Hi Hari,
From: "ext Kanigeri, Hari" <h-kanige...@ti.com>
Subject: [PATCH 4/4][v3] OMAP:iommu- add TLB preservation support
Date: Wed, 21 Apr 2010 00:57:10 +0200
> From b76cbee7bb442f61dd2f29ed64ccf98efd871791 Mon Sep 17 00:00:00 2001
> From: Hari Kanigeri <h-kanige...@ti.com>
> Date: Tue, 20 Apr 2010 12:11:16 -0500
> Subject: [PATCH 4/4] OMAP:iommu- add TLB preservation support
>
> This patch adds TLB preservation support to IOMMU module
>
> Signed-off-by: Hari Kanigeri <h-kanige...@ti.com>
> ---
> arch/arm/mach-omap2/iommu2.c | 4 +++-
> arch/arm/plat-omap/iommu.c | 12 ++++++++----
> 2 files changed, 11 insertions(+), 5 deletions(-)
>
> diff --git a/arch/arm/mach-omap2/iommu2.c b/arch/arm/mach-omap2/iommu2.c
> index f01f985..5ce76e4 100644
> --- a/arch/arm/mach-omap2/iommu2.c
> +++ b/arch/arm/mach-omap2/iommu2.c
> @@ -146,6 +146,7 @@ static u32 omap2_iommu_fault_isr(struct iommu *obj, u32
> *ra)
> printk("\n");
>
> iommu_write_reg(obj, stat, MMU_IRQSTATUS);
> + omap2_iommu_disable(obj);
> return stat;
> }
>
> @@ -211,7 +212,8 @@ static ssize_t omap2_dump_cr(struct iommu *obj, struct
> cr_regs *cr, char *buf)
> char *p = buf;
>
> /* FIXME: Need more detail analysis of cam/ram */
> - p += sprintf(p, "%08x %08x\n", cr->cam, cr->ram);
> + p += sprintf(p, "%08x %08x %01x\n", cr->cam, cr->ram,
> + (cr->cam & MMU_CAM_P) ? 1 : 0);
>
> return p - buf;
> }
> diff --git a/arch/arm/plat-omap/iommu.c b/arch/arm/plat-omap/iommu.c
> index 463d638..a1d4ccf 100644
> --- a/arch/arm/plat-omap/iommu.c
> +++ b/arch/arm/plat-omap/iommu.c
> @@ -171,15 +171,12 @@ static void iotlb_lock_get(struct iommu *obj, struct
> iotlb_lock *l)
> l->base = MMU_LOCK_BASE(val);
> l->vict = MMU_LOCK_VICT(val);
>
> - BUG_ON(l->base != 0); /* Currently no preservation is used */
> }
>
> static void iotlb_lock_set(struct iommu *obj, struct iotlb_lock *l)
> {
> u32 val;
>
> - BUG_ON(l->base != 0); /* Currently no preservation is used */
> -
> val = (l->base << MMU_LOCK_BASE_SHIFT);
> val |= (l->vict << MMU_LOCK_VICT_SHIFT);
>
> @@ -241,6 +238,11 @@ int load_iotlb_entry(struct iommu *obj, struct
> iotlb_entry *e)
> break;
> }
>
> + if (l.base == obj->nr_tlb_entries) {
> + dev_dbg(obj->dev, "%s: preserve entries full\n", __func__);
> + err = -EBUSY;
I am afraid that this doesn't work. There are two cases where TLB
entries are set/updated:
1) H/W TWL always updates TLB entries automatically, IOW, increments
"vict" implicitly.
2) A client module adds preservation TLB entry with explictly calling
"load_iotlb_entry()". IOW, increments both "vict" and "base"
intentionally.
Because of 1), "vict" gets full soon and this is normal, but even if
this "vict" is full, a preservation entry "base" should be able to be
set by a client module unless "base" itself rearches full. IOW, Even
if all TLB entries are valid, a new preservation TLB entry should be
set by replacing one of the existing normal entry.
In the case of "e" with perservation bit , "load_iotlb_entry()"
doesn't usually fail because it replaces the exisiting normal entry,
but it would fail only if all TLB entries are preserved.
> + goto out;
> + }
> if (i == obj->nr_tlb_entries) {
> dev_dbg(obj->dev, "%s: full: no entry\n", __func__);
> err = -EBUSY;
> @@ -256,9 +258,11 @@ int load_iotlb_entry(struct iommu *obj, struct
> iotlb_entry *e)
> iotlb_load_cr(obj, cr);
> kfree(cr);
>
> + if (e->prsvd)
> + l.base++;
> /* increment victim for next tlb load */
> if (++l.vict == obj->nr_tlb_entries)
> - l.vict = 0;
> + l.vict = l.base;
> iotlb_lock_set(obj, &l);
> out:
> clk_disable(obj->clk);
> --
> 1.7.0
>
>
> Thank you,
> Best regards,
> Hari
>
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
