[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <1443504379-31841-4-git-send-email-tfiga@chromium.org>
Date: Tue, 29 Sep 2015 14:25:26 +0900
From: Tomasz Figa <tfiga@...omium.org>
To: iommu@...ts.linux-foundation.org
Cc: Tomasz Figa <tfiga@...omium.org>, Vince Hsu <vinceh@...dia.com>,
Joerg Roedel <joro@...tes.org>,
Hiroshi Doyu <hdoyu@...dia.com>,
Stephen Warren <swarren@...dotorg.org>,
Thierry Reding <thierry.reding@...il.com>,
Alexandre Courbot <gnurou@...il.com>,
Vince Hsu <vince.h@...dia.com>, Paul Walmsley <paul@...an.com>,
Russell King <rmk+kernel@....linux.org.uk>,
Mikko Perttunen <mperttunen@...dia.com>,
Tomeu Vizoso <tomeu.vizoso@...labora.com>,
Will Deacon <will.deacon@....com>,
Alex Williamson <alex.williamson@...hat.com>,
Marek Szyprowski <m.szyprowski@...sung.com>,
Arnd Bergmann <arnd@...db.de>,
Antonios Motakis <a.motakis@...tualopensystems.com>,
Nicolas Iooss <nicolas.iooss_linux@....org>,
Olav Haugan <ohaugan@...eaurora.org>,
linux-kernel@...r.kernel.org, linux-tegra@...r.kernel.org
Subject: [RFC PATCH 3/3] iommu/tegra-smmu: Make the driver use out of band flushing
This patch modifies the tegra-smmu driver to perform PTC and TLB flushes
inside iommu_ops .flush() callback instead of map and unmap operations,
so that performance of large maps and unmaps is heavily optimized due to
elimination of page-by-page flushing.
Signed-off-by: Tomasz Figa <tfiga@...omium.org>
Signed-off-by: Vince Hsu <vinceh@...dia.com>
---
drivers/iommu/tegra-smmu.c | 91 ++++++++++++++++++++++++++++++++++++++++++----
1 file changed, 83 insertions(+), 8 deletions(-)
diff --git a/drivers/iommu/tegra-smmu.c b/drivers/iommu/tegra-smmu.c
index 9305964..92b46d2 100644
--- a/drivers/iommu/tegra-smmu.c
+++ b/drivers/iommu/tegra-smmu.c
@@ -614,18 +614,54 @@ static void tegra_smmu_pte_put_use(struct tegra_smmu_as *as, unsigned long iova)
}
}
+static void tegra_smmu_pte_put_use_range(struct tegra_smmu_as *as,
+ unsigned long iova, unsigned int len)
+{
+ unsigned int i;
+
+ for (i = 0; i < len; i++)
+ tegra_smmu_pte_put_use(as, iova + i * PAGE_SIZE);
+}
+
static void tegra_smmu_set_pte(struct tegra_smmu_as *as, unsigned long iova,
u32 *pte, dma_addr_t pte_dma, u32 val)
{
- struct tegra_smmu *smmu = as->smmu;
- unsigned long offset = offset_in_page(pte);
-
*pte = val;
+}
+
+static void tegra_smmu_flush_pte_range(struct tegra_smmu_as *as,
+ unsigned long iova, unsigned int num_ptes, u32 *pte,
+ dma_addr_t pt_dma)
+{
+ struct tegra_smmu *smmu = as->smmu;
+ unsigned int tlb_lines_per_atom;
+ unsigned int ptes_per_tlb_line;
+ unsigned int ptes_per_atom;
+ unsigned long offset;
+ unsigned long iova_end;
+ int i;
+
+ ptes_per_atom = smmu->mc->soc->atom_size / sizeof(*pte);
+ ptes_per_tlb_line = smmu->soc->tlb_line_size / sizeof(*pte);
+ tlb_lines_per_atom = smmu->mc->soc->atom_size
+ / smmu->soc->tlb_line_size;
+
+ offset = round_down(offset_in_page(pte), smmu->mc->soc->atom_size);
+
+ iova_end = iova + num_ptes * PAGE_SIZE;
+ iova = round_down(iova, ptes_per_atom * PAGE_SIZE);
+ iova_end = round_up(iova_end, ptes_per_atom * PAGE_SIZE);
+ num_ptes = (iova_end - iova) / PAGE_SIZE;
+ while (num_ptes) {
+ smmu_flush_ptc(smmu, pt_dma, offset);
+ for (i = 0; i < tlb_lines_per_atom; i++) {
+ smmu_flush_tlb_group(smmu, as->id, iova);
+ iova += ptes_per_tlb_line * PAGE_SIZE;
+ }
+ offset += smmu->mc->soc->atom_size;
+ num_ptes -= ptes_per_atom;
+ }
- dma_sync_single_range_for_device(smmu->dev, pte_dma, offset,
- 4, DMA_TO_DEVICE);
- smmu_flush_ptc(smmu, pte_dma, offset);
- smmu_flush_tlb_group(smmu, as->id, iova);
smmu_flush(smmu);
}
@@ -662,11 +698,49 @@ static size_t tegra_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
return 0;
tegra_smmu_set_pte(as, iova, pte, pte_dma, 0);
- tegra_smmu_pte_put_use(as, iova);
return size;
}
+static void tegra_smmu_flush(struct iommu_domain *domain, unsigned long iova,
+ size_t size)
+{
+ struct tegra_smmu_as *as = to_smmu_as(domain);
+ struct tegra_smmu *smmu = as->smmu;
+ u32 num = size >> PAGE_SHIFT;
+
+ might_sleep();
+
+ while (num) {
+ unsigned int pt_index = iova_pt_index(iova);
+ unsigned int len, end;
+ unsigned long offset;
+ dma_addr_t pte_dma;
+ u32 *pte;
+
+ end = pt_index + num;
+ if (end > SMMU_NUM_PTE)
+ end = SMMU_NUM_PTE;
+ len = end - pt_index;
+
+ pte = tegra_smmu_pte_lookup(as, iova, &pte_dma);
+ if (!pte)
+ goto next_pde;
+
+ offset = offset_in_page(pte);
+ dma_sync_single_range_for_device(smmu->dev, pte_dma, offset,
+ sizeof(*pte) * len, DMA_TO_DEVICE);
+
+ tegra_smmu_flush_pte_range(as, iova, len, pte, pte_dma);
+ if (*pte == 0)
+ tegra_smmu_pte_put_use_range(as, iova, len);
+
+next_pde:
+ num -= len;
+ iova += len << PAGE_SHIFT;
+ }
+}
+
static phys_addr_t tegra_smmu_iova_to_phys(struct iommu_domain *domain,
dma_addr_t iova)
{
@@ -743,6 +817,7 @@ static const struct iommu_ops tegra_smmu_ops = {
.map = tegra_smmu_map,
.unmap = tegra_smmu_unmap,
.map_sg = default_iommu_map_sg,
+ .flush = tegra_smmu_flush,
.iova_to_phys = tegra_smmu_iova_to_phys,
.pgsize_bitmap = SZ_4K,
--
2.6.0.rc2.230.g3dd15c0
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists