Merge branch 'pci/aspm' into next

* pci/aspm:
PCI/ASPM: Add comment about L1 substate latency
PCI/ASPM: Configure L1 substate settings
PCI/ASPM: Calculate and save the L1.2 timing parameters
PCI/ASPM: Read and set up L1 substate capabilities
PCI/ASPM: Add support for L1 substates
PCI/ASPM: Add L1 substate capability structure register definitions

drivers/pci/pcie/Kconfig

···
 	  Enable PCI Express ASPM L0s and L1 where possible, even if the
 	  BIOS did not.
 
+config PCIEASPM_POWER_SUPERSAVE
+	bool "Power Supersave"
+	depends on PCIEASPM
+	help
+	  Same as PCIEASPM_POWERSAVE, except it also enables L1 substates where
+	  possible. This would result in higher power savings while staying in L1
+	  where the components support it.
+
 config PCIEASPM_PERFORMANCE
 	bool "Performance"
 	depends on PCIEASPM

drivers/pci/pcie/aspm.c

···
 #define ASPM_STATE_L0S_UP	(1)	/* Upstream direction L0s state */
 #define ASPM_STATE_L0S_DW	(2)	/* Downstream direction L0s state */
 #define ASPM_STATE_L1		(4)	/* L1 state */
+#define ASPM_STATE_L1_1		(8)	/* ASPM L1.1 state */
+#define ASPM_STATE_L1_2		(0x10)	/* ASPM L1.2 state */
+#define ASPM_STATE_L1_1_PCIPM	(0x20)	/* PCI PM L1.1 state */
+#define ASPM_STATE_L1_2_PCIPM	(0x40)	/* PCI PM L1.2 state */
+#define ASPM_STATE_L1_SS_PCIPM	(ASPM_STATE_L1_1_PCIPM | ASPM_STATE_L1_2_PCIPM)
+#define ASPM_STATE_L1_2_MASK	(ASPM_STATE_L1_2 | ASPM_STATE_L1_2_PCIPM)
+#define ASPM_STATE_L1SS		(ASPM_STATE_L1_1 | ASPM_STATE_L1_1_PCIPM |\
+				 ASPM_STATE_L1_2_MASK)
 #define ASPM_STATE_L0S		(ASPM_STATE_L0S_UP | ASPM_STATE_L0S_DW)
-#define ASPM_STATE_ALL		(ASPM_STATE_L0S | ASPM_STATE_L1)
+#define ASPM_STATE_ALL		(ASPM_STATE_L0S | ASPM_STATE_L1 |	\
+				 ASPM_STATE_L1SS)
+
+/*
+ * When L1 substates are enabled, the LTR L1.2 threshold is a timing parameter
+ * that decides whether L1.1 or L1.2 is entered (Refer PCIe spec for details).
+ * Not sure is there is a way to "calculate" this on the fly, but maybe we
+ * could turn it into a parameter in future.  This value has been taken from
+ * the following files from Intel's coreboot (which is the only code I found
+ * to have used this):
+ * https://www.coreboot.org/pipermail/coreboot-gerrit/2015-March/021134.html
+ * https://review.coreboot.org/#/c/8832/
+ */
+#define LTR_L1_2_THRESHOLD_BITS	((1 << 21) | (1 << 23) | (1 << 30))
 
 struct aspm_latency {
 	u32 l0s;			/* L0s latency (nsec) */
···
 
 struct pcie_link_state {
 	struct pci_dev *pdev;		/* Upstream component of the Link */
+	struct pci_dev *downstream;	/* Downstream component, function 0 */
 	struct pcie_link_state *root;	/* pointer to the root port link */
 	struct pcie_link_state *parent;	/* pointer to the parent Link state */
 	struct list_head sibling;	/* node in link_list */
···
 	struct list_head link;		/* node in parent's children list */
 
 	/* ASPM state */
-	u32 aspm_support:3;		/* Supported ASPM state */
-	u32 aspm_enabled:3;		/* Enabled ASPM state */
-	u32 aspm_capable:3;		/* Capable ASPM state with latency */
-	u32 aspm_default:3;		/* Default ASPM state by BIOS */
-	u32 aspm_disable:3;		/* Disabled ASPM state */
+	u32 aspm_support:7;		/* Supported ASPM state */
+	u32 aspm_enabled:7;		/* Enabled ASPM state */
+	u32 aspm_capable:7;		/* Capable ASPM state with latency */
+	u32 aspm_default:7;		/* Default ASPM state by BIOS */
+	u32 aspm_disable:7;		/* Disabled ASPM state */
 
 	/* Clock PM state */
 	u32 clkpm_capable:1;		/* Clock PM capable? */
···
 	 * has one slot under it, so at most there are 8 functions.
 	 */
 	struct aspm_latency acceptable[8];
+
+	/* L1 PM Substate info */
+	struct {
+		u32 up_cap_ptr;		/* L1SS cap ptr in upstream dev */
+		u32 dw_cap_ptr;		/* L1SS cap ptr in downstream dev */
+		u32 ctl1;		/* value to be programmed in ctl1 */
+		u32 ctl2;		/* value to be programmed in ctl2 */
+	} l1ss;
 };
 
 static int aspm_disabled, aspm_force;
···
 #define POLICY_DEFAULT 0	/* BIOS default setting */
 #define POLICY_PERFORMANCE 1	/* high performance */
 #define POLICY_POWERSAVE 2	/* high power saving */
+#define POLICY_POWER_SUPERSAVE 3 /* possibly even more power saving */
 
 #ifdef CONFIG_PCIEASPM_PERFORMANCE
 static int aspm_policy = POLICY_PERFORMANCE;
 #elif defined CONFIG_PCIEASPM_POWERSAVE
 static int aspm_policy = POLICY_POWERSAVE;
+#elif defined CONFIG_PCIEASPM_POWER_SUPERSAVE
+static int aspm_policy = POLICY_POWER_SUPERSAVE;
 #else
 static int aspm_policy;
 #endif
···
 static const char *policy_str[] = {
 	[POLICY_DEFAULT] = "default",
 	[POLICY_PERFORMANCE] = "performance",
-	[POLICY_POWERSAVE] = "powersave"
+	[POLICY_POWERSAVE] = "powersave",
+	[POLICY_POWER_SUPERSAVE] = "powersupersave"
 };
 
 #define LINK_RETRAIN_TIMEOUT HZ
···
 		return 0;
 	case POLICY_POWERSAVE:
 		/* Enable ASPM L0s/L1 */
+		return (ASPM_STATE_L0S | ASPM_STATE_L1);
+	case POLICY_POWER_SUPERSAVE:
+		/* Enable Everything */
 		return ASPM_STATE_ALL;
 	case POLICY_DEFAULT:
 		return link->aspm_default;
···
 		/* Disable ASPM and Clock PM */
 		return 0;
 	case POLICY_POWERSAVE:
-		/* Disable Clock PM */
+	case POLICY_POWER_SUPERSAVE:
+		/* Enable Clock PM */
 		return 1;
 	case POLICY_DEFAULT:
 		return link->clkpm_default;
···
 	return (1000 << encoding);
 }
 
+/* Convert L1SS T_pwr encoding to usec */
+static u32 calc_l1ss_pwron(struct pci_dev *pdev, u32 scale, u32 val)
+{
+	switch (scale) {
+	case 0:
+		return val * 2;
+	case 1:
+		return val * 10;
+	case 2:
+		return val * 100;
+	}
+	dev_err(&pdev->dev, "%s: Invalid T_PwrOn scale: %u\n",
+		__func__, scale);
+	return 0;
+}
+
 struct aspm_register_info {
 	u32 support:2;
 	u32 enabled:2;
 	u32 latency_encoding_l0s;
 	u32 latency_encoding_l1;
+
+	/* L1 substates */
+	u32 l1ss_cap_ptr;
+	u32 l1ss_cap;
+	u32 l1ss_ctl1;
+	u32 l1ss_ctl2;
 };
 
 static void pcie_get_aspm_reg(struct pci_dev *pdev,
···
 	info->latency_encoding_l1  = (reg32 & PCI_EXP_LNKCAP_L1EL) >> 15;
 	pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &reg16);
 	info->enabled = reg16 & PCI_EXP_LNKCTL_ASPMC;
+
+	/* Read L1 PM substate capabilities */
+	info->l1ss_cap = info->l1ss_ctl1 = info->l1ss_ctl2 = 0;
+	info->l1ss_cap_ptr = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_L1SS);
+	if (!info->l1ss_cap_ptr)
+		return;
+	pci_read_config_dword(pdev, info->l1ss_cap_ptr + PCI_L1SS_CAP,
+			      &info->l1ss_cap);
+	if (!(info->l1ss_cap & PCI_L1SS_CAP_L1_PM_SS)) {
+		info->l1ss_cap = 0;
+		return;
+	}
+	pci_read_config_dword(pdev, info->l1ss_cap_ptr + PCI_L1SS_CTL1,
+			      &info->l1ss_ctl1);
+	pci_read_config_dword(pdev, info->l1ss_cap_ptr + PCI_L1SS_CTL2,
+			      &info->l1ss_ctl2);
 }
 
 static void pcie_aspm_check_latency(struct pci_dev *endpoint)
···
 		 * Check L1 latency.
 		 * Every switch on the path to root complex need 1
 		 * more microsecond for L1. Spec doesn't mention L0s.
+		 *
+		 * The exit latencies for L1 substates are not advertised
+		 * by a device.  Since the spec also doesn't mention a way
+		 * to determine max latencies introduced by enabling L1
+		 * substates on the components, it is not clear how to do
+		 * a L1 substate exit latency check.  We assume that the
+		 * L1 exit latencies advertised by a device include L1
+		 * substate latencies (and hence do not do any check).
 		 */
 		latency = max_t(u32, link->latency_up.l1, link->latency_dw.l1);
 		if ((link->aspm_capable & ASPM_STATE_L1) &&
···
 
 		link = link->parent;
 	}
+}
+
+/*
+ * The L1 PM substate capability is only implemented in function 0 in a
+ * multi function device.
+ */
+static struct pci_dev *pci_function_0(struct pci_bus *linkbus)
+{
+	struct pci_dev *child;
+
+	list_for_each_entry(child, &linkbus->devices, bus_list)
+		if (PCI_FUNC(child->devfn) == 0)
+			return child;
+	return NULL;
+}
+
+/* Calculate L1.2 PM substate timing parameters */
+static void aspm_calc_l1ss_info(struct pcie_link_state *link,
+				struct aspm_register_info *upreg,
+				struct aspm_register_info *dwreg)
+{
+	u32 val1, val2, scale1, scale2;
+
+	link->l1ss.up_cap_ptr = upreg->l1ss_cap_ptr;
+	link->l1ss.dw_cap_ptr = dwreg->l1ss_cap_ptr;
+	link->l1ss.ctl1 = link->l1ss.ctl2 = 0;
+
+	if (!(link->aspm_support & ASPM_STATE_L1_2_MASK))
+		return;
+
+	/* Choose the greater of the two T_cmn_mode_rstr_time */
+	val1 = (upreg->l1ss_cap >> 8) & 0xFF;
+	val2 = (upreg->l1ss_cap >> 8) & 0xFF;
+	if (val1 > val2)
+		link->l1ss.ctl1 |= val1 << 8;
+	else
+		link->l1ss.ctl1 |= val2 << 8;
+	/*
+	 * We currently use LTR L1.2 threshold to be fixed constant picked from
+	 * Intel's coreboot.
+	 */
+	link->l1ss.ctl1 |= LTR_L1_2_THRESHOLD_BITS;
+
+	/* Choose the greater of the two T_pwr_on */
+	val1 = (upreg->l1ss_cap >> 19) & 0x1F;
+	scale1 = (upreg->l1ss_cap >> 16) & 0x03;
+	val2 = (dwreg->l1ss_cap >> 19) & 0x1F;
+	scale2 = (dwreg->l1ss_cap >> 16) & 0x03;
+
+	if (calc_l1ss_pwron(link->pdev, scale1, val1) >
+	    calc_l1ss_pwron(link->downstream, scale2, val2))
+		link->l1ss.ctl2 |= scale1 | (val1 << 3);
+	else
+		link->l1ss.ctl2 |= scale2 | (val2 << 3);
 }
 
 static void pcie_aspm_cap_init(struct pcie_link_state *link, int blacklist)
···
 
 	/* Get upstream/downstream components' register state */
 	pcie_get_aspm_reg(parent, &upreg);
-	child = list_entry(linkbus->devices.next, struct pci_dev, bus_list);
+	child = pci_function_0(linkbus);
 	pcie_get_aspm_reg(child, &dwreg);
+	link->downstream = child;
 
 	/*
 	 * If ASPM not supported, don't mess with the clocks and link,
···
 	link->latency_up.l1 = calc_l1_latency(upreg.latency_encoding_l1);
 	link->latency_dw.l1 = calc_l1_latency(dwreg.latency_encoding_l1);
 
+	/* Setup L1 substate */
+	if (upreg.l1ss_cap & dwreg.l1ss_cap & PCI_L1SS_CAP_ASPM_L1_1)
+		link->aspm_support |= ASPM_STATE_L1_1;
+	if (upreg.l1ss_cap & dwreg.l1ss_cap & PCI_L1SS_CAP_ASPM_L1_2)
+		link->aspm_support |= ASPM_STATE_L1_2;
+	if (upreg.l1ss_cap & dwreg.l1ss_cap & PCI_L1SS_CAP_PCIPM_L1_1)
+		link->aspm_support |= ASPM_STATE_L1_1_PCIPM;
+	if (upreg.l1ss_cap & dwreg.l1ss_cap & PCI_L1SS_CAP_PCIPM_L1_2)
+		link->aspm_support |= ASPM_STATE_L1_2_PCIPM;
+
+	if (upreg.l1ss_ctl1 & dwreg.l1ss_ctl1 & PCI_L1SS_CTL1_ASPM_L1_1)
+		link->aspm_enabled |= ASPM_STATE_L1_1;
+	if (upreg.l1ss_ctl1 & dwreg.l1ss_ctl1 & PCI_L1SS_CTL1_ASPM_L1_2)
+		link->aspm_enabled |= ASPM_STATE_L1_2;
+	if (upreg.l1ss_ctl1 & dwreg.l1ss_ctl1 & PCI_L1SS_CTL1_PCIPM_L1_1)
+		link->aspm_enabled |= ASPM_STATE_L1_1_PCIPM;
+	if (upreg.l1ss_ctl1 & dwreg.l1ss_ctl1 & PCI_L1SS_CTL1_PCIPM_L1_2)
+		link->aspm_enabled |= ASPM_STATE_L1_2_PCIPM;
+
+	if (link->aspm_support & ASPM_STATE_L1SS)
+		aspm_calc_l1ss_info(link, &upreg, &dwreg);
+
 	/* Save default state */
 	link->aspm_default = link->aspm_enabled;
 
···
 	}
 }
 
+static void pci_clear_and_set_dword(struct pci_dev *pdev, int pos,
+				    u32 clear, u32 set)
+{
+	u32 val;
+
+	pci_read_config_dword(pdev, pos, &val);
+	val &= ~clear;
+	val |= set;
+	pci_write_config_dword(pdev, pos, val);
+}
+
+/* Configure the ASPM L1 substates */
+static void pcie_config_aspm_l1ss(struct pcie_link_state *link, u32 state)
+{
+	u32 val, enable_req;
+	struct pci_dev *child = link->downstream, *parent = link->pdev;
+	u32 up_cap_ptr = link->l1ss.up_cap_ptr;
+	u32 dw_cap_ptr = link->l1ss.dw_cap_ptr;
+
+	enable_req = (link->aspm_enabled ^ state) & state;
+
+	/*
+	 * Here are the rules specified in the PCIe spec for enabling L1SS:
+	 * - When enabling L1.x, enable bit at parent first, then at child
+	 * - When disabling L1.x, disable bit at child first, then at parent
+	 * - When enabling ASPM L1.x, need to disable L1
+	 *   (at child followed by parent).
+	 * - The ASPM/PCIPM L1.2 must be disabled while programming timing
+	 *   parameters
+	 *
+	 * To keep it simple, disable all L1SS bits first, and later enable
+	 * what is needed.
+	 */
+
+	/* Disable all L1 substates */
+	pci_clear_and_set_dword(child, dw_cap_ptr + PCI_L1SS_CTL1,
+				PCI_L1SS_CTL1_L1SS_MASK, 0);
+	pci_clear_and_set_dword(parent, up_cap_ptr + PCI_L1SS_CTL1,
+				PCI_L1SS_CTL1_L1SS_MASK, 0);
+	/*
+	 * If needed, disable L1, and it gets enabled later
+	 * in pcie_config_aspm_link().
+	 */
+	if (enable_req & (ASPM_STATE_L1_1 | ASPM_STATE_L1_2)) {
+		pcie_capability_clear_and_set_word(child, PCI_EXP_LNKCTL,
+						   PCI_EXP_LNKCTL_ASPM_L1, 0);
+		pcie_capability_clear_and_set_word(parent, PCI_EXP_LNKCTL,
+						   PCI_EXP_LNKCTL_ASPM_L1, 0);
+	}
+
+	if (enable_req & ASPM_STATE_L1_2_MASK) {
+
+		/* Program T_pwr_on in both ports */
+		pci_write_config_dword(parent, up_cap_ptr + PCI_L1SS_CTL2,
+				       link->l1ss.ctl2);
+		pci_write_config_dword(child, dw_cap_ptr + PCI_L1SS_CTL2,
+				       link->l1ss.ctl2);
+
+		/* Program T_cmn_mode in parent */
+		pci_clear_and_set_dword(parent, up_cap_ptr + PCI_L1SS_CTL1,
+					0xFF00, link->l1ss.ctl1);
+
+		/* Program LTR L1.2 threshold in both ports */
+		pci_clear_and_set_dword(parent,	dw_cap_ptr + PCI_L1SS_CTL1,
+					0xE3FF0000, link->l1ss.ctl1);
+		pci_clear_and_set_dword(child, dw_cap_ptr + PCI_L1SS_CTL1,
+					0xE3FF0000, link->l1ss.ctl1);
+	}
+
+	val = 0;
+	if (state & ASPM_STATE_L1_1)
+		val |= PCI_L1SS_CTL1_ASPM_L1_1;
+	if (state & ASPM_STATE_L1_2)
+		val |= PCI_L1SS_CTL1_ASPM_L1_2;
+	if (state & ASPM_STATE_L1_1_PCIPM)
+		val |= PCI_L1SS_CTL1_PCIPM_L1_1;
+	if (state & ASPM_STATE_L1_2_PCIPM)
+		val |= PCI_L1SS_CTL1_PCIPM_L1_2;
+
+	/* Enable what we need to enable */
+	pci_clear_and_set_dword(parent, up_cap_ptr + PCI_L1SS_CTL1,
+				PCI_L1SS_CAP_L1_PM_SS, val);
+	pci_clear_and_set_dword(child, dw_cap_ptr + PCI_L1SS_CTL1,
+				PCI_L1SS_CAP_L1_PM_SS, val);
+}
+
 static void pcie_config_aspm_dev(struct pci_dev *pdev, u32 val)
 {
 	pcie_capability_clear_and_set_word(pdev, PCI_EXP_LNKCTL,
···
 static void pcie_config_aspm_link(struct pcie_link_state *link, u32 state)
 {
 	u32 upstream = 0, dwstream = 0;
-	struct pci_dev *child, *parent = link->pdev;
+	struct pci_dev *child = link->downstream, *parent = link->pdev;
 	struct pci_bus *linkbus = parent->subordinate;
 
-	/* Nothing to do if the link is already in the requested state */
+	/* Enable only the states that were not explicitly disabled */
 	state &= (link->aspm_capable & ~link->aspm_disable);
+
+	/* Can't enable any substates if L1 is not enabled */
+	if (!(state & ASPM_STATE_L1))
+		state &= ~ASPM_STATE_L1SS;
+
+	/* Spec says both ports must be in D0 before enabling PCI PM substates*/
+	if (parent->current_state != PCI_D0 || child->current_state != PCI_D0) {
+		state &= ~ASPM_STATE_L1_SS_PCIPM;
+		state |= (link->aspm_enabled & ASPM_STATE_L1_SS_PCIPM);
+	}
+
+	/* Nothing to do if the link is already in the requested state */
 	if (link->aspm_enabled == state)
 		return;
 	/* Convert ASPM state to upstream/downstream ASPM register state */
···
 		upstream |= PCI_EXP_LNKCTL_ASPM_L1;
 		dwstream |= PCI_EXP_LNKCTL_ASPM_L1;
 	}
+
+	if (link->aspm_capable & ASPM_STATE_L1SS)
+		pcie_config_aspm_l1ss(link, state);
+
 	/*
 	 * Spec 2.0 suggests all functions should be configured the
 	 * same setting for ASPM. Enabling ASPM L1 should be done in
···
 	 * the BIOS's expectation, we'll do so once pci_enable_device() is
 	 * called.
 	 */
-	if (aspm_policy != POLICY_POWERSAVE) {
+	if (aspm_policy != POLICY_POWERSAVE &&
+	    aspm_policy != POLICY_POWER_SUPERSAVE) {
 		pcie_config_aspm_path(link);
 		pcie_set_clkpm(link, policy_to_clkpm_state(link));
 	}
···
 	if (aspm_disabled || !link)
 		return;
 
-	if (aspm_policy != POLICY_POWERSAVE)
+	if (aspm_policy != POLICY_POWERSAVE &&
+	    aspm_policy != POLICY_POWER_SUPERSAVE)
 		return;
 
 	down_read(&pci_bus_sem);

include/uapi/linux/pci_regs.h

···
 #define PCI_EXT_CAP_ID_PMUX	0x1A	/* Protocol Multiplexing */
 #define PCI_EXT_CAP_ID_PASID	0x1B	/* Process Address Space ID */
 #define PCI_EXT_CAP_ID_DPC	0x1D	/* Downstream Port Containment */
+#define PCI_EXT_CAP_ID_L1SS	0x1E	/* L1 PM Substates */
 #define PCI_EXT_CAP_ID_PTM	0x1F	/* Precision Time Measurement */
 #define PCI_EXT_CAP_ID_MAX	PCI_EXT_CAP_ID_PTM
 
···
 #define PCI_PTM_CTRL			0x08	    /* PTM Control */
 #define  PCI_PTM_CTRL_ENABLE		0x00000001  /* PTM enable */
 #define  PCI_PTM_CTRL_ROOT		0x00000002  /* Root select */
+
+/* L1 PM Substates */
+#define PCI_L1SS_CAP		    4	/* capability register */
+#define  PCI_L1SS_CAP_PCIPM_L1_2	 1	/* PCI PM L1.2 Support */
+#define  PCI_L1SS_CAP_PCIPM_L1_1	 2	/* PCI PM L1.1 Support */
+#define  PCI_L1SS_CAP_ASPM_L1_2		 4	/* ASPM L1.2 Support */
+#define  PCI_L1SS_CAP_ASPM_L1_1		 8	/* ASPM L1.1 Support */
+#define  PCI_L1SS_CAP_L1_PM_SS		16	/* L1 PM Substates Support */
+#define PCI_L1SS_CTL1		    8	/* Control Register 1 */
+#define  PCI_L1SS_CTL1_PCIPM_L1_2	1	/* PCI PM L1.2 Enable */
+#define  PCI_L1SS_CTL1_PCIPM_L1_1	2	/* PCI PM L1.1 Support */
+#define  PCI_L1SS_CTL1_ASPM_L1_2	4	/* ASPM L1.2 Support */
+#define  PCI_L1SS_CTL1_ASPM_L1_1	8	/* ASPM L1.1 Support */
+#define  PCI_L1SS_CTL1_L1SS_MASK	0x0000000F
+#define PCI_L1SS_CTL2		    0xC	/* Control Register 2 */
 
 #endif /* LINUX_PCI_REGS_H */