cascardo/linux.git
7 years agoPCI: mvebu: Add local struct device pointers
Bjorn Helgaas [Thu, 6 Oct 2016 18:38:58 +0000 (13:38 -0500)]
PCI: mvebu: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: layerscape: Reorder struct ls_pcie
Bjorn Helgaas [Thu, 6 Oct 2016 18:38:05 +0000 (13:38 -0500)]
PCI: layerscape: Reorder struct ls_pcie

Reorder struct ls_pcie to put generic fields first.  No functional change
intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: layerscape: Remove unused ls_add_pcie_port() platform_device arg
Bjorn Helgaas [Thu, 6 Oct 2016 18:38:06 +0000 (13:38 -0500)]
PCI: layerscape: Remove unused ls_add_pcie_port() platform_device arg

ls_add_pcie_port() doesn't use the platform_device pointer passed to it, so
remove it.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: layerscape: Move struct pcie_port setup to probe function
Bjorn Helgaas [Thu, 6 Oct 2016 18:38:06 +0000 (13:38 -0500)]
PCI: layerscape: Move struct pcie_port setup to probe function

Do the basic pcie_port setup in the probe function for consistency with
other drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: layerscape: Pass device-specific struct to internal functions
Bjorn Helgaas [Thu, 6 Oct 2016 18:38:05 +0000 (13:38 -0500)]
PCI: layerscape: Pass device-specific struct to internal functions

Only interfaces used from outside the driver, e.g., those called by the
DesignWare core, need to accept pointers to the generic struct pcie_port.
Internal interfaces can accept pointers to the device-specific struct,
which makes them more straightforward.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: layerscape: Remove redundant struct ls_pcie.dbi
Bjorn Helgaas [Thu, 6 Oct 2016 18:38:05 +0000 (13:38 -0500)]
PCI: layerscape: Remove redundant struct ls_pcie.dbi

Remove the struct ls_pcie.dbi member, which is a duplicate of the generic
pp.dbi_base member.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: layerscape: Remove unused platform data
Bjorn Helgaas [Wed, 12 Oct 2016 04:13:24 +0000 (23:13 -0500)]
PCI: layerscape: Remove unused platform data

The layerscape driver never uses the platform drvdata pointer, so don't
bother setting it.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: layerscape: Add local struct device pointers
Bjorn Helgaas [Thu, 6 Oct 2016 18:38:05 +0000 (13:38 -0500)]
PCI: layerscape: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: keystone: Reorder struct keystone_pcie
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:57 +0000 (13:36 -0500)]
PCI: keystone: Reorder struct keystone_pcie

Reorder struct keystone_pcie to put generic fields first.  No functional
change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: keystone: Add app register accessors
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:57 +0000 (13:36 -0500)]
PCI: keystone: Add app register accessors

Add device-specific register accessors for consistency across host drivers.
No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: keystone: Pass keystone_pcie, not va_app_base, to DBI functions
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:57 +0000 (13:36 -0500)]
PCI: keystone: Pass keystone_pcie, not va_app_base, to DBI functions

Instead of passing ks_pcie->va_app_base to DBI mode functions,
pass the struct keystone_pcie.  This will allow them to use register
accessors.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: keystone: Pass keystone_pcie, not address, to IRQ functions
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:56 +0000 (13:36 -0500)]
PCI: keystone: Pass keystone_pcie, not address, to IRQ functions

Instead of passing the application register base to IRQ functions,
pass the struct keystone_pcie.  This will allow them to use register
accessors.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: keystone: Use generic DesignWare accessors
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:57 +0000 (13:36 -0500)]
PCI: keystone: Use generic DesignWare accessors

The dw_pcie_readl_rc() and dw_pcie_writel_rc() interfaces already add in
pp->dbi_base, so use those instead of doing it ourselves in the keystone
driver.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: keystone: Add local struct device pointers
Bjorn Helgaas [Wed, 12 Oct 2016 03:48:42 +0000 (22:48 -0500)]
PCI: keystone: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: iproc: Hard-code PCIe capability offset instead of searching
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:07 +0000 (13:36 -0500)]
PCI: iproc: Hard-code PCIe capability offset instead of searching

We know where the PCIe capability lives in the host bridge's config space;
in fact, we already hard-coded the offset of the Link Control 2 register.

The hard-coded Link Control 2 offset was 0xdc.  Link Control 2 is at offset
0x30 into the PCIe capability, so the capability itself must be at
0xdc - 0x30 = 0xac.

Hard-code the PCIe capability offset, which means we don't have to search
for it and we can use the standard definitions for registers within the
capability.

No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: iproc: Remove redundant null pointer checking
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:08 +0000 (13:36 -0500)]
PCI: iproc: Remove redundant null pointer checking

The callers never pass a null "pcie" pointer (they check for kzalloc
failure), so we don't need to check here.  The bus driver should never call
the probe function with a null ->dev pointer, so we don't need to check
that either.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: iproc: Validate CSR base in BCMA setup code
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:07 +0000 (13:36 -0500)]
PCI: iproc: Validate CSR base in BCMA setup code

Validate iproc_pcie->base for BCMA devices just like we already do for
platform devices in iproc_pcie_pltfm_probe().  No functional change
intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: iproc: Set drvdata at end of probe function
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:08 +0000 (13:36 -0500)]
PCI: iproc: Set drvdata at end of probe function

Set the drvdata pointer at the end of probe function for consistency with
other drivers.  We don't need the drvdata until after the probe completes,
and we don't need it at all if the probe fails.  No functional change
intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: iproc: Add local struct device pointers
Bjorn Helgaas [Thu, 6 Oct 2016 18:36:08 +0000 (13:36 -0500)]
PCI: iproc: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: imx6: Remove unused return values
Bjorn Helgaas [Thu, 6 Oct 2016 18:35:17 +0000 (13:35 -0500)]
PCI: imx6: Remove unused return values

Remove unused return values.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: imx6: Reorder struct imx6_pcie
Bjorn Helgaas [Thu, 6 Oct 2016 18:35:17 +0000 (13:35 -0500)]
PCI: imx6: Reorder struct imx6_pcie

Reorder struct imx6_pcie to put generic fields first.  No functional change
intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: imx6: Use generic DesignWare accessors
Bjorn Helgaas [Wed, 12 Oct 2016 03:18:26 +0000 (22:18 -0500)]
PCI: imx6: Use generic DesignWare accessors

The dw_pcie_readl_rc() and dw_pcie_writel_rc() interfaces already add in
pp->dbi_base, so use those instead of doing it ourselves in the imx6
driver.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: imx6: Pass device-specific struct to internal functions
Bjorn Helgaas [Wed, 12 Oct 2016 03:06:47 +0000 (22:06 -0500)]
PCI: imx6: Pass device-specific struct to internal functions

Only interfaces used from outside the driver, e.g., those called by the
DesignWare core, need to accept pointers to the generic struct pcie_port.
Internal interfaces can accept pointers to the device-specific struct,
which makes them more straightforward.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: imx6: Pass struct imx6_pcie to PHY accessors
Bjorn Helgaas [Wed, 12 Oct 2016 03:09:32 +0000 (22:09 -0500)]
PCI: imx6: Pass struct imx6_pcie to PHY accessors

Pass the struct imx6_pcie pointer, not dbi_base address, to PHY accessors.
This enables future simplifications.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: imx6: Removed unused struct imx6_pcie.mem_base
Bjorn Helgaas [Thu, 6 Oct 2016 18:35:17 +0000 (13:35 -0500)]
PCI: imx6: Removed unused struct imx6_pcie.mem_base

Removed the unused struct imx6_pcie.mem_base member.  No functional change
intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: imx6: Remove redundant of_node pointer
Bjorn Helgaas [Thu, 6 Oct 2016 18:35:18 +0000 (13:35 -0500)]
PCI: imx6: Remove redundant of_node pointer

"np" and "node" are redundant copies of the of_node pointer.  Remove "np"
and use "node" instead.  Replace the "fsl,max-link-speed" use with "node"
as well.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: imx6: Add local struct device pointers
Bjorn Helgaas [Thu, 6 Oct 2016 18:35:18 +0000 (13:35 -0500)]
PCI: imx6: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: hisi: Use generic DesignWare accessors
Bjorn Helgaas [Thu, 6 Oct 2016 18:34:24 +0000 (13:34 -0500)]
PCI: hisi: Use generic DesignWare accessors

The dw_pcie_readl_rc() and dw_pcie_writel_rc() interfaces already add in
pp->dbi_base, so use those instead of doing it ourselves in the hisi
driver.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: hisi: Remove redundant struct hisi_pcie.reg_base
Bjorn Helgaas [Thu, 6 Oct 2016 18:34:23 +0000 (13:34 -0500)]
PCI: hisi: Remove redundant struct hisi_pcie.reg_base

Remove the struct hisi_pcie.reg_base member, which is a duplicate of the
generic pp.dbi_base member.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: hisi: Name private struct pointer "hisi_pcie" consistently
Bjorn Helgaas [Wed, 12 Oct 2016 02:40:32 +0000 (21:40 -0500)]
PCI: hisi: Name private struct pointer "hisi_pcie" consistently

Most struct hisi_pcie pointers are already called "hisi_pcie".  Change
the rest of them to match.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: hisi: Remove unused platform data
Bjorn Helgaas [Thu, 6 Oct 2016 18:34:25 +0000 (13:34 -0500)]
PCI: hisi: Remove unused platform data

The hisi driver never uses the platform drvdata pointer, so don't bother
setting it.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: hisi: Add local struct device pointers
Bjorn Helgaas [Thu, 6 Oct 2016 18:34:25 +0000 (13:34 -0500)]
PCI: hisi: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: exynos: Reorder struct exynos_pcie
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:40 +0000 (13:33 -0500)]
PCI: exynos: Reorder struct exynos_pcie

Reorder struct exynos_pcie to put generic fields first.  No functional
change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: exynos: Pass device-specific struct to internal functions
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:39 +0000 (13:33 -0500)]
PCI: exynos: Pass device-specific struct to internal functions

Only interfaces used from outside the driver, e.g., those called by the
DesignWare core, need to accept pointers to the generic struct pcie_port.
Internal interfaces can accept pointers to the device-specific struct,
which makes them more straightforward.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: exynos: Name private struct pointer "exynos_pcie" consistently
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:39 +0000 (13:33 -0500)]
PCI: exynos: Name private struct pointer "exynos_pcie" consistently

Most struct exynos_pcie pointers are already called "exynos_pcie".  Change
the rest of them to match.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: exynos: Uninline register accessors
Bjorn Helgaas [Mon, 10 Oct 2016 12:50:07 +0000 (07:50 -0500)]
PCI: exynos: Uninline register accessors

The register accessors are not performance critical and are small enough
that the compiler can inline them itself if it makes sense.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: exynos: Add local struct device pointers
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:41 +0000 (13:33 -0500)]
PCI: exynos: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: dra7xx: Move struct pcie_port setup to probe function
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:06 +0000 (13:33 -0500)]
PCI: dra7xx: Move struct pcie_port setup to probe function

Do the basic pcie_port setup in the probe function for consistency with
other drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: dra7xx: Pass device-specific struct to internal functions
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:05 +0000 (13:33 -0500)]
PCI: dra7xx: Pass device-specific struct to internal functions

Only interfaces used from outside the driver, e.g., those called by the
DesignWare core, need to accept pointers to the generic struct pcie_port.
Internal interfaces can accept pointers to the device-specific struct,
which makes them more straightforward.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: dra7xx: Use generic DesignWare accessors
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:06 +0000 (13:33 -0500)]
PCI: dra7xx: Use generic DesignWare accessors

The dw_pcie_readl_rc() and dw_pcie_writel_rc() interfaces already add in
pp->dbi_base, so use those instead of doing it ourselves in the dra7xx
driver.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: dra7xx: Set drvdata at end of probe function
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:07 +0000 (13:33 -0500)]
PCI: dra7xx: Set drvdata at end of probe function

Set the drvdata pointer at the end of probe function for consistency with
other drivers.  We don't need the drvdata until after the probe completes,
and we don't need it at all if the probe fails.  No functional change
intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: dra7xx: Remove redundant struct device pointer from dra7xx_pcie
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:05 +0000 (13:33 -0500)]
PCI: dra7xx: Remove redundant struct device pointer from dra7xx_pcie

The DesignWare core already stores the struct device pointer in struct
pcie_port.  Remove the redundant copy from struct dra7xx_pcie.dev.  No
functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: dra7xx: Add local struct device pointers
Bjorn Helgaas [Thu, 6 Oct 2016 18:33:06 +0000 (13:33 -0500)]
PCI: dra7xx: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: artpec6: Add resource name comments
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:56 +0000 (13:30 -0500)]
PCI: artpec6: Add resource name comments

Add comments about the Device Tree source of resources.  No functional
change.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
7 years agoPCI: artpec6: Pass device-specific struct to internal functions
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:56 +0000 (13:30 -0500)]
PCI: artpec6: Pass device-specific struct to internal functions

Only interfaces used from outside the driver, e.g., those called by the
DesignWare core, need to accept pointers to the generic struct pcie_port.
Internal interfaces can accept pointers to the device-specific struct,
which makes them more straightforward.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
7 years agoPCI: artpec6: Remove unnecessary artpec6_pcie_link_up()
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:56 +0000 (13:30 -0500)]
PCI: artpec6: Remove unnecessary artpec6_pcie_link_up()

Remove artpec6_pcie_link_up(); the generic dw_pcie_link_up() does the same
thing, so we don't need a device-specific version.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
7 years agoPCI: artpec6: Use generic DesignWare accessors
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:57 +0000 (13:30 -0500)]
PCI: artpec6: Use generic DesignWare accessors

The dw_pcie_readl_rc() and dw_pcie_writel_rc() interfaces already add in
pp->dbi_base, so use those instead of doing it ourselves in the armada8k
driver.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
7 years agoPCI: artpec6: Add register accessors
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:56 +0000 (13:30 -0500)]
PCI: artpec6: Add register accessors

Add device-specific register accessors for consistency across host
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
7 years agoPCI: artpec6: Remove unused platform data
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:57 +0000 (13:30 -0500)]
PCI: artpec6: Remove unused platform data

The artpec6 driver never uses the platform drvdata pointer, so don't
bother setting it.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
7 years agoPCI: artpec6: Add local struct device pointers
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:57 +0000 (13:30 -0500)]
PCI: artpec6: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
7 years agoPCI: armada: Reorder struct armada8k_pcie
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:00 +0000 (13:30 -0500)]
PCI: armada: Reorder struct armada8k_pcie

Reorder the device-specific struct to put the DesignWare generic struct
pcie_port first.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
7 years agoPCI: armada: Pass device-specific struct to internal functions
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:00 +0000 (13:30 -0500)]
PCI: armada: Pass device-specific struct to internal functions

Only interfaces used from outside the driver, e.g., those called by the
DesignWare core, need to accept pointers to the generic struct pcie_port.
Internal interfaces can accept pointers to the device-specific struct,
which makes them more straightforward.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
7 years agoPCI: armada: Use generic DesignWare accessors
Bjorn Helgaas [Wed, 12 Oct 2016 01:15:11 +0000 (20:15 -0500)]
PCI: armada: Use generic DesignWare accessors

The dw_pcie_readl_rc() and dw_pcie_writel_rc() interfaces already add in
pp->dbi_base, so use those instead of doing it ourselves in the armada8k
driver.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: armada: Remove redundant struct armada8k_pcie.base
Bjorn Helgaas [Thu, 6 Oct 2016 18:29:59 +0000 (13:29 -0500)]
PCI: armada: Remove redundant struct armada8k_pcie.base

The struct armada8k_pcie.base pointer is always a constant offset from
struct pcie_port.dbi_base.  Encode that offset in the register macros so we
don't need to maintain the armada8k_pcie.base pointer.  No functional
change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
7 years agoMerge tag 'drm-for-v4.9' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Wed, 12 Oct 2016 01:12:22 +0000 (18:12 -0700)]
Merge tag 'drm-for-v4.9' of git://people.freedesktop.org/~airlied/linux

Pull drm updates from Dave Airlie:
 "Core:
   - Fence destaging work
   - DRIVER_LEGACY to split off legacy drm drivers
   - drm_mm refactoring
   - Splitting drm_crtc.c into chunks and documenting better
   - Display info fixes
   - rbtree support for prime buffer lookup
   - Simple VGA DAC driver

  Panel:
   - Add Nexus 7 panel
   - More simple panels

  i915:
   - Refactoring GEM naming
   - Refactored vma/active tracking
   - Lockless request lookups
   - Better stolen memory support
   - FBC fixes
   - SKL watermark fixes
   - VGPU improvements
   - dma-buf fencing support
   - Better DP dongle support

  amdgpu:
   - Powerplay for Iceland asics
   - Improved GPU reset support
   - UVD/VEC powergating support for CZ/ST
   - Preinitialised VRAM buffer support
   - Virtual display support
   - Initial SI support
   - GTT rework
   - PCI shutdown callback support
   - HPD IRQ storm fixes

  amdkfd:
   - bugfixes

  tilcdc:
   - Atomic modesetting support

  mediatek:
   - AAL + GAMMA engine support
   - Hook up gamma LUT
   - Temporal dithering support

  imx:
   - Pixel clock from devicetree
   - drm bridge support for LVDS bridges
   - active plane reconfiguration
   - VDIC deinterlacer support
   - Frame synchronisation unit support
   - Color space conversion support

  analogix:
   - PSR support
   - Better panel on/off support

  rockchip:
   - rk3399 vop/crtc support
   - PSR support

  vc4:
   - Interlaced vblank timing
   - 3D rendering CPU overhead reduction
   - HDMI output fixes

  tda998x:
   - HDMI audio ASoC support

  sunxi:
   - Allwinner A33 support
   - better TCON support

  msm:
   - DT binding cleanups
   - Explicit fence-fd support

  sti:
   - remove sti415/416 support

  etnaviv:
   - MMUv2 refactoring
   - GC3000 support

  exynos:
   - Refactoring HDMI DCC/PHY
   - G2D pm regression fix
   - Page fault issues with wait for vblank

  There is no nouveau work in this tree, as Ben didn't get a pull
  request in, and he was fighting moving to atomic and adding mst
  support, so maybe best it waits for a cycle"

* tag 'drm-for-v4.9' of git://people.freedesktop.org/~airlied/linux: (1412 commits)
  drm/crtc: constify drm_crtc_index parameter
  drm/i915: Fix conflict resolution from backmerge of v4.8-rc8 to drm-next
  drm/i915/guc: Unwind GuC workqueue reservation if request construction fails
  drm/i915: Reset the breadcrumbs IRQ more carefully
  drm/i915: Force relocations via cpu if we run out of idle aperture
  drm/i915: Distinguish last emitted request from last submitted request
  drm/i915: Allow DP to work w/o EDID
  drm/i915: Move long hpd handling into the hotplug work
  drm/i915/execlists: Reinitialise context image after GPU hang
  drm/i915: Use correct index for backtracking HUNG semaphores
  drm/i915: Unalias obj->phys_handle and obj->userptr
  drm/i915: Just clear the mmiodebug before a register access
  drm/i915/gen9: only add the planes actually affected by ddb changes
  drm/i915: Allow PCH DPLL sharing regardless of DPLL_SDVO_HIGH_SPEED
  drm/i915/bxt: Fix HDMI DPLL configuration
  drm/i915/gen9: fix the watermark res_blocks value
  drm/i915/gen9: fix plane_blocks_per_line on watermarks calculations
  drm/i915/gen9: minimum scanlines for Y tile is not always 4
  drm/i915/gen9: fix the WaWmMemoryReadLatency implementation
  drm/i915/kbl: KBL also needs to run the SAGV code
  ...

7 years agoPCI: armada: Add local base pointer
Bjorn Helgaas [Wed, 12 Oct 2016 01:11:23 +0000 (20:11 -0500)]
PCI: armada: Add local base pointer

Add a local "base" pointer, as is done for other uses, to simplify a
subsequent patch.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: armada: Remove unused platform data
Bjorn Helgaas [Thu, 6 Oct 2016 18:30:01 +0000 (13:30 -0500)]
PCI: armada: Remove unused platform data

The armada driver never uses the platform drvdata pointer, so don't bother
setting it.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
7 years agoPCI: altera: Simplify TLP_CFG_DW1 usage
Bjorn Helgaas [Thu, 6 Oct 2016 18:29:02 +0000 (13:29 -0500)]
PCI: altera: Simplify TLP_CFG_DW1 usage

TLP_CFG_DW1() was only used with altera->root_bus_nr and RP_DEVFN, so
encode that directly into the macro so we don't have to clutter the uses
with the TLP_REQ_ID() usage.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: altera: Simplify TLB_CFG_DW0 usage
Bjorn Helgaas [Thu, 6 Oct 2016 18:29:01 +0000 (13:29 -0500)]
PCI: altera: Simplify TLB_CFG_DW0 usage

All TLP_CFG_DW0() uses follow the same pattern based on the root bus
number, so pull that into the macro itself to declutter the users.  No
functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: altera: Rename altera_pcie_valid_config() to altera_pcie_valid_device()
Bjorn Helgaas [Thu, 6 Oct 2016 18:29:03 +0000 (13:29 -0500)]
PCI: altera: Rename altera_pcie_valid_config() to altera_pcie_valid_device()

Rename altera_pcie_valid_config() to altera_pcie_valid_device().
No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: altera: Remove redundant platform_get_resource() return value check
Bjorn Helgaas [Thu, 6 Oct 2016 18:29:02 +0000 (13:29 -0500)]
PCI: altera: Remove redundant platform_get_resource() return value check

devm_ioremap_resource() fails gracefully when given a NULL resource
pointer, so we don't need to check separately for failure from
platform_get_resource_byname().  Remove the redundant check.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: altera: Remove unused platform data
Bjorn Helgaas [Thu, 6 Oct 2016 18:29:03 +0000 (13:29 -0500)]
PCI: altera: Remove unused platform data

The altera driver never uses the platform drvdata pointer, so don't bother
setting it.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: altera: Add local struct device pointers
Bjorn Helgaas [Wed, 12 Oct 2016 00:50:52 +0000 (19:50 -0500)]
PCI: altera: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
7 years agoPCI: aardvark: Remove unused platform data
Bjorn Helgaas [Thu, 6 Oct 2016 18:27:46 +0000 (13:27 -0500)]
PCI: aardvark: Remove unused platform data

The aardvark driver never uses the platform drvdata pointer, so don't
bother setting it.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
7 years agoPCI: aardvark: Add local struct device pointers
Bjorn Helgaas [Thu, 6 Oct 2016 18:27:46 +0000 (13:27 -0500)]
PCI: aardvark: Add local struct device pointers

Use a local "struct device *dev" for brevity and consistency with other
drivers.  No functional change intended.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
7 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Wed, 12 Oct 2016 00:34:10 +0000 (17:34 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge more updates from Andrew Morton:

 - a few block updates that fell in my lap

 - lib/ updates

 - checkpatch

 - autofs

 - ipc

 - a ton of misc other things

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (100 commits)
  mm: split gfp_mask and mapping flags into separate fields
  fs: use mapping_set_error instead of opencoded set_bit
  treewide: remove redundant #include <linux/kconfig.h>
  hung_task: allow hung_task_panic when hung_task_warnings is 0
  kthread: add kerneldoc for kthread_create()
  kthread: better support freezable kthread workers
  kthread: allow to modify delayed kthread work
  kthread: allow to cancel kthread work
  kthread: initial support for delayed kthread work
  kthread: detect when a kthread work is used by more workers
  kthread: add kthread_destroy_worker()
  kthread: add kthread_create_worker*()
  kthread: allow to call __kthread_create_on_node() with va_list args
  kthread/smpboot: do not park in kthread_create_on_cpu()
  kthread: kthread worker API cleanup
  kthread: rename probe_kthread_data() to kthread_probe_data()
  scripts/tags.sh: enable code completion in VIM
  mm: kmemleak: avoid using __va() on addresses that don't have a lowmem mapping
  kdump, vmcoreinfo: report memory sections virtual addresses
  ipc/sem.c: add cond_resched in exit_sme
  ...

7 years agoInput: i8042 - skip selftest on ASUS laptops
Marcos Paulo de Souza [Sat, 1 Oct 2016 19:07:35 +0000 (12:07 -0700)]
Input: i8042 - skip selftest on ASUS laptops

On suspend/resume cycle, selftest is executed to reset i8042 controller.
But when this is done in Asus devices, subsequent calls to detect/init
functions to elantech driver fails. Skipping selftest fixes this problem.

An easier step to reproduce this problem is adding i8042.reset=1 as a
kernel parameter. On Asus laptops, it'll make the system to start with the
touchpad already stuck, since psmouse_probe forcibly calls the selftest
function.

This patch was inspired by John Hiesey's change[1], but, since this problem
affects a lot of models of Asus, let's avoid running selftests on them.

All models affected by this problem:
A455LD
K401LB
K501LB
K501LX
R409L
V502LX
X302LA
X450LCP
X450LD
X455LAB
X455LDB
X455LF
Z450LA

[1]: https://marc.info/?l=linux-input&m=144312209020616&w=2

Fixes: "ETPS/2 Elantech Touchpad dies after resume from suspend"
(https://bugzilla.kernel.org/show_bug.cgi?id=107971)

Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
7 years agoInput: melfas_mip4 - add ic_name sysfs attribute
Sangwon Jee [Fri, 30 Sep 2016 22:38:47 +0000 (15:38 -0700)]
Input: melfas_mip4 - add ic_name sysfs attribute

Add ic_name sysfs attribute for retrieving IC model name.

Signed-off-by: Sangwon Jee <jeesw@melfas.com>
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
7 years agomm: split gfp_mask and mapping flags into separate fields
Michal Hocko [Tue, 11 Oct 2016 20:56:04 +0000 (13:56 -0700)]
mm: split gfp_mask and mapping flags into separate fields

mapping->flags currently encodes two different things into a single flag.
It contains sticky gfp_mask for page cache allocations and AS_ codes used
to report errors/enospace and other states which are mapping specific.
Condensing the two semantically unrelated things saves few bytes but it
also complicates other things.  For one thing the gfp flags space is
reduced and in fact we are already running out of available bits.  It can
be assumed that more gfp flags will be necessary later on.

To not introduce the address_space grow (at least on x86_64) we can stick
it right after private_lock because we have a hole there.

struct address_space {
        struct inode *             host;                 /*     0     8 */
        struct radix_tree_root     page_tree;            /*     8    16 */
        spinlock_t                 tree_lock;            /*    24     4 */
        atomic_t                   i_mmap_writable;      /*    28     4 */
        struct rb_root             i_mmap;               /*    32     8 */
        struct rw_semaphore        i_mmap_rwsem;         /*    40    40 */
        /* --- cacheline 1 boundary (64 bytes) was 16 bytes ago --- */
        long unsigned int          nrpages;              /*    80     8 */
        long unsigned int          nrexceptional;        /*    88     8 */
        long unsigned int          writeback_index;      /*    96     8 */
        const struct address_space_operations  * a_ops;  /*   104     8 */
        long unsigned int          flags;                /*   112     8 */
        spinlock_t                 private_lock;         /*   120     4 */

        /* XXX 4 bytes hole, try to pack */

        /* --- cacheline 2 boundary (128 bytes) --- */
        struct list_head           private_list;         /*   128    16 */
        void *                     private_data;         /*   144     8 */

        /* size: 152, cachelines: 3, members: 14 */
        /* sum members: 148, holes: 1, sum holes: 4 */
        /* last cacheline: 24 bytes */
};

Link: http://lkml.kernel.org/r/20160912114852.GI14524@dhcp22.suse.cz
Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agofs: use mapping_set_error instead of opencoded set_bit
Michal Hocko [Tue, 11 Oct 2016 20:56:01 +0000 (13:56 -0700)]
fs: use mapping_set_error instead of opencoded set_bit

The mapping_set_error() helper sets the correct AS_ flag for the mapping
so there is no reason to open code it.  Use the helper directly.

[akpm@linux-foundation.org: be honest about conversion from -ENXIO to -EIO]
Link: http://lkml.kernel.org/r/20160912111608.2588-2-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agotreewide: remove redundant #include <linux/kconfig.h>
Masahiro Yamada [Tue, 11 Oct 2016 20:55:58 +0000 (13:55 -0700)]
treewide: remove redundant #include <linux/kconfig.h>

Kernel source files need not include <linux/kconfig.h> explicitly
because the top Makefile forces to include it with:

  -include $(srctree)/include/linux/kconfig.h

This commit removes explicit includes except the following:

  * arch/s390/include/asm/facilities_src.h
  * tools/testing/radix-tree/linux/kernel.h

These two are used for host programs.

Link: http://lkml.kernel.org/r/1473656164-11929-1-git-send-email-yamada.masahiro@socionext.com
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agohung_task: allow hung_task_panic when hung_task_warnings is 0
John Siddle [Tue, 11 Oct 2016 20:55:56 +0000 (13:55 -0700)]
hung_task: allow hung_task_panic when hung_task_warnings is 0

Previously hung_task_panic would not be respected if enabled after
hung_task_warnings had already been decremented to 0.

Permit the kernel to panic if hung_task_panic is enabled after
hung_task_warnings has already been decremented to 0 and another task
hangs for hung_task_timeout_secs seconds.

Check if hung_task_panic is enabled so we don't return prematurely, and
check if hung_task_warnings is non-zero so we don't print the warning
unnecessarily.

[akpm@linux-foundation.org: fix off-by-one]
Link: http://lkml.kernel.org/r/1473450214-4049-1-git-send-email-jsiddle@redhat.com
Signed-off-by: John Siddle <jsiddle@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: add kerneldoc for kthread_create()
Jonathan Corbet [Tue, 11 Oct 2016 20:55:53 +0000 (13:55 -0700)]
kthread: add kerneldoc for kthread_create()

This macro is referenced in other kerneldoc comments, but lacks one of its
own; fix that.

Link: http://lkml.kernel.org/r/20160826072313.726a3485@lwn.net
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Reported-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: better support freezable kthread workers
Petr Mladek [Tue, 11 Oct 2016 20:55:50 +0000 (13:55 -0700)]
kthread: better support freezable kthread workers

This patch allows to make kthread worker freezable via a new @flags
parameter. It will allow to avoid an init work in some kthreads.

It currently does not affect the function of kthread_worker_fn()
but it might help to do some optimization or fixes eventually.

I currently do not know about any other use for the @flags
parameter but I believe that we will want more flags
in the future.

Finally, I hope that it will not cause confusion with @flags member
in struct kthread. Well, I guess that we will want to rework the
basic kthreads implementation once all kthreads are converted into
kthread workers or workqueues. It is possible that we will merge
the two structures.

Link: http://lkml.kernel.org/r/1470754545-17632-12-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: allow to modify delayed kthread work
Petr Mladek [Tue, 11 Oct 2016 20:55:46 +0000 (13:55 -0700)]
kthread: allow to modify delayed kthread work

There are situations when we need to modify the delay of a delayed kthread
work. For example, when the work depends on an event and the initial delay
means a timeout. Then we want to queue the work immediately when the event
happens.

This patch implements kthread_mod_delayed_work() as inspired workqueues.
It cancels the timer, removes the work from any worker list and queues it
again with the given timeout.

A very special case is when the work is being canceled at the same time.
It might happen because of the regular kthread_cancel_delayed_work_sync()
or by another kthread_mod_delayed_work(). In this case, we do nothing and
let the other operation win. This should not normally happen as the caller
is supposed to synchronize these operations a reasonable way.

Link: http://lkml.kernel.org/r/1470754545-17632-11-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: allow to cancel kthread work
Petr Mladek [Tue, 11 Oct 2016 20:55:43 +0000 (13:55 -0700)]
kthread: allow to cancel kthread work

We are going to use kthread workers more widely and sometimes we will need
to make sure that the work is neither pending nor running.

This patch implements cancel_*_sync() operations as inspired by
workqueues.  Well, we are synchronized against the other operations via
the worker lock, we use del_timer_sync() and a counter to count parallel
cancel operations.  Therefore the implementation might be easier.

First, we check if a worker is assigned.  If not, the work has newer been
queued after it was initialized.

Second, we take the worker lock.  It must be the right one.  The work must
not be assigned to another worker unless it is initialized in between.

Third, we try to cancel the timer when it exists.  The timer is deleted
synchronously to make sure that the timer call back is not running.  We
need to temporary release the worker->lock to avoid a possible deadlock
with the callback.  In the meantime, we set work->canceling counter to
avoid any queuing.

Fourth, we try to remove the work from a worker list. It might be
the list of either normal or delayed works.

Fifth, if the work is running, we call kthread_flush_work().  It might
take an arbitrary time.  We need to release the worker-lock again.  In the
meantime, we again block any queuing by the canceling counter.

As already mentioned, the check for a pending kthread work is done under a
lock.  In compare with workqueues, we do not need to fight for a single
PENDING bit to block other operations.  Therefore we do not suffer from
the thundering storm problem and all parallel canceling jobs might use
kthread_flush_work().  Any queuing is blocked until the counter gets zero.

Link: http://lkml.kernel.org/r/1470754545-17632-10-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: initial support for delayed kthread work
Petr Mladek [Tue, 11 Oct 2016 20:55:40 +0000 (13:55 -0700)]
kthread: initial support for delayed kthread work

We are going to use kthread_worker more widely and delayed works
will be pretty useful.

The implementation is inspired by workqueues.  It uses a timer to queue
the work after the requested delay.  If the delay is zero, the work is
queued immediately.

In compare with workqueues, each work is associated with a single worker
(kthread).  Therefore the implementation could be much easier.  In
particular, we use the worker->lock to synchronize all the operations with
the work.  We do not need any atomic operation with a flags variable.

In fact, we do not need any state variable at all.  Instead, we add a list
of delayed works into the worker.  Then the pending work is listed either
in the list of queued or delayed works.  And the existing check of pending
works is the same even for the delayed ones.

A work must not be assigned to another worker unless reinitialized.
Therefore the timer handler might expect that dwork->work->worker is valid
and it could simply take the lock.  We just add some sanity checks to help
with debugging a potential misuse.

Link: http://lkml.kernel.org/r/1470754545-17632-9-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: detect when a kthread work is used by more workers
Petr Mladek [Tue, 11 Oct 2016 20:55:36 +0000 (13:55 -0700)]
kthread: detect when a kthread work is used by more workers

Nothing currently prevents a work from queuing for a kthread worker when
it is already running on another one.  This means that the work might run
in parallel on more than one worker.  Also some operations are not
reliable, e.g.  flush.

This problem will be even more visible after we add kthread_cancel_work()
function.  It will only have "work" as the parameter and will use
worker->lock to synchronize with others.

Well, normally this is not a problem because the API users are sane.
But bugs might happen and users also might be crazy.

This patch adds a warning when we try to insert the work for another
worker.  It does not fully prevent the misuse because it would make the
code much more complicated without a big benefit.

It adds the same warning also into kthread_flush_work() instead of the
repeated attempts to get the right lock.

A side effect is that one needs to explicitly reinitialize the work if it
must be queued into another worker.  This is needed, for example, when the
worker is stopped and started again.  It is a bit inconvenient.  But it
looks like a good compromise between the stability and complexity.

I have double checked all existing users of the kthread worker API and
they all seems to initialize the work after the worker gets started.

Just for completeness, the patch adds a check that the work is not already
in a queue.

The patch also puts all the checks into a separate function.  It will be
reused when implementing delayed works.

Link: http://lkml.kernel.org/r/1470754545-17632-8-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: add kthread_destroy_worker()
Petr Mladek [Tue, 11 Oct 2016 20:55:33 +0000 (13:55 -0700)]
kthread: add kthread_destroy_worker()

The current kthread worker users call flush() and stop() explicitly.
This function does the same plus it frees the kthread_worker struct
in one call.

It is supposed to be used together with kthread_create_worker*() that
allocates struct kthread_worker.

Link: http://lkml.kernel.org/r/1470754545-17632-7-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: add kthread_create_worker*()
Petr Mladek [Tue, 11 Oct 2016 20:55:30 +0000 (13:55 -0700)]
kthread: add kthread_create_worker*()

Kthread workers are currently created using the classic kthread API,
namely kthread_run().  kthread_worker_fn() is passed as the @threadfn
parameter.

This patch defines kthread_create_worker() and
kthread_create_worker_on_cpu() functions that hide implementation details.

They enforce using kthread_worker_fn() for the main thread.  But I doubt
that there are any plans to create any alternative.  In fact, I think that
we do not want any alternative main thread because it would be hard to
support consistency with the rest of the kthread worker API.

The naming and function of kthread_create_worker() is inspired by the
workqueues API like the rest of the kthread worker API.

The kthread_create_worker_on_cpu() variant is motivated by the original
kthread_create_on_cpu().  Note that we need to bind per-CPU kthread
workers already when they are created.  It makes the life easier.
kthread_bind() could not be used later for an already running worker.

This patch does _not_ convert existing kthread workers.  The kthread
worker API need more improvements first, e.g.  a function to destroy the
worker.

IMPORTANT:

kthread_create_worker_on_cpu() allows to use any format of the worker
name, in compare with kthread_create_on_cpu().  The good thing is that it
is more generic.  The bad thing is that most users will need to pass the
cpu number in two parameters, e.g.  kthread_create_worker_on_cpu(cpu,
"helper/%d", cpu).

To be honest, the main motivation was to avoid the need for an empty
va_list.  The only legal way was to create a helper function that would be
called with an empty list.  Other attempts caused compilation warnings or
even errors on different architectures.

There were also other alternatives, for example, using #define or
splitting __kthread_create_worker().  The used solution looked like the
least ugly.

Link: http://lkml.kernel.org/r/1470754545-17632-6-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: allow to call __kthread_create_on_node() with va_list args
Petr Mladek [Tue, 11 Oct 2016 20:55:27 +0000 (13:55 -0700)]
kthread: allow to call __kthread_create_on_node() with va_list args

kthread_create_on_node() implements a bunch of logic to create the
kthread.  It is already called by kthread_create_on_cpu().

We are going to extend the kthread worker API and will need to call
kthread_create_on_node() with va_list args there.

This patch does only a refactoring and does not modify the existing
behavior.

Link: http://lkml.kernel.org/r/1470754545-17632-5-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread/smpboot: do not park in kthread_create_on_cpu()
Petr Mladek [Tue, 11 Oct 2016 20:55:23 +0000 (13:55 -0700)]
kthread/smpboot: do not park in kthread_create_on_cpu()

kthread_create_on_cpu() was added by the commit 2a1d446019f9a5983e
("kthread: Implement park/unpark facility").  It is currently used only
when enabling new CPU.  For this purpose, the newly created kthread has to
be parked.

The CPU binding is a bit tricky.  The kthread is parked when the CPU has
not been allowed yet.  And the CPU is bound when the kthread is unparked.

The function would be useful for more per-CPU kthreads, e.g.
bnx2fc_thread, fcoethread.  For this purpose, the newly created kthread
should stay in the uninterruptible state.

This patch moves the parking into smpboot.  It binds the thread already
when created.  Then the function might be used universally.  Also the
behavior is consistent with kthread_create() and kthread_create_on_node().

Link: http://lkml.kernel.org/r/1470754545-17632-4-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: kthread worker API cleanup
Petr Mladek [Tue, 11 Oct 2016 20:55:20 +0000 (13:55 -0700)]
kthread: kthread worker API cleanup

A good practice is to prefix the names of functions by the name
of the subsystem.

The kthread worker API is a mix of classic kthreads and workqueues.  Each
worker has a dedicated kthread.  It runs a generic function that process
queued works.  It is implemented as part of the kthread subsystem.

This patch renames the existing kthread worker API to use
the corresponding name from the workqueues API prefixed by
kthread_:

__init_kthread_worker() -> __kthread_init_worker()
init_kthread_worker() -> kthread_init_worker()
init_kthread_work() -> kthread_init_work()
insert_kthread_work() -> kthread_insert_work()
queue_kthread_work() -> kthread_queue_work()
flush_kthread_work() -> kthread_flush_work()
flush_kthread_worker() -> kthread_flush_worker()

Note that the names of DEFINE_KTHREAD_WORK*() macros stay
as they are. It is common that the "DEFINE_" prefix has
precedence over the subsystem names.

Note that INIT() macros and init() functions use different
naming scheme. There is no good solution. There are several
reasons for this solution:

  + "init" in the function names stands for the verb "initialize"
    aka "initialize worker". While "INIT" in the macro names
    stands for the noun "INITIALIZER" aka "worker initializer".

  + INIT() macros are used only in DEFINE() macros

  + init() functions are used close to the other kthread()
    functions. It looks much better if all the functions
    use the same scheme.

  + There will be also kthread_destroy_worker() that will
    be used close to kthread_cancel_work(). It is related
    to the init() function. Again it looks better if all
    functions use the same naming scheme.

  + there are several precedents for such init() function
    names, e.g. amd_iommu_init_device(), free_area_init_node(),
    jump_label_init_type(),  regmap_init_mmio_clk(),

  + It is not an argument but it was inconsistent even before.

[arnd@arndb.de: fix linux-next merge conflict]
Link: http://lkml.kernel.org/r/20160908135724.1311726-1-arnd@arndb.de
Link: http://lkml.kernel.org/r/1470754545-17632-3-git-send-email-pmladek@suse.com
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Petr Mladek <pmladek@suse.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokthread: rename probe_kthread_data() to kthread_probe_data()
Petr Mladek [Tue, 11 Oct 2016 20:55:17 +0000 (13:55 -0700)]
kthread: rename probe_kthread_data() to kthread_probe_data()

Patch series "kthread: Kthread worker API improvements"

The intention of this patchset is to make it easier to manipulate and
maintain kthreads.  Especially, I want to replace all the custom main
cycles with a generic one.  Also I want to make the kthreads sleep in a
consistent state in a common place when there is no work.

This patch (of 11):

A good practice is to prefix the names of functions by the name of the
subsystem.

This patch fixes the name of probe_kthread_data().  The other wrong
functions names are part of the kthread worker API and will be fixed
separately.

Link: http://lkml.kernel.org/r/1470754545-17632-2-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoscripts/tags.sh: enable code completion in VIM
Mathieu Maret [Tue, 11 Oct 2016 20:55:14 +0000 (13:55 -0700)]
scripts/tags.sh: enable code completion in VIM

Vim, with the omnicppcomplete(#1) plugin, can do code completion using
information build by ctags.  Add flags needed by omnicppcomplete(#2) to
have completion on member of structure.

1: https://github.com/vim-scripts/omnicppcomplete
2: https://github.com/vim-scripts/OmniCppComplete/blob/master/doc/omnicppcomplete.txt#L93

Link: http://lkml.kernel.org/r/20160830191546.4469-1-mathieu.maret@gmail.com
Signed-off-by: Mathieu Maret <mathieu.maret@gmail.com>
Cc: Michal Marek <mmarek@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: kmemleak: avoid using __va() on addresses that don't have a lowmem mapping
Catalin Marinas [Tue, 11 Oct 2016 20:55:11 +0000 (13:55 -0700)]
mm: kmemleak: avoid using __va() on addresses that don't have a lowmem mapping

Some of the kmemleak_*() callbacks in memblock, bootmem, CMA convert a
physical address to a virtual one using __va().  However, such physical
addresses may sometimes be located in highmem and using __va() is
incorrect, leading to inconsistent object tracking in kmemleak.

The following functions have been added to the kmemleak API and they take
a physical address as the object pointer.  They only perform the
corresponding action if the address has a lowmem mapping:

kmemleak_alloc_phys
kmemleak_free_part_phys
kmemleak_not_leak_phys
kmemleak_ignore_phys

The affected calling places have been updated to use the new kmemleak
API.

Link: http://lkml.kernel.org/r/1471531432-16503-1-git-send-email-catalin.marinas@arm.com
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Reported-by: Vignesh R <vigneshr@ti.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokdump, vmcoreinfo: report memory sections virtual addresses
Thomas Garnier [Tue, 11 Oct 2016 20:55:08 +0000 (13:55 -0700)]
kdump, vmcoreinfo: report memory sections virtual addresses

KASLR memory randomization can randomize the base of the physical memory
mapping (PAGE_OFFSET), vmalloc (VMALLOC_START) and vmemmap
(VMEMMAP_START).  Adding these variables on VMCOREINFO so tools can easily
identify the base of each memory section.

Link: http://lkml.kernel.org/r/1471531632-23003-1-git-send-email-thgarnie@google.com
Signed-off-by: Thomas Garnier <thgarnie@google.com>
Acked-by: Baoquan He <bhe@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H . Peter Anvin" <hpa@zytor.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Xunlei Pang <xlpang@redhat.com>
Cc: HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Eugene Surovegin <surovegin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoipc/sem.c: add cond_resched in exit_sme
Nikolay Borisov [Tue, 11 Oct 2016 20:55:05 +0000 (13:55 -0700)]
ipc/sem.c: add cond_resched in exit_sme

In CONFIG_PREEMPT=n kernel a softlockup was observed while the for loop in
exit_sem.  Apparently it's possible for the loop to take quite a long time
and it doesn't have a scheduling point in it.  Since the codes is
executing under an rcu read section this may also cause rcu stalls, which
in turn block synchronize_rcu operations, which more or less de-stabilises
the whole system.

Fix this by introducing a cond_resched() at the beginning of the loop.

So this patch fixes the following:

  NMI watchdog: BUG: soft lockup - CPU#10 stuck for 23s! [httpd:18119]
  CPU: 10 PID: 18119 Comm: httpd Tainted: G           O    4.4.20-clouder2 #6
  Hardware name: Supermicro X10DRi/X10DRi, BIOS 1.1 04/14/2015
  task: ffff88348d695280 ti: ffff881c95550000 task.ti: ffff881c95550000
  RIP: 0010:[<ffffffff81614bc7>]  [<ffffffff81614bc7>] _raw_spin_lock+0x17/0x30
  RSP: 0018:ffff881c95553e40  EFLAGS: 00000246
  RAX: 0000000000000000 RBX: ffff883161b1eea8 RCX: 000000000000000d
  RDX: 0000000000000001 RSI: 000000000000000e RDI: ffff883161b1eea4
  RBP: ffff881c95553ea0 R08: ffff881c95553e68 R09: ffff883fef376f88
  R10: ffff881fffb58c20 R11: ffffea0072556600 R12: ffff883161b1eea0
  R13: ffff88348d695280 R14: ffff883dec427000 R15: ffff8831621672a0
  FS:  0000000000000000(0000) GS:ffff881fffb40000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00007f3b3723e020 CR3: 0000000001c0a000 CR4: 00000000001406e0
  Call Trace:
    ? exit_sem+0x7c/0x280
    do_exit+0x338/0xb40
    do_group_exit+0x43/0xd0
    SyS_exit_group+0x14/0x20
    entry_SYSCALL_64_fastpath+0x16/0x6e

Link: http://lkml.kernel.org/r/1475154992-6363-1-git-send-email-kernel@kyup.com
Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Cc: Herton R. Krzesinski <herton@redhat.com>
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoipc/msg: avoid waking sender upon full queue
Davidlohr Bueso [Tue, 11 Oct 2016 20:55:02 +0000 (13:55 -0700)]
ipc/msg: avoid waking sender upon full queue

Blocked tasks queued in q_senders waiting for their message to fit in the
queue are blindly awoken every time we think there's a remote chance this
might happen.  This could cause numerous (and expensive -- thundering
herd-ish) bogus wakeups if the queue is still really full.  Adding to the
scheduling cost/overhead, there's also the fact that we need to take the
ipc object lock and requeue ourselves in the q_senders list.

By keeping track of the blocked sender's message size, we can know
previously if the wakeup ought to occur or not.  Otherwise, to maintain
the current wakeup order we just move it to the tail.  This is exactly
what occurs right now if the sender needs to go back to sleep.

The case of EIDRM is left completely untouched, as we need to wakeup all
the tasks, and shouldn't be playing games in the first place.

This patch was seen to save on the 'msgctl10' ltp testcase ~15% in context
switches (avg out of ten runs).  Although these tests are really about
functionality (as opposed to performance), is does show the direct
benefits of the optimization.

[akpm@linux-foundation.org: coding-style fixes]
Link: http://lkml.kernel.org/r/1469748819-19484-6-git-send-email-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Manfred Spraul <manfred@colorfullife.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoipc/msg: make ss_wakeup() kill arg boolean
Davidlohr Bueso [Tue, 11 Oct 2016 20:54:59 +0000 (13:54 -0700)]
ipc/msg: make ss_wakeup() kill arg boolean

... 'tis annoying.

Link: http://lkml.kernel.org/r/1469748819-19484-4-git-send-email-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dave@stgolabs.net>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Manfred Spraul <manfred@colorfullife.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoipc/msg: batch queue sender wakeups
Davidlohr Bueso [Tue, 11 Oct 2016 20:54:56 +0000 (13:54 -0700)]
ipc/msg: batch queue sender wakeups

Currently the use of wake_qs in sysv msg queues are only for the receiver
tasks that are blocked on the queue.  But blocked sender tasks (due to
queue size constraints) still are awoken with the ipc object lock held,
which can be a problem particularly for small sized queues and far from
gracious for -rt (just like it was for the receiver side).

The paths that actually wakeup a sender are obviously related to when we
are either getting rid of the queue or after (some) space is freed-up
after a receiver takes the msg (msgrcv).  Furthermore, with the exception
of msgrcv, we can always piggy-back on expunge_all that has its own tasks
lined-up for waking.  Finally, upon unlinking the message, it should be no
problem delaying the wakeups a bit until after we've released the lock.

Link: http://lkml.kernel.org/r/1469748819-19484-3-git-send-email-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Manfred Spraul <manfred@colorfullife.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoipc/msg: implement lockless pipelined wakeups
Sebastian Andrzej Siewior [Tue, 11 Oct 2016 20:54:53 +0000 (13:54 -0700)]
ipc/msg: implement lockless pipelined wakeups

This patch moves the wakeup_process() invocation so it is not done under
the ipc global lock by making use of a lockless wake_q.  With this change,
the waiter is woken up once the message has been assigned and it does not
need to loop on SMP if the message points to NULL.  In the signal case we
still need to check the pointer under the lock to verify the state.

This change should also avoid the introduction of preempt_disable() in -RT
which avoids a busy-loop which pools for the NULL -> !NULL change if the
waiter has a higher priority compared to the waker.

By making use of wake_qs, the logic of sysv msg queues is greatly
simplified (and very well suited as we can batch lockless wakeups),
particularly around the lockless receive algorithm.

This has been tested with Manred's pmsg-shared tool on a "AMD A10-7800
Radeon R7, 12 Compute Cores 4C+8G":

test             |   before   |   after    | diff
-----------------|------------|------------|----------
pmsg-shared 8 60 | 19,347,422 | 30,442,191 | + ~57.34 %
pmsg-shared 4 60 | 21,367,197 | 35,743,458 | + ~67.28 %
pmsg-shared 2 60 | 22,884,224 | 24,278,200 | +  ~6.09 %

Link: http://lkml.kernel.org/r/1469748819-19484-2-git-send-email-dave@stgolabs.net
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoipc/sem.c: fix complex_count vs. simple op race
Manfred Spraul [Tue, 11 Oct 2016 20:54:50 +0000 (13:54 -0700)]
ipc/sem.c: fix complex_count vs. simple op race

Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a
race:

sem_lock has a fast path that allows parallel simple operations.
There are two reasons why a simple operation cannot run in parallel:
 - a non-simple operations is ongoing (sma->sem_perm.lock held)
 - a complex operation is sleeping (sma->complex_count != 0)

As both facts are stored independently, a thread can bypass the current
checks by sleeping in the right positions.  See below for more details
(or kernel bugzilla 105651).

The patch fixes that by creating one variable (complex_mode)
that tracks both reasons why parallel operations are not possible.

The patch also updates stale documentation regarding the locking.

With regards to stable kernels:
The patch is required for all kernels that include the
commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?)

The alternative is to revert the patch that introduced the race.

The patch is safe for backporting, i.e. it makes no assumptions
about memory barriers in spin_unlock_wait().

Background:
Here is the race of the current implementation:

Thread A: (simple op)
- does the first "sma->complex_count == 0" test

Thread B: (complex op)
- does sem_lock(): This includes an array scan. But the scan can't
  find Thread A, because Thread A does not own sem->lock yet.
- the thread does the operation, increases complex_count,
  drops sem_lock, sleeps

Thread A:
- spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock)
- sleeps before the complex_count test

Thread C: (complex op)
- does sem_lock (no array scan, complex_count==1)
- wakes up Thread B.
- decrements complex_count

Thread A:
- does the complex_count test

Bug:
Now both thread A and thread C operate on the same array, without
any synchronization.

Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()")
Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com
Reported-by: <felixh@informatik.uni-bremen.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: <1vier1@web.de>
Cc: <stable@vger.kernel.org> [3.10+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agokcov: do not instrument lib/stackdepot.c
Alexander Potapenko [Tue, 11 Oct 2016 20:54:47 +0000 (13:54 -0700)]
kcov: do not instrument lib/stackdepot.c

There's no point in collecting coverage from lib/stackdepot.c, as it is
not a function of syscall inputs.  Disabling kcov instrumentation for that
file will reduce the coverage noise level.

Link: http://lkml.kernel.org/r/1474640972-104131-1-git-send-email-glider@google.com
Signed-off-by: Alexander Potapenko <glider@google.com>
Acked-by: Dmitry Vyukov <dvyukov@google.com>
Cc: Kostya Serebryany <kcc@google.com>
Cc: Andrey Konovalov <andreyknvl@google.com>
Cc: syzkaller <syzkaller@googlegroups.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoconfig: android: enable CONFIG_SECCOMP
Rob Herring [Tue, 11 Oct 2016 20:54:44 +0000 (13:54 -0700)]
config: android: enable CONFIG_SECCOMP

As of Android N, SECCOMP is required. Without it, we will get
mediaextractor error:

E /system/bin/mediaextractor: libminijail: prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER): Invalid argument

Link: http://lkml.kernel.org/r/20160908185934.18098-3-robh@kernel.org
Signed-off-by: Rob Herring <robh@kernel.org>
Acked-by: John Stultz <john.stultz@linaro.org>
Cc: Amit Pundir <amit.pundir@linaro.org>
Cc: Dmitry Shmidt <dimitrysh@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoconfig: android: set SELinux as default security mode
Rob Herring [Tue, 11 Oct 2016 20:54:41 +0000 (13:54 -0700)]
config: android: set SELinux as default security mode

Android won't boot without SELinux enabled, so make it the default.

Link: http://lkml.kernel.org/r/20160908185934.18098-2-robh@kernel.org
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoconfig: android: move device mapper options to recommended
Rob Herring [Tue, 11 Oct 2016 20:54:38 +0000 (13:54 -0700)]
config: android: move device mapper options to recommended

CONFIG_MD is in recommended, but other dependent options like DM_CRYPT and
DM_VERITY options are in base.  The result is the options in base don't
get enabled when applying both base and recommended fragments.  Move all
the options to recommended.

Link: http://lkml.kernel.org/r/20160908185934.18098-1-robh@kernel.org
Signed-off-by: Rob Herring <robh@kernel.org>
Acked-by: John Stultz <john.stultz@linaro.org>
Cc: Amit Pundir <amit.pundir@linaro.org>
Cc: Dmitry Shmidt <dimitrysh@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoconfig/android: Remove CONFIG_IPV6_PRIVACY
Borislav Petkov [Tue, 11 Oct 2016 20:54:36 +0000 (13:54 -0700)]
config/android: Remove CONFIG_IPV6_PRIVACY

Option is long gone, see commit 5d9efa7ee99e ("ipv6: Remove privacy
config option.")

Link: http://lkml.kernel.org/r/20160811170340.9859-1-bp@alien8.de
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Rob Herring <robh@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agorelay: Use irq_work instead of plain timer for deferred wakeup
Peter Zijlstra [Tue, 11 Oct 2016 20:54:33 +0000 (13:54 -0700)]
relay: Use irq_work instead of plain timer for deferred wakeup

Relay avoids calling wake_up_interruptible() for doing the wakeup of
readers/consumers, waiting for the generation of new data, from the
context of a process which produced the data.  This is apparently done to
prevent the possibility of a deadlock in case Scheduler itself is is
generating data for the relay, after acquiring rq->lock.

The following patch used a timer (to be scheduled at next jiffy), for
delegating the wakeup to another context.
commit 7c9cb38302e78d24e37f7d8a2ea7eed4ae5f2fa7
Author: Tom Zanussi <zanussi@comcast.net>
Date:   Wed May 9 02:34:01 2007 -0700

relay: use plain timer instead of delayed work

relay doesn't need to use schedule_delayed_work() for waking readers
when a simple timer will do.

Scheduling a plain timer, at next jiffies boundary, to do the wakeup
causes a significant wakeup latency for the Userspace client, which makes
relay less suitable for the high-frequency low-payload use cases where the
data gets generated at a very high rate, like multiple sub buffers getting
filled within a milli second.  Moreover the timer is re-scheduled on every
newly produced sub buffer so the timer keeps getting pushed out if sub
buffers are filled in a very quick succession (less than a jiffy gap
between filling of 2 sub buffers).  As a result relay runs out of sub
buffers to store the new data.

By using irq_work it is ensured that wakeup of userspace client, blocked
in the poll call, is done at earliest (through self IPI or next timer
tick) enabling it to always consume the data in time.  Also this makes
relay consistent with printk & ring buffers (trace), as they too use
irq_work for deferred wake up of readers.

[arnd@arndb.de: select CONFIG_IRQ_WORK]
Link: http://lkml.kernel.org/r/20160912154035.3222156-1-arnd@arndb.de
[akpm@linux-foundation.org: coding-style fixes]
Link: http://lkml.kernel.org/r/1472906487-1559-1-git-send-email-akash.goel@intel.com
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agopps: kc: fix non-tickless system config dependency
Maciej S. Szmigiero [Tue, 11 Oct 2016 20:54:30 +0000 (13:54 -0700)]
pps: kc: fix non-tickless system config dependency

CONFIG_NO_HZ currently only sets the default value of dynticks config so
if PPS kernel consumer needs periodic timer ticks it should depend on
!CONFIG_NO_HZ_COMMON instead of !CONFIG_NO_HZ.

Otherwise it is possible to enable it even on tickless system which has
CONFIG_NO_HZ not set and CONFIG_NO_HZ_IDLE (or CONFIG_NO_HZ_FULL) set.

Link: http://lkml.kernel.org/r/57E2B769.50202@maciej.szmigiero.name
Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name>
Acked-by: Rodolfo Giometti <giometti@enneenne.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>