Re: [PATCH v4 2/5] irqchip/gic-v3-its: add ability to save/restore ITS state

[dbasehore .]

On Tue, Feb 6, 2018 at 8:23 AM, Marc Zyngier <marc.zyngier@xxxxxxx> wrote:
> On 05/02/18 21:33, dbasehore . wrote:
>> On Mon, Feb 5, 2018 at 7:56 AM, Marc Zyngier <marc.zyngier@xxxxxxx> wrote:
>>> On 03/02/18 01:24, Derek Basehore wrote:
>>>> Some platforms power off GIC logic in suspend, so we need to
>>>> save/restore state. The distributor and redistributor registers need
>>>> to be handled in platform code due to access permissions on those
>>>> registers, but the ITS registers can be restored in the kernel.
>>>>
>>>> Signed-off-by: Derek Basehore <dbasehore@xxxxxxxxxxxx>
>>>> ---
>>>>  drivers/irqchip/irq-gic-v3-its.c | 101 +++++++++++++++++++++++++++++++++++++++
>>>>  1 file changed, 101 insertions(+)
>>>>
>>>> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
>>>> index 06f025fd5726..e13515cdb68f 100644
>>>> --- a/drivers/irqchip/irq-gic-v3-its.c
>>>> +++ b/drivers/irqchip/irq-gic-v3-its.c
>>>> @@ -33,6 +33,7 @@
>>>>  #include <linux/of_platform.h>
>>>>  #include <linux/percpu.h>
>>>>  #include <linux/slab.h>
>>>> +#include <linux/syscore_ops.h>
>>>>
>>>>  #include <linux/irqchip.h>
>>>>  #include <linux/irqchip/arm-gic-v3.h>
>>>> @@ -46,6 +47,7 @@
>>>>  #define ITS_FLAGS_CMDQ_NEEDS_FLUSHING                (1ULL << 0)
>>>>  #define ITS_FLAGS_WORKAROUND_CAVIUM_22375    (1ULL << 1)
>>>>  #define ITS_FLAGS_WORKAROUND_CAVIUM_23144    (1ULL << 2)
>>>> +#define ITS_FLAGS_SAVE_SUSPEND_STATE         (1ULL << 3)
>>>>
>>>>  #define RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING  (1 << 0)
>>>>
>>>> @@ -83,6 +85,15 @@ struct its_baser {
>>>>       u32             psz;
>>>>  };
>>>>
>>>> +/*
>>>> + * Saved ITS state - this is where saved state for the ITS is stored
>>>> + * when it's disabled during system suspend.
>>>> + */
>>>> +struct its_ctx {
>>>> +     u64                     cbaser;
>>>> +     u32                     ctlr;
>>>> +};
>>>
>>> Nit: This is pretty small for the ITS context. Given that its_node is
>>> the context, you can safely expand this in the its_node structure.
>>
>> Sounds reasonable. I think I just have it this way because I used to
>> have more in here.
>>
>>>
>>>> +
>>>>  struct its_device;
>>>>
>>>>  /*
>>>> @@ -101,6 +112,7 @@ struct its_node {
>>>>       struct its_collection   *collections;
>>>>       struct fwnode_handle    *fwnode_handle;
>>>>       u64                     (*get_msi_base)(struct its_device *its_dev);
>>>> +     struct its_ctx          its_ctx;
>>>>       struct list_head        its_device_list;
>>>>       u64                     flags;
>>>>       unsigned long           list_nr;
>>>> @@ -3042,6 +3054,90 @@ static void its_enable_quirks(struct its_node *its)
>>>>       gic_enable_quirks(iidr, its_quirks, its);
>>>>  }
>>>>
>>>> +static int its_save_disable(void)
>>>> +{
>>>> +     struct its_node *its;
>>>> +     int err = 0;
>>>> +
>>>> +     spin_lock(&its_lock);
>>>> +     list_for_each_entry(its, &its_nodes, entry) {
>>>> +             struct its_ctx *ctx;
>>>> +             void __iomem *base;
>>>> +
>>>> +             if (!(its->flags & ITS_FLAGS_SAVE_SUSPEND_STATE))
>>>> +                     continue;
>>>> +
>>>> +             ctx = &its->its_ctx;
>>>> +             base = its->base;
>>>> +             ctx->ctlr = readl_relaxed(base + GITS_CTLR);
>>>> +             err = its_force_quiescent(base);
>>>> +             if (err) {
>>>> +                     pr_err("ITS failed to quiesce\n");
>>>> +                     writel_relaxed(ctx->ctlr, base + GITS_CTLR);
>>>> +                     goto err;
>>>> +             }
>>>> +
>>>> +             ctx->cbaser = gits_read_cbaser(base + GITS_CBASER);
>>>> +     }
>>>> +
>>>> +err:
>>>> +     if (err) {
>>>> +             list_for_each_entry_continue_reverse(its, &its_nodes, entry) {
>>>> +                     if (its->flags & ITS_FLAGS_SAVE_SUSPEND_STATE) {
>>>> +                             struct its_ctx *ctx = &its->its_ctx;
>>>> +                             void __iomem *base = its->base;
>>>> +
>>>> +                             writel_relaxed(ctx->ctlr, base + GITS_CTLR);
>>>> +                     }
>>>> +             }
>>>> +     }
>>>> +
>>>> +     spin_unlock(&its_lock);
>>>> +
>>>> +     return err;
>>>> +}
>>>> +
>>>> +static void its_restore_enable(void)
>>>> +{
>>>> +     struct its_node *its;
>>>> +
>>>> +     spin_lock(&its_lock);
>>>> +     list_for_each_entry(its, &its_nodes, entry) {
>>>> +             if (its->flags & ITS_FLAGS_SAVE_SUSPEND_STATE) {
>>>> +                     struct its_ctx *ctx = &its->its_ctx;
>>>> +                     void __iomem *base = its->base;
>>>> +                     /*
>>>> +                      * Only the lower 32 bits matter here since the upper 32
>>>> +                      * don't include any of the offset.
>>>> +                      */
>>>> +                     u32 creader = readl_relaxed(base + GITS_CREADR);
>>>
>>> Accessor matching gits_write_cwriter and co?
>>>
>>>> +                     int i;
>>>> +
>>>> +                     /*
>>>> +                      * Reset the write location to where the ITS is
>>>> +                      * currently at.
>>>> +                      */
>>>> +                     gits_write_cbaser(ctx->cbaser, base + GITS_CBASER);
>>>> +                     gits_write_cwriter(creader, base + GITS_CWRITER);
>>>> +                     its->cmd_write = &its->cmd_base[
>>>> +                             creader / sizeof(struct its_cmd_block)];
>>>
>>> Nit: please do not split lines like this, this is unreadable. We both
>>> have screens that are wide enough for this to fit on a single line.
>>>
>>> More importantly: Why isn't it sufficient to reset both CREADR and
>>> CWRITER to zero? Is there any case where you can suspend whilst having
>>> anything in flight?
>>
>> CREADR is RO and we need to handle the non-zero case. I was planning
>> on getting rid of the write to CWRITER since it shouldn't be needed.
>> Either CREADR and CWRITER have the prior values, or both are reset to
>> 0.
>
> You're writing GITS_CBASER, which has for consequence: "When this
> register is successfully written, the value of GITS_CREADR is set to
> zero.". Ergo, none of that is necessary and you *must* set CWRITER to 0.

Huh, I missed that. It at least makes this a lot simpler. That also
means that the current code has a potential bug if suspend is
preempted.

>
>>
>>>
>>>> +                     /* Restore GITS_BASER from the value cache. */
>>>> +                     for (i = 0; i < GITS_BASER_NR_REGS; i++) {
>>>> +                             struct its_baser *baser = &its->tables[i];
>>>> +
>>>> +                             its_write_baser(its, baser, baser->val);
>>>
>>> You may want to first test that this BASER register is actually
>>> requiring something before writing to it. Yes, this is normally safe.
>>> But HW is also normally broken.
>>>
>>>> +                     }
>>>> +                     writel_relaxed(ctx->ctlr, base + GITS_CTLR);
>>>
>>> Before restoring all of this, shouldn't you first test that the ITS is
>>> actually in a disabled state?
>>
>> The save_disable code put it in a disabled state. The reset state is> also disabled. I don't expect PSCI to change the GITS_CTLR register at
>> all. Are you expecting something on the other side of the PSCI layer
>> to poke this register? We'd probably want to force disable at the
>> start of restore_enable if so.
>
> I expect firmware to do its worse, because even if mainline ATF is close
> to perfection, it will be hacked to death by platform people who usually
> do not have a clue (that's definitely my experience). So I expect this
> code to be written defensively.

Fair enough. I'll disable the ITS at the start of restore_enable.

>
> Thanks,
>
>         M.
> --
> Jazz is not dead. It just smells funny...