The definition of ahci_port_standby() and ahci_port_spinup()
PATCH 3/6 The definition of ahci_port_standby() and ahci_port_spinup()
Put port power management[1], staggered-spinup[2], cold presence detection[3] related operations during the power state transition into ahci_port_standby() and ahci_port_spinup().
In particular ahci_port_standby() does the [1] and [2] when power transits to a lower level; ahci_port_spinup() does [1], [2] and [3] when power transits to a higher level.
---
drivers/scsi/ahci.c | 105 +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 files changed, 105 insertions(+), 0 deletions(-)
3d52b79eccec86284b92a074ae364d6576ed7c0a
drivers/scsi/ahci.c +105 -0
@@ -92,7 +92,9 @@ enum {HOST_AHCI_EN = (1 <<31 ),/* AHCI enabled */ /* HOST_CAP bits */ +HOST_CAP_SSC = (1 <<14 ),/* Slumber capable */ HOST_CAP_CLO = (1 <<24 ),/* Command List Override support */ +HOST_CAP_SSS = (1 <<27 ),/* Staggered Spin-up */ HOST_CAP_NCQ = (1 <<30 ),/* Native Command Queueing */ HOST_CAP_64 = (1 << 31), /* PCI DAC (64-bit DMA) support */
@@ -146,6 +148,7 @@ enum {/* PORT_CMD bits */ PORT_CMD_ATAPI = (1 <<24 ),/* Device is ATAPI */ +PORT_CMD_CPD = (1 <<20 ),/* Cold presence detection */ PORT_CMD_LIST_ON = (1 <<15 ),/* cmd list DMA engine running */ PORT_CMD_FIS_ON = (1 <<14 ),/* FIS DMA engine running */ PORT_CMD_FIS_RX = (1 << 4), /* Enable FIS receive DMA engine */
@@ -154,6 +157,7 @@ enum {PORT_CMD_SPIN_UP = (1 <<1 ),/* Spin up device */ PORT_CMD_START = (1 <<0 ),/* Enable port DMA engine */ +PORT_CMD_ICC_MASK = (0xf <<28 ),/* i/f ICC state mask */ PORT_CMD_ICC_ACTIVE = (0x1 <<28 ),/* Put i/f in active state */ PORT_CMD_ICC_PARTIAL = (0x2 <<28 ),/* Put i/f in partial state */ PORT_CMD_ICC_SLUMBER = (0x6 << 28), /* Put i/f in slumber state */
@@ -210,6 +214,8 @@ static int ahci_probe_reset(struct ata_pstatic void ahci_irq_clear (struct ata_port *ap);static int ahci_port_start (struct ata_port *ap);static void ahci_port_stop (struct ata_port *ap); +static int ahci_port_standby (void __iomem *port_mmio,u32 cap); +static int ahci_port_spinup (void __iomem *port_mmio,u32 cap);static void ahci_tf_read (struct ata_port *ap,struct ata_taskfile *tf);static void ahci_qc_prep (struct ata_queued_cmd *qc);static u8 ahci_check_status (struct ata_port *ap);
@@ -638,6 +644,105 @@ static void ahci_start_fis_rx(void __iomreadl (port_mmio +PORT_CMD );/* flush */ } +static int ahci_port_standby (void __iomem *port_mmio,u32 cap) +{ +u32 tmp, scontrol, sstatus; + + tmp =readl (port_mmio +PORT_CMD ); +/* +* AHCI Rev1.1 Section 5.3.2.3: +* Software is only allowed to program the PxCMD.FRE, +* PxCMD.POD, PxSCTL.DET, and PxCMD.SUD register bits +* when PxCMD.ST is set to '0' +*/ +if (tmp &PORT_CMD_START ) +return -EBUSY ; + +if (cap &HOST_CAP_SSC ) { +/* +* Enable transitions to slumber mode +*/ + scontrol =readl (port_mmio +PORT_SCR_CTL ); +if ((scontrol &0x0f00 ) >0x100 ) { + scontrol &= ~0xf00 ; +writel (scontrol, port_mmio +PORT_SCR_CTL ); + } +/* +* Put device into slumber mode +*/ + tmp |=PORT_CMD_ICC_SLUMBER ; +writel (tmp, port_mmio +PORT_CMD ); + tmp =readl (port_mmio +PORT_CMD ); + +/* +* Actually, we should wait for the device to +* enter slumber mode by checking +* sstatus & 0xf00 == 6 +*/ + sstatus =readl (port_mmio +PORT_SCR_STAT ); + } + +/* +* Put device into listen mode +*/ + scontrol =readl (port_mmio +PORT_SCR_CTL ); + scontrol &= ~0xf ; +writel (scontrol, port_mmio +PORT_SCR_CTL ); + + tmp =readl (port_mmio +PORT_CMD ); +if (cap &HOST_CAP_SSS ) { +/* +* Spin down the device for staggered spin-up support +*/ + tmp &= ~PORT_CMD_SPIN_UP ; +writel (tmp, port_mmio +PORT_CMD ); +readl (port_mmio +PORT_CMD );/* flush */ + } + +return 0 ; +} + +static int ahci_port_spinup (void __iomem *port_mmio,u32 cap) +{ +u32 tmp; + + tmp =readl (port_mmio +PORT_CMD ); +/* +* AHCI Rev1.1 Section 5.3.2.3: +* Software is only allowed to program the PxCMD.FRE, +* PxCMD.POD, PxSCTL.DET, and PxCMD.SUD register bits +* when PxCMD.ST is set to '0' +*/ +if (tmp &PORT_CMD_START ) +return -EBUSY ; + +/* +* Power on device if supported +*/ +if (tmp &PORT_CMD_CPD ) { + tmp |=PORT_CMD_POWER_ON ; +writel (tmp, port_mmio +PORT_CMD ); + tmp =readl (port_mmio +PORT_CMD ); + } + +/* +* Spin up device +*/ +if (cap &HOST_CAP_SSS ) { + tmp |=PORT_CMD_SPIN_UP ; +writel (tmp, port_mmio +PORT_CMD ); + tmp =readl (port_mmio +PORT_CMD ); + } + +if ((tmp &PORT_CMD_ICC_MASK ) !=PORT_CMD_ICC_ACTIVE ) { + tmp |=PORT_CMD_ICC_ACTIVE ; +writel (tmp, port_mmio +PORT_CMD ); + tmp =readl (port_mmio +PORT_CMD ); + } + +return 0 ; +} +static unsigned int ahci_dev_classify (struct ata_port *ap) {void __iomem *port_mmio = (void __iomem *) ap->ioaddr.cmd_addr;
Signature · 1 line
1.2.6
Quoted 8 lines · forrest
Put port power management[1], staggered-spinup[2], cold presence detection[3] related operations during the power state transition into
ahci_port_standby()andahci_port_spinup().In particular
ahci_port_standby()does the [1] and [2] when power transits to a lower level;ahci_port_spinup()does [1], [2] and [3] when power transits to a higher level.
Did you actually test with staggered-spinup and/or cold-presence detection? I just added them per spec but never got a chance to test them here.
And: it might be an idea to rename the HOST_CAP bits to be consistent with those printed during initialisation. It's bad enough as it is (ie the spec uses different abbrevs than the implementation), but we should try to be consistent within the implementation.
I would rename them to HOST_CAP_STAG and HOST_CAP_SLUM.
Add slumber mode: libata-core used to issue a port reset with an implicit SLUMBER/PARTIAL mode disable. So after a reset we might not be able to put the device into SLUMBER mode even if supported. What was the reasoning for this?
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke [email protected] SuSE Linux Products GmbH S390 & zSeries Maxfeldstraße 5 +49 911 74053 688 90409 Nürnberg http://www.suse.de
Quoted 3 lines · Hannes Reinecke
Did you actually test with staggered-spinup and/or cold-presence detection? I just added them per spec but never got a chance to test them here.
We have no mobile platform in our lab, which support staggered-spinup or cold presence detection. I would test it when the newest chipset is available.
Quoted 5 lines
And: it might be an idea to rename the
HOST_CAPbits to be consistent with those printed during initialisation. It's bad enough as it is (ie the spec uses different abbrevs than the implementation), but we should try to be consistent within the implementation.
I would rename them toHOST_CAP_STAGandHOST_CAP_SLUM.
Agree. When reading code, I need to translate the MACRO into the abbrevs in spec to understand the meaning.
Quoted 4 lines
Add slumber mode: libata-core used to issue a port reset with an implicit SLUMBER/PARTIAL mode disable. So after a reset we might not be able to put the device into SLUMBER mode even if supported. What was the reasoning for this?
I have the same question. Maybe the original author of libata-core.c can help answer it.
drivers/scsi/ahci.c static void ahci_start_fis_rx(void __iom · line 659
Quoted 13 lines · forrest
drivers/scsi/ahci.c +13
-0+static int ahci_port_standby (void __iomem *port_mmio,u32 cap) +{ +u32 tmp, scontrol, sstatus; + + tmp =readl (port_mmio +PORT_CMD ); +/* +* AHCI Rev1.1 Section 5.3.2.3: +* Software is only allowed to program the PxCMD.FRE, +* PxCMD.POD, PxSCTL.DET, and PxCMD.SUD register bits +* when PxCMD.ST is set to '0' +*/ +if (tmp &PORT_CMD_START ) +return -EBUSY ;
Same here.
Quoted 13 lines
drivers/scsi/ahci.c +13
-0+if (cap &HOST_CAP_SSC ) { +/* +* Enable transitions to slumber mode +*/ + scontrol =readl (port_mmio +PORT_SCR_CTL ); +if ((scontrol &0x0f00 ) >0x100 ) { + scontrol &= ~0xf00 ; +writel (scontrol, port_mmio +PORT_SCR_CTL ); + } +/* +* Put device into slumber mode +*/ + tmp |=PORT_CMD_ICC_SLUMBER ;
drivers/scsi/ahci.c static void ahci_start_fis_rx(void __iom · line 683
Quoted 10 lines
drivers/scsi/ahci.c +10
-0+writel (tmp, port_mmio +PORT_CMD ); + tmp =readl (port_mmio +PORT_CMD ); + +/* +* Actually, we should wait for the device to +* enter slumber mode by checking +* sstatus & 0xf00 == 6 +*/ + sstatus =readl (port_mmio +PORT_SCR_STAT ); + }
AHCI rev1.1 says at the beginning of 8.3.3 HBA D3 state,
"After the interface and device have been put into a low power state, the HBA may be put into a low power state."
But, it also says the following at the end of 8.2
"Note: The Phy is not required to be in a Slumber state when the device is in a D1, D2 or D3 state, nor is it required to be in a Slumber state when the HBA is in a D3 state. While this may be the likely condition of the interface when the devices connected to the interface are in a low power state, it is not requirement, and the interface shall break out of these states on a power management event."
My gut feeling is that above SLUMBER/SCR listen code is not strictly necessary, but I don't know whether we should keep or drop it.
drivers/scsi/ahci.c static void ahci_start_fis_rx(void __iom · line 690
Quoted 7 lines
+
drivers/scsi/ahci.c +6
-0+/* +* Put device into listen mode +*/ + scontrol =readl (port_mmio +PORT_SCR_CTL ); + scontrol &= ~0xf ; +writel (scontrol, port_mmio +PORT_SCR_CTL );
Above code is to put the HBA into listening mode, right? If so, please put above code into the following if block and update comment.
Review
Quoted 3 lines
+
(quoted diff) +2
-0+ tmp = readl(port_mmio + PORT_CMD); + if (cap & HOST_CAP_SSS) {
Why not use hpriv->cap?
drivers/scsi/ahci.c static void ahci_start_fis_rx(void __iom · line 699
Quoted 6 lines
drivers/scsi/ahci.c +6
-0+/* +* Spin down the device for staggered spin-up support +*/ + tmp &= ~PORT_CMD_SPIN_UP ; +writel (tmp, port_mmio +PORT_CMD ); +readl (port_mmio +PORT_CMD );/* flush */
Clearing PORT_CMD_SPIN_UP does not spin down the device. Please correct the comment.
Quoted 8 lines
drivers/scsi/ahci.c +8
-0+ } + +return 0 ; +} + +static int ahci_port_spinup (void __iomem *port_mmio,u32 cap) +{ +u32 tmp;
drivers/scsi/ahci.c static void ahci_start_fis_rx(void __iom · line 717
Quoted 10 lines
drivers/scsi/ahci.c +10
-0+ + tmp =readl (port_mmio +PORT_CMD ); +/* +* AHCI Rev1.1 Section 5.3.2.3: +* Software is only allowed to program the PxCMD.FRE, +* PxCMD.POD, PxSCTL.DET, and PxCMD.SUD register bits +* when PxCMD.ST is set to '0' +*/ +if (tmp &PORT_CMD_START ) +return -EBUSY ;
Ditto...
drivers/scsi/ahci.c static void ahci_start_fis_rx(void __iom · line 726
Quoted 8 lines
drivers/scsi/ahci.c +8
-0+/* +* Power on device if supported +*/ +if (tmp &PORT_CMD_CPD ) { + tmp |=PORT_CMD_POWER_ON ; +writel (tmp, port_mmio +PORT_CMD ); + tmp =readl (port_mmio +PORT_CMD ); + }
IMGO, it would be better to add cold presence detection support after we have such hardware and test the code on it.
Quoted 4 lines
drivers/scsi/ahci.c +4
-0+/* +* Spin up device +*/ +if (cap &HOST_CAP_SSS ) {
drivers/scsi/ahci.c static void ahci_start_fis_rx(void __iom · line 741
Quoted 10 lines
drivers/scsi/ahci.c +10
-0+ tmp |=PORT_CMD_SPIN_UP ; +writel (tmp, port_mmio +PORT_CMD ); + tmp =readl (port_mmio +PORT_CMD ); + } + +if ((tmp &PORT_CMD_ICC_MASK ) !=PORT_CMD_ICC_ACTIVE ) { + tmp |=PORT_CMD_ICC_ACTIVE ; +writel (tmp, port_mmio +PORT_CMD ); + tmp =readl (port_mmio +PORT_CMD ); + }
Again, I think it's better to do things like above unconditionally. ie. Just tell the controller to transit to active.
Signature · 1 line
tejun
General comment:
I don't have code+specs+testbox in front of me, so I can only offer general [from the airport :)] advice:
Under AHCI particularly, -not- disabling some power management modes has the effect of making runtime power management _too_ aggressive, leading to either a too-often parking of heads, or aggressively blinking the link, which sometimes causes COMRESET to be issued too often.
So libata has followed the general policy of always disabling runtime power management. This is _not_ a rule, just a starting point. Changes are welcome...... after those changes are verified to not have negative effects.
Jeff