Re: Fwd: [PATCH v7 2/3] iommu/arm-smmu-v3: Add workaround for Cavium ThunderX2 erratum #74

[Jayachandran C <jnair-M3mlKVOIwJVv6pq1l3V1OdBPR1lH4CV8@public.gmane.org>]

On Fri, Jun 09, 2017 at 04:43:07PM +0100, Robin Murphy wrote:
> On 09/06/17 12:38, Jayachandran C wrote:
> > On Fri, Jun 09, 2017 Robin Murphy wrote:
> >>
> >> On 30/05/17 13:03, Geetha sowjanya wrote:
> >>> From: Linu Cherian <linu.cherian-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
> >>>
> >>> Cavium ThunderX2 SMMU implementation doesn't support page 1 register space
> >>> and PAGE0_REGS_ONLY option is enabled as an errata workaround.
> >>> This option when turned on, replaces all page 1 offsets used for
> >>> EVTQ_PROD/CONS, PRIQ_PROD/CONS register access with page 0 offsets.
> >>>
> >>> SMMU resource size checks are now based on SMMU option PAGE0_REGS_ONLY,
> >>> since resource size can be either 64k/128k.
> >>> For this, arm_smmu_device_dt_probe/acpi_probe has been moved before
> >>> platform_get_resource call, so that SMMU options are set beforehand.
> >>>
> >>> Signed-off-by: Linu Cherian <linu.cherian-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
> >>> Signed-off-by: Geetha Sowjanya <geethasowjanya.akula-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
> >>> ---
> >>>  Documentation/arm64/silicon-errata.txt             |    1 +
> >>>  .../devicetree/bindings/iommu/arm,smmu-v3.txt      |    6 ++
> >>>  drivers/iommu/arm-smmu-v3.c                        |   64 +++++++++++++++-----
> >>>  3 files changed, 56 insertions(+), 15 deletions(-)
> >>>
> >>> diff --git a/Documentation/arm64/silicon-errata.txt b/Documentation/arm64/silicon-errata.txt
> >>> index 10f2ddd..4693a32 100644
> >>> --- a/Documentation/arm64/silicon-errata.txt
> >>> +++ b/Documentation/arm64/silicon-errata.txt
> >>> @@ -62,6 +62,7 @@ stable kernels.
> >>>  | Cavium         | ThunderX GICv3  | #23154          | CAVIUM_ERRATUM_23154        |
> >>>  | Cavium         | ThunderX Core   | #27456          | CAVIUM_ERRATUM_27456        |
> >>>  | Cavium         | ThunderX SMMUv2 | #27704          | N/A                         |
> >>> +| Cavium         | ThunderX2 SMMUv3| #74             | N/A                         |
> >>>  |                |                 |                 |                             |
> >>>  | Freescale/NXP  | LS2080A/LS1043A | A-008585        | FSL_ERRATUM_A008585         |
> >>>  |                |                 |                 |                             |
> >>> diff --git a/Documentation/devicetree/bindings/iommu/arm,smmu-v3.txt b/Documentation/devicetree/bindings/iommu/arm,smmu-v3.txt
> >>> index be57550..607e270 100644
> >>> --- a/Documentation/devicetree/bindings/iommu/arm,smmu-v3.txt
> >>> +++ b/Documentation/devicetree/bindings/iommu/arm,smmu-v3.txt
> >>> @@ -49,6 +49,12 @@ the PCIe specification.
> >>>  - hisilicon,broken-prefetch-cmd
> >>>                      : Avoid sending CMD_PREFETCH_* commands to the SMMU.
> >>>
> >>> +- cavium,cn9900-broken-page1-regspace
> >>> +                    : Replaces all page 1 offsets used for EVTQ_PROD/CONS,
> >>> +                                             PRIQ_PROD/CONS register access with page 0 offsets.
> >>> +                                             Set for Caviun ThunderX2 silicon that doesn't support
> >>> +                                             SMMU page1 register space.
> >>
> >> The indentation's a bit funky here - the rest of this file is actually
> >> indented with spaces, but either way it's clear your editor isn't set to
> >> 8-space tabs ;)
> >>
> >>> +
> >>>  ** Example
> >>>
> >>>          smmu@2b400000 {
> >>> diff --git a/drivers/iommu/arm-smmu-v3.c b/drivers/iommu/arm-smmu-v3.c
> >>> index 380969a..4e80205 100644
> >>> --- a/drivers/iommu/arm-smmu-v3.c
> >>> +++ b/drivers/iommu/arm-smmu-v3.c
> >>> @@ -412,6 +412,9 @@
> >>>  #define MSI_IOVA_BASE                        0x8000000
> >>>  #define MSI_IOVA_LENGTH                      0x100000
> >>>
> >>> +#define ARM_SMMU_PAGE0_REGS_ONLY(smmu)               \
> >>> +     ((smmu)->options & ARM_SMMU_OPT_PAGE0_REGS_ONLY)
> >>
> >> At the two places we use this macro, frankly I think it would be clearer
> >> to just reference smmu->options directly, as we currently do for
> >> SKIP_PREFETCH. The abstraction also adds more lines than it saves...
> >>
> >>> +
> >>>  static bool disable_bypass;
> >>>  module_param_named(disable_bypass, disable_bypass, bool, S_IRUGO);
> >>>  MODULE_PARM_DESC(disable_bypass,
> >>> @@ -597,6 +600,7 @@ struct arm_smmu_device {
> >>>       u32                             features;
> >>>
> >>>  #define ARM_SMMU_OPT_SKIP_PREFETCH   (1 << 0)
> >>> +#define ARM_SMMU_OPT_PAGE0_REGS_ONLY    (1 << 1)
> >>
> >> Whitespace again, although this time it's spaces where there should be a
> >> tab.
> >>
> >>>       u32                             options;
> >>>
> >>>       struct arm_smmu_cmdq            cmdq;
> >>> @@ -663,9 +667,19 @@ struct arm_smmu_option_prop {
> >>>
> >>>  static struct arm_smmu_option_prop arm_smmu_options[] = {
> >>>       { ARM_SMMU_OPT_SKIP_PREFETCH, "hisilicon,broken-prefetch-cmd" },
> >>> +     { ARM_SMMU_OPT_PAGE0_REGS_ONLY, "cavium,cn9900-broken-page1-regspace"},
> >>>       { 0, NULL},
> >>>  };
> >>>
> >>> +static inline void __iomem *arm_smmu_page1_fixup(unsigned long offset,
> >>> +                                              struct arm_smmu_device *smmu)
> >>> +{
> >>> +     if (offset > SZ_64K && ARM_SMMU_PAGE0_REGS_ONLY(smmu))
> >>> +             offset -= SZ_64K;
> >>> +
> >>> +     return smmu->base + offset;
> >>> +}
> >>> +
> >>>  static struct arm_smmu_domain *to_smmu_domain(struct iommu_domain *dom)
> >>>  {
> >>>       return container_of(dom, struct arm_smmu_domain, domain);
> >>> @@ -1961,8 +1975,8 @@ static int arm_smmu_init_one_queue(struct arm_smmu_device *smmu,
> >>>               return -ENOMEM;
> >>>       }
> >>>
> >>> -     q->prod_reg     = smmu->base + prod_off;
> >>> -     q->cons_reg     = smmu->base + cons_off;
> >>> +     q->prod_reg     = arm_smmu_page1_fixup(prod_off, smmu);
> >>> +     q->cons_reg     = arm_smmu_page1_fixup(cons_off, smmu);
> >>>       q->ent_dwords   = dwords;
> >>>
> >>>       q->q_base  = Q_BASE_RWA;
> >>> @@ -2363,8 +2377,10 @@ static int arm_smmu_device_reset(struct arm_smmu_device *smmu, bool bypass)
> >>>
> >>>       /* Event queue */
> >>>       writeq_relaxed(smmu->evtq.q.q_base, smmu->base + ARM_SMMU_EVTQ_BASE);
> >>> -     writel_relaxed(smmu->evtq.q.prod, smmu->base + ARM_SMMU_EVTQ_PROD);
> >>> -     writel_relaxed(smmu->evtq.q.cons, smmu->base + ARM_SMMU_EVTQ_CONS);
> >>> +     writel_relaxed(smmu->evtq.q.prod,
> >>> +                    arm_smmu_page1_fixup(ARM_SMMU_EVTQ_PROD, smmu));
> >>> +     writel_relaxed(smmu->evtq.q.cons,
> >>> +                    arm_smmu_page1_fixup(ARM_SMMU_EVTQ_CONS, smmu));
> > 
> > This sequence and the arm_smmu_page1_fixup() call is repeated in quite a few
> > places. I think this errata code is messy because the original driver does not
> > make the alias page usage explicit.
> 
> It *is* explicit - the architecture says the event queue and PRI queue
> pointers exist only on page 1, and that is the offset we define for
> them. The architecture also says "The equivalent Page 0 offsets of
> registers that are defined on Page 1 are Reserved and ARM recommends
> that they are not accessed. Access to these offsets is CONSTRAINED
> UNPREDICTABLE..."
 
Ok. The patch makes the page used for producer/consumer queue registers
explicit fo cmdq (page 0) and the eventq/priq(page1). There is no suggestion
here to use page 0 address for eventq/priq.

> This workaround is a bodge dependent on a specific implementation always
> having a specific CONSTRAINED UNPREDICTABLE behaviour, and I see no
> point in trying to dress it up as anything else. Yes, it could be
> considered a little bit messy, but messy is what you get when you step
> outside the spec. The fixup is invoked a grand total of 6 times, over 3
> locations, and there's no way of factoring it out further that doesn't
> just add significantly more code and complexity than it would save.
 
With the changes below, the fixup are simpler and will be needed only in
2 places - and it gets the hardware init repeated 3 times in the driver
into a single place. That bikeshed will look real nice :)

JC.

> 
> > A patch like the one below (no functional changes) would clean up the original
> > driver and make the errata change much simpler - any comments?
> > 
> > -- >8 --
> > 
> > Date: Tue, 30 May 2017 15:43:29 +0000
> > Subject: [PATCH] iommu: arm-smmu-v3: make alias page usage explicit
> > 
> > ---
> >  drivers/iommu/arm-smmu-v3.c | 76 +++++++++++++++++++++++++++------------------
> >  1 file changed, 46 insertions(+), 30 deletions(-)
> > 
> > diff --git a/drivers/iommu/arm-smmu-v3.c b/drivers/iommu/arm-smmu-v3.c
> > index 380969a..11fdb4f 100644
> > --- a/drivers/iommu/arm-smmu-v3.c
> > +++ b/drivers/iommu/arm-smmu-v3.c
> > @@ -171,20 +171,19 @@
> >  #define STRTAB_BASE_CFG_FMT_LINEAR	(0 << STRTAB_BASE_CFG_FMT_SHIFT)
> >  #define STRTAB_BASE_CFG_FMT_2LVL	(1 << STRTAB_BASE_CFG_FMT_SHIFT)
> >  
> > +#define ARM_SMMU_Q_PROD(qbase)		((qbase) + 0x8)
> > +#define ARM_SMMU_Q_PROD_PAGE1(qbase)	((qbase) + 0x10008)
> > +#define ARM_SMMU_Q_CONS(qbase)		((qbase) + 0xc)
> > +#define ARM_SMMU_Q_CONS_PAGE1(qbase)	((qbase) + 0x1000c)
> > +
> >  #define ARM_SMMU_CMDQ_BASE		0x90
> > -#define ARM_SMMU_CMDQ_PROD		0x98
> > -#define ARM_SMMU_CMDQ_CONS		0x9c
> >  
> >  #define ARM_SMMU_EVTQ_BASE		0xa0
> > -#define ARM_SMMU_EVTQ_PROD		0x100a8
> > -#define ARM_SMMU_EVTQ_CONS		0x100ac
> >  #define ARM_SMMU_EVTQ_IRQ_CFG0		0xb0
> >  #define ARM_SMMU_EVTQ_IRQ_CFG1		0xb8
> >  #define ARM_SMMU_EVTQ_IRQ_CFG2		0xbc
> >  
> >  #define ARM_SMMU_PRIQ_BASE		0xc0
> > -#define ARM_SMMU_PRIQ_PROD		0x100c8
> > -#define ARM_SMMU_PRIQ_CONS		0x100cc
> >  #define ARM_SMMU_PRIQ_IRQ_CFG0		0xd0
> >  #define ARM_SMMU_PRIQ_IRQ_CFG1		0xd8
> >  #define ARM_SMMU_PRIQ_IRQ_CFG2		0xdc
> > @@ -1946,11 +1945,30 @@ static struct iommu_ops arm_smmu_ops = {
> >  };
> >  
> >  /* Probing and initialisation functions */
> > +static int arm_smmu_reset_one_queue(struct arm_smmu_device *smmu,
> > +				    struct arm_smmu_queue *q,
> > +				    unsigned long qoffset,
> > +				    int page_to_use)
> > +{
> > +	unsigned long prod, cons;
> > +
> > +	writeq_relaxed(q->q_base, smmu->base + qoffset);
> > +	if (page_to_use == 1) {
> > +		prod = ARM_SMMU_Q_PROD_PAGE1(qoffset);
> > +		cons = ARM_SMMU_Q_CONS_PAGE1(qoffset);
> > +	} else {
> > +		prod = ARM_SMMU_Q_PROD(qoffset);
> > +		cons = ARM_SMMU_Q_CONS(qoffset);
> > +	}
> > +	writeq_relaxed(q->prod, smmu->base + prod);
> > +	writeq_relaxed(q->cons, smmu->base + cons);
> > +}
> > +
> >  static int arm_smmu_init_one_queue(struct arm_smmu_device *smmu,
> >  				   struct arm_smmu_queue *q,
> > -				   unsigned long prod_off,
> > -				   unsigned long cons_off,
> > -				   size_t dwords)
> > +				   unsigned long qoffset,
> > +				   size_t dwords,
> > +				   int page_to_use)
> >  {
> >  	size_t qsz = ((1 << q->max_n_shift) * dwords) << 3;
> >  
> > @@ -1961,8 +1979,13 @@ static int arm_smmu_init_one_queue(struct arm_smmu_device *smmu,
> >  		return -ENOMEM;
> >  	}
> >  
> > -	q->prod_reg	= smmu->base + prod_off;
> > -	q->cons_reg	= smmu->base + cons_off;
> > +	if (page_to_use == 1) {
> > +		q->prod_reg	= smmu->base + ARM_SMMU_Q_PROD_PAGE1(qoffset);
> > +		q->cons_reg	= smmu->base + ARM_SMMU_Q_CONS_PAGE1(qoffset);
> > +	} else {
> > +		q->prod_reg	= smmu->base + ARM_SMMU_Q_PROD(qoffset);
> > +		q->cons_reg	= smmu->base + ARM_SMMU_Q_CONS(qoffset);
> > +	}
> >  	q->ent_dwords	= dwords;
> >  
> >  	q->q_base  = Q_BASE_RWA;
> > @@ -1980,14 +2003,14 @@ static int arm_smmu_init_queues(struct arm_smmu_device *smmu)
> >  
> >  	/* cmdq */
> >  	spin_lock_init(&smmu->cmdq.lock);
> > -	ret = arm_smmu_init_one_queue(smmu, &smmu->cmdq.q, ARM_SMMU_CMDQ_PROD,
> > -				      ARM_SMMU_CMDQ_CONS, CMDQ_ENT_DWORDS);
> > +	ret = arm_smmu_init_one_queue(smmu, &smmu->cmdq.q, ARM_SMMU_CMDQ_BASE,
> > +				      CMDQ_ENT_DWORDS, 0);
> >  	if (ret)
> >  		return ret;
> >  
> >  	/* evtq */
> > -	ret = arm_smmu_init_one_queue(smmu, &smmu->evtq.q, ARM_SMMU_EVTQ_PROD,
> > -				      ARM_SMMU_EVTQ_CONS, EVTQ_ENT_DWORDS);
> > +	ret = arm_smmu_init_one_queue(smmu, &smmu->evtq.q, ARM_SMMU_EVTQ_BASE,
> > +				      EVTQ_ENT_DWORDS, USE_PAGE1);
> >  	if (ret)
> >  		return ret;
> >  
> > @@ -1995,8 +2018,8 @@ static int arm_smmu_init_queues(struct arm_smmu_device *smmu)
> >  	if (!(smmu->features & ARM_SMMU_FEAT_PRI))
> >  		return 0;
> >  
> > -	return arm_smmu_init_one_queue(smmu, &smmu->priq.q, ARM_SMMU_PRIQ_PROD,
> > -				       ARM_SMMU_PRIQ_CONS, PRIQ_ENT_DWORDS);
> > +	return arm_smmu_init_one_queue(smmu, &smmu->priq.q, ARM_SMMU_PRIQ_BASE,
> > +				       PRIQ_ENT_DWORDS, 1);
> >  }
> >  
> >  static int arm_smmu_init_l1_strtab(struct arm_smmu_device *smmu)
> > @@ -2332,9 +2355,8 @@ static int arm_smmu_device_reset(struct arm_smmu_device *smmu, bool bypass)
> >  		       smmu->base + ARM_SMMU_STRTAB_BASE_CFG);
> >  
> >  	/* Command queue */
> > -	writeq_relaxed(smmu->cmdq.q.q_base, smmu->base + ARM_SMMU_CMDQ_BASE);
> > -	writel_relaxed(smmu->cmdq.q.prod, smmu->base + ARM_SMMU_CMDQ_PROD);
> > -	writel_relaxed(smmu->cmdq.q.cons, smmu->base + ARM_SMMU_CMDQ_CONS);
> > +	arm_smmu_reset_one_queue(smmu, &smmu->evtq.q,
> > +				 ARM_SMMU_CMDQ_BASE, 0);
> >  
> >  	enables = CR0_CMDQEN;
> >  	ret = arm_smmu_write_reg_sync(smmu, enables, ARM_SMMU_CR0,
> > @@ -2362,9 +2384,8 @@ static int arm_smmu_device_reset(struct arm_smmu_device *smmu, bool bypass)
> >  	arm_smmu_cmdq_issue_cmd(smmu, &cmd);
> >  
> >  	/* Event queue */
> > -	writeq_relaxed(smmu->evtq.q.q_base, smmu->base + ARM_SMMU_EVTQ_BASE);
> > -	writel_relaxed(smmu->evtq.q.prod, smmu->base + ARM_SMMU_EVTQ_PROD);
> > -	writel_relaxed(smmu->evtq.q.cons, smmu->base + ARM_SMMU_EVTQ_CONS);
> > +	arm_smmu_reset_one_queue(smmu, &smmu->evtq.q,
> > +				 ARM_SMMU_EVTQ_BASE, 1);
> >  
> >  	enables |= CR0_EVTQEN;
> >  	ret = arm_smmu_write_reg_sync(smmu, enables, ARM_SMMU_CR0,
> > @@ -2376,13 +2397,8 @@ static int arm_smmu_device_reset(struct arm_smmu_device *smmu, bool bypass)
> >  
> >  	/* PRI queue */
> >  	if (smmu->features & ARM_SMMU_FEAT_PRI) {
> > -		writeq_relaxed(smmu->priq.q.q_base,
> > -			       smmu->base + ARM_SMMU_PRIQ_BASE);
> > -		writel_relaxed(smmu->priq.q.prod,
> > -			       smmu->base + ARM_SMMU_PRIQ_PROD);
> > -		writel_relaxed(smmu->priq.q.cons,
> > -			       smmu->base + ARM_SMMU_PRIQ_CONS);
> > -
> > +		arm_smmu_reset_one_queue(smmu, &smmu->priq.q,
> > +					 ARM_SMMU_PRIQ_BASE, 1);
> >  		enables |= CR0_PRIQEN;
> >  		ret = arm_smmu_write_reg_sync(smmu, enables, ARM_SMMU_CR0,
> >  					      ARM_SMMU_CR0ACK);
> > 
>