More information on ATI IXP failure in git9
The system uses sata_sil with
00:11.0 IDE interface: ATI Technologies Inc ATI 437A Serial ATA Controller 00:12.0 IDE interface: ATI Technologies Inc ATI 4379 Serial ATA Controller
I added some printks and it fails here:
static void sil_host_intr(struct ata_port *ap, u32 bmdma2)
{
struct ata_queued_cmd *qc = ata_qc_from_tag(ap, ap->active_tag);
u8 status;
if (unlikely(bmdma2 & SIL_DMA_SATA_IRQ)) {
u32 serror;
...
<------ hits two times,
goto freeze;Appended a successfull boot from 2.6.16.
libata version 1.20 loaded.sata_sil 0000:00:11.0: version 0.9
GSI 19 sharing vector 0xC9 and IRQ 19
ACPI: PCI Interrupt 0000:00:11.0[A] -> GSI 23 (level, low) -> IRQ 201
ata1: SATA max UDMA/100 cmd 0xFFFFC20000012080 ctl 0xFFFFC2000001208A bmdma 0xFFFFC20000012000 irq 201
ata2: SATA max UDMA/100 cmd 0xFFFFC200000120C0 ctl 0xFFFFC200000120CA bmdma 0xFFFFC20000012008 irq 201
ata1: SATA link down (SStatus 0)
scsi0 : sata_sil
ata2: SATA link up 1.5 Gbps (SStatus 113)
ata2: dev 0 cfg 00:0040 49:2f00 82:746b 83:7f01 84:4023 85:7469 86:3c01 87:4023 88:40ff 93:0000
ata2: dev 0 ATA-7, max UDMA7, 488397168 sectors: LBA48sata_get_dev_handle: SATA dev addr=0x110000, handle=0x0000000000000000
ata2: dev 0 configured for UDMA/100sata_get_dev_handle: SATA dev addr=0x110000, handle=0x0000000000000000
scsi1 : sata_sil
Vendor: ATA Model: SAMSUNG SP2504C Rev: VT10
Type: Direct-Access ANSI SCSI revision: 05
ACPI: PCI Interrupt 0000:00:12.0[A] -> GSI 22 (level, low) -> IRQ 177
ata3: SATA max UDMA/100 cmd 0xFFFFC20000014080 ctl 0xFFFFC2000001408A bmdma 0xFFFFC20000014000 irq 177
ata4: SATA max UDMA/100 cmd 0xFFFFC200000140C0 ctl 0xFFFFC200000140CA bmdma 0xFFFFC20000014008 irq 177
ata3: SATA link down (SStatus 0)
scsi2 : sata_sil
ata4: SATA link down (SStatus 0)
scsi3 : sata_sil
Driver 'sd' needs updating - please use bus_type methods
SCSI device sda: 488397168 512-byte hdwr sectors (250059 MB)
sda: Write Protect is off
sda: Mode Sense: 00 3a 00 00
SCSI device sda: drive cache: write back
SCSI device sda: 488397168 512-byte hdwr sectors (250059 MB)
sda: Write Protect is off
sda: Mode Sense: 00 3a 00 00
SCSI device sda: drive cache: write back
sda: sda1 sda2 sda3
sd 1:0:0:0: Attached scsi disk sda
-Andi
Quoted 9 lines · Andi Kleen
The system uses
sata_silwith00:11.0 IDE interface: ATI Technologies Inc ATI 437A Serial ATA Controller 00:12.0 IDE interface: ATI Technologies Inc ATI 4379 Serial ATA Controller
I added some printks and it fails here:
static void
sil_host_intr(struct ata_port *ap, u32 bmdma2)
Quoted 12 lines
{
struct ata_queued_cmd *qc = ata_qc_from_tag(ap, ap->active_tag);
u8 status;
if (unlikely(bmdma2 & SIL_DMA_SATA_IRQ)) {
u32 serror;
...
<------ hits two times,
goto freeze;Kernel 2.6.17's sata_sil always masks the SATA phy interrupts, whereas 2.6.17-git sata_sil does not.
Can you print the hex value of 'serror' variable for us?
Jeff
Quoted 2 lines
Can you print the hex value of 'serror' variable for us?
It's always 0
-Andi
Andi created a patch at my suggestion:
http://one.firstfloor.org/~andi/sil-debug
and that got him mounting the device, but the EH was still giving him much pain:
http://one.firstfloor.org/~andi/sata-error-loop
It looks like some condition isn't getting ack'd...
Jeff
Quoted 8 lines · Jeff Garzik
Andi created a patch at my suggestion:
http://one.firstfloor.org/~andi/sil-debugand that got him mounting the device, but the EH was still giving him much pain:
http://one.firstfloor.org/~andi/sata-error-loopIt looks like some condition isn't getting ack'd...
Weird. To me, it almost seems like BMDMA2 is lying about whether it generated interrupt or not. e.g. It's sharing IRQ w/ another device, but when the other device raises interrupt, BMDMA2 says it generated the IRQ.
Once frozen, new EH does pretty good job of clearing status and the interrupts seem too sporadic to be unacked IRQ.
Andi, can you pleaes try the following patch? It makes sata_sil print more data on errors and disables SATA PHY IRQ.
Thanks.
drivers/scsi/sata_sil.c +18 -3
@@ -341,8 +341,11 @@ static void sil_scr_write (struct ata_postatic void sil_host_intr (struct ata_port *ap,u32 bmdma2) {struct ata_queued_cmd *qc =ata_qc_from_tag (ap, ap->active_tag); +u32 bmdma;u8 status; + bmdma =readl (ap->host_set->mmio_base + sil_port[ap->port_no].bmdma); +if (unlikely (bmdma2 &SIL_DMA_SATA_IRQ )) {u32 serror;
@@ -363,11 +366,18 @@ static void sil_host_intr(struct ata_por ap->eh_info.serror |= serror; } +ata_port_printk (ap,KERN_ERR , +"SIL freeze 1, bmdma=%x bmdma2=%x serr=%x\n" , + bmdma, bmdma2, serror); goto freeze; } -if (unlikely (!qc || qc->tf.ctl &ATA_NIEN )) +if (unlikely (!qc || qc->tf.ctl &ATA_NIEN )) { +ata_port_printk (ap,KERN_ERR , +"SIL freeze 2, bmdma=%x bmdma2=%x\n" , + bmdma, bmdma2); goto freeze; + }/* Check whether we are expecting interrupt in this state */ switch (ap->hsm_task_state) {
@@ -398,13 +408,18 @@ static void sil_host_intr(struct ata_porcase HSM_ST :break ;default : +ata_port_printk (ap,KERN_ERR ,"SIL hsm 1, bmdma=%x bmdma2=%x\n" , + bmdma, bmdma2); goto err_hsm; }/* check main status, clearing INTRQ */ status =ata_chk_status (ap); -if (unlikely (status &ATA_BUSY )) +if (unlikely (status &ATA_BUSY )) { +ata_port_printk (ap,KERN_ERR ,"SIL hsm 2, bmdma=%x bmdma2=%x\n" , + bmdma, bmdma2); goto err_hsm; + }/* ack bmdma irq events */ ata_bmdma_irq_clear (ap);
@@ -475,7 +490,7 @@ static void sil_thaw(struct ata_port *apata_bmdma_irq_clear (ap);/* turn on SATA IRQ */ -writel (SIL_SIEN_N , mmio_base + sil_port[ap->port_no].sien); +/*writel(SIL_SIEN_N, mmio_base + sil_port[ap->port_no].sien);*/ /* turn on IRQ */ tmp =readl (mmio_base +SIL_SYSCFG );
Quoted 4 lines · Tejun Heo
Quoted 3 lines · Jeff Garzik
Andi created a patch at my suggestion:
http://one.firstfloor.org/~andi/sil-debug
Quoted 8 lines
Quoted 5 lines
and that got him mounting the device, but the EH was still giving him much pain:
http://one.firstfloor.org/~andi/sata-error-loopIt looks like some condition isn't getting ack'd...
Weird. To me, it almost seems like BMDMA2 is lying about whether it generated interrupt or not. e.g. It's sharing IRQ w/ another device,
It shares the interrupt with a 8139 NIC
Quoted 8 lines
but when the other device raises interrupt, BMDMA2 says it generated the IRQ.
Once frozen, new EH does pretty good job of clearing status and the interrupts seem too sporadic to be unacked IRQ.
Andi, can you pleaes try the following patch? It makes
sata_silprint more data on errors and disables SATA PHY IRQ.
Still doesn't work with it - see http://one.firstfloor.org/~andi/sata3.jpg
For reference a working 2.6.16-suse boot again (I think the ports must be mislabelled on the mobo - i especially connected the disk to port 1 as printed on the PCB)
ata1: SATA max UDMA/100 cmd 0xFFFFC20000012080 ctl 0xFFFFC2000001208A bmdma 0xFFFFC20000012000 irq 201
ata2: SATA max UDMA/100 cmd 0xFFFFC200000120C0 ctl 0xFFFFC200000120CA bmdma 0xFFFFC20000012008 irq 201
ata1: SATA link down (SStatus 0)
scsi0 : sata_sil
ata2: SATA link down (SStatus 0)
scsi1 : sata_sil
ACPI: PCI Interrupt 0000:00:12.0[A] -> GSI 22 (level, low) -> IRQ 177
ata3: SATA max UDMA/100 cmd 0xFFFFC20000014080 ctl 0xFFFFC2000001408A bmdma 0xFFFFC20000014000 irq 177
ata4: SATA max UDMA/100 cmd 0xFFFFC200000140C0 ctl 0xFFFFC200000140CA bmdma 0xFFFFC20000014008 irq 177
ata3: SATA link up 1.5 Gbps (SStatus 113)
ata3: dev 0 cfg 00:0040 49:2f00 82:746b 83:7f01 84:4023 85:7469 86:3c01 87:4023 88:40ff 93:0000
ata3: dev 0 ATA-7, max UDMA7, 488397168 sectors: LBA48sata_get_dev_handle: SATA dev addr=0x120000, handle=0x0000000000000000
ata3: dev 0 configured for UDMA/100sata_get_dev_handle: SATA dev addr=0x120000, handle=0x0000000000000000
scsi2 : sata_sil
ata4: SATA link down (SStatus 0)
scsi3 : sata_sil
Vendor: ATA Model: SAMSUNG SP2504C Rev: VT10
Type: Direct-Access ANSI SCSI revision: 05
Driver 'sd' needs updating - please use bus_type methods
SCSI device sda: 488397168 512-byte hdwr sectors (250059 MB)
sda: Write Protect is off
sda: Mode Sense: 00 3a 00 00
SCSI device sda: drive cache: write back
SCSI device sda: 488397168 512-byte hdwr sectors (250059 MB)
sda: Write Protect is off
sda: Mode Sense: 00 3a 00 00
SCSI device sda: drive cache: write back
sda: sda1 sda2 sda3
sd 2:0:0:0: Attached scsi disk sdaehci_hcd: block sizes: qh 160 qtd 96 itd 192 sitd 96
-Andi
Hello, Andi.
Quoted 1 line · Andi Kleen
Still doesn't work with it - see http://one.firstfloor.org/~andi/sata3.jpg
* SATA_IRQ bit is stuck (on all other sil controllers, this gets cleared when SError is cleared)
* whenever interrupt occurs (from itself or from any other ones sharing the IRQ), irq handler sees SATA_IRQ bit set and thus thinks it just received phy status changed IRQ.
* phy status change aborts the active command, so no command gets completed.
I'm attaching two patches. The first one tries to clear SATA_IRQ by writing 1 to it in thaw() assuming the bit is implemented as W1C (which BTW is out of spec). The second one kills SATA_IRQ handling completely.
I wish the first one works but if not I'll update the second one such
that it applies only to ATI IXP.
Thanks.
Signature · 1 line
tejun
drivers/scsi/sata_sil.c +3 -0
@@ -474,6 +474,9 @@ static void sil_thaw(struct ata_port *apata_chk_status (ap);ata_bmdma_irq_clear (ap); +/* SATA_IRQ has to be cleared manually on some controllers */ +writel (SIL_DMA_SATA_IRQ , mmio_base + sil_port[ap->port_no].bmdma2); +/* turn on SATA IRQ */ writel (SIL_SIEN_N , mmio_base + sil_port[ap->port_no].sien);
drivers/scsi/sata_sil.c +4 -2
@@ -343,6 +343,7 @@ static void sil_host_intr(struct ata_porstruct ata_queued_cmd *qc =ata_qc_from_tag (ap, ap->active_tag);u8 status; +#if 0 if (unlikely (bmdma2 &SIL_DMA_SATA_IRQ )) {u32 serror;
@@ -365,6 +366,7 @@ static void sil_host_intr(struct ata_por goto freeze; } +#endif if (unlikely (!qc || qc->tf.ctl &ATA_NIEN )) goto freeze;
@@ -438,7 +440,7 @@ static irqreturn_t sil_interrupt(int irqcontinue ;if (bmdma2 ==0xffffffff || - !(bmdma2 & (SIL_DMA_COMPLETE |SIL_DMA_SATA_IRQ ))) + !(bmdma2 & (SIL_DMA_COMPLETE /* | SIL_DMA_SATA_IRQ*/ )))continue ;sil_host_intr (ap, bmdma2);
@@ -475,7 +477,7 @@ static void sil_thaw(struct ata_port *apata_bmdma_irq_clear (ap);/* turn on SATA IRQ */ -writel (SIL_SIEN_N , mmio_base + sil_port[ap->port_no].sien); +/*writel(SIL_SIEN_N, mmio_base + sil_port[ap->port_no].sien);*/ /* turn on IRQ */ tmp =readl (mmio_base +SIL_SYSCFG );
Quoted 7 lines · Tejun Heo
Hello, Andi.
Quoted 1 line · Andi Kleen
Still doesn't work with it - see http://one.firstfloor.org/~andi/sata3.jpg
*
SATA_IRQbit is stuck (on all other sil controllers, this gets cleared when SError is cleared)
Quoted 7 lines
* whenever interrupt occurs (from itself or from any other ones sharing the IRQ), irq handler sees
SATA_IRQbit set and thus thinks it just received phy status changed IRQ.* phy status change aborts the active command, so no command gets completed.
I'm attaching two patches. The first one tries to clear
SATA_IRQby writing 1 to it inthaw()assuming the bit is implemented as W1C (which BTW is out of spec). The second one killsSATA_IRQhandling completely.
I wish the first one works but if not I'll update the second one such
that it applies only to ATI IXP.
And another one, just in case they were crazy enough to implement the bit as RW.
Signature · 1 line
tejun
drivers/scsi/sata_sil.c +4 -0
@@ -474,6 +474,10 @@ static void sil_thaw(struct ata_port *apata_chk_status (ap);ata_bmdma_irq_clear (ap); +/* SATA_IRQ has to be cleared manually on some controllers */ +writel (SIL_DMA_SATA_IRQ , mmio_base + sil_port[ap->port_no].bmdma2); +writel (0 , mmio_base + sil_port[ap->port_no].bmdma2); +/* turn on SATA IRQ */ writel (SIL_SIEN_N , mmio_base + sil_port[ap->port_no].sien);
Quoted 17 lines · Tejun Heo
Quoted 16 lines · Tejun Heo
Hello, Andi.
Quoted 2 lines · Andi Kleen
Still doesn't work with it - see
http://one.firstfloor.org/~andi/sata3.jpg*
SATA_IRQbit is stuck (on all other sil controllers, this gets cleared when SError is cleared)* whenever interrupt occurs (from itself or from any other ones sharing the IRQ), irq handler sees
SATA_IRQbit set and thus thinks it just received phy status changed IRQ.* phy status change aborts the active command, so no command gets completed.
Quoted 8 lines
Quoted 5 lines
I'm attaching two patches. The first one tries to clear
SATA_IRQby writing 1 to it inthaw()assuming the bit is implemented as W1C (which BTW is out of spec). The second one killsSATA_IRQhandling completely.
I wish the first one works but if not I'll update the second one such
that it applies only to ATI IXP.And another one, just in case they were crazy enough to implement the bit as RW.
patch0 + this one also don't work (zero err_mask etc.)
-Andi
Quoted 12 lines · Tejun Heo
Hello, Andi.
Quoted 1 line · Andi Kleen
Still doesn't work with it - see http://one.firstfloor.org/~andi/sata3.jpg
*
SATA_IRQbit is stuck (on all other sil controllers, this gets cleared when SError is cleared)* whenever interrupt occurs (from itself or from any other ones sharing the IRQ), irq handler sees
SATA_IRQbit set and thus thinks it just received phy status changed IRQ.
Quoted 7 lines
* phy status change aborts the active command, so no command gets completed.
I'm attaching two patches. The first one tries to clear
SATA_IRQby writing 1 to it inthaw()assuming the bit is implemented as W1C (which BTW is out of spec). The second one killsSATA_IRQhandling completely.
I wish the first one works but if not I'll update the second one such
that it applies only to ATI IXP.
Patch 0 and 1 together work, patch 0 only runs into the usual zero err_mask problem and doesn't find root.
-Andi
PATCH 1/1 sata_sil: disable hotplug interrupts on two ATI IXPs
Some SATA controllers embedded in ATI IXPs seem to have broken SATA_IRQ bit in their bmdma2 registers which is always stuck at 1. This makes the driver believe that there has been a hotplug event and freeze the port whenever there's an interrupt thus failing all commands.
This patch disables SATA_IRQ for those controllers.
---
Andi, can you please verify this patch works? Jeff, if Andi confirms it, please push this to Andrew & Linus. I'll contact SIMG and try to find another way around this. Also, can somebody test 0x437a?
Thanks.
drivers/scsi/sata_sil.c +23 -6
@@ -52,6 +52,7 @@ enum {/* * host flags */ +SIL_FLAG_NO_SATA_IRQ = (1 <<28 ),SIL_FLAG_RERR_ON_DMA_ACT = (1 <<29 ), SIL_FLAG_MOD15WRITE = (1 << 30),
@@ -62,8 +63,9 @@ enum {* Controller IDs */ sil_3112 =0 , - sil_3512 =1 , - sil_3114 = 2, + sil_3112_no_sata_irq =1 , + sil_3512 =2 , + sil_3114 = 3,/* * Register offsets
@@ -123,8 +125,8 @@ static const struct pci_device_id sil_pc { 0x1095, 0x3512, PCI_ANY_ID, PCI_ANY_ID, 0, 0, sil_3512 }, { 0x1095, 0x3114, PCI_ANY_ID, PCI_ANY_ID, 0, 0, sil_3114 }, { 0x1002, 0x436e, PCI_ANY_ID, PCI_ANY_ID, 0, 0, sil_3112 }, - { 0x1002, 0x4379, PCI_ANY_ID, PCI_ANY_ID, 0, 0, sil_3112 }, - { 0x1002, 0x437a, PCI_ANY_ID, PCI_ANY_ID, 0, 0, sil_3112 }, + { 0x1002, 0x4379, PCI_ANY_ID, PCI_ANY_ID, 0, 0, sil_3112_no_sata_irq }, + { 0x1002, 0x437a, PCI_ANY_ID, PCI_ANY_ID, 0, 0, sil_3112_no_sata_irq }, { } /* terminate list */ };
@@ -217,6 +219,16 @@ static const struct ata_port_info sil_po .udma_mask = 0x3f, /* udma0-5 */ .port_ops = &sil_ops, }, +/* sil_3112_no_sata_irq */ + { + .sht = &sil_sht, + .host_flags = SIL_DFL_HOST_FLAGS | SIL_FLAG_MOD15WRITE | + SIL_FLAG_NO_SATA_IRQ, + .pio_mask = 0x1f, /* pio0-4 */ + .mwdma_mask = 0x07, /* mwdma0-2 */ + .udma_mask = 0x3f, /* udma0-5 */ + .port_ops = &sil_ops, + },/* sil_3512 */ { .sht = &sil_sht,
@@ -437,6 +449,10 @@ static irqreturn_t sil_interrupt(int irqif (unlikely (!ap || ap->flags &ATA_FLAG_DISABLED ))continue ; +/* turn off SATA_IRQ if not supported */ +if (ap->flags &SIL_FLAG_NO_SATA_IRQ ) + bmdma2 &= ~SIL_DMA_SATA_IRQ ; +if (bmdma2 ==0xffffffff || !(bmdma2 & (SIL_DMA_COMPLETE |SIL_DMA_SATA_IRQ )))continue ;
@@ -474,8 +490,9 @@ static void sil_thaw(struct ata_port *apata_chk_status (ap);ata_bmdma_irq_clear (ap); -/* turn on SATA IRQ */ -writel (SIL_SIEN_N , mmio_base + sil_port[ap->port_no].sien); +/* turn on SATA IRQ if supported */ +if (!(ap->flags &SIL_FLAG_NO_SATA_IRQ )) +writel (SIL_SIEN_N , mmio_base + sil_port[ap->port_no].sien);/* turn on IRQ */ tmp =readl (mmio_base +SIL_SYSCFG );
Re: [PATCH] sata_sil: disable hotplug interrupts on two ATI IXPs
Quoted 6 lines · Tejun Heo
Some SATA controllers embedded in ATI IXPs seem to have broken
SATA_IRQbit in their bmdma2 registers which is always stuck at 1. This makes the driver believe that there has been a hotplug event and freeze the port whenever there's an interrupt thus failing all commands.
Quoted 8 lines
This patch disables
SATA_IRQfor those controllers.Signed-off-by Tejun Heo <[email protected]>---
Andi, can you please verify this patch works? Jeff, if Andi confirms it, please push this to Andrew & Linus. I'll contact SIMG and try to find another way around this.
Works thanks.
-Andi
libata version 1.30 loaded.sata_sil 0000:00:11.0: version 1.0
GSI 19 sharing vector 0xC9 and IRQ 19
ACPI: PCI Interrupt 0000:00:11.0[A] -> GSI 23 (level, low) -> IRQ 201
ata1: SATA max UDMA/100 cmd 0xFFFFC2001008E080 ctl 0xFFFFC2001008E08A bmdma 0xFF FFC2001008E000 irq 201
ata2: SATA max UDMA/100 cmd 0xFFFFC2001008E0C0 ctl 0xFFFFC2001008E0CA bmdma 0xFF FFC2001008E008 irq 201
scsi0 : sata_sil
ata1: SATA link down (SStatus 0 SControl 300)
scsi1 : sata_sil
ata2: SATA link down (SStatus 0 SControl 300)
ACPI: PCI Interrupt 0000:00:12.0[A] -> GSI 22 (level, low) -> IRQ 177
ata3: SATA max UDMA/100 cmd 0xFFFFC20010090080 ctl 0xFFFFC2001009008A bmdma 0xFF FFC20010090000 irq 177
ata4: SATA max UDMA/100 cmd 0xFFFFC200100900C0 ctl 0xFFFFC200100900CA bmdma 0xFF FFC20010090008 irq 177
scsi2 : sata_sil
ata3: SATA link up 1.5 Gbps (SStatus 113 SControl 300)
ata3.00: configured for UDMA/100
scsi3 : sata_sil
ata4: SATA link down (SStatus 0 SControl 300)
Vendor: ATA Model: SAMSUNG SP2504C Rev: VT10
Type: Direct-Access ANSI SCSI revision: 05
SCSI device sda: 488397168 512-byte hdwr sectors (250059 MB)
sda: Write Protect is off
sda: Mode Sense: 00 3a 00 00
SCSI device sda: drive cache: write back
SCSI device sda: 488397168 512-byte hdwr sectors (250059 MB)
sda: Write Protect is off
sda: Mode Sense: 00 3a 00 00
SCSI device sda: drive cache: write back
sda: sda1 sda2 sda3
sd 2:0:0:0: Attached scsi disk sda
Jeff Garzik
AppliedRe: [PATCH] sata_sil: disable hotplug interrupts on two ATI IXPs
In reply to Tejun Heoapplied