[PATCHv3 00/16] Suspend to RAM support for Armada XP

Thomas Petazzoni thomas.petazzoni at free-electrons.com
Fri Nov 21 07:59:57 PST 2014


Hello,

This set of patches implement suspend to RAM for the Marvell Armada XP
platform, and specifically for the Armada XP GP board. For now, it
does not implement suspend/resume for all device drivers, it only
implements the core of the suspend/resume functionality: enough to
make sure that the system can enter suspend and resume from suspend,
restart all CPUs, and get back to a shell prompt. Additional work in
individual device drivers will be needed as follow-up patches.

There are two important things to understand about the hardware before
reading this patch series:

 - The Armada XP SoC cannot put itself into a suspend to RAM state. It
   requires an external circuit, which in most cases, is a PIC
   micro-controller, to turn off the SoC. This PIC is connected to the
   SoC using a set of GPIOs, which allow the SoC to talk to the PIC
   and request certain things to be done.

   This means that the Armada XP suspend/resume logic has some
   board-specific code, which is fairly uncommon for suspend/resume
   code in the ARM world.

   We have chosen to represent the PIC and the fact that it's
   connected to the SoC using 3 GPIOs in the Device
   Tree.

 - When exiting from suspend, the SoC is actually powered up again
   completely from scratch, so it goes through the bootloader. The
   kernel has to fill a bootloader-specific structure at a fixed
   physical address to pass information to the bootloader that will
   tell the bootloader to do a resume and therefore jump back directly
   into the kernel resume entry point, instead of doing a normal boot.

 - The bootloader has to reconfigure the DRAM controller, which
   involves executing some DDR3 training code that has the unfortunate
   side effect of overwriting the first 10 KB of each DRAM
   chip-select. We therefore have to ensure that the kernel does not
   use the first 10 KB of each DRAM chip select.

So, in sequence:

 * PATCH 1 and 2 are mainly preparation patches: Device Tree binding
   documentation, fixing an errata of the Armada XP.

 * PATCH 3 to 8 add suspend/resume support to some core drivers:
   irqchip, clocksource, gpio, mvebu-mbus and clock.

   Note that the gpio suspend/resume patch was said to be applied by
   Linus Walleij in
   http://lists.infradead.org/pipermail/linux-arm-kernel/2014-November/299201.html,
   but I don't see it present in
   https://git.kernel.org/cgit/linux/kernel/git/linusw/linux-gpio.git/log/?h=for-next,
   so I've kept it for now in this patch series.

 * PATCH 9 implements the core of the SoC suspend/resume logic.

 * PATCH 10 makes sure the first 10 KB of each DRAM chip select isn't
   used by the kernel.

 * PATCH 11 implements the board-specific suspend/resume logic,
   especially talking with the PIC micro-controller over GPIOs.

 * PATCH 12 and 13 make some other additional changes to the Armada XP
   SoC support to make suspend/resume work properly.

 * PATCH 14 to 16 modify the Armada XP and Armada XP GP Device Tree
   description to add the relevant hardware blocks that are needed for
   suspend/resume: description of the PIC micro-controller and the 3
   GPIOs connecting it to the SoC, an additional set of the register
   for mvebu-mbus, and the registers for the SDRAM controller.

Changes since v2:

 - Add Acked-by from Daniel Lezcano on "clocksource:
   time-armada-370-xp: add suspend/resume support".

 - In "clk: mvebu: add suspend/resume for gatable clocks", switch to a
   global syscore_ops structure instead of a per gatable clock
   controller instance. Suggested by Mike Turquette.

Changes since v1:

 - Add Reviewed-by from Grégory Clement on

   Documentation: dt-bindings: minimal documentation for MVEBU SDRAM controller
   gpio: mvebu: add suspend/resume support
   bus: mvebu-mbus: provide a mechanism to save SDRAM window configuration

 - Added Acked-by from Grégory Clement on

   ARM: mvebu: enable strex backoff delay
   ARM: mvebu: reserve the first 10 KB of each memory bank for suspend/resume
   ARM: mvebu: make sure MMU is disabled in armada_370_xp_cpu_resume
   ARM: mvebu: synchronize secondary CPU clocks on resume
   ARM: mvebu: add suspend/resume DT information for Armada XP GP
   ARM: mvebu: adjust mbus controller description on Armada 370/XP
   ARM: mvebu: add SDRAM controller description for Armada XP
   clk: mvebu: add suspend/resume for gatable clocks

 - Add Acked-by from Alexandre Courbot on
   gpio: mvebu: add suspend/resume support

 - Drop patch "irqchip: irq-armada-370-xp: use proper return value for
   ->set_affinity()" as it was applied by Jason Cooper.

 - Fix typo in commit log of "clocksource: time-armada-370-xp: add
   suspend/resume support", noticed by Grégory Clement.

 - Adjust the name of the new variables used in "clocksource:
   time-armada-370-xp: add suspend/resume support" to save the state
   of timer 0, to make it clear only the state of timer 0 is
   preserved. Suggested by Grégory Clement.

 - In the suspend/resume code of the mvebu-mbus driver, make sure to
   fail (and therefore abort the suspend) if the MBus bridge registers
   were not mapped, instead of continuing with the suspend procedure
   and later on have a system that we know cannot resume
   properly. Suggested by Grégory Clement.

 - Remove cpu_pm_cluster_{enter,exit} calls from
   mach-mvebu/pm.c. Suggested by Grégory Clement.

 - In patch "ARM: mvebu: Armada XP GP specific suspend/resume code",
   fixup the wait loop of 100 cycles before entering suspend to
   actually initialize the r1 register before entering the wait
   loop. Before that, r1 was uninitialized, and a JTAG has shown that
   it was generally defined to 0, so the loop started by decrementing
   it by 1, making r1 equal 0xFFFFFFFF. So we had 2^32 iterations of a
   two instructions loop, at 1.2 Ghz, about a 7 seconds
   delay. Switching to the 100 cycles loop that is needed showed that
   the previous mdelay of 250 ms was way too short: the PIC needs
   several seconds before the two GPIO toggles. The Marvell vendor
   kernel uses a bit more than 3 seconds delay, and experience has
   shown that a 3 seconds delay works fine, while shorter delays
   sometimes lead to a system that doesn't enter suspend. Noticed by
   Grégory Clement.

Best regards,

Thomas

Nadav Haklai (1):
  ARM: mvebu: enable strex backoff delay

Thomas Petazzoni (15):
  Documentation: dt-bindings: minimal documentation for MVEBU SDRAM
    controller
  irqchip: irq-armada-370-xp: suspend/resume support
  clocksource: time-armada-370-xp: add suspend/resume support
  gpio: mvebu: add suspend/resume support
  bus: mvebu-mbus: suspend/resume support
  bus: mvebu-mbus: provide a mechanism to save SDRAM window
    configuration
  clk: mvebu: add suspend/resume for gatable clocks
  ARM: mvebu: implement suspend/resume support for Armada XP
  ARM: mvebu: reserve the first 10 KB of each memory bank for
    suspend/resume
  ARM: mvebu: Armada XP GP specific suspend/resume code
  ARM: mvebu: make sure MMU is disabled in armada_370_xp_cpu_resume
  ARM: mvebu: synchronize secondary CPU clocks on resume
  ARM: mvebu: add suspend/resume DT information for Armada XP GP
  ARM: mvebu: adjust mbus controller description on Armada 370/XP
  ARM: mvebu: add SDRAM controller description for Armada XP

 .../devicetree/bindings/bus/mvebu-mbus.txt         |  17 +-
 .../memory-controllers/mvebu-sdram-controller.txt  |  21 ++
 arch/arm/boot/dts/armada-370-xp.dtsi               |   3 +-
 arch/arm/boot/dts/armada-xp-gp.dts                 |  19 +-
 arch/arm/boot/dts/armada-xp.dtsi                   |   5 +
 arch/arm/mach-mvebu/Makefile                       |   2 +-
 arch/arm/mach-mvebu/board-v7.c                     |  51 +++++
 arch/arm/mach-mvebu/common.h                       |   2 +
 arch/arm/mach-mvebu/platsmp.c                      |  31 ++-
 arch/arm/mach-mvebu/pm-board.c                     | 141 +++++++++++++
 arch/arm/mach-mvebu/pm.c                           | 218 +++++++++++++++++++++
 arch/arm/mach-mvebu/pmsu.h                         |   1 +
 arch/arm/mach-mvebu/pmsu_ll.S                      |   8 +
 arch/arm/mm/proc-v7.S                              |   2 -
 drivers/bus/mvebu-mbus.c                           | 180 ++++++++++++++++-
 drivers/clk/mvebu/common.c                         |  32 ++-
 drivers/clocksource/time-armada-370-xp.c           |  25 +++
 drivers/gpio/gpio-mvebu.c                          |  99 ++++++++++
 drivers/irqchip/irq-armada-370-xp.c                |  52 +++++
 include/linux/mbus.h                               |   1 +
 20 files changed, 876 insertions(+), 34 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/memory-controllers/mvebu-sdram-controller.txt
 create mode 100644 arch/arm/mach-mvebu/pm-board.c
 create mode 100644 arch/arm/mach-mvebu/pm.c

-- 
2.1.0




More information about the linux-arm-kernel mailing list