thr3ads.net - Nouveau - [Nouveau] [PATCH 0/4] nouveau fixes for RPM/Optimus-related hangs [May 2016]

If this information is useful, please help other people find it:
Share via:

Peter Wu

2016-May-24 22:52 UTC

[Nouveau] [PATCH 0/4] nouveau fixes for RPM/Optimus-related hangs

Hi,

Here are two patches to fix an issue reported on kernel bugzilla (infinite loop
due to unchecked function) and a more important fix to fix hanging Optimus
machines when runtime PM is enabled (with pm/pci patches).

An older (obsolete) patch for the first issue was tested by the reporter:
https://bugzilla.kernel.org/show_bug.cgi?id=104791#c11
(it is replaced by "check for function 0x1B before using it").

The second issue will occur when:
 - A modern Optimus laptop is in use (designed for Windows 8 or newer).
 - nouveau runtime PM is enabled (1 or the default -1).
 - The patch "PCI: Add runtime PM support for PCIe ports" from Mika is
pulled
   into v4.7 (or v4.8[1]?) via the pci/pm branch,
  
https://git.kernel.org/cgit/linux/kernel/git/helgaas/pci.git/commit/?h=pci/pm&id=8b71f5652eeac561acf883da01ab4810f763ee42
(see also the discussion for "[PATCH] PCI: Power on bridges before scanning
new
devices" at
http://article.gmane.org/gmane.linux.power-management.general/76411)

The first two patches are just refactoring to reduce code duplication (and
scratch an itch) and make the following patches possible. The next two patches
fix the problems reported above.

I intend to get these patches in 4.7 (or the first version where pci/pm gets
merged) to avoid a lockup when runpm is enabled. Note:
 - If the fourth patch is merged before/without Mika's PCIe port patch, then
   those modern Optimus machines above will not be put into D3cold.
 - If the fourth patch is not merged (or merged after Mika's patch), then
under
   the above conditions the affected machine can lock up.
 - The three other patches are unrelated to this issue and can safely be merged.

Tested with:
 - Linux v4.6 + pci/pm + these four patches
 - Hardware: Clevo P651RA with acpi_osi="!Windows 2015" (the latter is
a
   workaround for another PCIe issue).
 - Card is asleep, woke up with lspci, waited a bit and retried/suspended:
   - # lspci -xxxxnnvvvv >/dev/null; sleep 5
   - # lspci -xxxxnnvvvv >/dev/null; sleep 5; systemctl suspend
   - # lspci -xxxxnnvvvv >/dev/null; systemctl suspend

Kind regards,
Peter

 [1]: https://lkml.kernel.org/r/20160524211309.GH1789 at lahna.fi.intel.com

Peter Wu (4):
  drm/nouveau/acpi: ensure matching ACPI handle and supported functions
  drm/nouveau/acpi: return supported DSM functions
  drm/nouveau/acpi: check for function 0x1B before using it
  drm/nouveau/acpi: fix lockup with PCIe runtime PM

 drivers/gpu/drm/nouveau/nouveau_acpi.c | 100 +++++++++++++++++++++------------
 1 file changed, 63 insertions(+), 37 deletions(-)

-- 
2.8.2

Peter Wu

2016-May-24 22:52 UTC

head link

[Nouveau] [PATCH 1/4] drm/nouveau/acpi: ensure matching ACPI handle and supported functions

Ensure that the returned set of supported DSM functions (MUX, Optimus)
match the ACPI handle that is set in nouveau_dsm_pci_probe.

As there are no machines with a MUX function on just one PCI device and
an Optimus on another, there should not be a functional impact. This
change however makes this implicit assumption more obvious.

Convert int to bool and rename has_dsm to has_mux while at it.

Signed-off-by: Peter Wu <peter at lekensteyn.nl>
---
 drivers/gpu/drm/nouveau/nouveau_acpi.c | 55 ++++++++++++++--------------------
 1 file changed, 23 insertions(+), 32 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_acpi.c
b/drivers/gpu/drm/nouveau/nouveau_acpi.c
index cdf5227..45fa9b2 100644
--- a/drivers/gpu/drm/nouveau/nouveau_acpi.c
+++ b/drivers/gpu/drm/nouveau/nouveau_acpi.c
@@ -57,9 +57,6 @@ bool nouveau_is_v1_dsm(void) {
 	return nouveau_dsm_priv.dsm_detected;
 }
 
-#define NOUVEAU_DSM_HAS_MUX 0x1
-#define NOUVEAU_DSM_HAS_OPT 0x2
-
 #ifdef CONFIG_VGA_SWITCHEROO
 static const char nouveau_dsm_muid[] = {
 	0xA0, 0xA0, 0x95, 0x9D, 0x60, 0x00, 0x48, 0x4D,
@@ -212,26 +209,33 @@ static const struct vga_switcheroo_handler
nouveau_dsm_handler = {
 	.get_client_id = nouveau_dsm_get_client_id,
 };
 
-static int nouveau_dsm_pci_probe(struct pci_dev *pdev)
+static void nouveau_dsm_pci_probe(struct pci_dev *pdev, bool *has_mux,
+				  bool *has_opt)
 {
 	acpi_handle dhandle;
-	int retval = 0;
+	bool supports_mux;
+	bool supports_opt;
 
 	dhandle = ACPI_HANDLE(&pdev->dev);
 	if (!dhandle)
-		return false;
+		return;
 
 	if (!acpi_has_method(dhandle, "_DSM"))
-		return false;
+		return;
+
+	supports_mux = acpi_check_dsm(dhandle, nouveau_dsm_muid, 0x00000102,
+				      1 << NOUVEAU_DSM_POWER);
+	supports_opt = nouveau_check_optimus_dsm(dhandle);
 
-	if (acpi_check_dsm(dhandle, nouveau_dsm_muid, 0x00000102,
-			   1 << NOUVEAU_DSM_POWER))
-		retval |= NOUVEAU_DSM_HAS_MUX;
+	/* Does not look like a Nvidia device. */
+	if (!supports_mux && !supports_opt)
+		return;
 
-	if (nouveau_check_optimus_dsm(dhandle))
-		retval |= NOUVEAU_DSM_HAS_OPT;
+	nouveau_dsm_priv.dhandle = dhandle;
+	*has_mux = supports_mux;
+	*has_opt = supports_opt;
 
-	if (retval & NOUVEAU_DSM_HAS_OPT) {
+	if (supports_opt) {
 		uint32_t result;
 		nouveau_optimus_dsm(dhandle, NOUVEAU_DSM_OPTIMUS_CAPS, 0,
 				    &result);
@@ -240,10 +244,6 @@ static int nouveau_dsm_pci_probe(struct pci_dev *pdev)
 			 (result & OPTIMUS_DYNAMIC_PWR_CAP) ? "dynamic power, " :
"",
 			 (result & OPTIMUS_HDA_CODEC_MASK) ? "hda bios codec
supported" : "");
 	}
-	if (retval)
-		nouveau_dsm_priv.dhandle = dhandle;
-
-	return retval;
 }
 
 static bool nouveau_dsm_detect(void)
@@ -251,11 +251,10 @@ static bool nouveau_dsm_detect(void)
 	char acpi_method_name[255] = { 0 };
 	struct acpi_buffer buffer = {sizeof(acpi_method_name), acpi_method_name};
 	struct pci_dev *pdev = NULL;
-	int has_dsm = 0;
-	int has_optimus = 0;
+	bool has_mux = false;
+	bool has_optimus = false;
 	int vga_count = 0;
 	bool guid_valid;
-	int retval;
 	bool ret = false;
 
 	/* lookup the MXM GUID */
@@ -268,32 +267,24 @@ static bool nouveau_dsm_detect(void)
 	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev)) != NULL)
{
 		vga_count++;
 
-		retval = nouveau_dsm_pci_probe(pdev);
-		if (retval & NOUVEAU_DSM_HAS_MUX)
-			has_dsm |= 1;
-		if (retval & NOUVEAU_DSM_HAS_OPT)
-			has_optimus = 1;
+		nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus);
 	}
 
 	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_3D << 8, pdev)) != NULL)
{
 		vga_count++;
 
-		retval = nouveau_dsm_pci_probe(pdev);
-		if (retval & NOUVEAU_DSM_HAS_MUX)
-			has_dsm |= 1;
-		if (retval & NOUVEAU_DSM_HAS_OPT)
-			has_optimus = 1;
+		nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus);
 	}
 
 	/* find the optimus DSM or the old v1 DSM */
-	if (has_optimus == 1) {
+	if (has_optimus) {
 		acpi_get_name(nouveau_dsm_priv.dhandle, ACPI_FULL_PATHNAME,
 			&buffer);
 		printk(KERN_INFO "VGA switcheroo: detected Optimus DSM method %s
handle\n",
 			acpi_method_name);
 		nouveau_dsm_priv.optimus_detected = true;
 		ret = true;
-	} else if (vga_count == 2 && has_dsm && guid_valid) {
+	} else if (vga_count == 2 && has_mux && guid_valid) {
 		acpi_get_name(nouveau_dsm_priv.dhandle, ACPI_FULL_PATHNAME,
 			&buffer);
 		printk(KERN_INFO "VGA switcheroo: detected DSM switching method %s
handle\n",
-- 
2.8.2

Peter Wu

2016-May-24 22:52 UTC

head link

[Nouveau] [PATCH 2/4] drm/nouveau/acpi: return supported DSM functions

Return the set of supported functions to the caller. No functional
changes.

Signed-off-by: Peter Wu <peter at lekensteyn.nl>
---
 drivers/gpu/drm/nouveau/nouveau_acpi.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_acpi.c
b/drivers/gpu/drm/nouveau/nouveau_acpi.c
index 45fa9b2..71d5e6a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_acpi.c
+++ b/drivers/gpu/drm/nouveau/nouveau_acpi.c
@@ -107,7 +107,7 @@ static int nouveau_optimus_dsm(acpi_handle handle, int func,
int arg, uint32_t *
  * requirements on the fourth parameter, so a private implementation
  * instead of using acpi_check_dsm().
  */
-static int nouveau_check_optimus_dsm(acpi_handle handle)
+static int nouveau_dsm_get_optimus_functions(acpi_handle handle)
 {
 	int result;
 
@@ -122,7 +122,9 @@ static int nouveau_check_optimus_dsm(acpi_handle handle)
 	 * ACPI Spec v4 9.14.1: if bit 0 is zero, no function is supported.
 	 * If the n-th bit is enabled, function n is supported
 	 */
-	return result & 1 && result & (1 <<
NOUVEAU_DSM_OPTIMUS_CAPS);
+	if (result & 1 && result & (1 <<
NOUVEAU_DSM_OPTIMUS_CAPS))
+		return result;
+	return 0;
 }
 
 static int nouveau_dsm(acpi_handle handle, int func, int arg)
@@ -214,7 +216,7 @@ static void nouveau_dsm_pci_probe(struct pci_dev *pdev, bool
*has_mux,
 {
 	acpi_handle dhandle;
 	bool supports_mux;
-	bool supports_opt;
+	int optimus_funcs;
 
 	dhandle = ACPI_HANDLE(&pdev->dev);
 	if (!dhandle)
@@ -225,17 +227,17 @@ static void nouveau_dsm_pci_probe(struct pci_dev *pdev,
bool *has_mux,
 
 	supports_mux = acpi_check_dsm(dhandle, nouveau_dsm_muid, 0x00000102,
 				      1 << NOUVEAU_DSM_POWER);
-	supports_opt = nouveau_check_optimus_dsm(dhandle);
+	optimus_funcs = nouveau_dsm_get_optimus_functions(dhandle);
 
 	/* Does not look like a Nvidia device. */
-	if (!supports_mux && !supports_opt)
+	if (!supports_mux && !optimus_funcs)
 		return;
 
 	nouveau_dsm_priv.dhandle = dhandle;
 	*has_mux = supports_mux;
-	*has_opt = supports_opt;
+	*has_opt = !!optimus_funcs;
 
-	if (supports_opt) {
+	if (optimus_funcs) {
 		uint32_t result;
 		nouveau_optimus_dsm(dhandle, NOUVEAU_DSM_OPTIMUS_CAPS, 0,
 				    &result);
-- 
2.8.2

Peter Wu

2016-May-24 22:53 UTC

head link

[Nouveau] [PATCH 3/4] drm/nouveau/acpi: check for function 0x1B before using it

Do not unconditionally invoke function 0x1B without checking for its
availability, it leads to an infinite loop on some firmware.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=104791
Fixes: 5addcf0a5f0fad ("nouveau: add runtime PM support (v0.9)")
Signed-off-by: Peter Wu <peter at lekensteyn.nl>
---
 drivers/gpu/drm/nouveau/nouveau_acpi.c | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_acpi.c
b/drivers/gpu/drm/nouveau/nouveau_acpi.c
index 71d5e6a..df9f73e 100644
--- a/drivers/gpu/drm/nouveau/nouveau_acpi.c
+++ b/drivers/gpu/drm/nouveau/nouveau_acpi.c
@@ -45,6 +45,7 @@
 static struct nouveau_dsm_priv {
 	bool dsm_detected;
 	bool optimus_detected;
+	bool optimus_flags_detected;
 	acpi_handle dhandle;
 	acpi_handle rom_handle;
 } nouveau_dsm_priv;
@@ -212,7 +213,7 @@ static const struct vga_switcheroo_handler
nouveau_dsm_handler = {
 };
 
 static void nouveau_dsm_pci_probe(struct pci_dev *pdev, bool *has_mux,
-				  bool *has_opt)
+				  bool *has_opt, bool *has_opt_flags)
 {
 	acpi_handle dhandle;
 	bool supports_mux;
@@ -236,6 +237,7 @@ static void nouveau_dsm_pci_probe(struct pci_dev *pdev, bool
*has_mux,
 	nouveau_dsm_priv.dhandle = dhandle;
 	*has_mux = supports_mux;
 	*has_opt = !!optimus_funcs;
+	*has_opt_flags = optimus_funcs & (1 << NOUVEAU_DSM_OPTIMUS_FLAGS);
 
 	if (optimus_funcs) {
 		uint32_t result;
@@ -255,6 +257,7 @@ static bool nouveau_dsm_detect(void)
 	struct pci_dev *pdev = NULL;
 	bool has_mux = false;
 	bool has_optimus = false;
+	bool has_optimus_flags = false;
 	int vga_count = 0;
 	bool guid_valid;
 	bool ret = false;
@@ -269,13 +272,15 @@ static bool nouveau_dsm_detect(void)
 	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev)) != NULL)
{
 		vga_count++;
 
-		nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus);
+		nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus,
+				      &has_optimus_flags);
 	}
 
 	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_3D << 8, pdev)) != NULL)
{
 		vga_count++;
 
-		nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus);
+		nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus,
+				      &has_optimus_flags);
 	}
 
 	/* find the optimus DSM or the old v1 DSM */
@@ -285,6 +290,7 @@ static bool nouveau_dsm_detect(void)
 		printk(KERN_INFO "VGA switcheroo: detected Optimus DSM method %s
handle\n",
 			acpi_method_name);
 		nouveau_dsm_priv.optimus_detected = true;
+		nouveau_dsm_priv.optimus_flags_detected = has_optimus_flags;
 		ret = true;
 	} else if (vga_count == 2 && has_mux && guid_valid) {
 		acpi_get_name(nouveau_dsm_priv.dhandle, ACPI_FULL_PATHNAME,
@@ -317,8 +323,9 @@ void nouveau_switcheroo_optimus_dsm(void)
 	if (!nouveau_dsm_priv.optimus_detected)
 		return;
 
-	nouveau_optimus_dsm(nouveau_dsm_priv.dhandle, NOUVEAU_DSM_OPTIMUS_FLAGS,
-			    0x3, &result);
+	if (nouveau_dsm_priv.optimus_flags_detected)
+		nouveau_optimus_dsm(nouveau_dsm_priv.dhandle, NOUVEAU_DSM_OPTIMUS_FLAGS,
+				    0x3, &result);
 
 	nouveau_optimus_dsm(nouveau_dsm_priv.dhandle, NOUVEAU_DSM_OPTIMUS_CAPS,
 		NOUVEAU_DSM_OPTIMUS_SET_POWERDOWN, &result);
-- 
2.8.2

Peter Wu

2016-May-24 22:53 UTC

head link

[Nouveau] [PATCH 4/4] drm/nouveau/acpi: fix lockup with PCIe runtime PM

Since "PCI: Add runtime PM support for PCIe ports", the parent PCIe
port
can be runtime-suspended which disables power resources via ACPI. This
is incompatible with DSM, resulting in a GPU device which is still in D3
and locks up the kernel on resume.

Mirror the behavior of Windows 8 and newer[1] (as observed via an AMLi
debugger trace) and stop using the DSM functions for D3cold when power
resources are available on the parent PCIe port.

 [1]:
https://msdn.microsoft.com/windows/hardware/drivers/bringup/firmware-requirements-for-d3cold

Signed-off-by: Peter Wu <peter at lekensteyn.nl>
---
 drivers/gpu/drm/nouveau/nouveau_acpi.c | 34 ++++++++++++++++++++++++++++++----
 1 file changed, 30 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_acpi.c
b/drivers/gpu/drm/nouveau/nouveau_acpi.c
index df9f73e..e469df7 100644
--- a/drivers/gpu/drm/nouveau/nouveau_acpi.c
+++ b/drivers/gpu/drm/nouveau/nouveau_acpi.c
@@ -46,6 +46,7 @@ static struct nouveau_dsm_priv {
 	bool dsm_detected;
 	bool optimus_detected;
 	bool optimus_flags_detected;
+	bool optimus_skip_dsm;
 	acpi_handle dhandle;
 	acpi_handle rom_handle;
 } nouveau_dsm_priv;
@@ -212,8 +213,26 @@ static const struct vga_switcheroo_handler
nouveau_dsm_handler = {
 	.get_client_id = nouveau_dsm_get_client_id,
 };
 
+/* Firmware supporting Windows 8 or later do not use _DSM to put the device
into
+ * D3cold, they instead rely on disabling power resources on the parent. */
+static bool nouveau_pr3_present(struct pci_dev *pdev)
+{
+	struct pci_dev *parent_pdev = pci_upstream_bridge(pdev);
+	struct acpi_device *ad;
+
+	if (!parent_pdev)
+		return false;
+
+	ad = ACPI_COMPANION(&parent_pdev->dev);
+	if (!ad)
+		return false;
+
+	return ad->power.flags.power_resources;
+}
+
 static void nouveau_dsm_pci_probe(struct pci_dev *pdev, bool *has_mux,
-				  bool *has_opt, bool *has_opt_flags)
+				  bool *has_opt, bool *has_opt_flags,
+				  bool *has_power_resources)
 {
 	acpi_handle dhandle;
 	bool supports_mux;
@@ -238,6 +257,7 @@ static void nouveau_dsm_pci_probe(struct pci_dev *pdev, bool
*has_mux,
 	*has_mux = supports_mux;
 	*has_opt = !!optimus_funcs;
 	*has_opt_flags = optimus_funcs & (1 << NOUVEAU_DSM_OPTIMUS_FLAGS);
+	*has_power_resources = false;
 
 	if (optimus_funcs) {
 		uint32_t result;
@@ -247,6 +267,8 @@ static void nouveau_dsm_pci_probe(struct pci_dev *pdev, bool
*has_mux,
 			 (result & OPTIMUS_ENABLED) ? "enabled" :
"disabled",
 			 (result & OPTIMUS_DYNAMIC_PWR_CAP) ? "dynamic power, " :
"",
 			 (result & OPTIMUS_HDA_CODEC_MASK) ? "hda bios codec
supported" : "");
+
+		*has_power_resources = nouveau_pr3_present(pdev);
 	}
 }
 
@@ -258,6 +280,7 @@ static bool nouveau_dsm_detect(void)
 	bool has_mux = false;
 	bool has_optimus = false;
 	bool has_optimus_flags = false;
+	bool has_power_resources = false;
 	int vga_count = 0;
 	bool guid_valid;
 	bool ret = false;
@@ -273,14 +296,14 @@ static bool nouveau_dsm_detect(void)
 		vga_count++;
 
 		nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus,
-				      &has_optimus_flags);
+				      &has_optimus_flags, &has_power_resources);
 	}
 
 	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_3D << 8, pdev)) != NULL)
{
 		vga_count++;
 
 		nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus,
-				      &has_optimus_flags);
+				      &has_optimus_flags, &has_power_resources);
 	}
 
 	/* find the optimus DSM or the old v1 DSM */
@@ -289,8 +312,11 @@ static bool nouveau_dsm_detect(void)
 			&buffer);
 		printk(KERN_INFO "VGA switcheroo: detected Optimus DSM method %s
handle\n",
 			acpi_method_name);
+		if (has_power_resources)
+			pr_info("nouveau: detected PR support, will not use DSM\n");
 		nouveau_dsm_priv.optimus_detected = true;
 		nouveau_dsm_priv.optimus_flags_detected = has_optimus_flags;
+		nouveau_dsm_priv.optimus_skip_dsm = has_power_resources;
 		ret = true;
 	} else if (vga_count == 2 && has_mux && guid_valid) {
 		acpi_get_name(nouveau_dsm_priv.dhandle, ACPI_FULL_PATHNAME,
@@ -320,7 +346,7 @@ void nouveau_register_dsm_handler(void)
 void nouveau_switcheroo_optimus_dsm(void)
 {
 	u32 result = 0;
-	if (!nouveau_dsm_priv.optimus_detected)
+	if (!nouveau_dsm_priv.optimus_detected || nouveau_dsm_priv.optimus_skip_dsm)
 		return;
 
 	if (nouveau_dsm_priv.optimus_flags_detected)
-- 
2.8.2

Hans de Goede

2016-May-25 09:08 UTC

head link

[Nouveau] [PATCH 0/4] nouveau fixes for RPM/Optimus-related hangs

Hi,

On 25-05-16 00:52, Peter Wu wrote:> Hi,
>
> Here are two patches to fix an issue reported on kernel bugzilla (infinite
loop
> due to unchecked function) and a more important fix to fix hanging Optimus
> machines when runtime PM is enabled (with pm/pci patches).
>
> An older (obsolete) patch for the first issue was tested by the reporter:
> https://bugzilla.kernel.org/show_bug.cgi?id=104791#c11
> (it is replaced by "check for function 0x1B before using it").
>
> The second issue will occur when:
>  - A modern Optimus laptop is in use (designed for Windows 8 or newer).
>  - nouveau runtime PM is enabled (1 or the default -1).
>  - The patch "PCI: Add runtime PM support for PCIe ports" from
Mika is pulled
>    into v4.7 (or v4.8[1]?) via the pci/pm branch,
>   
https://git.kernel.org/cgit/linux/kernel/git/helgaas/pci.git/commit/?h=pci/pm&id=8b71f5652eeac561acf883da01ab4810f763ee42
> (see also the discussion for "[PATCH] PCI: Power on bridges before
scanning new
> devices" at
http://article.gmane.org/gmane.linux.power-management.general/76411)
>
> The first two patches are just refactoring to reduce code duplication (and
> scratch an itch) and make the following patches possible. The next two
patches
> fix the problems reported above.
>
> I intend to get these patches in 4.7 (or the first version where pci/pm
gets
> merged) to avoid a lockup when runpm is enabled. Note:
>  - If the fourth patch is merged before/without Mika's PCIe port patch,
then
>    those modern Optimus machines above will not be put into D3cold.
>  - If the fourth patch is not merged (or merged after Mika's patch),
then under
>    the above conditions the affected machine can lock up.
>  - The three other patches are unrelated to this issue and can safely be
merged.
>
> Tested with:
>  - Linux v4.6 + pci/pm + these four patches
>  - Hardware: Clevo P651RA with acpi_osi="!Windows 2015" (the
latter is a
>    workaround for another PCIe issue).
>  - Card is asleep, woke up with lspci, waited a bit and retried/suspended:
>    - # lspci -xxxxnnvvvv >/dev/null; sleep 5
>    - # lspci -xxxxnnvvvv >/dev/null; sleep 5; systemctl suspend
>    - # lspci -xxxxnnvvvv >/dev/null; systemctl suspend
Series looks good to me:

Reviewed-by: Hans de Goede <hdegoede at redhat.com>

Regards,

Hans




>
> Kind regards,
> Peter
>
>  [1]: https://lkml.kernel.org/r/20160524211309.GH1789 at lahna.fi.intel.com
>
> Peter Wu (4):
>   drm/nouveau/acpi: ensure matching ACPI handle and supported functions
>   drm/nouveau/acpi: return supported DSM functions
>   drm/nouveau/acpi: check for function 0x1B before using it
>   drm/nouveau/acpi: fix lockup with PCIe runtime PM
>
>  drivers/gpu/drm/nouveau/nouveau_acpi.c | 100
+++++++++++++++++++++------------
>  1 file changed, 63 insertions(+), 37 deletions(-)
>

Mika Westerberg

2016-May-25 13:55 UTC

head link

[Nouveau] [PATCH 4/4] drm/nouveau/acpi: fix lockup with PCIe runtime PM

On Wed, May 25, 2016 at 12:53:01AM +0200, Peter Wu
wrote:> Since "PCI: Add runtime PM support for PCIe ports", the parent
PCIe port
> can be runtime-suspended which disables power resources via ACPI. This
> is incompatible with DSM, resulting in a GPU device which is still in D3
> and locks up the kernel on resume.
> 
> Mirror the behavior of Windows 8 and newer[1] (as observed via an AMLi
> debugger trace) and stop using the DSM functions for D3cold when power
> resources are available on the parent PCIe port.
> 
>  [1]:
https://msdn.microsoft.com/windows/hardware/drivers/bringup/firmware-requirements-for-d3cold
> 
> Signed-off-by: Peter Wu <peter at lekensteyn.nl>
> ---
>  drivers/gpu/drm/nouveau/nouveau_acpi.c | 34
++++++++++++++++++++++++++++++----
>  1 file changed, 30 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/nouveau/nouveau_acpi.c
b/drivers/gpu/drm/nouveau/nouveau_acpi.c
> index df9f73e..e469df7 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_acpi.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_acpi.c
> @@ -46,6 +46,7 @@ static struct nouveau_dsm_priv {
>  	bool dsm_detected;
>  	bool optimus_detected;
>  	bool optimus_flags_detected;
> +	bool optimus_skip_dsm;
>  	acpi_handle dhandle;
>  	acpi_handle rom_handle;
>  } nouveau_dsm_priv;
> @@ -212,8 +213,26 @@ static const struct vga_switcheroo_handler
nouveau_dsm_handler = {
>  	.get_client_id = nouveau_dsm_get_client_id,
>  };
>  
> +/* Firmware supporting Windows 8 or later do not use _DSM to put the
device into
> + * D3cold, they instead rely on disabling power resources on the parent.
*/
> +static bool nouveau_pr3_present(struct pci_dev *pdev)
> +{
> +	struct pci_dev *parent_pdev = pci_upstream_bridge(pdev);
> +	struct acpi_device *ad;
Nit: please call this adev instead of ad.
> +
> +	if (!parent_pdev)
> +		return false;
> +
> +	ad = ACPI_COMPANION(&parent_pdev->dev);
> +	if (!ad)
> +		return false;
> +
> +	return ad->power.flags.power_resources;
Is this sufficient to tell if the parent device has _PR3? I thought it
returns true if it has power resources in general, not necessarily _PR3.

Otherwise this looks okay to me.

Emil Velikov

2016-May-27 13:01 UTC

head link

[Nouveau] [PATCH 4/4] drm/nouveau/acpi: fix lockup with PCIe runtime PM

Hi Peter,

On 24 May 2016 at 23:53, Peter Wu <peter at lekensteyn.nl>
wrote:> Since "PCI: Add runtime PM support for PCIe ports", the parent
PCIe port
> can be runtime-suspended which disables power resources via ACPI. This
> is incompatible with DSM, resulting in a GPU device which is still in D3
> and locks up the kernel on resume.
>
> Mirror the behavior of Windows 8 and newer[1] (as observed via an AMLi
> debugger trace) and stop using the DSM functions for D3cold when power
> resources are available on the parent PCIe port.
>
>  [1]:
https://msdn.microsoft.com/windows/hardware/drivers/bringup/firmware-requirements-for-d3cold
>
> Signed-off-by: Peter Wu <peter at lekensteyn.nl>
> ---
>  drivers/gpu/drm/nouveau/nouveau_acpi.c | 34
++++++++++++++++++++++++++++++----
>  1 file changed, 30 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/nouveau/nouveau_acpi.c
b/drivers/gpu/drm/nouveau/nouveau_acpi.c
> index df9f73e..e469df7 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_acpi.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_acpi.c
> @@ -46,6 +46,7 @@ static struct nouveau_dsm_priv {
>         bool dsm_detected;
>         bool optimus_detected;
>         bool optimus_flags_detected;
> +       bool optimus_skip_dsm;
>         acpi_handle dhandle;
>         acpi_handle rom_handle;
>  } nouveau_dsm_priv;
> @@ -212,8 +213,26 @@ static const struct vga_switcheroo_handler
nouveau_dsm_handler = {
>         .get_client_id = nouveau_dsm_get_client_id,
>  };
>
> +/* Firmware supporting Windows 8 or later do not use _DSM to put the
device into
> + * D3cold, they instead rely on disabling power resources on the parent.
*/
> +static bool nouveau_pr3_present(struct pci_dev *pdev)
> +{
> +       struct pci_dev *parent_pdev = pci_upstream_bridge(pdev);
> +       struct acpi_device *ad;
> +
> +       if (!parent_pdev)
> +               return false;
> +
> +       ad = ACPI_COMPANION(&parent_pdev->dev);
> +       if (!ad)
> +               return false;
> +
> +       return ad->power.flags.power_resources;
> +}
> +
>  static void nouveau_dsm_pci_probe(struct pci_dev *pdev, bool *has_mux,
> -                                 bool *has_opt, bool *has_opt_flags)
> +                                 bool *has_opt, bool *has_opt_flags,
> +                                 bool *has_power_resources)
>  {
>         acpi_handle dhandle;
>         bool supports_mux;
> @@ -238,6 +257,7 @@ static void nouveau_dsm_pci_probe(struct pci_dev *pdev,
bool *has_mux,
>         *has_mux = supports_mux;
>         *has_opt = !!optimus_funcs;
>         *has_opt_flags = optimus_funcs & (1 <<
NOUVEAU_DSM_OPTIMUS_FLAGS);
> +       *has_power_resources = false;
>
>         if (optimus_funcs) {
>                 uint32_t result;
> @@ -247,6 +267,8 @@ static void nouveau_dsm_pci_probe(struct pci_dev *pdev,
bool *has_mux,
>                          (result & OPTIMUS_ENABLED) ?
"enabled" : "disabled",
>                          (result & OPTIMUS_DYNAMIC_PWR_CAP) ?
"dynamic power, " : "",
>                          (result & OPTIMUS_HDA_CODEC_MASK) ? "hda
bios codec supported" : "");
> +
> +               *has_power_resources = nouveau_pr3_present(pdev);
>         }
>  }
>
> @@ -258,6 +280,7 @@ static bool nouveau_dsm_detect(void)
>         bool has_mux = false;
>         bool has_optimus = false;
>         bool has_optimus_flags = false;
> +       bool has_power_resources = false;
>         int vga_count = 0;
>         bool guid_valid;
>         bool ret = false;
> @@ -273,14 +296,14 @@ static bool nouveau_dsm_detect(void)
>                 vga_count++;
>
>                 nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus,
> -                                     &has_optimus_flags);
> +                                     &has_optimus_flags,
&has_power_resources);
>         }
>
>         while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_3D << 8,
pdev)) != NULL) {
>                 vga_count++;
>
>                 nouveau_dsm_pci_probe(pdev, &has_mux, &has_optimus,
> -                                     &has_optimus_flags);
> +                                     &has_optimus_flags,
&has_power_resources);
>         }
>This and earlier patch break things in a subtle way.

Namely: upon the second (and any later) call into the
nouveau_dsm_pci_probe() function, the had_foo flags are reset. Thus
only the specifics of the _final_ device are being used (at a later
stage). IMHO one should change that to "_any_ device", which will
match the original code and the actual intent further down in the
file.

Regards,
Emil

Reasonably Related Threads

Search for more possibly parallel threads

Nouveau - May 2016 - [PATCH 0/4] nouveau fixes for RPM/Optimus-related hangs

[Nouveau] [PATCH 0/4] nouveau fixes for RPM/Optimus-related hangs

[Nouveau] [PATCH 1/4] drm/nouveau/acpi: ensure matching ACPI handle and supported functions

[Nouveau] [PATCH 2/4] drm/nouveau/acpi: return supported DSM functions

[Nouveau] [PATCH 3/4] drm/nouveau/acpi: check for function 0x1B before using it

[Nouveau] [PATCH 4/4] drm/nouveau/acpi: fix lockup with PCIe runtime PM

[Nouveau] [PATCH 0/4] nouveau fixes for RPM/Optimus-related hangs

[Nouveau] [PATCH 4/4] drm/nouveau/acpi: fix lockup with PCIe runtime PM

[Nouveau] [PATCH 4/4] drm/nouveau/acpi: fix lockup with PCIe runtime PM

Reasonably Related Threads