AHCI updates: Marvell AHCI PATA works; pata_marvell fate?
cover AHCI updates: Marvell AHCI PATA works; pata_marvell fate?
This is an update of the recent libahci patchset, currently available on the 'libahci' branch ofgit://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev.git
These patches move a large amount of code from the currently self-contained ahci driver to a new kernel module, libahci, in preparation for supporting some "98% AHCI clones" from Marvell and ACard.
This also has the nice side effect of isolating the generic AHCI controller code from the grungy vendor-specific details, and opens the way for supporting AHCI on non-PCI buses (at least one such example exists, amba).
Changes since last posting: Marvell PATA started working
Marvell 6145 has successfully been stress-tested with the following modern, high performance 2GB hard drive:
ATA device, with non-removable media
Model Number: WDC AC22100H
Firmware Revision: 10.07H12
Standards:
Likely used: 2
Marvell PATA on 6121 and similar chips should also work, but that hasn't yet been tested. Here's the full list of what still needs testing:
- master + slave devices
- ATAPI
- 6121 (6123? others?)
Also, there is the open question of how to deal with pata_marvell co-existence (hello, Alan :)), now that we have two noticeable driver changes in this area:
- Marvell AHCI-alikes moved from 'ahci' to 'mv-ahci'
- We can now fully support SATA+PATA configurations in 'mv-ahci'
The mv-ahci code is obviously still a bit raw with "#if 0" in places, but it is nonetheless ready for testing.
PATCH 1/3 ahci: move core code into new libahci module
commit f3e4ee18ce6c6f538f1ae5d2c6321695898e77d9
Author: Jeff Garzik <[email protected]> Date: Sun Apr 12 10:24:42 2009 -0400
libata: Move Marvell AHCI to separate driver 'mv-ahci'. Create 'libahci'.
Move AHCI common code into new kernel module "libahci".
Move Marvell-specific code into new driver "mv-ahci".
drivers/ata/Kconfig | 13 drivers/ata/Makefile | 3 drivers/ata/ahci.c | 1992 -------------------------------------------------- drivers/ata/ahci.h | 283 +++++++ drivers/ata/libahci.c | 1737 +++++++++++++++++++++++++++++++++++++++++++ drivers/ata/mv-ahci.c | 375 +++++++++ 6 files changed, 2424 insertions(+), 1979 deletions(-)
drivers/ata/Kconfig +13 -0
@@ -111,6 +111,15 @@ config ATA_PIIX If unsure, say N. +config SATA_MV_AHCI + tristate "Marvell AHCI support (6121, 6145)" + depends on PCI + help + This option enables support for AHCI-like chips + from Marvell: 6121, 6145. + + If unsure, say N. + config SATA_MV tristate "Marvell SATA support" help
@@ -118,6 +127,10 @@ config SATA_MV Currently supports 88SX[56]0[48][01] PCI(-X) chips, as well as the newer [67]042 PCI-X/PCIe and SOC devices. + This does NOT include support for AHCI-like chips from + Marvell, notably: 6121, 6145. Use the mv-ahci (SATA_MV_AHCI) + driver for that. + If unsure, say N. config SATA_NV
drivers/ata/Makefile +2 -1
@@ -1,7 +1,7 @@ obj-$(CONFIG_ATA) += libata.o -obj-$(CONFIG_SATA_AHCI) += ahci.o +obj-$(CONFIG_SATA_AHCI) += libahci.o ahci.o obj-$(CONFIG_SATA_SVW) += sata_svw.o obj-$(CONFIG_ATA_PIIX) += ata_piix.o obj-$(CONFIG_SATA_PROMISE) += sata_promise.o
@@ -14,6 +14,7 @@ obj-$(CONFIG_SATA_SIS) += sata_sis.o obj-$(CONFIG_SATA_SX4) += sata_sx4.o obj-$(CONFIG_SATA_NV) += sata_nv.o obj-$(CONFIG_SATA_ULI) += sata_uli.o +obj-$(CONFIG_SATA_MV_AHCI) += libahci.o mv-ahci.o obj-$(CONFIG_SATA_MV) += sata_mv.o obj-$(CONFIG_SATA_INIC162X) += sata_inic162x.o obj-$(CONFIG_PDC_ADMA) += pdc_adma.o
drivers/ata/ahci.c +14 -1978
2,114 lines not on this page · raw message
drivers/ata/ahci.h new file +283 -0
283 lines not on this page · raw message
drivers/ata/libahci.c new file +1737 -0
563 lines not on this page · raw message
drivers/ata/mv-ahci.c new file +375 -0
375 lines not on this page · raw message
The rest of this diff is not shown. View the raw message.
Part of the diff is not shown here. View the raw message.
PATCH 2/3 mv-ahci: Add Marvell PATA support
In reply to Jeff Garzikcommit b789e972e23509ab27312b9cf7273b611feb25cb
Author: Jeff Garzik <[email protected]> Date: Thu Apr 16 22:13:27 2009 -0400
[libata] mv-ahci: Add Marvell PATA support
drivers/ata/ahci.h | 12 drivers/ata/libahci.c | 92 ++---- drivers/ata/mv-ahci.c | 748 +++++++++++++++++++++++++++++++++++++++++++++++++- 3 files changed, 782 insertions(+), 70 deletions(-)
drivers/ata/ahci.h +11 -1
@@ -129,7 +129,6 @@ enum {AHCI_HFLAG_IGN_IRQ_IF_ERR = (1 <<1 ),/* ignore IRQ_IF_ERR */ AHCI_HFLAG_IGN_SERR_INTERNAL = (1 <<2 ),/* ignore SERR_INTERNAL */ AHCI_HFLAG_32BIT_ONLY = (1 <<3 ),/* force 32bit */ -AHCI_HFLAG_MV_PATA = (1 <<4 ),/* PATA port */ AHCI_HFLAG_NO_MSI = (1 <<5 ),/* no PCI MSI */ AHCI_HFLAG_NO_PMP = (1 <<6 ),/* no PMP */ AHCI_HFLAG_NO_HOTPLUG = (1 << 7), /* ignore PxSERR.DIAG.N */
@@ -149,6 +148,8 @@ enum {EM_CTL_RST = (1 <<9 ),/* Reset */ EM_CTL_TM = (1 <<8 ),/* Transmit Message */ EM_CTL_ALHD = (1 <<26 ),/* Activity LED */ + + AHCI_FLAG_MV_PATA = (1 << 30), }; enum {
@@ -211,6 +212,9 @@ struct ahci_port_priv { unsigned int ncq_saw_dmas:1; unsigned int ncq_saw_sdb:1;u32 intr_mask;/* interrupts to enable */ + +int mvp_selected; +struct ahci_em_priv em_priv[AHCI_MAX_SLOTS ];/* enclosure management info per PM slot */ };
@@ -273,6 +277,12 @@ extern ssize_t ahci_activity_store(struct ata_device *dev, enum sw_activity val)extern ssize_t ahci_activity_show (struct ata_device *dev,char *buf);extern void ahci_print_info (struct ata_host *host);extern void ahci_port_intr (struct ata_port *ap); +extern void ahci_fill_cmd_slot (struct ahci_port_priv *pp,unsigned int tag,u32 opts); +extern unsigned int ahci_fill_sg (struct ata_queued_cmd *qc,void *cmd_tbl); +extern int ahci_deinit_port (struct ata_port *ap,const char **emsg); +extern void ahci_port_init (struct pci_dev *pdev,struct ata_port *ap, +int port_no,void __iomem *mmio, +void __iomem *port_mmio);#ifdef CONFIG_PM extern int ahci_port_suspend (struct ata_port *ap,pm_message_t mesg);
drivers/ata/libahci.c +33 -59
198 lines not on this page · raw message
drivers/ata/mv-ahci.c +738 -10
830 lines not on this page · raw message
PATCH 3/3 acard-ahci: Add new ACard ATP AHCI driver
In reply to Jeff Garzikcommit d67b67d0b5ec6a9ccd79fbc4082e84337a59dbf4
Author: Jeff Garzik <[email protected]> Date: Sun Apr 12 13:14:19 2009 -0400
libata: Add new 'acard-ahci' driver for ACard ATP8620 chips
drivers/ata/Kconfig | 9 + drivers/ata/Makefile | 1 drivers/ata/acard-ahci.c | 350 +++++++++++++++++++++++++++++++++++++++++++++++ drivers/ata/ahci.c | 94 ------------ drivers/ata/ahci.h | 3 drivers/ata/libahci.c | 98 +++++++++++++ 6 files changed, 462 insertions(+), 93 deletions(-)
drivers/ata/Kconfig +9 -0
@@ -55,6 +55,15 @@ config SATA_AHCI If unsure, say N. +config SATA_ACARD_AHCI + tristate "ACard AHCI variant (ATP 8620)" + depends on PCI + help + This option enables support for AHCI-like chips + from ACard: 8620 + + If unsure, say N. + config SATA_SIL24 tristate "Silicon Image 3124/3132 SATA support" depends on PCI
drivers/ata/Makefile +1 -0
@@ -2,6 +2,7 @@ obj-$(CONFIG_ATA) += libata.o obj-$(CONFIG_SATA_AHCI) += libahci.o ahci.o +obj-$(CONFIG_SATA_ACARD_AHCI) += libahci.o acard-ahci.o obj-$(CONFIG_SATA_SVW) += sata_svw.o obj-$(CONFIG_ATA_PIIX) += ata_piix.o obj-$(CONFIG_SATA_PROMISE) += sata_promise.o
drivers/ata/acard-ahci.c new file +350 -0
350 lines not on this page · raw message
drivers/ata/ahci.c +1 -93
107 lines not on this page · raw message
drivers/ata/ahci.h +3 -0
@@ -237,6 +237,8 @@ static inline int ahci_nr_ports(u32 cap)return (cap &0x1f ) +1 ; } +extern struct ata_port_operations ahci_base_ops; +extern void ahci_postreset (struct ata_link *link,unsigned int *class);extern void ahci_qc_prep (struct ata_queued_cmd *qc);extern void ahci_freeze (struct ata_port *ap);
@@ -277,6 +279,7 @@ extern ssize_t ahci_activity_store(struct ata_device *dev, enum sw_activity val)extern ssize_t ahci_activity_show (struct ata_device *dev,char *buf);extern void ahci_print_info (struct ata_host *host);extern void ahci_port_intr (struct ata_port *ap); +extern irqreturn_t ahci_interrupt (int irq,void *dev_instance);extern void ahci_fill_cmd_slot (struct ahci_port_priv *pp,unsigned int tag,u32 opts);extern unsigned int ahci_fill_sg (struct ata_queued_cmd *qc,void *cmd_tbl);extern int ahci_deinit_port (struct ata_port *ap,const char **emsg);
drivers/ata/libahci.c +98 -0
101 lines not on this page · raw message
Quoted 6 lines
Marvell PATA on 6121 and similar chips should also work, but that hasn't yet been tested. Here's the full list of what still needs testing:
- master + slave devices
- ATAPI
- 6121 (6123? others?)
LBA48
Quoted 3 lines
Also, there is the open question of how to deal with
pata_marvellco-existence
I think for the moment make PATA support by AHCI a config option and as soon as we trust it we can get rid of the option and of pata_marvell.
Alan
Quoted 8 lines · Alan Cox
Quoted 6 lines
Marvell PATA on 6121 and similar chips should also work, but that hasn't yet been tested. Here's the full list of what still needs testing:
- master + slave devices
- ATAPI
- 6121 (6123? others?)LBA48
And 80-pin cables... I cannot find where it permits cable detection or where DMA/PIO mode in AHCI mode, complicating things :(
Quoted 5 lines
Quoted 2 lines
Also, there is the open question of how to deal with
pata_marvellco-existenceI think for the moment make PATA support by AHCI a config option and as soon as we trust it we can get rid of the option and of
pata_marvell.
I'm not sure I see the worth of making PATA, solely, a config option?
Today, the choice is either pata_marvell (PATA) or ahci w/ module option (SATA).
My patches morph that choice into: pata_marvell (PATA) or mv-ahci (PATA + SATA).
Thus, considering the hardware, if the user disables PATA support in mv-ahci, they still have the problem of two drivers competing for the same PCI IDs - no matter that the drivers would each probe different bus types (PATA, or SATA).
We could skip SATA_MV_AHCI if PATA_MARVELL is set in .config... Now that mv-ahci solely contains Marvell PCI IDs, that would have the effect at compile-time that a module option would have at runtime.
Jeff
Quoted 11 lines · Alan Cox
Quoted 6 lines
Marvell PATA on 6121 and similar chips should also work, but that hasn't yet been tested. Here's the full list of what still needs testing:
- master + slave devices
- ATAPI
- 6121 (6123? others?)LBA48
Quoted 1 line
Quoted 5 lines
Quoted 2 lines
Also, there is the open question of how to deal with
pata_marvellco-existenceI think for the moment make PATA support by AHCI a config option and as soon as we trust it we can get rid of the option and of
pata_marvell.
How does this relate to 6101 which is a PATA only controller? Would that continue to use pata_marvell or does this new stuff support such things too?
Signature · 3 lines
Ville Syrjälä [email protected] http://www.sci.fi/~syrjala/
Quoted 10 lines · Ville Syrjälä
Quoted 9 lines · Alan Cox
Quoted 6 lines
Marvell PATA on 6121 and similar chips should also work, but that hasn't yet been tested. Here's the full list of what still needs testing:
- master + slave devices
- ATAPI
- 6121 (6123? others?)LBA48
Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
Also, there is the open question of how to deal with
pata_marvellco-existenceI think for the moment make PATA support by AHCI a config option and as soon as we trust it we can get rid of the option and of
pata_marvell.How does this relate to 6101 which is a PATA only controller? Would that continue to use
pata_marvellor does this new stuff support such things too?
I'm pretty sure 6101 will behave similarly to 6121, but have not verified.
Jeff
Quoted 6 lines · Jeff Garzik <jeff <at> garzik.org> writes:
This is an update of the recent libahci patchset, currently available on the 'libahci' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev.git
[...]
The mv-ahci code is obviously still a bit raw with "#if 0" in places, but it is nonetheless ready for testing.
I tested the libata-dev kernel as of commit c3939cc09. I had to comment out a bunch of code (for example, code related to the sntf board in ahci.c and some other stuff) and add a cap2 variable to the ahci_host_priv struct in ahci.h in order to successfully compile the branch.
I also had to comment out the pata_enable check in mv-ahci.c:mv_ahci_init_one in order to get the mv_ahci module to initialize my hardware, even though I have the pata_marvell module blacklisted.
My test hardware:
Intel D75XBX2 motherboard with 2 onboard SATA controllers
- ICH7 controller in AHCI mode
- 3 disks (2 x ST3320620AS, 1 x ST3500418AS)
- Marvell 88SE6145 controller, using AHCI (ahci.marvell_enable=1)
- 3 disks (3 x ST3500418AS)
I used "fio surface-scan" for my test (is that a good way to test this?) and I found no regressions from the stock Fedora 12 kernel 2.6.31.6-166.fc12.x86_64. However, I was hoping the latest libata-dev branch resolved this issue:
https://bugzilla.redhat.com/show_bug.cgi?id=549981
It did not. I continue to get the same error as described in my bugzilla report with libata-dev.
I hope that helps. I am willing to help out with other testing on this hardware, as necessary. Please cc me on replies.
Cheers,
Raman Gupta
Quoted 25 lines · Raman Gupta
Quoted 6 lines · Jeff Garzik<jeff<at> garzik.org> writes:
This is an update of the recent libahci patchset, currently available on the 'libahci' branch of
git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev.git
[...]
The mv-ahci code is obviously still a bit raw with "#if 0" in places, but it is nonetheless ready for testing.I tested the libata-dev kernel as of commit
c3939cc09. I had to comment out a bunch of code (for example, code related to the sntf board inahci.cand some other stuff) and add a cap2 variable to theahci_host_privstruct inahci.hin order to successfully compile the branch.I also had to comment out the
pata_enablecheck in mv-ahci.c:mv_ahci_init_one in order to get themv_ahcimodule to initialize my hardware, even though I have thepata_marvellmodule blacklisted.My test hardware:
Intel D75XBX2 motherboard with 2 onboard SATA controllers
- ICH7 controller in AHCI mode
- 3 disks (2 x ST3320620AS, 1 x ST3500418AS)
- Marvell 88SE6145 controller, using AHCI (ahci.marvell_enable=1)
- 3 disks (3 x ST3500418AS)
Quoted 8 lines
I used "fio surface-scan" for my test (is that a good way to test this?) and I found no regressions from the stock Fedora 12 kernel 2.6.31.6-166.fc12.x86_64. However, I was hoping the latest libata-dev branch resolved this issue:
https://bugzilla.redhat.com/show_bug.cgi?id=549981
It did not. I continue to get the same error as described in my bugzilla report with libata-dev.
I wouldn't expect any bug fixes to be in that branch, it's just a code reorganization.
From your last post on the Bugzilla report, it looks like all 3 drives basically stopped talking to the point they wouldn't respond to IDENTIFY commands. That seems really strange to me. You mentioned you were doing a surface scan at the time, which presumably would involve all disks being accessed simultaneously. I'd have a good look at the hardware on that system, specifically the power supply. We've seen a number of cases where running multiple HDs on a system can trigger such problems with SATA links because of voltage sags (even momentary). HDs draw much higher peak power under certain conditions than when idle so such problems may not be obvious unless you stress multiple drives at once.
I'm not sure if that's related to the SMART issue you were seeing or not..
Quoted 6 lines · Robert Hancock
Quoted 5 lines · Raman Gupta
I used "fio surface-scan" for my test (is that a good way to test this?) and I found no regressions from the stock Fedora 12 kernel 2.6.31.6-166.fc12.x86_64. However, I was hoping the latest libata-dev branch resolved this issue:
Quoted 7 lines
Quoted 4 lines
https://bugzilla.redhat.com/show_bug.cgi?id=549981
It did not. I continue to get the same error as described in my bugzilla report with libata-dev.
I wouldn't expect any bug fixes to be in that branch, it's just a code reorganization.
Ok. It looked to me like it contained various fixes plus the code reorg on top of a very recent kernel version (2.6.33-rc1). Is there another branch to test that is more likely to contain bug fixes that may solve my problem? I trolled around git.kernel.org and didn't see any.
Quoted 5 lines
From your last post on the Bugzilla report, it looks like all 3 drives
basically stopped talking to the point they wouldn't respond to IDENTIFY commands. That seems really strange to me. You mentioned you were doing a surface scan at the time, which presumably would involve all disks being accessed simultaneously.
Yes, all drives were being access simultaneously, as they are part of a 4-disk md RAID-5 array.
However, note that I can make the "exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x6 frozen" happen even with the RAID array stopped and no filesystems mounted. All I have to do is run the smartctl -a /dev/sdd command (sdd is attached to the Marvell controller) repeatedly until this exception occurs:
Dec 27 18:59:30 x kernel: ata6.00: exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x6 frozen
Dec 27 18:59:30 x kernel: ata6.00: cmd ec/00:01:00:00:00/00:00:00:00:00/00 tag 0 pio 512 in
Dec 27 18:59:30 x kernel: res 40/00:00:00:4f:c2/00:00:00:00:00/40 Emask 0x4 (timeout)
Dec 27 18:59:30 x kernel: ata6.00: status: { DRDY }
Dec 27 18:59:30 x kernel: ata6: hard resetting link
Dec 27 18:59:30 x kernel: ata6: SATA link up 3.0 Gbps (SStatus 123 SControl 300)
Dec 27 18:59:30 x kernel: ata6.00: configured for UDMA/133
Dec 27 18:59:30 x kernel: ata6: EH completeUsually 10-15 executions is sufficient to replicate the issue.
Quoted 6 lines
I'd have a good look at the hardware on that system, specifically the power supply. We've seen a number of cases where running multiple HDs on a system can trigger such problems with SATA links because of voltage sags (even momentary). HDs draw much higher peak power under certain conditions than when idle so such problems may not be obvious unless you stress multiple drives at once.
I used a multimeter to monitor the 5v and 12v output from the power supply for about two minutes, while the RAID array was under load, and didn't see the needle budge. Also note that I have three drives on the ICH7 controller and they have demonstrated no problems, nor has any other hardware in the system. I guess it could be a problem with the Marvell controller on this motherboard. I hope not.
Quoted 1 line
I'm not sure if that's related to the SMART issue you were seeing or not..
The timing is certainly suspicious. For now, given that I don't see any obvious hardware problems, I'm assuming the IDENTIFY error is related to the previous exception. If the primary issue can be solved then I'll retest to make sure the IDENTIFY error doesn't occur independently.
Is there some debugging I can turn on to get more information?
Cheers,
Raman
Quoted 9 lines · Raman Gupta
Quoted 5 lines · Robert Hancock
From your last post on the Bugzilla report, it looks like all 3 drives basically stopped talking to the point they wouldn't respond to IDENTIFY commands. That seems really strange to me. You mentioned you were doing a surface scan at the time, which presumably would involve all disks being accessed simultaneously.
Yes, all drives were being access simultaneously, as they are part of a 4-disk md RAID-5 array.
Very interesting: I re-ran the fio test, but this time I configured the smartd and hddtemp daemons to ignore the drives attached to the Marvell controller. The test completed with no errors.
It certainly seems as if disk commands executed by the smartd or hddtemp daemons are causing the exception.
Cheers,
Raman
Quoted 11 lines · Raman Gupta
However, note that I can make the "exception Emask
0x0SAct0x0SErr0x0action0x6frozen" happen even with the RAID array stopped and no filesystems mounted. All I have to do is run the smartctl -a/dev/sddcommand (sdd is attached to the Marvell controller) repeatedly until this exception occurs:Dec 27 18:59:30 x kernel: ata6.00: exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x6 frozen Dec 27 18:59:30 x kernel: ata6.00: cmd ec/00:01:00:00:00/00:00:00:00:00/00 tag 0 pio 512 in Dec 27 18:59:30 x kernel: res 40/00:00:00:4f:c2/00:00:00:00:00/40 Emask 0x4 (timeout) Dec 27 18:59:30 x kernel: ata6.00: status: { DRDY } Dec 27 18:59:30 x kernel: ata6: hard resetting link Dec 27 18:59:30 x kernel: ata6: SATA link up 3.0 Gbps (SStatus 123 SControl 300) Dec 27 18:59:30 x kernel: ata6.00: configured for UDMA/133 Dec 27 18:59:30 x kernel: ata6: EH complete
Quoted 1 line
Usually 10-15 executions is sufficient to replicate the issue.
Hmm. I wonder what running this script from this bug:
http://bugzilla.kernel.org/show_bug.cgi?id=14831
against drives attached to other controllers would do? It doesn't do anything particularly special - just runs smartctl in a loop while also writing to the same drive (via fs using dd).
Out of interest have you tried drives from other manufacturers? Would also be interested to see what happens if you run the script against the same drive, but attached to the ICH7?
Cheers,
Tim.
Marvell exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x6 frozen (was Re: [PATCH 0/3] AHCI updates: Marvell AHCI PATA works; pata_marvell fate?)
Quoted 22 lines · Tim Small
Quoted 19 lines · Raman Gupta
However, note that I can make the "exception Emask
0x0SAct0x0SErr0x0action0x6frozen" happen even with the RAID array stopped and no filesystems mounted. All I have to do is run the smartctl -a/dev/sddcommand (sdd is attached to the Marvell controller) repeatedly until this exception occurs:Dec 27 18:59:30 x kernel: ata6.00: exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x6 frozen Dec 27 18:59:30 x kernel: ata6.00: cmd ec/00:01:00:00:00/00:00:00:00:00/00 tag 0 pio 512 in Dec 27 18:59:30 x kernel: res 40/00:00:00:4f:c2/00:00:00:00:00/40 Emask 0x4 (timeout) Dec 27 18:59:30 x kernel: ata6.00: status: { DRDY } Dec 27 18:59:30 x kernel: ata6: hard resetting link Dec 27 18:59:30 x kernel: ata6: SATA link up 3.0 Gbps (SStatus 123 SControl 300) Dec 27 18:59:30 x kernel: ata6.00: configured for UDMA/133 Dec 27 18:59:30 x kernel: ata6: EH completeQuoted 1 line
Usually 10-15 executions is sufficient to replicate the issue.
Quoted 7 lines
Hmm. I wonder what running this script from this bug:
http://bugzilla.kernel.org/show_bug.cgi?id=14831
against drives attached to other controllers would do? It doesn't do anything particularly special - just runs smartctl in a loop while also writing to the same drive (via fs using dd).
Against a Seagate ST3500418AS on the Marvell controller, the script produced the first "smartctl failed" error in 55 seconds. Within about 8 minutes, everything went to pot and all drives on that controller were completely inaccessible (all filesystem writes failed and the kernel could not IDENTIFY the drives). As far as I can tell with my multimeter, voltages were stable.
Quoted 1 line
Out of interest have you tried drives from other manufacturers?
Unfortunately, at the moment I don't have any non-Seagate drives available.
Quoted 2 lines
Would also be interested to see what happens if you run the script against the same drive, but attached to the ICH7?
The problem occurs against any of the three Seagate ST3500418AS drives I have attached to the Marvell. Against the same model of drive attached to my ICH7 controller, I canceled the script after it ran for 1.5 hours without any problems. So the problem appears to be exclusive to the Marvell -- either the hardware or the driver.
Furthermore, over the last few days, I've had smartd and hddtemp turned off for the Marvell drives, and they have been stable and error-free.
Cheers,
Raman