Skip to contentSkip to conversation

x86: atom-punit/-pmc s2idle device state checks

PATCH In review Hans de Goede 2 participants 7 messages Mar 5, 2024 – Mar 6, 2024

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

v4 cover x86: atom-punit/-pmc s2idle device state checks

Hi All,

Here is v4 of this patch series to add S0ix related s2idle checks to the pmc_atom and punit_atom code.

Changes in v4:
- Add pmc_s2idle_check_register() helper which is a stub when CONFIG_SUSPEND is not set to avoid needing more then 1 ifdef
- Add a couple of Acked-by and Reviewed-by tags

Patches 1/5 and 5/5 have Acked-by-s from the clk resp. x86 maintainers to be merged through the pdx86 tree together with the rest of the series.

Ilpo, I believe that this is ready for merging now. It would be great if you can still pick this up for the 6.9 cycle. OTOH there is no big rush to get these debug patches for older platforms merged. So postponing them to the next cycle is fine too.

Below is the previous cover-letter of v3 with some more background info:

These patches are an upstream submission of a patch titled:
"Intel Atom suspend: add debug check for S0ix blockers"

Which I have been carrying in my personal kernel tree for years now. This code originally comes from the latte-l-oss branch of:
https://github.com/MiCode/Xiaomi_Kernel_OpenSource

And has been posted on upstream mailinglists before by Johannes Stezenbach, whose authorship I have kept for the 2 base patches and has been reposted by Takashi Iwai and at one point in time I picked this up from Takashi's reposting as can be seen from the S-o-b lines. Unfortunately I cannot find the original postings, so I have no link to those.

The original version of this added some ugly hooks into the intel_idle driver which I presume is why these patches never got anywhere upstream.

With the new acpi_s2idle_dev_ops and acpi_register_lps0_dev() functionality this functionality can now be implemented cleanly and that is what this patch-series does.

Changes in v3:
- Reword commit message of patch 3/5 and 5/5
- Drop confusing /* Low Part */ and /* High Part */ comments in pmc_atom.c
- Add punit_s2idle_check_[un]register() helper functions

Changes in v2:
- Move CLK reg defines to include/linux/platform_data/x86/pmc_atom.h
- Drop duplicated "pmc_atom: " prefix from logged messages

Regards,

Hans

Hans de Goede (3):
clk: x86: Move clk-pmc-atom register defines to
include/linux/platform_data/x86/pmc_atom.h
platform/x86: pmc_atom: Annotate d3_sts register bit defines platform/x86: pmc_atom: Check state of PMC clocks on s2idle

Johannes Stezenbach (2):
platform/x86: pmc_atom: Check state of PMC managed devices on s2idle x86/platform/atom: Check state of Punit managed devices on s2idle

 arch/x86/platform/atom/punit_atom_debug.c  | 54 ++++++++++++++-
 drivers/clk/x86/clk-pmc-atom.c             | 13 +---
 drivers/platform/x86/pmc_atom.c            | 79 ++++++++++++++++++++++
 include/linux/platform_data/x86/pmc_atom.h | 25 +++++--
 4 files changed, 152 insertions(+), 19 deletions(-)
Signature · 1 line
2.43.2
#1 lore raw

PATCH v4 1/5 clk: x86: Move clk-pmc-atom register defines to include/linux/platform_data/x86/pmc_atom.h

Move the register defines for the Atom (Bay Trail, Cherry Trail) PMC clocks to include/linux/platform_data/x86/pmc_atom.h.

This is a preparation patch to extend the S0i3 readiness checks in drivers/platform/x86/pmc_atom.c with checking that the PMC clocks are off on suspend entry.

Note these are added to include/linux/platform_data/x86/pmc_atom.h rather then to include/linux/platform_data/x86/clk-pmc-atom.h because the former already has all the other Atom PMC register defines.

---
Changes in v2:
- This is a new patch in v2 of this series

---
 drivers/clk/x86/clk-pmc-atom.c             | 13 +------------
 include/linux/platform_data/x86/pmc_atom.h | 13 +++++++++++++
 2 files changed, 14 insertions(+), 12 deletions(-)
drivers/clk/x86/clk-pmc-atom.c +1 -12
index 2974dd0ec6f4..5ec9255e33fa 100644
@@ -11,23 +11,12 @@
 #include <linux/err.h>
 #include <linux/io.h>
 #include <linux/platform_data/x86/clk-pmc-atom.h>
+#include <linux/platform_data/x86/pmc_atom.h>
 #include <linux/platform_device.h>
 #include <linux/slab.h>
 
 #define PLT_CLK_NAME_BASE	"pmc_plt_clk"
 
-#define PMC_CLK_CTL_OFFSET		0x60
-#define PMC_CLK_CTL_SIZE		4
-#define PMC_CLK_NUM			6
-#define PMC_CLK_CTL_GATED_ON_D3		0x0
-#define PMC_CLK_CTL_FORCE_ON		0x1
-#define PMC_CLK_CTL_FORCE_OFF		0x2
-#define PMC_CLK_CTL_RESERVED		0x3
-#define PMC_MASK_CLK_CTL		GENMASK(1, 0)
-#define PMC_MASK_CLK_FREQ		BIT(2)
-#define PMC_CLK_FREQ_XTAL		(0 << 2)	/* 25 MHz */
-#define PMC_CLK_FREQ_PLL		(1 << 2)	/* 19.2 MHz */
-
 struct clk_plt_fixed {
 	struct clk_hw *clk;
 	struct clk_lookup *lookup;
include/linux/platform_data/x86/pmc_atom.h +13 -0
index b8a701c77fd0..557622ef0390 100644
@@ -43,6 +43,19 @@
 				BIT_ORED_DEDICATED_IRQ_GPSC | \
 				BIT_SHARED_IRQ_GPSS)
 
+/* External clk generator settings */
+#define PMC_CLK_CTL_OFFSET		0x60
+#define PMC_CLK_CTL_SIZE		4
+#define PMC_CLK_NUM			6
+#define PMC_CLK_CTL_GATED_ON_D3		0x0
+#define PMC_CLK_CTL_FORCE_ON		0x1
+#define PMC_CLK_CTL_FORCE_OFF		0x2
+#define PMC_CLK_CTL_RESERVED		0x3
+#define PMC_MASK_CLK_CTL		GENMASK(1, 0)
+#define PMC_MASK_CLK_FREQ		BIT(2)
+#define PMC_CLK_FREQ_XTAL		(0 << 2)	/* 25 MHz */
+#define PMC_CLK_FREQ_PLL		(1 << 2)	/* 19.2 MHz */
+
 /* The timers accumulate time spent in sleep state */
 #define	PMC_S0IR_TMR		0x80
 #define	PMC_S0I1_TMR		0x84
Signature · 1 line
2.43.2
#2 lore raw

PATCH v4 2/5 platform/x86: pmc_atom: Annotate d3_sts register bit defines

In reply to Hans de Goede

The include/linux/platform_data/x86/pmc_atom.h d3_sts register bit defines are named after how these bits are used on Bay Trail devices.

On Cherry Trail (CHT) devices some of these bits have a different meaning according to the datasheet.

At a comment to the defines for bits which have a different meaning on Cherry Trail devices.

---
 include/linux/platform_data/x86/pmc_atom.h | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
include/linux/platform_data/x86/pmc_atom.h +6 -6
index 557622ef0390..161e4bc1c9ee 100644
@@ -117,14 +117,14 @@
 #define	BIT_SCC_SDIO		BIT(9)
 #define	BIT_SCC_SDCARD		BIT(10)
 #define	BIT_SCC_MIPI		BIT(11)
-#define	BIT_HDA			BIT(12)
+#define	BIT_HDA			BIT(12) /* CHT datasheet: reserved */
 #define	BIT_LPE			BIT(13)
 #define	BIT_OTG			BIT(14)
-#define	BIT_USH			BIT(15)
-#define	BIT_GBE			BIT(16)
-#define	BIT_SATA		BIT(17)
-#define	BIT_USB_EHCI		BIT(18)
-#define	BIT_SEC			BIT(19)
+#define	BIT_USH			BIT(15) /* CHT datasheet: reserved */
+#define	BIT_GBE			BIT(16) /* CHT datasheet: reserved */
+#define	BIT_SATA		BIT(17) /* CHT datasheet: reserved */
+#define	BIT_USB_EHCI		BIT(18) /* CHT datasheet: XHCI!    */
+#define	BIT_SEC			BIT(19) /* BYT datasheet: reserved */
 #define	BIT_PCIE_PORT0		BIT(20)
 #define	BIT_PCIE_PORT1		BIT(21)
 #define	BIT_PCIE_PORT2		BIT(22)
Signature · 1 line
2.43.2
#3 lore raw

PATCH v4 3/5 platform/x86: pmc_atom: Check state of PMC managed devices on s2idle

In reply to Hans de Goede

From: Johannes Stezenbach <[email protected]>

For the Bay Trail or Cherry Trail SoC to enter the S0i3 power-level at s2idle suspend requires most of the hw-blocks / devices in the SoC to be in D3 when entering s2idle suspend.

If some devices are not in D3 then the SoC will stay in a higher power state, consuming much more power from the battery then in S0i3.

Use the new acpi_s2idle_dev_ops and acpi_register_lps0_dev() functionality to register a new s2idle check function which checks that all hardware blocks in the South complex (controlled by the PMC) are in a state that allows the SoC to enter S0i3 and prints an error message for any device in D0.

Some blocks are not used on lower-featured versions of the SoC and these blocks will always report being in D0 on SoCs were they are not used. A false-positive mask is used to identify these blocks and for blocks in this mask the error is turned into a debug message to avoid false-positive error messages.

Note the pmc_atom code is enabled by CONFIG_X86_INTEL_LPSS which already depends on ACPI.

[hdegoede: Use acpi_s2idle_dev_ops, ignore fused off blocks, PMIC I2C]

---
Changes in v4:
- Add pmc_s2idle_check_register() helper which is a stub when CONFIG_SUSPEND is not set to avoid needing more then 1 ifdef

Changes in v3:
- Reword commit message
- Drop confusing /* Low Part */ and /* High Part */ comments

Changes in v2:
- Drop duplicated "pmc_atom: " prefix from pr_err() / pr_dbg() messages

---
 drivers/platform/x86/pmc_atom.c | 68 +++++++++++++++++++++++++++++++++
 1 file changed, 68 insertions(+)
drivers/platform/x86/pmc_atom.c +68 -0
index 93a6414c6611..7d391bd36eae 100644
@@ -6,6 +6,7 @@
 
 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 
+#include <linux/acpi.h>
 #include <linux/debugfs.h>
 #include <linux/device.h>
 #include <linux/dmi.h>
@@ -17,6 +18,7 @@
 #include <linux/platform_device.h>
 #include <linux/pci.h>
 #include <linux/seq_file.h>
+#include <linux/suspend.h>
 
 struct pmc_bit_map {
 	const char *name;
@@ -448,6 +450,71 @@ static int pmc_setup_clks(struct pci_dev *pdev, void __iomem *pmc_regmap,
 	return 0;
 }
 
+#ifdef CONFIG_SUSPEND
+static void pmc_dev_state_check(u32 sts, const struct pmc_bit_map *sts_map,
+				u32 fd, const struct pmc_bit_map *fd_map,
+				u32 sts_possible_false_pos)
+{
+	int index;
+
+	for (index = 0; sts_map[index].name; index++) {
+		if (!(fd_map[index].bit_mask & fd) &&
+		    !(sts_map[index].bit_mask & sts)) {
+			if (sts_map[index].bit_mask & sts_possible_false_pos)
+				pm_pr_dbg("%s is in D0 prior to s2idle\n",
+					  sts_map[index].name);
+			else
+				pr_err("%s is in D0 prior to s2idle\n",
+				       sts_map[index].name);
+		}
+	}
+}
+
+static void pmc_s2idle_check(void)
+{
+	struct pmc_dev *pmc = &pmc_device;
+	const struct pmc_reg_map *m = pmc->map;
+	u32 func_dis, func_dis_2;
+	u32 d3_sts_0, d3_sts_1;
+	u32 false_pos_sts_0, false_pos_sts_1;
+
+	func_dis = pmc_reg_read(pmc, PMC_FUNC_DIS);
+	func_dis_2 = pmc_reg_read(pmc, PMC_FUNC_DIS_2);
+	d3_sts_0 = pmc_reg_read(pmc, PMC_D3_STS_0);
+	d3_sts_1 = pmc_reg_read(pmc, PMC_D3_STS_1);
+
+	/*
+	 * Some blocks are not used on lower-featured versions of the SoC and
+	 * always report D0, add these to false_pos mask to log at debug level.
+	 */
+	if (m->d3_sts_1	== byt_d3_sts_1_map) {
+		/* Bay Trail */
+		false_pos_sts_0 = BIT_GBE | BIT_SATA | BIT_PCIE_PORT0 |
+			BIT_PCIE_PORT1 | BIT_PCIE_PORT2 | BIT_PCIE_PORT3 |
+			BIT_LPSS2_F5_I2C5;
+		false_pos_sts_1 = BIT_SMB | BIT_USH_SS_PHY | BIT_DFX;
+	} else {
+		/* Cherry Trail */
+		false_pos_sts_0 = BIT_GBE | BIT_SATA | BIT_LPSS2_F7_I2C7;
+		false_pos_sts_1 = BIT_SMB | BIT_STS_ISH;
+	}
+
+	pmc_dev_state_check(d3_sts_0, m->d3_sts_0, func_dis, m->func_dis, false_pos_sts_0);
+	pmc_dev_state_check(d3_sts_1, m->d3_sts_1, func_dis_2, m->func_dis_2, false_pos_sts_1);
+}
+
+static struct acpi_s2idle_dev_ops pmc_s2idle_ops = {
+	.check = pmc_s2idle_check,
+};
+
+static void pmc_s2idle_check_register(void)
+{
+	acpi_register_lps0_dev(&pmc_s2idle_ops);
+}
+#else
+static void pmc_s2idle_check_register(void) {}
+#endif
+
 static int pmc_setup_dev(struct pci_dev *pdev, const struct pci_device_id *ent)
 {
 	struct pmc_dev *pmc = &pmc_device;
@@ -485,6 +552,7 @@ static int pmc_setup_dev(struct pci_dev *pdev, const struct pci_device_id *ent)
 		dev_warn(&pdev->dev, "platform clocks register failed: %d\n",
 			 ret);
 
+	pmc_s2idle_check_register();
 	pmc->init = true;
 	return ret;
 }
Signature · 1 line
2.43.2
#4 lore raw

PATCH v4 4/5 platform/x86: pmc_atom: Check state of PMC clocks on s2idle

In reply to Hans de Goede

Extend the s2idle check with checking that none of the PMC clocks is in the forced-on state. If one of the clocks is in forced on state then S0i3 cannot be reached.

---
Changes in v3:
- Improve comment for clocks check

Changes in v2:
- Drop the PMC_CLK_* defines these are defined in include/linux/platform_data/x86/pmc_atom.h now
- Drop duplicated "pmc_atom: " prefix from pr_err() message

---
 drivers/platform/x86/pmc_atom.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
drivers/platform/x86/pmc_atom.c +11 -0
index 7d391bd36eae..0aa7076bc9cc 100644
@@ -477,6 +477,7 @@ static void pmc_s2idle_check(void)
 	u32 func_dis, func_dis_2;
 	u32 d3_sts_0, d3_sts_1;
 	u32 false_pos_sts_0, false_pos_sts_1;
+	int i;
 
 	func_dis = pmc_reg_read(pmc, PMC_FUNC_DIS);
 	func_dis_2 = pmc_reg_read(pmc, PMC_FUNC_DIS_2);
@@ -501,6 +502,16 @@ static void pmc_s2idle_check(void)
 
 	pmc_dev_state_check(d3_sts_0, m->d3_sts_0, func_dis, m->func_dis, false_pos_sts_0);
 	pmc_dev_state_check(d3_sts_1, m->d3_sts_1, func_dis_2, m->func_dis_2, false_pos_sts_1);
+
+	/* Forced-on PMC clocks prevent S0i3 */
+	for (i = 0; i < PMC_CLK_NUM; i++) {
+		u32 ctl = pmc_reg_read(pmc, PMC_CLK_CTL_OFFSET + 4 * i);
+
+		if ((ctl & PMC_MASK_CLK_CTL) != PMC_CLK_CTL_FORCE_ON)
+			continue;
+
+		pr_err("clock %d is ON prior to freeze (ctl 0x%08x)\n", i, ctl);
+	}
 }
 
 static struct acpi_s2idle_dev_ops pmc_s2idle_ops = {
Signature · 1 line
2.43.2
#5 lore raw

PATCH v4 5/5 x86/platform/atom: Check state of Punit managed devices on s2idle

In reply to Hans de Goede

From: Johannes Stezenbach <[email protected]>

For the Bay Trail or Cherry Trail SoC to enter the S0i3 power-level at s2idle suspend requires most of the hw-blocks / devices in the SoC to be in D3 when entering s2idle suspend.

If some devices are not in D3 then the SoC will stay in a higher power state, consuming much more power from the battery then in S0i3.

Use the new acpi_s2idle_dev_ops and acpi_register_lps0_dev() functionality to register a new s2idle check function which checks that all hardware blocks in the North complex (controlled by Punit) are in a state that allows the SoC to enter S0i3 and prints an error message for any device in D0.

---
Changes in v3:
- Reword commit message
- Add punit_s2idle_check_[un]register() helper functions

---
 arch/x86/platform/atom/punit_atom_debug.c | 54 ++++++++++++++++++++++-
 1 file changed, 53 insertions(+), 1 deletion(-)
arch/x86/platform/atom/punit_atom_debug.c +53 -1
index f8ed5f66cd20..6b9c6deca8ba 100644
@@ -7,6 +7,9 @@
  * Copyright (c) 2015, Intel Corporation.
  */
 
+#define pr_fmt(fmt) "punit_atom: " fmt
+
+#include <linux/acpi.h>
 #include <linux/module.h>
 #include <linux/init.h>
 #include <linux/device.h>
@@ -117,6 +120,51 @@ static void punit_dbgfs_unregister(void)
 	debugfs_remove_recursive(punit_dbg_file);
 }
 
+#if defined(CONFIG_ACPI) && defined(CONFIG_SUSPEND)
+static const struct punit_device *punit_dev;
+
+static void punit_s2idle_check(void)
+{
+	const struct punit_device *punit_devp;
+	u32 punit_pwr_status, dstate;
+	int status;
+
+	for (punit_devp = punit_dev; punit_devp->name; punit_devp++) {
+		/* Skip MIO, it is on till the very last moment */
+		if (punit_devp->reg == MIO_SS_PM)
+			continue;
+
+		status = iosf_mbi_read(BT_MBI_UNIT_PMC, MBI_REG_READ,
+				       punit_devp->reg, &punit_pwr_status);
+		if (status) {
+			pr_err("%s read failed\n", punit_devp->name);
+		} else  {
+			dstate = (punit_pwr_status >> punit_devp->sss_pos) & 3;
+			if (!dstate)
+				pr_err("%s is in D0 prior to s2idle\n", punit_devp->name);
+		}
+	}
+}
+
+static struct acpi_s2idle_dev_ops punit_s2idle_ops = {
+	.check = punit_s2idle_check,
+};
+
+static void punit_s2idle_check_register(struct punit_device *punit_device)
+{
+	punit_dev = punit_device;
+	acpi_register_lps0_dev(&punit_s2idle_ops);
+}
+
+static void punit_s2idle_check_unregister(void)
+{
+	acpi_unregister_lps0_dev(&punit_s2idle_ops);
+}
+#else
+static void punit_s2idle_check_register(struct punit_device *punit_device) {}
+static void punit_s2idle_check_unregister(void) {}
+#endif
+
 #define X86_MATCH(model, data)						 \
 	X86_MATCH_VENDOR_FAM_MODEL_FEATURE(INTEL, 6, INTEL_FAM6_##model, \
 					   X86_FEATURE_MWAIT, data)
@@ -131,19 +179,23 @@ MODULE_DEVICE_TABLE(x86cpu, intel_punit_cpu_ids);
 
 static int __init punit_atom_debug_init(void)
 {
+	struct punit_device *punit_device;
 	const struct x86_cpu_id *id;
 
 	id = x86_match_cpu(intel_punit_cpu_ids);
 	if (!id)
 		return -ENODEV;
 
-	punit_dbgfs_register((struct punit_device *)id->driver_data);
+	punit_device = (struct punit_device *)id->driver_data;
+	punit_dbgfs_register(punit_device);
+	punit_s2idle_check_register(punit_device);
 
 	return 0;
 }
 
 static void __exit punit_atom_debug_exit(void)
 {
+	punit_s2idle_check_unregister();
 	punit_dbgfs_unregister();
 }
 
Signature · 1 line
2.43.2
maintainer
#6 lore raw
In reply to Hans de Goede
Quoted 9 lines · Hans de Goede

Here is v4 of this patch series to add S0ix related s2idle checks to the pmc_atom and punit_atom code.

Changes in v4:
- Add pmc_s2idle_check_register() helper which is a stub when CONFIG_SUSPEND is not set to avoid needing more then 1 ifdef
- Add a couple of Acked-by and Reviewed-by tags

[...]

Thank you for your contribution, it has been applied to my local review-ilpo branch. Note it will show up in the public platform-drivers-x86/review-ilpo branch only once I've pushed my local branch there, which might take a while.

The list of commits applied:
[1/5] clk: x86: Move clk-pmc-atom register defines to include/linux/platform_data/x86/pmc_atom.h

      commit: 2ad8bda51c6069fa2d616ed57dc3ba8a3ec26ca4
[2/5] platform/x86: pmc_atom: Annotate d3_sts register bit defines
      commit: 5c3fc0ff939b0390128dfa852fc56b89016c0884
[3/5] platform/x86: pmc_atom: Check state of PMC managed devices on s2idle
      commit: cc7922fff7ee652577165c4fe6035429ccc175cc
[4/5] platform/x86: pmc_atom: Check state of PMC clocks on s2idle
      commit: cbb1c5f2b02da78169b6bacf506777f99ed07596
[5/5] x86/platform/atom: Check state of Punit managed devices on s2idle
      commit: ea2b3b2292fed64ed9b8451c03d638d61902badc

--
i.