an hour or two?

Linus

#4 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 9 lines · you
Quoted 2 lines · Michael S. Zick

Well, that one isn't anything to brag about"

So what's the problem on that machine? Is the VIA C7-M (or some random device in it) just buggy and it eventually hangs, or what? Has _any_ kernel ever worked on that machine for longer than an hour or two?

It is not (directly) the C7-M - it might be the chipset CX700.

I have a small pool (5 or so actively reporting) testers -

On the Everex Cloudbook (C7-M/CX700) I have gotten a record of 4h45 minutes of up-time on some, locally patched, builds. The same is reported on other Everex Cloudbooks.

On the Sylvania gBook (also C7-M/CX700) the user reports 10s of minutes as their record up-time.

H.W. got hold of one of the companies demo-boards but I haven't heard back from him yet on his experiences.

On the HP-2133 (C7-M/CN896) the uptime is unknown - - only greater than 12 hours (it has never/ever deadlocked).

I also have a couple of technical questions pending with H.W. for the silicon grower's department. No response yet.

- - - -

But you asked for "testing" - hence this post.

Mike

Quoted 3 lines

Linus

#5 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Hi Michael,

Quoted 6 lines · Michael S. Zick

On the Everex Cloudbook (C7-M/CX700) I have gotten a record of 4h45 minutes of up-time on some, locally patched, builds. The same is reported on other Everex Cloudbooks.

On the Sylvania gBook (also C7-M/CX700) the user reports 10s of minutes as their record up-time.

It seems like I need to obtain one of those. I'll see if VIA can help, but it might help if you have a known-bad unit that I could get for debugging.

Just to re-confirm: Those kind of problems are you seeing with any kernel, or only specifically with later (2.6.30-rcX) kernels? If it's not a regression, then I think there is no impact of the bug on the 2.6.30 release.

Quoted 2 lines

H.W. got hold of one of the companies demo-boards but I haven't heard back from him yet on his experiences.

I don't really have your test case. I've personally been running a CX700 based mainboard (some EPIA board) as my digital video recording device for > one year here, without any lock ups at all. I have not been running any recent kernels on that system, though.

Regards,

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#6 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Michael S. Zick
Quoted 19 lines · Michael S. Zick
Quoted 9 lines · you
Quoted 2 lines · Michael S. Zick

Well, that one isn't anything to brag about"

So what's the problem on that machine? Is the VIA C7-M (or some random device in it) just buggy and it eventually hangs, or what? Has _any_ kernel ever worked on that machine for longer than an hour or two?

It is not (directly) the C7-M - it might be the chipset CX700.

I have a small pool (5 or so actively reporting) testers -

On the Everex Cloudbook (C7-M/CX700) I have gotten a record of 4h45 minutes of up-time on some, locally patched, builds. The same is reported on other Everex Cloudbooks.

Quoted 8 lines

On the Sylvania gBook (also C7-M/CX700) the user reports 10s of minutes as their record up-time.

H.W. got hold of one of the companies demo-boards but I haven't heard back from him yet on his experiences.

On the HP-2133 (C7-M/CN896) the uptime is unknown - - only greater than 12 hours (it has never/ever deadlocked).

IIRC I did way longer tests than 12 hours on hp2133... it worked ok.

The other 'via from hell' machine would last overnight with cca 50% probability.

Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#7 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Linus Torvalds

2009/6/4 Linus Torvalds <[email protected]>:

Quoted 6 lines · Michael S. Zick
Quoted 1 line

Well, that one isn't anything to brag about"

So what's the problem on that machine? Is the VIA C7-M (or some random device in it) just buggy and it eventually hangs, or what? Has _any_ kernel ever worked on that machine for longer than an hour or two?

There have been reports of hangs on various VIA C7 machines going back a year now. The version of the kernel doesn't seem to matter, but the version of glibc does. Unfortunately there hasn't been much progress in getting to the bottom of it.

See here (and other linked reports):
http://bugs.gentoo.org/show_bug.cgi?id=228263

Cheers,
Duane.

Signature · 1 line
"I never could learn to drink that blood and call it wine" - Bob Dylan
#8 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 8 lines · Duane Griffin

There have been reports of hangs on various VIA C7 machines going back a year now. The version of the kernel doesn't seem to matter, but the version of glibc does. Unfortunately there hasn't been much progress in getting to the bottom of it.

See here (and other linked reports):
http://bugs.gentoo.org/show_bug.cgi?id=228263

Hmm. That looks like a CPU problem, but hey, it might be that the glibc version thing is just coincidence, and just changes timings or whatever, and the problem is in the chipsets.

So at least from that particular report it smells very much non-kernel-related.

That said, even if it isn't kernel-related, it might be fixable with some kernel patch that changes the setup of the CPU/chipset. But we'd need VIA to help with anythign like that.

Linus

#9 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 15 lines · Linus Torvalds
Quoted 8 lines · Duane Griffin

There have been reports of hangs on various VIA C7 machines going back a year now. The version of the kernel doesn't seem to matter, but the version of glibc does. Unfortunately there hasn't been much progress in getting to the bottom of it.

See here (and other linked reports):
http://bugs.gentoo.org/show_bug.cgi?id=228263

Hmm. That looks like a CPU problem, but hey, it might be that the glibc version thing is just coincidence, and just changes timings or whatever, and the problem is in the chipsets.

Quoted 7 lines

So at least from that particular report it smells very much non-kernel-related.

That said, even if it isn't kernel-related, it might be fixable with some kernel patch that changes the setup of the CPU/chipset. But we'd need VIA to help with anythign like that.

That is one of my pending questions - -
(It is included as a comment at the appropriate point in my patchset.)

The VIA processors have MCR's not MTRR's - -
The C7-M processor uses "in-order retirement" not "out-of-order" - - I think the MCR's **should not** be set for "weak ordered writes" -

_But_ until I hear the recommendation of the Silicon Growers department...

Mike

Quoted 3 lines

Linus

#10 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 3 lines · Michael S. Zick

The C7-M processor uses "in-order retirement" not "out-of-order" - - I think the MCR's **should not** be set for "weak ordered writes" -

In-order retirement does not really imply anything at all about how write ordering works out. In most CPU parlance, you'd say that you've "retired" a write instruction when it has completed in the write queue - but it would not mean anything in particular for memory ordering.

Of course, I don't think the C7 is just in-order retirement, I think it's pretty much in-order everything. Usually you only specify that "retirement" part when there is some out-of-order execution in other parts of the pipeline, but I think the C7 is entirely in-order pipeline, although I suspect the FPU side is likely somewhat separated.

But still, with a write buffer (and _no_ sane x86 does not have a write buffer), that doesn't actually mean that the cache and memory accesses are necessarily entirely in-order.

That said, how all the internal CPU registers are set is all black magic.

Linus

#11 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Side note: is it more stable if you disable the VIA speedstep thing (whatever it's called (ok, google tells me it's called "TwinTurbo" and "Advanced PowerSaver")?

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

Linus

#12 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 5 lines · Linus Torvalds

Side note: is it more stable if you disable the VIA speedstep thing (whatever it's called (ok, google tells me it's called "TwinTurbo" and "Advanced PowerSaver")?

The e_powersave code.
Yes, I build test cases with and without - -
It was a fixed-speed kernel build that first hit the 4 hour up-time mark. I just reposted that build today (the -09143lk).

Quoted 5 lines

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

There is also a possible thermal issue with these machines - - I doubt that VIA runs their qualification testing in bake ovens; which is what NetBook cases amount too. ;)

Before the week's end - I am going to run a few of these test builds with the machine in the freezer - a 20C decrease in ambient should show if there is any thermal relationship with the problem.

I just need to run my lan into the freezer compartment - one of the places in this house without a lan connection. ;)

For want of a documented JTAG port...

Mike

Quoted 9 lines

Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#13 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 5 lines · Michael S. Zick

Yes, I build test cases with and without - -
It was a fixed-speed kernel build that first hit the 4 hour up-time mark. I just reposted that build today (the -09143lk).

Quoted 8 lines
Quoted 4 lines

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

There is also a possible thermal issue with these machines - - I doubt that VIA runs their qualification testing in bake ovens; which is what NetBook cases amount too. ;)

If the fixed-speed case runs for longer, it's not likely to be a thermal issue. The fixed speed case should be the higher-power one.

So it can easily be a weak power setup (insufficient grounding, bad capacitors etc). But it could also be external bus issues, in case VIA power management also impact the external bus (eg "stopclock" like behavior on the CPU<->chipset bus).

One thing you could try is to avoid using the "halt" instruction. It will obviously increase power use (and thus higher temperatures), but again, current fluctuations are much more likely to cause problems than higher, but fairly constant, power draw.

Think about all the light-bulbs you've seen that burn out just when you turn them on.

Use "idle=poll" on the kernel command line to avoid the idle loop using the "halt" or "mwait" instructions to save power.

(That polling idle loop can also end up hiding cache coherency issues with DMA, so if that works better, it doesn't necessarily prove it's power-related. Shutting down the CPU core can have interesting implications for external events, and you can have various races - maybe you shut down the core just as a chipset event happened, and the chipset _thinks_ the core is now awake, but the core went to sleep. End result:
hung machine).

Linus

#14 lore raw

VIA PowerSaver (Re: Linux 2.6.30-rc8 [also: VIA Support])

Hi Linus and Michael,

Quoted 15 lines · Linus Torvalds
Quoted 5 lines · Michael S. Zick

Yes, I build test cases with and without - -
It was a fixed-speed kernel build that first hit the 4 hour up-time mark. I just reposted that build today (the -09143lk).

Quoted 8 lines
Quoted 4 lines

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

There is also a possible thermal issue with these machines - - I doubt that VIA runs their qualification testing in bake ovens; which is what NetBook cases amount too. ;)

Quoted 7 lines

If the fixed-speed case runs for longer, it's not likely to be a thermal issue. The fixed speed case should be the higher-power one.

So it can easily be a weak power setup (insufficient grounding, bad capacitors etc). But it could also be external bus issues, in case VIA power management also impact the external bus (eg "stopclock" like behavior on the CPU<->chipset bus).

I'm not intending to disagree with you, I just wanted to quote from a not [yet] public document on the C7-M. This quote describes model A (family 6, model 10(hex A), stepping 0-15):
===============
Enhanced PowerSaver technology allows the dynamic adjustment of the operating frequency and operating voltage. The VIA C7-M can only change from the highest supported performance state to the lowest supported performance state:
intermediate performance states are not guaranteed to work and are not offi- cially supported. System software can use Enhanced PowerSaver to request the sufficient amount of performance. Each individual performance state (P-State) is described in the system bios according to 8.4.4 of the ACPI 3.0 specification.

The VIA C7-M processor incorporates two on-chip core clock PLLs. This allows the processor to ping-pong between two frequencies instantaneously. In the simplest scenario, where there are only two clock frequencies of interest and no voltage changes, the transition can be instantaneous with no latency. In more complex scenarios, where there are multiple clock frequencies of interest, the "old" frequency can continue to be used while the new frequency is ramped up. The transition is still instantaneous from a software point of view (code still executes), but there is a latency associated with switching to the ramp- ing "new" frequency.

VIA C7-M allows for a clean hardware approach to processor operating point transitions. The transitions are performed instantaneously from a software and functional point of view. Snoops and interrupts, for example, are unaffected by transitions.
===============

A C7-M model D (family 6, model 13(hex D), stepping 0-15) has advanced performance states, they use an inflection ratio, as well as adaptive-p-state control and adaptive overclocking, as well as iteravie P-state transitions and adaptive thermal control. I'm not yet aware of all the details, but have requested them.

In any case, the problems that have been reported by Michael were "Model A", so those particular deatils shouldn't matter at this point.

Regards,

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#15 lore raw

Re: VIA PowerSaver (Re: Linux 2.6.30-rc8 [also: VIA Support])

Quoted 40 lines · Harald Welte

Hi Linus and Michael,

Quoted 15 lines · Linus Torvalds
Quoted 5 lines · Michael S. Zick

Yes, I build test cases with and without - -
It was a fixed-speed kernel build that first hit the 4 hour up-time mark. I just reposted that build today (the -09143lk).

Quoted 8 lines
Quoted 4 lines

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

There is also a possible thermal issue with these machines - - I doubt that VIA runs their qualification testing in bake ovens; which is what NetBook cases amount too. ;)

Quoted 7 lines

If the fixed-speed case runs for longer, it's not likely to be a thermal issue. The fixed speed case should be the higher-power one.

So it can easily be a weak power setup (insufficient grounding, bad capacitors etc). But it could also be external bus issues, in case VIA power management also impact the external bus (eg "stopclock" like behavior on the CPU<->chipset bus).

I'm not intending to disagree with you, I just wanted to quote from a not [yet] public document on the C7-M. This quote describes model A (family 6, model 10(hex A), stepping 0-15):
===============
Enhanced PowerSaver technology allows the dynamic adjustment of the operating frequency and operating voltage. The VIA C7-M can only change from the highest supported performance state to the lowest supported performance state:
intermediate performance states are not guaranteed to work and are not offi- cially supported. System software can use Enhanced PowerSaver to request the sufficient amount of performance. Each individual performance state (P-State) is described in the system bios according to 8.4.4 of the ACPI 3.0 specification.

Quoted 10 lines

The VIA C7-M processor incorporates two on-chip core clock PLLs. This allows the processor to ping-pong between two frequencies instantaneously. In the simplest scenario, where there are only two clock frequencies of interest and no voltage changes, the transition can be instantaneous with no latency. In more complex scenarios, where there are multiple clock frequencies of interest, the "old" frequency can continue to be used while the new frequency is ramped up. The transition is still instantaneous from a software point of view (code still executes), but there is a latency associated with switching to the ramp- ing "new" frequency.

This appears to be what the e_powersaver is trying to do - - It just needs to do it better. ;)

The current behavior ends up as 9 speeds, (I.E: 8 steps) of twice the FSB frequency. I do show "stats/time_in_state" for all of them.

It is one heck of a system, It will be nice if I can make it work - - If it has to be turned into a two-speed system, so be it.

= = = =

Note: This is one of VIA's claims to fame - and a good one - - and different than what the cpufreq stuff was probably tested with.

Most other brands of CPU will "stall" (or some such) internally while re-syncing the core clock chain (there-by "stalling" the code progress) - The VIA processors do not - they keep on computing - something the general code may not account for. ;)

Since I have a machine that is sensitive to this, I can test other ways of doing something other than the two speed solution.

Quoted 7 lines

VIA C7-M allows for a clean hardware approach to processor operating point transitions. The transitions are performed instantaneously from a software and functional point of view. Snoops and interrupts, for example, are unaffected by transitions.
===============

Quoted 8 lines

A C7-M model D (family 6, model 13(hex D), stepping 0-15) has advanced performance states, they use an inflection ratio, as well as adaptive-p-state control and adaptive overclocking, as well as iteravie P-state transitions and adaptive thermal control. I'm not yet aware of all the details, but have requested them.

In any case, the problems that have been reported by Michael were "Model A", so those particular deatils shouldn't matter at this point.

dmesg is reporting it as an "Model D" - I will check if that reporting is correct. Tell the Silicon Grower's department "Thanks" for the recommendation of what to look for.

Mike

Quoted 1 line

Regards,

#16 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Linus Torvalds
Quoted 12 lines · Linus Torvalds
Quoted 10 lines · Michael S. Zick

Yes, I build test cases with and without - -
It was a fixed-speed kernel build that first hit the 4 hour up-time mark. I just reposted that build today (the -09143lk).

Quoted 4 lines

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

Quoted 7 lines
Quoted 3 lines

There is also a possible thermal issue with these machines - - I doubt that VIA runs their qualification testing in bake ovens; which is what NetBook cases amount too. ;)

If the fixed-speed case runs for longer, it's not likely to be a thermal issue. The fixed speed case should be the higher-power one.

I can respond to that point now; VIA Tech has answered some of my questions -

The mainstream kernel, e_powersaver, is *under-clocking* my machine -

The cpuid instruction provides the minimum and maximum GSF values (Guaranteed Stable Frequency) for that processor mask run - Passing that on as the lower and upper limits to e_powersaver should stop that problem. Will be testing this RSN.

Once you start operating the processor outside of the reported-by-silicon mask limits - quote: "there is no quarantee of stable operation" - -

To that I add my opinion:
For a severely under-clocked machine (already possibly un-stable) - - Thermal effects are almost certain to be present.

To paraphrase VIA Tech once again: "Don't do that." ;)

Mike

#17 lore raw

e_powersaver / underclocking (was Re: Linux 2.6.30-rc8 [also: VIA Support])

Quoted 8 lines · Michael S. Zick

I can respond to that point now; VIA Tech has answered some of my questions -

The mainstream kernel, e_powersaver, is *under-clocking* my machine -

The cpuid instruction provides the minimum and maximum GSF values (Guaranteed Stable Frequency) for that processor mask run - Passing that on as the lower and upper limits to e_powersaver should stop that problem. Will be testing this RSN.

It's really surprising to me that none of this seems to be handled correct so far, I'll talk to Centaur and try to find out how we could have ended up in this situation.

My assumption is that e_powersavre is no longer supposd to do any of those low-level bits - rather the ACPI code is expected to get it right, hiding the details from the OS. But in this case, there needs to be some run-time detection whether the ACPI cpufreq should be used, or e_powersaver. And I don't see any of that right now.

Also note that now with OLPC XO1.5 going for the C7-M (on a VX855 chipset, though), many of those issues should soon receive much more attention - especially on the power management front. And as you know, they don't use any legacy BIOS...

Regards,

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#18 lore raw

Re: e_powersaver / underclocking (was Re: Linux 2.6.30-rc8 [also: VIA Support])

Quoted 6 lines · Harald Welte
Quoted 4 lines · Michael S. Zick

I can respond to that point now; VIA Tech has answered some of my questions -

The mainstream kernel, e_powersaver, is *under-clocking* my machine -

Quoted 9 lines
Quoted 4 lines

The cpuid instruction provides the minimum and maximum GSF values (Guaranteed Stable Frequency) for that processor mask run - Passing that on as the lower and upper limits to e_powersaver should stop that problem. Will be testing this RSN.

It's really surprising to me that none of this seems to be handled correct so far, I'll talk to Centaur and try to find out how we could have ended up in this situation.

Ah, but we are talking here of the *second* NetBook ever produced. If one is to believe the dmidecode output - it is using the VIA demo board BIOS.

I bet the demo board BIOS is intended to demo the features of the product - not the correctness or completeness of the ACPI support. ;)

If I where shipping demo boards - they would be demonstrating **my** product's features. Maybe I am just projecting what I would do.

Quoted 6 lines

My assumption is that e_powersavre is no longer supposd to do any of those low-level bits - rather the ACPI code is expected to get it right, hiding the details from the OS. But in this case, there needs to be some run-time detection whether the ACPI cpufreq should be used, or e_powersaver. And I don't see any of that right now.

I can keep my eyes open for a way to do that -
First, I want to get the machine running **with-in** the specs it can provide. The one I have is running at 2/3rds of the reported *minimum* clockspeed. I must have gotten a high quality "mask/process run" for it to be running at all.

Quoted 5 lines

Also note that now with OLPC XO1.5 going for the C7-M (on a VX855 chipset, though), many of those issues should soon receive much more attention - especially on the power management front. And as you know, they don't use any legacy BIOS...

I'll keep my eyes open on that subject also when looking at the e_powersaver code - The OLPC project will probably be requesting chip runs that **do** run at the minimums the design is capable of and it will **have to** be stable for OLPC.

Mike

Quoted 1 line

Regards,

#19 lore raw

Re: e_powersaver / underclocking (was Re: Linux 2.6.30-rc8 [also: VIA Support])

Quoted 35 lines · Michael S. Zick
Quoted 6 lines · Harald Welte
Quoted 4 lines · Michael S. Zick

I can respond to that point now; VIA Tech has answered some of my questions -

The mainstream kernel, e_powersaver, is *under-clocking* my machine -

Quoted 9 lines
Quoted 4 lines

The cpuid instruction provides the minimum and maximum GSF values (Guaranteed Stable Frequency) for that processor mask run - Passing that on as the lower and upper limits to e_powersaver should stop that problem. Will be testing this RSN.

It's really surprising to me that none of this seems to be handled correct so far, I'll talk to Centaur and try to find out how we could have ended up in this situation.

Ah, but we are talking here of the *second* NetBook ever produced. If one is to believe the dmidecode output - it is using the VIA demo board BIOS.

I bet the demo board BIOS is intended to demo the features of the product - not the correctness or completeness of the ACPI support. ;)

If I where shipping demo boards - they would be demonstrating **my** product's features. Maybe I am just projecting what I would do.

Quoted 6 lines

My assumption is that e_powersavre is no longer supposd to do any of those low-level bits - rather the ACPI code is expected to get it right, hiding the details from the OS. But in this case, there needs to be some run-time detection whether the ACPI cpufreq should be used, or e_powersaver. And I don't see any of that right now.

Quoted 5 lines

I can keep my eyes open for a way to do that -
First, I want to get the machine running **with-in** the specs it can provide. The one I have is running at 2/3rds of the reported *minimum* clockspeed. I must have gotten a high quality "mask/process run" for it to be running at all.

If any readers have noticed, the links I post go to what was originally named "Cloudbookuser" - the first forum for the second machine - -

Ever since day 3 or 4 - people have been writing scripts to change the on-demand governor's minimum from 400Mhz to 600Mhz - without knowing "why" other than it was required to keep the machine running (even on distribution kernels).

_Now_ I know why. Tell the VIA Tech folks "thanks" for the information.

Mike

Quoted 21 lines
Quoted 5 lines

Also note that now with OLPC XO1.5 going for the C7-M (on a VX855 chipset, though), many of those issues should soon receive much more attention - especially on the power management front. And as you know, they don't use any legacy BIOS...

I'll keep my eyes open on that subject also when looking at the e_powersaver code - The OLPC project will probably be requesting chip runs that **do** run at the minimums the design is capable of and it will **have to** be stable for OLPC.

Mike

Quoted 1 line

Regards,

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#20 lore raw

e_powersaver driver considered DANGEROUS (was Re: Linux 2.6.30-rc8 [also: VIA Support])

Dear Michael and others,

Quoted 12 lines · Michael S. Zick
Quoted 12 lines
Quoted 11 lines
Quoted 6 lines

The mainstream kernel, e_powersaver, is *under-clocking* my machine -

The cpuid instruction provides the minimum and maximum GSF values (Guaranteed Stable Frequency) for that processor mask run - Passing that on as the lower and upper limits to e_powersaver should stop that problem. Will be testing this RSN.

It's really surprising to me that none of this seems to be handled correct so far, I'll talk to Centaur and try to find out how we could have ended up in this situation.

Quoted 6 lines
Quoted 6 lines

Ah, but we are talking here of the *second* NetBook ever produced. If one is to believe the dmidecode output - it is using the VIA demo board BIOS.

I bet the demo board BIOS is intended to demo the features of the product - not the correctness or completeness of the ACPI support. ;)

So I've just been told the VIA reference BIOS has full support for the processor p-state support. I'd therefore suppose every production BIOS contains that code, too. The kernel should never use a native driver such as e_powersaver on any C7 or Nano system, but use the ACPI provided tables/methods, which are intel compatible. A native driver would only be needed on really old C3 systems.

The e_powersaver.c driver neither respects the maximum/minimum frequency constraints specified in the MSR's, nor does it take care of the inflection ratio, parallax and other advanced stuff that C7 and Nano are doing in this area.

I would consider the e_powersaver driver DANGEROUS on any C7 or Nano system, i.e. on every system it supports. It is bound to cause system instability, as it will most likely operate the CPU out of spec on virtually every system.

I'll do some actual testing on getting the intel-compatible ACPI method for p-states working on my c7 and nano demo boards, and submit any patches (if required).

I'll also meanwhile submit a patch to mark e_powersaver as DANGEROUS and EXPERIMENTAL.

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#21 lore raw

Re: e_powersaver driver considered DANGEROUS (was Re: Linux 2.6.30-rc8 [also: VIA Support])

Hi!

Quoted 14 lines
Quoted 6 lines
Quoted 6 lines

Ah, but we are talking here of the *second* NetBook ever produced. If one is to believe the dmidecode output - it is using the VIA demo board BIOS.

I bet the demo board BIOS is intended to demo the features of the product - not the correctness or completeness of the ACPI support. ;)

So I've just been told the VIA reference BIOS has full support for the processor p-state support. I'd therefore suppose every production BIOS contains that code, too. The kernel should never use a native driver such as e_powersaver on any C7 or Nano system, but use the ACPI provided tables/methods, which are intel compatible. A native driver would only be needed on really old C3 systems.

Quoted 4 lines

The e_powersaver.c driver neither respects the maximum/minimum frequency constraints specified in the MSR's, nor does it take care of the inflection ratio, parallax and other advanced stuff that C7 and Nano are doing in this area.

Uhuh, what is inflection ratio/parallax?

I did play with cpufreq on intel/amd systems, and never heard about those...

Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#22 lore raw

PATCH 1/2 CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

In reply to Michael S. Zick

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +2 -1
index 208ecf6..ee03585 100644
@@ -90,7 +90,8 @@ static int check_est_cpu(unsigned int cpuid)
 {
 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 
-	if (cpu->x86_vendor != X86_VENDOR_INTEL ||
+	if ((cpu->x86_vendor != X86_VENDOR_INTEL &&
+	     cpu->x86_vendor != X86_VENDOR_CENTAUR) ||
 	    !cpu_has(cpu, X86_FEATURE_EST))
 		return 0;
 
Signature · 1 line
1.6.2.4
#23 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 3 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

Hm. Is the MSR interface compatible with the Intel one? I'm guessing so based on the est flag being present. Maybe we should just drop the intel check and only look for CPUs which claim est support.

Signature · 1 line
Matthew Garrett | [email protected]
#24 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 6 lines · Matthew Garrett
Quoted 3 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

Hm. Is the MSR interface compatible with the Intel one?

Yes, it is intended to be. If you know of any public intel spec on the actual MSR values, I'd be happy to have a look and compare it with Centaur's specs.

VIA/Centaur specs are being prepared for public release, but it will probably take longer than we can responsibly delay fixing this bug.

Quoted 2 lines

I'm guessing so based on the est flag being present. Maybe we should just drop the intel check and only look for CPUs which claim est support.

I think that might be the right decision.

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#25 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

In reply to Harald Welte
Quoted 4 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

It looks like we should modify (conditional on ...MVIAC7 at build, model='d' runtime) the acpi-cpufreq controls to deal properly with the Model-D adaptive controller.

What probably needs to be done is test if it has been set-up and turn it off before handing control over to the ACPI - -
The information I have from VIA Tech. says not to use external controls while the adaptive controller is enabled.

Perhaps the same sort of thing in the 'resume' path - some BIOS may be enabling the adaptive controller.

Mike

Quoted 19 lines
---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +2 -1
index 208ecf6..ee03585 100644
@@ -90,7 +90,8 @@ static int check_est_cpu(unsigned int cpuid)
 {
 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 
-	if (cpu->x86_vendor != X86_VENDOR_INTEL ||
+	if ((cpu->x86_vendor != X86_VENDOR_INTEL &&
+	     cpu->x86_vendor != X86_VENDOR_CENTAUR) ||
 	    !cpu_has(cpu, X86_FEATURE_EST))
 		return 0;
 
#26 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 8 lines · Michael S. Zick
Quoted 4 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

It looks like we should modify (conditional on ...MVIAC7 at build, model='d' runtime) the acpi-cpufreq controls to deal properly with the Model-D adaptive controller.

Can't make it build-time dependent - distribution kernels may not explicitly support the C7. It's valid to have a vendor=centaur conditional that turns off any adaptive control if appropriate ACPI methods are present.

Signature · 1 line
Matthew Garrett | [email protected]
#27 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 7 lines · Matthew Garrett
Quoted 6 lines · Michael S. Zick
Quoted 4 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

Quoted 8 lines
Quoted 2 lines

It looks like we should modify (conditional on ...MVIAC7 at build, model='d' runtime) the acpi-cpufreq controls to deal properly with the Model-D adaptive controller.

Can't make it build-time dependent - distribution kernels may not explicitly support the C7. It's valid to have a vendor=centaur conditional that turns off any adaptive control if appropriate ACPI methods are present.

A valid point.

I haven't looked yet, but I think we have advanced to the point where the 'VIA hack' for cache-line size can also go away.
Now that the pci-cache-line-size setting is being done differently. (currently a proposed change).

The C7(xxxx) is 99 44/100% a Pentium-M with minor differences. (Like the power/thermal/freq adaptive controller on the "D" models.)

Mike

#28 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

In reply to Harald Welte
Quoted 15 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +0 -0
index 208ecf6..ee03585 100644

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c static int check_est_cpu(unsigned int cpuid) · line 95

Quoted 7 lines
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +2 -1
@@ -90,7 +90,8 @@ static int check_est_cpu(unsigned int cpuid)
 {
 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 
-	if (cpu->x86_vendor != X86_VENDOR_INTEL ||
+	if ((cpu->x86_vendor != X86_VENDOR_INTEL &&
+	     cpu->x86_vendor != X86_VENDOR_CENTAUR) ||
 	    !cpu_has(cpu, X86_FEATURE_EST))

Hmm. This all really should be just

	static int check_est_cpu(unsigned int cpuid)
	{
		struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
		return cpu_has(cpu, X86_FEATURE_EST);
	}

I suspect, with no vendor tests. That's the whole _point_ of CPU features, after all.

If some vendor claims EST but doesn't actually support the EST interfaces, we should just have fixups to clear the bit in the per-vendor cpuinfo code, not in some random driver.

The only thing that makes me nervous about this is how close to 2.6.30 we are. I'd be happier if this was resolved by doing this as a patch post-2.6.30, and then adding '[email protected]' as a Cc: tag, and backporting it to 2.6.30.1 if no problems appear.

It's not like this is a regression, I think.

Does that sound like a reasonable plan?

Linus

#29 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 40 lines · Linus Torvalds
Quoted 15 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +0 -0
index 208ecf6..ee03585 100644

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c static int check_est_cpu(unsigned int cpuid) · line 95

Quoted 7 lines
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +2 -1
@@ -90,7 +90,8 @@ static int check_est_cpu(unsigned int cpuid)
 {
 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 
-	if (cpu->x86_vendor != X86_VENDOR_INTEL ||
+	if ((cpu->x86_vendor != X86_VENDOR_INTEL &&
+	     cpu->x86_vendor != X86_VENDOR_CENTAUR) ||
 	    !cpu_has(cpu, X86_FEATURE_EST))

Hmm. This all really should be just

	static int check_est_cpu(unsigned int cpuid)
	{
		struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
		return cpu_has(cpu, X86_FEATURE_EST);
	}

I suspect, with no vendor tests. That's the whole _point_ of CPU features, after all.

If some vendor claims EST but doesn't actually support the EST interfaces, we should just have fixups to clear the bit in the per-vendor cpuinfo code, not in some random driver.

Quoted 9 lines

The only thing that makes me nervous about this is how close to 2.6.30 we are. I'd be happier if this was resolved by doing this as a patch post-2.6.30, and then adding '[email protected]' as a Cc: tag, and backporting it to 2.6.30.1 if no problems appear.

It's not like this is a regression, I think.

Does that sound like a reasonable plan?

Sounds like a plan to me - it has been broke a long time - a little longer... I can continue to carry any change we code up as a local patch until it gets done at the mainline level.
There are only about 20,000 of these old Everex machines ever produced.

Mike

Quoted 3 lines

Linus

#30 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

In reply to Linus Torvalds
Quoted 33 lines · Linus Torvalds
Quoted 15 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +0 -0
index 208ecf6..ee03585 100644

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c static int check_est_cpu(unsigned int cpuid) · line 95

Quoted 7 lines
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +2 -1
@@ -90,7 +90,8 @@ static int check_est_cpu(unsigned int cpuid)
 {
 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 
-	if (cpu->x86_vendor != X86_VENDOR_INTEL ||
+	if ((cpu->x86_vendor != X86_VENDOR_INTEL &&
+	     cpu->x86_vendor != X86_VENDOR_CENTAUR) ||
 	    !cpu_has(cpu, X86_FEATURE_EST))

Hmm. This all really should be just

	static int check_est_cpu(unsigned int cpuid)
	{
		struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
		return cpu_has(cpu, X86_FEATURE_EST);
	}
Quoted 7 lines

I suspect, with no vendor tests. That's the whole _point_ of CPU features, after all.

If some vendor claims EST but doesn't actually support the EST interfaces, we should just have fixups to clear the bit in the per-vendor cpuinfo code, not in some random driver.

Sounds like a plan to me - -

Shall we define a kernel 'cpu-feature' for the internal, adaptive thermal/power/freq controller?

There are at least two cpufreq drivers that need to be able to check for the feature
_and_
perhaps the clock/timing routines would need to know if it was there/enabled -

Since it is internal, on-chip, it will not be sending any notifications as it adapts the core clock speed. (so much for 'loops/xSec').

Knowing that it is present and enabled would let the timing routines establish non-loop counting measurements.

Mike

Quoted 17 lines

The only thing that makes me nervous about this is how close to 2.6.30 we are. I'd be happier if this was resolved by doing this as a patch post-2.6.30, and then adding '[email protected]' as a Cc: tag, and backporting it to 2.6.30.1 if no problems appear.

It's not like this is a regression, I think.

Does that sound like a reasonable plan?

Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#31 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

In reply to Linus Torvalds
Quoted 10 lines · Linus Torvalds

Hmm. This all really should be just

	static int check_est_cpu(unsigned int cpuid)
	{
		struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
		return cpu_has(cpu, X86_FEATURE_EST);
	}

I suspect, with no vendor tests. That's the whole _point_ of CPU features, after all.

That's what I was thinking, too. If there was no such vendor test, it would have worked ever since the code was written (the C7 is by far not a new component, it's around for years).

Quoted 3 lines

If some vendor claims EST but doesn't actually support the EST interfaces, we should just have fixups to clear the bit in the per-vendor cpuinfo code, not in some random driver.

agreed.

Quoted 8 lines

The only thing that makes me nervous about this is how close to 2.6.30 we are. I'd be happier if this was resolved by doing this as a patch post-2.6.30, and then adding '[email protected]' as a Cc: tag, and backporting it to 2.6.30.1 if no problems appear.

It's not like this is a regression, I think.

Does that sound like a reasonable plan?

Sounds fine with me. But what I would definitely suggest merging before 2.6.30 is the marking e_powersaver EXPERIMENTAL + DANGEROUS patch.

Regards,

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#32 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 27 lines · Harald Welte
Quoted 10 lines · Linus Torvalds

Hmm. This all really should be just

	static int check_est_cpu(unsigned int cpuid)
	{
		struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
		return cpu_has(cpu, X86_FEATURE_EST);
	}

I suspect, with no vendor tests. That's the whole _point_ of CPU features, after all.

That's what I was thinking, too. If there was no such vendor test, it would have worked ever since the code was written (the C7 is by far not a new component, it's around for years).

Quoted 3 lines

If some vendor claims EST but doesn't actually support the EST interfaces, we should just have fixups to clear the bit in the per-vendor cpuinfo code, not in some random driver.

agreed.

Quoted 5 lines

The only thing that makes me nervous about this is how close to 2.6.30 we are. I'd be happier if this was resolved by doing this as a patch post-2.6.30, and then adding '[email protected]' as a Cc: tag, and backporting it to 2.6.30.1 if no problems appear.

Quoted 7 lines
Quoted 3 lines

It's not like this is a regression, I think.

Does that sound like a reasonable plan?

Sounds fine with me. But what I would definitely suggest merging before 2.6.30 is the marking e_powersaver EXPERIMENTAL + DANGEROUS patch.

As posted somewhere in this thread,
the acpi-cpufreq controller appears to work on my machine in initial tests.

A few tests on only one machine is not much to go on and it isn't enough to call it "tested" but is forward progress.

I was only risking my "throw-away" machine yesterday, will see what happens on the "good" one today (the HP-2133).

@H.W. - are you running any of these changes on your HP-2133?

Mike

Quoted 1 line

Regards,

#33 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 6 lines · On Tue, 9 Jun 2009 07:26:11 -0500 "Michael S. Zick" <[email protected]> wrote:
Quoted 3 lines

Sounds fine with me. But what I would definitely suggest merging before 2.6.30 is the marking e_powersaver EXPERIMENTAL + DANGEROUS patch.

As posted somewhere in this thread,
the acpi-cpufreq controller appears to work on my machine in initial tests.

It works great on the Samsung NC20 with an x86_64 kernel.

#34 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 12 lines · Chuck Ebbert
Quoted 6 lines · On Tue, 9 Jun 2009 07:26:11 -0500 "Michael S. Zick" <[email protected]> wrote:
Quoted 3 lines

Sounds fine with me. But what I would definitely suggest merging before 2.6.30 is the marking e_powersaver EXPERIMENTAL + DANGEROUS patch.

As posted somewhere in this thread,
the acpi-cpufreq controller appears to work on my machine in initial tests.

It works great on the Samsung NC20 with an x86_64 kernel.

For non-NetBook aware readers, that means the:
VIA Nano U2250 processor with the VX800 system chipset.

Mike

Quoted 7 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#35 lore raw

TSC features, was: Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

In reply to Linus Torvalds
Quoted 27 lines · Linus Torvalds
Quoted 15 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +0 -0
index 208ecf6..ee03585 100644

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c static int check_est_cpu(unsigned int cpuid) · line 95

Quoted 7 lines
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +2 -1
@@ -90,7 +90,8 @@ static int check_est_cpu(unsigned int cpuid)
 {
 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 
-	if (cpu->x86_vendor != X86_VENDOR_INTEL ||
+	if ((cpu->x86_vendor != X86_VENDOR_INTEL &&
+	     cpu->x86_vendor != X86_VENDOR_CENTAUR) ||
 	    !cpu_has(cpu, X86_FEATURE_EST))

Hmm. This all really should be just

Quoted 9 lines
	static int check_est_cpu(unsigned int cpuid)
	{
		struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
		return cpu_has(cpu, X86_FEATURE_EST);
	}

I suspect, with no vendor tests. That's the whole _point_ of CPU features, after all.

Following that same logic, shouldn't this chunk be based on CPU features also? (Spotted while tracking down why the non-stop TSC wasn't being used on VIA):

#if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86) static void tsc_check_state(int state)

{
        switch (boot_cpu_data.x86_vendor) {
        case X86_VENDOR_AMD:
        case X86_VENDOR_INTEL:
                /*
                 * AMD Fam10h TSC will tick in all
                 * C/P/S0/S1 states when this bit is set.
                 */
                if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
                        return;

                /*FALL THROUGH*/
        default:
                /* TSC could halt in idle, so notify users */
                if (state > ACPI_STATE_C1)
                        mark_tsc_unstable("TSC halts in idle");
        }
}

#else
static void tsc_check_state(int state) { return; } #endif

Mike

#36 lore raw

[PATCH, RFC] Re: TSC features, ...

Quoted 10 lines · Michael S. Zick
Quoted 9 lines · Linus Torvalds

Hmm. This all really should be just

	static int check_est_cpu(unsigned int cpuid)
	{
		struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
		return cpu_has(cpu, X86_FEATURE_EST);
	}
Quoted 4 lines
Quoted 3 lines

I suspect, with no vendor tests. That's the whole _point_ of CPU features, after all.

Like this;
Let the vendor's cpu setup code be responsible for getting the feature flag correct:

drivers/acpi/processor_idle.c +5 -15
index 10a2d91..daea6a9 100644
@@ -244,22 +244,12 @@ int acpi_processor_resume(struct acpi_device * device)
 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 static void tsc_check_state(int state)
 {
-       switch (boot_cpu_data.x86_vendor) {
-       case X86_VENDOR_AMD:
-       case X86_VENDOR_INTEL:
-               /*
-                * AMD Fam10h TSC will tick in all
-                * C/P/S0/S1 states when this bit is set.
-                */
-               if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
-                       return;
+       if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
+               return;
 
-               /*FALL THROUGH*/
-       default:
-               /* TSC could halt in idle, so notify users */
-               if (state > ACPI_STATE_C1)
-                       mark_tsc_unstable("TSC halts in idle");
-       }
+       /* TSC could halt in idle, so notify users */
+       if (state > ACPI_STATE_C1)
+               mark_tsc_unstable("TSC halts in idle");
 }
 #else
 static void tsc_check_state(int state) { return; }

Mike

#37 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

In reply to Harald Welte
Quoted 14 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +0 -0
index 208ecf6..ee03585 100644

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c static int check_est_cpu(unsigned int cpuid) · line 97

Quoted 9 lines
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +2 -1
@@ -90,7 +90,8 @@ static int check_est_cpu(unsigned int cpuid)
 {
 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 
-	if (cpu->x86_vendor != X86_VENDOR_INTEL ||
+	if ((cpu->x86_vendor != X86_VENDOR_INTEL &&
+	     cpu->x86_vendor != X86_VENDOR_CENTAUR) ||
 	    !cpu_has(cpu, X86_FEATURE_EST))
 		return 0;
 

For my own internal testing, based on subsequent posts, I simplified that to:

static int check_est_cpu(unsigned int cpuid)

{
        struct cpuinfo_x86 *cpu = &cpu_data(cpuid);

        if (!cpu_has(cpu, X86_FEATURE_EST))
                return 0;

        return 1;
}

Have not done anything (yet) to deal with the adaptive controller, **which is safe to (not) do on my test system because I know it isn't enabled**

Note: This machine is known to run common, proprietary operating systems, so I am also assuming its BIOS ACPI is complete enough.

Note 2: Don't try this unless you have a machine to waste. ;)

Mike

#38 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 38 lines · Michael S. Zick
Quoted 14 lines · Harald Welte

The VIA/Centaur C7, C7-M and Nano CPU's all support ACPI based cpu p-states using a MSR interface. The Linux driver just never made use of it, since in addition to the check for the EST flag it also checked if the vendor is Intel.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +0 -0
index 208ecf6..ee03585 100644

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c static int check_est_cpu(unsigned int cpuid) · line 97

Quoted 9 lines
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +2 -1
@@ -90,7 +90,8 @@ static int check_est_cpu(unsigned int cpuid)
 {
 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 
-	if (cpu->x86_vendor != X86_VENDOR_INTEL ||
+	if ((cpu->x86_vendor != X86_VENDOR_INTEL &&
+	     cpu->x86_vendor != X86_VENDOR_CENTAUR) ||
 	    !cpu_has(cpu, X86_FEATURE_EST))
 		return 0;
 

For my own internal testing, based on subsequent posts, I simplified that to:

static int check_est_cpu(unsigned int cpuid)

{
        struct cpuinfo_x86 *cpu = &cpu_data(cpuid);

        if (!cpu_has(cpu, X86_FEATURE_EST))
                return 0;

        return 1;
}
Quoted 8 lines

Have not done anything (yet) to deal with the adaptive controller, **which is safe to (not) do on my test system because I know it isn't enabled**

Note: This machine is known to run common, proprietary operating systems, so I am also assuming its BIOS ACPI is complete enough.

Note 2: Don't try this unless you have a machine to waste. ;)

Phooey, close but no cigar - - -

root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_available_frequencies
1200000 1000000 800000 600000 400000
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat stats/time_in_state
1200000 3281
1000000 120
800000 130
600000 234
400000 52980
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq

It loads, it appears to work, but that lowest 400Mhz has ACPI entries even though this is a 600Mhz..1200Mhz chip.

Should be fixable -
The code looks like it is "asking" the BIOS for the table ranges - It needs to sanity check against the limits etched in the silicon for this particular [mask/process run].

I am thinking, maybe put the min/max GSF into cpu_boot_data. Will probably need it there to fix e_powersaver anyway.

Mike

Quoted 8 lines

Mike
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#39 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 12 lines · Michael S. Zick

Phooey, close but no cigar - - -

root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_available_frequencies
1200000 1000000 800000 600000 400000
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat stats/time_in_state
1200000 3281
1000000 120
800000 130
600000 234
400000 52980
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq
Quoted 2 lines

It loads, it appears to work, but that lowest 400Mhz has ACPI entries even though this is a 600Mhz..1200Mhz chip.

If the BIOS tables claim it then it's likely Windows uses it on the same hardware. What's the stability like with the ACPI code?

Signature · 1 line
Matthew Garrett | [email protected]
#40 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 14 lines · Matthew Garrett
Quoted 12 lines · Michael S. Zick

Phooey, close but no cigar - - -

root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_available_frequencies
1200000 1000000 800000 600000 400000
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat stats/time_in_state
1200000 3281
1000000 120
800000 130
600000 234
400000 52980
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq
Quoted 6 lines
Quoted 2 lines

It loads, it appears to work, but that lowest 400Mhz has ACPI entries even though this is a 600Mhz..1200Mhz chip.

If the BIOS tables claim it then it's likely Windows uses it on the same hardware. What's the stability like with the ACPI code?

Too soon to say - those where first results (about par with "it boots"). ;)

Will be poking at this machine more today -

* * * *

I do not have a copy of any proprietary operating system to compare with on this machine.

You could use the search features of this forum:
http://forum.netbookuser.com/index.php
to look for windows problem reports/discussions.

That forum was active since the machine was announced and included members from Everex. If there where windows problems, they are in there somewhere.

Mike

#41 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

Quoted 18 lines · Michael S. Zick
Quoted 17 lines · Matthew Garrett
Quoted 14 lines · Michael S. Zick

Phooey, close but no cigar - - -

root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_available_frequencies
1200000 1000000 800000 600000 400000
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat stats/time_in_state
1200000 3281
1000000 120
800000 130
600000 234
400000 52980
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq

It loads, it appears to work, but that lowest 400Mhz has ACPI entries even though this is a 600Mhz..1200Mhz chip.

Quoted 8 lines
Quoted 3 lines

If the BIOS tables claim it then it's likely Windows uses it on the same hardware. What's the stability like with the ACPI code?

Too soon to say - those where first results (about par with "it boots"). ;)

Will be poking at this machine more today -

An interesting note - not sure if it is cosmetic or an indication of something significant [sub-title: computer voodoo]:

You see above that acpi-cpufreq has the frequencies listed in decreasing order - -
The e_powersaver module lists them in increasing order. [????]

Mike

Quoted 21 lines

* * * *

I do not have a copy of any proprietary operating system to compare with on this machine.

You could use the search features of this forum:
http://forum.netbookuser.com/index.php
to look for windows problem reports/discussions.

That forum was active since the machine was announced and included members from Everex. If there where windows problems, they are in there somewhere.

Mike
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#42 lore raw

Re: [PATCH 1/2] CPUFREQ: Enable acpi-cpufreq driver for VIA/Centaur CPUs

In reply to Michael S. Zick
Quoted 18 lines · Michael S. Zick
Quoted 17 lines · Matthew Garrett
Quoted 14 lines · Michael S. Zick

Phooey, close but no cigar - - -

root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_available_frequencies
1200000 1000000 800000 600000 400000
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq# cat stats/time_in_state
1200000 3281
1000000 120
800000 130
600000 234
400000 52980
root@cb01:/sys/devices/system/cpu/cpu0/cpufreq

It loads, it appears to work, but that lowest 400Mhz has ACPI entries even though this is a 600Mhz..1200Mhz chip.

Quoted 8 lines
Quoted 3 lines

If the BIOS tables claim it then it's likely Windows uses it on the same hardware. What's the stability like with the ACPI code?

Too soon to say - those where first results (about par with "it boots"). ;)

Will be poking at this machine more today -

Naw - e_powersaver does a better job and consistently provides nearly an order of magnitude better up-time.

Guess I'll go back to reading/auditing e_powersaver. ;)

Mike

Quoted 21 lines

* * * *

I do not have a copy of any proprietary operating system to compare with on this machine.

You could use the search features of this forum:
http://forum.netbookuser.com/index.php
to look for windows problem reports/discussions.

That forum was active since the machine was announced and included members from Everex. If there where windows problems, they are in there somewhere.

Mike
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#43 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Michael S. Zick
Quoted 6 lines · Michael S. Zick

The mainstream kernel, e_powersaver, is *under-clocking* my machine -

The cpuid instruction provides the minimum and maximum GSF values (Guaranteed Stable Frequency) for that processor mask run - Passing that on as the lower and upper limits to e_powersaver should stop that problem. Will be testing this RSN.

Can you provide more information on where you think you are getting the GSF values from? I am not aware of such information.

The internal documentation always only refres to the minimum VID/FID values from the MSR. Also, according to the specifications of all C7 models listed, there are some that require a minimum of 400MHz, others require a minimum of
800. I have not found any that would require a minimum of 600MHz.

So I don't really know if your system really operates outside any specified range.

The only thing I know of that you can read from the CPUID vendor string is the maximum clock frequency in ASCII.

So it seems, e_powersaver is not really at all as dangerous as it appeared!

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#44 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Michael S. Zick
Quoted 8 lines · Michael S. Zick
Quoted 4 lines

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

There is also a possible thermal issue with these machines - - I doubt that VIA runs their qualification testing in bake ovens; which is what NetBook cases amount too. ;)

FWIW, they do, I've toured their facility and seen those ovens ;) I forget just how high they bake them at, but it was something ridiculous that even netbooks won't come close to.

Dave

#45 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Michael S. Zick
Quoted 8 lines · Michael S. Zick
Quoted 5 lines · Linus Torvalds

Side note: is it more stable if you disable the VIA speedstep thing (whatever it's called (ok, google tells me it's called "TwinTurbo" and "Advanced PowerSaver")?

The e_powersave code.

that is only for the early C7, as I've been told by Centaur. C7-M should definitely support the ACPI states. So better to disable both.

Quoted 2 lines

It was a fixed-speed kernel build that first hit the 4 hour up-time mark. I just reposted that build today (the -09143lk).

did you actually check if the power transitions are on longer happening?

Quoted 1 line

For want of a documented JTAG port...

there is JTAG on the CPU, but only for EXTEST/INTEST, i.e. access to the actual signal lanes for making sure all the BGA pads are soldered and actually connect to the right signals of the next component on the board. There is no embedded ICE or anything of that sort. Also, your board most likely doesn't have JTAG on an accessible connector, since it's out of mass production.

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#46 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 10 lines · Harald Welte
Quoted 8 lines · Michael S. Zick
Quoted 5 lines · Linus Torvalds

Side note: is it more stable if you disable the VIA speedstep thing (whatever it's called (ok, google tells me it's called "TwinTurbo" and "Advanced PowerSaver")?

The e_powersave code.

Quoted 8 lines

that is only for the early C7, as I've been told by Centaur. C7-M should definitely support the ACPI states. So better to disable both.

Quoted 2 lines

It was a fixed-speed kernel build that first hit the 4 hour up-time mark. I just reposted that build today (the -09143lk).

did you actually check if the power transitions are on longer happening?

No - without a /sys...cpufreq directory I don't know how to do that. But the machine is performing exactly as if it was at a fixed speed of 600Mhz.

Much different than when e_powersaver is included - -

Also, the e_powersaver stats look reasonable when compared with test loads running, so although this may be an acpi controlled system, e_powersaver "thinks" it has control (and does make a difference in performance). It is not that hard to tell subjectively the difference between 0.4Ghz and 1.2Ghz.

Mike

Quoted 8 lines
Quoted 1 line

For want of a documented JTAG port...

there is JTAG on the CPU, but only for EXTEST/INTEST, i.e. access to the actual signal lanes for making sure all the BGA pads are soldered and actually connect to the right signals of the next component on the board. There is no embedded ICE or anything of that sort. Also, your board most likely doesn't have JTAG on an accessible connector, since it's out of mass production.

#47 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Linus Torvalds
Quoted 10 lines · Linus Torvalds

Side note: is it more stable if you disable the VIA speedstep thing (whatever it's called (ok, google tells me it's called "TwinTurbo" and "Advanced PowerSaver")?

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

I agree, that might be an interesting test.

As a side note: The C7-M actually uses the same software interface as Intel CPU
for doing the power transitions (ACPI Performance States). Only very early C7
models used the cpufreq/e_powersaver.c driver.

So if you want to disable it, make sure you disable X86_ACPI_CPUFREQ.

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#48 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 12 lines · Harald Welte
Quoted 10 lines · Linus Torvalds

Side note: is it more stable if you disable the VIA speedstep thing (whatever it's called (ok, google tells me it's called "TwinTurbo" and "Advanced PowerSaver")?

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

Quoted 6 lines

I agree, that might be an interesting test.

As a side note: The C7-M actually uses the same software interface as Intel CPU
for doing the power transitions (ACPI Performance States). Only very early C7
models used the cpufreq/e_powersaver.c driver.

What is "very early C7 models" in terms of cpuid results?

If you can get permission to reveal those field details - we can test for that (this build reports "stepping 0" but I don't trust the report).

Mike

Quoted 2 lines

So if you want to disable it, make sure you disable X86_ACPI_CPUFREQ.

#49 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Hi Michael,

Quoted 5 lines · Michael S. Zick
Quoted 3 lines

As a side note: The C7-M actually uses the same software interface as Intel CPU
for doing the power transitions (ACPI Performance States). Only very early C7
models used the cpufreq/e_powersaver.c driver.

What is "very early C7 models" in terms of cpuid results?

I've inquired about that.

Looking at he code, the driver only seems to checking for a minimum CPUID value, but doesn't contain a maximum.

Regards,

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#50 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 8 lines · Harald Welte

Hi Michael,

Quoted 4 lines · Michael S. Zick
Quoted 3 lines

As a side note: The C7-M actually uses the same software interface as Intel CPU
for doing the power transitions (ACPI Performance States). Only very early C7
models used the cpufreq/e_powersaver.c driver.

Quoted 7 lines
Quoted 1 line

What is "very early C7 models" in terms of cpuid results?

I've inquired about that.

Looking at he code, the driver only seems to checking for a minimum CPUID value, but doesn't contain a maximum.

That is the way I read it also.

Mike

Quoted 1 line

Regards,

#51 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Linus Torvalds
Quoted 10 lines · Linus Torvalds

Side note: is it more stable if you disable the VIA speedstep thing (whatever it's called (ok, google tells me it's called "TwinTurbo" and "Advanced PowerSaver")?

Features like that easily put a huge stress on power regulators etc, if they result in sudden changes in current draw. Underspecced capacitors etc can cause CPU "brown-outs", which in turn can easily cause total failure.

If you try that, do idle=poll too. Hides class of bugs related to deep idle states.

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#52 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Linus Torvalds
Quoted 3 lines · Linus Torvalds

That said, how all the internal CPU registers are set is all black magic.

That is the situation as I find it -

I have been practicing Computer Black Magic (CBM) not Computer Science (CS) these past few months. <<deleted>> comment on un-documented processors.

Mike

Quoted 3 lines

Linus

#53 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Michael S. Zick
Quoted 4 lines · Michael S. Zick

That is one of my pending questions - -
(It is included as a comment at the appropriate point in my patchset.)

The VIA processors have MCR's not MTRR's - -

AFAIK, that was true for processors like the Winhcip / C6, i.e. earlier than the C3. The C3, C7 and later support 8 intel-style MTRR's.

Quoted 2 lines

The C7-M processor uses "in-order retirement" not "out-of-order" - - I think the MCR's **should not** be set for "weak ordered writes" -

why would it matter on UP? as indicated, I'm not the expert here, but I thought memory ordering issues only arise in SMP systems [or possibly with regard to DMA, but as we already explored much earlier in this thread, drivers that access DMA buffers whil the hardware owns them are buggy and need to be fixed]

Regards,

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#54 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 10 lines · Harald Welte
Quoted 4 lines · Michael S. Zick

That is one of my pending questions - -
(It is included as a comment at the appropriate point in my patchset.)

The VIA processors have MCR's not MTRR's - -

AFAIK, that was true for processors like the Winhcip / C6, i.e. earlier than the C3. The C3, C7 and later support 8 intel-style MTRR's.

Super! A specific breakage!

The c7 setup code is re-using the c6 setup code (MCR's) - - Will "if 0" out the appropriate parts and arrange for the MTRR setup.

@Linus - -
The Debian/Ubuntu distribution kernels require irqpoll (2.6.28+) - I took that out very early in my testing, when that problem got fixed - I will also try putting that back in, it might be needed for some of its side-effects on the processor/chipset.

Quoted 8 lines
Quoted 2 lines

The C7-M processor uses "in-order retirement" not "out-of-order" - - I think the MCR's **should not** be set for "weak ordered writes" -

why would it matter on UP? as indicated, I'm not the expert here, but I thought memory ordering issues only arise in SMP systems [or possibly with regard to DMA, but as we already explored much earlier in this thread, drivers that access DMA buffers whil the hardware owns them are buggy and need to be fixed]

I just recall the problems with the pa-risc port (not all machines have coherent I/O); some have consistent I/O only. I have one of those also, lucky me. ;)

Mike

Quoted 1 line

Regards,

#55 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 10 lines · Michael S. Zick
Quoted 2 lines · Harald Welte

On Thu, Jun 04, 2009 at 11:21:28AM -0500, Michael S. Zick wrote:

@Linus - -
The Debian/Ubuntu distribution kernels require irqpoll (2.6.28+) - I took that out very early in my testing, when that problem got fixed - I will also try putting that back in, it might be needed for some of its side-effects on the processor/chipset.

It looks like I stopped using irqpoll on 4/10 (-rc1)
http://forum.netbookuser.com/viewtopic.php?pid=6708#p6708

Am re-testing the "virgin" kernel build with both command options now:
http://forum.netbookuser.com/viewtopic.php?pid=7037#p7037

Will post if it has less than 6 months of up-time. ;)

Mike

#56 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 12 lines · Michael S. Zick
Quoted 10 lines · Michael S. Zick
Quoted 2 lines · Harald Welte

On Thu, Jun 04, 2009 at 11:21:28AM -0500, Michael S. Zick wrote:

@Linus - -
The Debian/Ubuntu distribution kernels require irqpoll (2.6.28+) - I took that out very early in my testing, when that problem got fixed - I will also try putting that back in, it might be needed for some of its side-effects on the processor/chipset.

Quoted 8 lines

It looks like I stopped using irqpoll on 4/10 (-rc1)
http://forum.netbookuser.com/viewtopic.php?pid=6708#p6708

Am re-testing the "virgin" kernel build with both command options now:
http://forum.netbookuser.com/viewtopic.php?pid=7037#p7037

Will post if it has less than 6 months of up-time. ;)

About 6 months less than 6 months up-time ;)

2.6.30-rc8 - no special cmd line options: 1h45m
2.6.30-rc8 - idle=poll irqpoll options: 1h08m

Summary:
Phooey!

Mike

Quoted 9 lines

Mike

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#57 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 19 lines · Michael S. Zick
Quoted 18 lines · Michael S. Zick
Quoted 10 lines · Michael S. Zick
Quoted 2 lines · Harald Welte

On Thu, Jun 04, 2009 at 11:21:28AM -0500, Michael S. Zick wrote:

@Linus - -
The Debian/Ubuntu distribution kernels require irqpoll (2.6.28+) - I took that out very early in my testing, when that problem got fixed - I will also try putting that back in, it might be needed for some of its side-effects on the processor/chipset.

It looks like I stopped using irqpoll on 4/10 (-rc1)
http://forum.netbookuser.com/viewtopic.php?pid=6708#p6708

Am re-testing the "virgin" kernel build with both command options now:
http://forum.netbookuser.com/viewtopic.php?pid=7037#p7037

Quoted 8 lines
Quoted 2 lines

Will post if it has less than 6 months of up-time. ;)

About 6 months less than 6 months up-time ;)

2.6.30-rc8 - no special cmd line options: 1h45m
2.6.30-rc8 - idle=poll irqpoll options: 1h08m

2.6.30-rc8-ce1200v-09155 - uptime: 5 hours+, still running. Ship it.

Details:
http://forum.netbookuser.com/viewtopic.php?pid=7039#p7039

Mike

Quoted 23 lines

Summary:
Phooey!

Mike

Quoted 9 lines

Mike

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#58 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 23 lines · Michael S. Zick
Quoted 22 lines · Michael S. Zick
Quoted 20 lines · Michael S. Zick
Quoted 10 lines · Michael S. Zick
Quoted 2 lines · Harald Welte

On Thu, Jun 04, 2009 at 11:21:28AM -0500, Michael S. Zick wrote:

@Linus - -
The Debian/Ubuntu distribution kernels require irqpoll (2.6.28+) - I took that out very early in my testing, when that problem got fixed - I will also try putting that back in, it might be needed for some of its side-effects on the processor/chipset.

It looks like I stopped using irqpoll on 4/10 (-rc1)
http://forum.netbookuser.com/viewtopic.php?pid=6708#p6708

Am re-testing the "virgin" kernel build with both command options now:
http://forum.netbookuser.com/viewtopic.php?pid=7037#p7037

Will post if it has less than 6 months of up-time. ;)

Quoted 9 lines
Quoted 5 lines

About 6 months less than 6 months up-time ;)

2.6.30-rc8 - no special cmd line options: 1h45m
2.6.30-rc8 - idle=poll irqpoll options: 1h08m

2.6.30-rc8-ce1200v-09155 - uptime: 5 hours+, still running. Ship it.

New record: 9h15m

But that is bandaid'd enough I can stop with the Black Magic and Computer Voodoo - -

Now just a matter of cutting each problem out of the herd and solving each of them. . . ;)

A likely first candidate is this USB driver with its 28Mbyte cry for help in the syslog file.

Mike

Quoted 37 lines

Details:
http://forum.netbookuser.com/viewtopic.php?pid=7039#p7039

Mike

Quoted 16 lines

Summary:
Phooey!

Mike

Quoted 9 lines

Mike

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Quoted 7 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#59 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Michael S. Zick
Quoted 8 lines · Michael S. Zick
Quoted 7 lines · Harald Welte
Quoted 4 lines · Michael S. Zick

That is one of my pending questions - -
(It is included as a comment at the appropriate point in my patchset.)

The VIA processors have MCR's not MTRR's - -

Quoted 8 lines
Quoted 3 lines

AFAIK, that was true for processors like the Winhcip / C6, i.e. earlier than the C3. The C3, C7 and later support 8 intel-style MTRR's.

Super! A specific breakage!

The c7 setup code is re-using the c6 setup code (MCR's) - - Will "if 0" out the appropriate parts and arrange for the MTRR setup.

It's not touching the MCRs. C7's are family 6. The MCR code is only
called for family 5. (Winchips) Look at the switch statement
in init_centaur()

Dave

#60 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 13 lines · Dave Jones
Quoted 12 lines · Michael S. Zick
Quoted 10 lines · Harald Welte
Quoted 4 lines · Michael S. Zick

That is one of my pending questions - -
(It is included as a comment at the appropriate point in my patchset.)

The VIA processors have MCR's not MTRR's - -

AFAIK, that was true for processors like the Winhcip / C6, i.e. earlier than the C3. The C3, C7 and later support 8 intel-style MTRR's.

Quoted 9 lines
Quoted 4 lines

Super! A specific breakage!

The c7 setup code is re-using the c6 setup code (MCR's) - - Will "if 0" out the appropriate parts and arrange for the MTRR setup.

It's not touching the MCRs. C7's are family 6. The MCR code is only
called for family 5. (Winchips) Look at the switch statement
in init_centaur()

Yup, mis-read that one - -

        case 6:
                init_c3(c);
                break;

= = = =

But while your here, what is your opinion on this one,
in: int __init pcibios_init(void)

- - - - -

        pci_cache_line_size = 32 >> 2;
        if (c->x86 >= 6
        && (c->x86_vendor == X86_VENDOR_AMD) || (c->x86_vendor == X86_VENDOR_CENTAUR))
                pci_cache_line_size = 64 >> 2;  /* K7 & K8 and VIA C7-M */
        else if (c->x86 > 6 && c->x86_vendor == X86_VENDOR_INTEL)
                pci_cache_line_size = 128 >> 2; /* P4 */

Mike

#61 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 4 lines · Michael S. Zick

But while your here, what is your opinion on this one,
in: int __init pcibios_init(void)

- - - - -

Quoted 8 lines
        pci_cache_line_size = 32 >> 2;
        if (c->x86 >= 6
        && (c->x86_vendor == X86_VENDOR_AMD) || (c->x86_vendor == X86_VENDOR_CENTAUR))
                pci_cache_line_size = 64 >> 2;  /* K7 & K8 and VIA C7-M */
        else if (c->x86 > 6 && c->x86_vendor == X86_VENDOR_INTEL)
                pci_cache_line_size = 128 >> 2; /* P4 */

Mike

C7's L1 cachelines are 64 bytes, so it's right in that case,
but the earlier Centaur CPUs are 32 bytes, so it should be checking steppings.

Or better yet, why not just set it to boot_cpu_data->x86_clflush_size ?

Dave

#62 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 13 lines · Dave Jones
Quoted 11 lines · Michael S. Zick

But while your here, what is your opinion on this one,
in: int __init pcibios_init(void)

- - - - -

        pci_cache_line_size = 32 >> 2;
        if (c->x86 >= 6
        && (c->x86_vendor == X86_VENDOR_AMD) || (c->x86_vendor == X86_VENDOR_CENTAUR))
                pci_cache_line_size = 64 >> 2;  /* K7 & K8 and VIA C7-M */
        else if (c->x86 > 6 && c->x86_vendor == X86_VENDOR_INTEL)
                pci_cache_line_size = 128 >> 2; /* P4 */
Quoted 7 lines
Quoted 1 line

Mike

C7's L1 cachelines are 64 bytes, so it's right in that case,
but the earlier Centaur CPUs are 32 bytes, so it should be checking steppings.

Or better yet, why not just set it to boot_cpu_data->x86_clflush_size ?

I have already preceded that chunk of code with a printk and confirmed that x86_clflush_size is properly set to 64 bytes (somewhere else).

So your suggestion is the obvious one for the C7-M, I don't know about any other makes/models.

This machine's C7-M is being reported as a "stepping 0" ?? What is earlier than a stepping 0 ??

Mike

Quoted 4 lines

Dave

#63 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 8 lines · Michael S. Zick

I have already preceded that chunk of code with a printk and confirmed that x86_clflush_size is properly set to 64 bytes (somewhere else).

So your suggestion is the obvious one for the C7-M, I don't know about any other makes/models.

This machine's C7-M is being reported as a "stepping 0" ?? What is earlier than a stepping 0 ??

Earlier model numbers.
(Stepping gets reset to 0 every time they bump the model)

Dave

#64 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 4 lines · Dave Jones
Quoted 3 lines · Michael S. Zick

I have already preceded that chunk of code with a printk and confirmed that x86_clflush_size is properly set to 64 bytes (somewhere else).

Quoted 9 lines
Quoted 5 lines

So your suggestion is the obvious one for the C7-M, I don't know about any other makes/models.

This machine's C7-M is being reported as a "stepping 0" ?? What is earlier than a stepping 0 ??

Earlier model numbers.
(Stepping gets reset to 0 every time they bump the model)

So 6.model=13.stepping=0 should do it for the (or this) C7-M, correct? Your mention of "earlier models" will not apply until there is a stepping >0 for a model=13, correct?

Would a good practice be to put a "WARN_ON" in there in case a stepping greater than =0 happens to execute the code?
In addition to checking specificly for 6.13.0 (rather than just 6.13).

Hmmm...
But that is making a big mess of a small mess - Why not just do the assignment you suggest for 6.model>=13?

Models earlier than 13 are evidently working (due to lack of bug reports or other noise on this list).

Mike

Quoted 9 lines

Dave

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#65 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 3 lines · Michael S. Zick
Quoted 1 line · Dave Jones

On Thu, Jun 04, 2009 at 04:38:13PM -0500, Michael S. Zick wrote:

I have it now, you where speaking of the "product vendor Centaur" not a specific model name "Centaur".

Which I translate into: that statement block needs to be converted into (possibly nested) switch statement(s).
Since there is not a "model check" in it only a "Series" and "Vendor" check.

Yuck.
Mike

Quoted 47 lines
Quoted 12 lines
Quoted 8 lines

I have already preceded that chunk of code with a printk and confirmed that x86_clflush_size is properly set to 64 bytes (somewhere else).

So your suggestion is the obvious one for the C7-M, I don't know about any other makes/models.

This machine's C7-M is being reported as a "stepping 0" ?? What is earlier than a stepping 0 ??

Earlier model numbers.
(Stepping gets reset to 0 every time they bump the model)

So 6.model=13.stepping=0 should do it for the (or this) C7-M, correct? Your mention of "earlier models" will not apply until there is a stepping >0 for a model=13, correct?

Would a good practice be to put a "WARN_ON" in there in case a stepping greater than =0 happens to execute the code?
In addition to checking specificly for 6.13.0 (rather than just 6.13).

Hmmm...
But that is making a big mess of a small mess - Why not just do the assignment you suggest for 6.model>=13?

Models earlier than 13 are evidently working (due to lack of bug reports or other noise on this list).

Mike

Quoted 9 lines

Dave

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#66 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 5 lines · Michael S. Zick
Quoted 3 lines · Michael S. Zick
Quoted 1 line · Dave Jones

On Thu, Jun 04, 2009 at 04:38:13PM -0500, Michael S. Zick wrote:

Quoted 8 lines

I have it now, you where speaking of the "product vendor Centaur" not a specific model name "Centaur".

Which I translate into: that statement block needs to be converted into (possibly nested) switch statement(s).
Since there is not a "model check" in it only a "Series" and "Vendor" check.

Yuck.

I meant just doing something like this..

(untested)

I'm not sure if the clflush_size==0 case can happen, which is why I left the fallback. Maybe on ancient cpus?

Dave

arch/x86/pci/common.c +8 -5
index 2202b62..a293c71 100644
@@ -426,11 +426,14 @@ int __init pcibios_init(void)
 	 * and P4. It's also good for 386/486s (which actually have 16)
 	 * as quite a few PCI devices do not support smaller values.
 	 */
-	pci_cache_line_size = 32 >> 2;
-	if (c->x86 >= 6 && c->x86_vendor == X86_VENDOR_AMD)
-		pci_cache_line_size = 64 >> 2;	/* K7 & K8 */
-	else if (c->x86 > 6 && c->x86_vendor == X86_VENDOR_INTEL)
-		pci_cache_line_size = 128 >> 2;	/* P4 */
+
+	if (c->x86_clflush_size > 0)
+		pci_cache_line_size = c->x86_clflush_size >> 2;
+	else
+		pci_cache_line_size = 32 >> 2;
+
+	printk(KERN_DEBUG "PCI: pci_cache_line_size set to %d bytes\n",
+		pci_cache_line_size >> 2);
 
 	pcibios_resource_survey();
 
#67 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 38 lines · Dave Jones
Quoted 5 lines · Michael S. Zick
Quoted 3 lines · Michael S. Zick
Quoted 1 line · Dave Jones

On Thu, Jun 04, 2009 at 04:38:13PM -0500, Michael S. Zick wrote:

Quoted 8 lines

I have it now, you where speaking of the "product vendor Centaur" not a specific model name "Centaur".

Which I translate into: that statement block needs to be converted into (possibly nested) switch statement(s).
Since there is not a "model check" in it only a "Series" and "Vendor" check.

Yuck.

I meant just doing something like this..

(untested)

I'm not sure if the clflush_size==0 case can happen, which is why I left the fallback. Maybe on ancient cpus?

Dave

arch/x86/pci/common.c +1 -5
index 2202b62..a293c71 100644
@@ -426,11 +426,14 @@ int __init pcibios_init(void)
 	 * and P4. It's also good for 386/486s (which actually have 16)
 	 * as quite a few PCI devices do not support smaller values.
 	 */
-	pci_cache_line_size = 32 >> 2;
-	if (c->x86 >= 6 && c->x86_vendor == X86_VENDOR_AMD)
-		pci_cache_line_size = 64 >> 2;	/* K7 & K8 */
-	else if (c->x86 > 6 && c->x86_vendor == X86_VENDOR_INTEL)
-		pci_cache_line_size = 128 >> 2;	/* P4 */
+

arch/x86/pci/common.c int __init pcibios_init(void) · line 439

Quoted 10 lines
arch/x86/pci/common.c +7 -0
+	if (c->x86_clflush_size > 0)
+		pci_cache_line_size = c->x86_clflush_size >> 2;
+	else
+		pci_cache_line_size = 32 >> 2;
+
+	printk(KERN_DEBUG "PCI: pci_cache_line_size set to %d bytes\n",
+		pci_cache_line_size >> 2);
 
 	pcibios_resource_survey();
 

That would clean up the code and deal with the situation. Sorry I was so slow about following what you meant. Age.

Will put that in tomorrows build instead of my idea.

Mike

Quoted 2 lines
#68 lore raw

VIA CPU PCI cache line size (Re: Linux 2.6.30-rc8 [also: VIA Support])

In reply to Dave Jones
Quoted 6 lines · Dave Jones

I meant just doing something like this..

(untested)

I'm not sure if the clflush_size==0 case can happen, which is why I left the fallback. Maybe on ancient cpus?

this patch looks fine with me. I didn't do any testing yet either, though.

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#69 lore raw

Re: VIA CPU PCI cache line size (Re: Linux 2.6.30-rc8 [also: VIA Support])

Quoted 11 lines · Harald Welte
Quoted 6 lines · Dave Jones

I meant just doing something like this..

(untested)

I'm not sure if the clflush_size==0 case can happen, which is why I left the fallback. Maybe on ancient cpus?

this patch looks fine with me. I didn't do any testing yet either, though.

Although not coded as Dave J. suggests - I have been re-setting the that value to the cpu cache line size for a week or two now.

It doesn't break anything in an obvious way - does seem to help.

Mike

#70 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Harald Welte
Quoted 5 lines · Harald Welte

why would it matter on UP? as indicated, I'm not the expert here, but I thought memory ordering issues only arise in SMP systems [or possibly with regard to DMA, but as we already explored much earlier in this thread, drivers that access DMA buffers whil the hardware owns them are buggy and need to be fixed]

Sorry we didn't establish that. Accessing data structures that are also accessed by DMA hardware is pretty common in fact and memory ordering issues also come up regularly (e.g. all the infamous PCI posting bugs)

What we established is that the drivers don't use LOCK for it (or at least we think that's very unlikely)

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only.
#71 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 7 lines · Andi Kleen
Quoted 5 lines · Harald Welte

why would it matter on UP? as indicated, I'm not the expert here, but I thought memory ordering issues only arise in SMP systems [or possibly with regard to DMA, but as we already explored much earlier in this thread, drivers that access DMA buffers whil the hardware owns them are buggy and need to be fixed]

Quoted 8 lines

Sorry we didn't establish that. Accessing data structures that are also accessed by DMA hardware is pretty common in fact and memory ordering issues also come up regularly (e.g. all the infamous PCI posting bugs)

What we established is that the drivers don't use LOCK for it (or at least we think that's very unlikely)

It was a real headache in the pa-risc port - -
Even went so far as to build some experimental kernels where all the spin-lock structures where in a separate loader section.

That was to avoid in-direct interference - I.E: Both DMA and the processor handling the locking **both** invalidating the same cache line at the same time (only one can win).

Things might get that deep with this processor/chip-set combination; but pa-risc has some very unusual hardware in some older models.

My favorite still is a human coding error somewhere, not an architectural or structural problem.

Mike

Quoted 2 lines

-Andi

#72 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 19 lines · Michael S. Zick
Quoted 5 lines · Andi Kleen
Quoted 4 lines · Harald Welte

why would it matter on UP? as indicated, I'm not the expert here, but I thought memory ordering issues only arise in SMP systems [or possibly with regard to DMA, but as we already explored much earlier in this thread, drivers that access DMA buffers whil the hardware owns them are buggy and need to be fixed]

Quoted 8 lines

Sorry we didn't establish that. Accessing data structures that are also accessed by DMA hardware is pretty common in fact and memory ordering issues also come up regularly (e.g. all the infamous PCI posting bugs)

What we established is that the drivers don't use LOCK for it (or at least we think that's very unlikely)

It was a real headache in the pa-risc port - -
Even went so far as to build some experimental kernels where all the spin-lock structures where in a separate loader section.

Quoted 6 lines

That was to avoid in-direct interference - I.E: Both DMA and the processor handling the locking **both** invalidating the same cache line at the same time (only one can win).

Things might get that deep with this processor/chip-set combination; but pa-risc has some very unusual hardware in some older models.

That sort of thing should be architecturally impossible on x86. In order for something to invalidate the cache line, it first has to own it (except maybe for some unusual cases like Memory Write and Invalidate where the writer promises to overwrite the entire cache line).

#73 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 24 lines · Robert Hancock
Quoted 23 lines · Michael S. Zick
Quoted 5 lines · Andi Kleen
Quoted 4 lines · Harald Welte

why would it matter on UP? as indicated, I'm not the expert here, but I thought memory ordering issues only arise in SMP systems [or possibly with regard to DMA, but as we already explored much earlier in this thread, drivers that access DMA buffers whil the hardware owns them are buggy and need to be fixed]

Quoted 8 lines

Sorry we didn't establish that. Accessing data structures that are also accessed by DMA hardware is pretty common in fact and memory ordering issues also come up regularly (e.g. all the infamous PCI posting bugs)

What we established is that the drivers don't use LOCK for it (or at least we think that's very unlikely)

It was a real headache in the pa-risc port - -
Even went so far as to build some experimental kernels where all the spin-lock structures where in a separate loader section.

That was to avoid in-direct interference - I.E: Both DMA and the processor handling the locking **both** invalidating the same cache line at the same time (only one can win).

Quoted 9 lines
Quoted 2 lines

Things might get that deep with this processor/chip-set combination; but pa-risc has some very unusual hardware in some older models.

That sort of thing should be architecturally impossible on x86. In order for something to invalidate the cache line, it first has to own it (except maybe for some unusual cases like Memory Write and Invalidate where the writer promises to overwrite the entire cache line).

VIA has not publicly published sufficient technical information to presume that the cache coherency control protocols are the same as Intel's.

These are cpu/chipset pairs - Think System On 2 Chips. SoS2C.

Mike
Mike

#74 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 4 lines · Michael S. Zick

VIA has not publicly published sufficient technical information to presume that the cache coherency control protocols are the same as Intel's.

These are cpu/chipset pairs - Think System On 2 Chips. SoS2C.

At the low level, maybe not, but functionally it has to be equivalent. Otherwise the chip can't really be considered x86-compatible.

#75 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Andi Kleen
Quoted 7 lines · Andi Kleen
Quoted 5 lines · Harald Welte

why would it matter on UP? as indicated, I'm not the expert here, but I thought memory ordering issues only arise in SMP systems [or possibly with regard to DMA, but as we already explored much earlier in this thread, drivers that access DMA buffers whil the hardware owns them are buggy and need to be fixed]

Quoted 7 lines

Sorry we didn't establish that. Accessing data structures that are also accessed by DMA hardware is pretty common in fact and memory ordering issues also come up regularly (e.g. all the infamous PCI posting bugs)

What we established is that the drivers don't use LOCK for it (or at least we think that's very unlikely)

Yes, sorry for not mentioning that. I was implicitly thinking of "not using an appropriate locking method"

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#76 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Linus Torvalds

Dear Linus and others,

Quoted 12 lines · Linus Torvalds
Quoted 7 lines

There have been reports of hangs on various VIA C7 machines going back a year now. The version of the kernel doesn't seem to matter, but the version of glibc does. Unfortunately there hasn't been much progress in getting to the bottom of it.

See here (and other linked reports):
http://bugs.gentoo.org/show_bug.cgi?id=228263

Hmm. That looks like a CPU problem, but hey, it might be that the glibc version thing is just coincidence, and just changes timings or whatever, and the problem is in the chipsets.

Quoted 6 lines

So at least from that particular report it smells very much non-kernel-related.

That said, even if it isn't kernel-related, it might be fixable with some kernel patch that changes the setup of the CPU/chipset. But we'd need VIA to help with anythign like that.

So far, inside VIA there is no well-known issue/bug about such hangs / locks at all.

I have seen a number (probably between 5 or 10) of sporadic reports from a number of people on a variety of systems. Some from actual commercial vendors of VIA+Linux based appliances, and some from the wider community of end users. So far, to the best of my knowledge, none of those isseus has been narrowed down to a sufficiently easy to reproduce test case. Also, none of the bug reporters has so far been able to reproduce the problem on a genuine VIA mainboard, i.e. it could be issues introduced by the actual board hardware or how the speicfic BIOS initializes the low-level hardware.

Especially when SMI/SMM based debugging no longer works (i.e. something that appears to be a bus lockup), the actual bug needs to be reproduced on a reference board that can be hooked up to a logic/protocol analyzer.

On the other hand, VIA's CPU division (CentaurLabs) is performing extensive testing on their CPUs with a large codebase of x86 code, AFAIK based on more than 40 operating systems. Also, there are large quantities of VIA CPU+chipset systems that run without any problem, especially in 24/7 embedded x86 worloads on Linux...

I'm more than determined to help resolving those sporadic Linux lock-up problems. It feels like there is some problem out there, given the fact that there is a number of independent reporters who talk about some kind of hard system hang without oops that even prevents the NMI watchdog to kick in.

However, unless we can somehow narrow down at least one of those reports into something that is easier to reproduce, and which can actuall be reproduced on a VIA board. Triggering in 1-4 hours is already very good, I have reports where 1 of 30 system exposes a lock once within 5 days of continuous full application workload.

Sure, third party BIOS/board vendors selling products that randomly produce locks are obviously also not a particularly great advertisement for VIA... but debigging on such a board is much more difficult due to the lack of access to BIOS sources, schematics and hardware debugging interfaces.

In any case, if somebody can ship me a system that exposes one of those lock-ups, together with a pre-installed test case that exposes the problem within let's say less than one day, plus the full kernel sources used in that particular system: I'm happy to spend time to investigate the issue, try to run the same test case on a VIA board, etc.

Any additional help is much appreciated.

Regards,

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#77 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 61 lines · Harald Welte

Dear Linus and others,

Quoted 12 lines · Linus Torvalds
Quoted 7 lines

There have been reports of hangs on various VIA C7 machines going back a year now. The version of the kernel doesn't seem to matter, but the version of glibc does. Unfortunately there hasn't been much progress in getting to the bottom of it.

See here (and other linked reports):
http://bugs.gentoo.org/show_bug.cgi?id=228263

Hmm. That looks like a CPU problem, but hey, it might be that the glibc version thing is just coincidence, and just changes timings or whatever, and the problem is in the chipsets.

Quoted 6 lines

So at least from that particular report it smells very much non-kernel-related.

That said, even if it isn't kernel-related, it might be fixable with some kernel patch that changes the setup of the CPU/chipset. But we'd need VIA to help with anythign like that.

So far, inside VIA there is no well-known issue/bug about such hangs / locks at all.

I have seen a number (probably between 5 or 10) of sporadic reports from a number of people on a variety of systems. Some from actual commercial vendors of VIA+Linux based appliances, and some from the wider community of end users. So far, to the best of my knowledge, none of those isseus has been narrowed down to a sufficiently easy to reproduce test case. Also, none of the bug reporters has so far been able to reproduce the problem on a genuine VIA mainboard, i.e. it could be issues introduced by the actual board hardware or how the speicfic BIOS initializes the low-level hardware.

Especially when SMI/SMM based debugging no longer works (i.e. something that appears to be a bus lockup), the actual bug needs to be reproduced on a reference board that can be hooked up to a logic/protocol analyzer.

On the other hand, VIA's CPU division (CentaurLabs) is performing extensive testing on their CPUs with a large codebase of x86 code, AFAIK based on more than 40 operating systems. Also, there are large quantities of VIA CPU+chipset systems that run without any problem, especially in 24/7 embedded x86 worloads on Linux...

I'm more than determined to help resolving those sporadic Linux lock-up problems. It feels like there is some problem out there, given the fact that there is a number of independent reporters who talk about some kind of hard system hang without oops that even prevents the NMI watchdog to kick in.

However, unless we can somehow narrow down at least one of those reports into something that is easier to reproduce, and which can actuall be reproduced on a VIA board. Triggering in 1-4 hours is already very good, I have reports where 1 of 30 system exposes a lock once within 5 days of continuous full application workload.

Sure, third party BIOS/board vendors selling products that randomly produce locks are obviously also not a particularly great advertisement for VIA... but debigging on such a board is much more difficult due to the lack of access to BIOS sources, schematics and hardware debugging interfaces.

Quoted 6 lines

In any case, if somebody can ship me a system that exposes one of those lock-ups, together with a pre-installed test case that exposes the problem within let's say less than one day, plus the full kernel sources used in that particular system: I'm happy to spend time to investigate the issue, try to run the same test case on a VIA board, etc.

I am about at my wits end with this Everex product -

Give me a couple more weeks at the problem and if I haven't solved it; I'll give you this machine if you promise to update LKML with any fix.

Mike

Quoted 3 lines

Any additional help is much appreciated.

Regards,

#78 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 10 lines · Michael S. Zick
Quoted 5 lines

In any case, if somebody can ship me a system that exposes one of those lock-ups, together with a pre-installed test case that exposes the problem within let's say less than one day, plus the full kernel sources used in that particular system: I'm happy to spend time to investigate the issue, try to run the same test case on a VIA board, etc.

I am about at my wits end with this Everex product -

Give me a couple more weeks at the problem and if I haven't solved it; I'll give you this machine if you promise to update LKML with any fix.

Well, I cannnot promise that I will actually find the problem and find a fix. But I can promise to look into the issue and potentially ask other people at VIA/Centaur to help with investigating it.

And as soon as the problem can be reproduced, or we have a fix, or I give up, you would get the system back, obviously.

I'll try to figure out how we can do the shipping while billing it to VIA's account.

In any case, let's wait until I hear from VIA if they can provide one of those machines to me.

Signature · 3 lines
- Harald Welte <[email protected]>	    http://linux.via.com.tw/
============================================================================
VIA Free and Open Source Software Liaison
#79 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 12 lines · Harald Welte
Quoted 10 lines · Michael S. Zick
Quoted 5 lines

In any case, if somebody can ship me a system that exposes one of those lock-ups, together with a pre-installed test case that exposes the problem within let's say less than one day, plus the full kernel sources used in that particular system: I'm happy to spend time to investigate the issue, try to run the same test case on a VIA board, etc.

I am about at my wits end with this Everex product -

Give me a couple more weeks at the problem and if I haven't solved it; I'll give you this machine if you promise to update LKML with any fix.

Quoted 7 lines

Well, I cannnot promise that I will actually find the problem and find a fix. But I can promise to look into the issue and potentially ask other people at VIA/Centaur to help with investigating it.

And as soon as the problem can be reproduced, or we have a fix, or I give up, you would get the system back, obviously.

I would not want it back, obviously.

Quoted 3 lines

I'll try to figure out how we can do the shipping while billing it to VIA's account.

Collect?

Quoted 3 lines

In any case, let's wait until I hear from VIA if they can provide one of those machines to me.

Just ask for an Everex CrapBook - they will know what you mean. The company went out of the NetBook business for a good reason.

Mike

#80 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

Quoted 27 lines · Michael S. Zick
Quoted 12 lines · Harald Welte
Quoted 10 lines · Michael S. Zick
Quoted 5 lines

In any case, if somebody can ship me a system that exposes one of those lock-ups, together with a pre-installed test case that exposes the problem within let's say less than one day, plus the full kernel sources used in that particular system: I'm happy to spend time to investigate the issue, try to run the same test case on a VIA board, etc.

I am about at my wits end with this Everex product -

Give me a couple more weeks at the problem and if I haven't solved it; I'll give you this machine if you promise to update LKML with any fix.

Quoted 7 lines

Well, I cannnot promise that I will actually find the problem and find a fix. But I can promise to look into the issue and potentially ask other people at VIA/Centaur to help with investigating it.

And as soon as the problem can be reproduced, or we have a fix, or I give up, you would get the system back, obviously.

I would not want it back, obviously.

Quoted 3 lines

I'll try to figure out how we can do the shipping while billing it to VIA's account.

Quoted 9 lines

Collect?

Quoted 3 lines

In any case, let's wait until I hear from VIA if they can provide one of those machines to me.

Just ask for an Everex CrapBook - they will know what you mean. The company went out of the NetBook business for a good reason.

Oh, I am within a hundred miles of offices of the company that FIC sold the 75% interest in Everex to - I even recognize a few of the officer's names - not that I have seen them for years.

Mike

Quoted 8 lines

Mike
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#81 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Harald Welte

Hi,

Quoted 4 lines

I'm more than determined to help resolving those sporadic Linux lock-up problems. It feels like there is some problem out there, given the fact that there is a number of independent reporters who talk about some kind of hard system hang without oops that even prevents the NMI watchdog to kick in.

Data point: I have trouble with an (oldish) via box too.

processor : 0
vendor_id : CentaurHauls
cpu family : 6
model : 9
model name : VIA Nehemiah
stepping : 3

It has a quite interesting error pattern: It runs rock solid in summer, it has stability issues in winter. It is located in a server rack in a basement, so the weather out there shouldn't make a big difference to the machine. I'm really curious what this might be ...

"stability issues" means it does either lock up or reboot, usually without anything on the serial console. Seems to be more likely under load. The periods of time the machine runs can range from hours to weeks.

cheers,
Gerd

#82 lore raw

Re: Linux 2.6.30-rc8 [also: VIA Support]

In reply to Duane Griffin
Quoted 4 lines · Duane Griffin

2009/6/4 Linus Torvalds <[email protected]>:

Quoted 3 lines · Michael S. Zick
Quoted 1 line

Well, that one isn't anything to brag about"

Quoted 9 lines
Quoted 3 lines

So what's the problem on that machine? Is the VIA C7-M (or some random device in it) just buggy and it eventually hangs, or what? Has _any_ kernel ever worked on that machine for longer than an hour or two?

There have been reports of hangs on various VIA C7 machines going back a year now. The version of the kernel doesn't seem to matter, but the version of glibc does. Unfortunately there hasn't been much progress in getting to the bottom of it.

There have been recent tweaks to the futex service in the kernel, this might interact with how a particular glibc expects them to work.

I also add a few tweaks in my local patches to futex.etc, so it may need more in mainline (un-proven at this point).

It seems to make a difference with pulse-audio - - which is my "test program".

Mike

Quoted 6 lines

See here (and other linked reports):
http://bugs.gentoo.org/show_bug.cgi?id=228263

Cheers,
Duane.

#83 lore raw
In reply to Linus Torvalds
Quoted 10 lines · Linus Torvalds

This is almost certainly the last -rc, and I debated even doing it. But rather than just do 2.6.30, I decided that I'd be better off doing a last -rc8 and then a real release probably this weekend.

This has mostly driver and arch updates, with perhaps the intel drm/kms and network driver changes standing out, but there's powerpc, blackfin and arm updates too. Admittedly, the two biggest parts of the powerpc update are a revert and a defconfig update.

Quoted 5 lines

A lot of small stuff, fixing a few regressions (and at least one bugzilla entry going back to 2.6.24). The small stuff does matter. Please test.

Linus

Today's test build for VIA powered NetBooks is posted (as -09156):
Details at:
http://forum.netbookuser.com/viewtopic.php?pid=7043#p7043

Summary:
Yesterday's (last night's, this morning's) record holder (9+ hours) plus the changes proposed by Dave Jones and Bob Copeland.

Mike