| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <1272058093-20914-12-git-send-email-corbet@lwn.net>
Date: Fri, 23 Apr 2010 15:28:11 -0600
From: Jonathan Corbet <corbet@....net>
To: linux-kernel@...r.kernel.org
Cc: Harald Welte <laforge@...monks.org>,
Deepak Saxena <dsaxena@...top.org>,
linux-fbdev@...r.kernel.org, JosephChan@....com.tw,
ScottFang@...tech.com.cn,
Florian Tobias Schandinat <FlorianSchandinat@....de>
Subject: [PATCH 11/13] viafb: Add a simple VX855 DMA engine driver
This code provides a minimal amount of access to the DMA engine as
needed by the camera driver. VX855 only; it's guaranteed not to work
on other chipsets, so it won't try.
Signed-off-by: Jonathan Corbet <corbet@....net>
---
drivers/video/via/via-core.c | 232 ++++++++++++++++++++++++++++++++++++++++++
drivers/video/via/via-core.h | 29 +++++
2 files changed, 261 insertions(+), 0 deletions(-)
diff --git a/drivers/video/via/via-core.c b/drivers/video/via/via-core.c
index d368f4f..957af6b 100644
--- a/drivers/video/via/via-core.c
+++ b/drivers/video/via/via-core.c
@@ -13,6 +13,7 @@
#include "global.h"
#include <linux/module.h>
+#include <linux/interrupt.h>
#include <linux/platform_device.h>
/*
@@ -92,7 +93,238 @@ void viafb_irq_disable(u32 mask)
}
EXPORT_SYMBOL_GPL(viafb_irq_disable);
+/* ---------------------------------------------------------------------- */
+/*
+ * Access to the DMA engine. This currently provides what the camera
+ * driver needs (i.e. outgoing only) but is easily expandable if need
+ * be.
+ */
+
+/*
+ * There are four DMA channels in the vx855. For now, we only
+ * use one of them, though. Most of the time, the DMA channel
+ * will be idle, so we keep the IRQ handler unregistered except
+ * when some subsystem has indicated an interest.
+ */
+static int viafb_dma_users;
+static DECLARE_COMPLETION(viafb_dma_completion);
+/*
+ * This mutex protects viafb_dma_users and our global interrupt
+ * registration state; it also serializes access to the DMA
+ * engine.
+ */
+static DEFINE_MUTEX(viafb_dma_lock);
+
+/*
+ * The VX855 DMA descriptor (used for s/g transfers) looks
+ * like this.
+ */
+struct viafb_vx855_dma_descr {
+ u32 addr_low; /* Low part of phys addr */
+ u32 addr_high; /* High 12 bits of addr */
+ u32 fb_offset; /* Offset into FB memory */
+ u32 seg_size; /* Size, 16-byte units */
+ u32 tile_mode; /* "tile mode" setting */
+ u32 next_desc_low; /* Next descriptor addr */
+ u32 next_desc_high;
+ u32 pad; /* Fill out to 64 bytes */
+};
+
+/*
+ * Flags added to the "next descriptor low" pointers
+ */
+#define VIAFB_DMA_MAGIC 0x01 /* ??? Just has to be there */
+#define VIAFB_DMA_FINAL_SEGMENT 0x02 /* Final segment */
+
+/*
+ * The completion IRQ handler.
+ */
+static irqreturn_t viafb_dma_irq(int irq, void *data)
+{
+ int csr;
+ irqreturn_t ret = IRQ_NONE;
+
+ spin_lock(&global_dev.reg_lock);
+ csr = viafb_mmio_read(VDMA_CSR0);
+ if (csr & VDMA_C_DONE) {
+ viafb_mmio_write(VDMA_CSR0, VDMA_C_DONE);
+ complete(&viafb_dma_completion);
+ ret = IRQ_HANDLED;
+ }
+ spin_unlock(&global_dev.reg_lock);
+ return ret;
+}
+
+/*
+ * Indicate a need for DMA functionality.
+ */
+int viafb_request_dma(void)
+{
+ int ret = 0;
+
+ /*
+ * Only VX855 is supported currently.
+ */
+ if (global_dev.chip_type != UNICHROME_VX855)
+ return -ENODEV;
+ /*
+ * Note the new user and set up our interrupt handler
+ * if need be.
+ */
+ mutex_lock(&viafb_dma_lock);
+ viafb_dma_users++;
+ if (viafb_dma_users == 1) {
+ ret = request_irq(global_dev.pdev->irq, viafb_dma_irq,
+ IRQF_SHARED, "via-dma", &viafb_dma_users);
+ if (ret)
+ viafb_dma_users--;
+ else
+ viafb_irq_enable(VDE_I_DMA0TDEN);
+ }
+ mutex_unlock(&viafb_dma_lock);
+ return ret;
+}
+EXPORT_SYMBOL_GPL(viafb_request_dma);
+
+void viafb_release_dma(void)
+{
+ mutex_lock(&viafb_dma_lock);
+ viafb_dma_users--;
+ if (viafb_dma_users == 0) {
+ viafb_irq_disable(VDE_I_DMA0TDEN);
+ free_irq(global_dev.pdev->irq, &viafb_dma_users);
+ }
+ mutex_unlock(&viafb_dma_lock);
+}
+EXPORT_SYMBOL_GPL(viafb_release_dma);
+
+
+#if 0
+/*
+ * Copy a single buffer from FB memory, synchronously. This code works
+ * but is not currently used.
+ */
+void viafb_dma_copy_out(unsigned int offset, dma_addr_t paddr, int len)
+{
+ unsigned long flags;
+ int csr;
+
+ mutex_lock(&viafb_dma_lock);
+ init_completion(&viafb_dma_completion);
+ /*
+ * Program the controller.
+ */
+ spin_lock_irqsave(&global_dev.reg_lock, flags);
+ viafb_mmio_write(VDMA_CSR0, VDMA_C_ENABLE|VDMA_C_DONE);
+ /* Enable ints; must happen after CSR0 write! */
+ viafb_mmio_write(VDMA_MR0, VDMA_MR_TDIE);
+ viafb_mmio_write(VDMA_MARL0, (int) (paddr & 0xfffffff0));
+ viafb_mmio_write(VDMA_MARH0, (int) ((paddr >> 28) & 0xfff));
+ /* Data sheet suggests DAR0 should be <<4, but it lies */
+ viafb_mmio_write(VDMA_DAR0, offset);
+ viafb_mmio_write(VDMA_DQWCR0, len >> 4);
+ viafb_mmio_write(VDMA_TMR0, 0);
+ viafb_mmio_write(VDMA_DPRL0, 0);
+ viafb_mmio_write(VDMA_DPRH0, 0);
+ viafb_mmio_write(VDMA_PMR0, 0);
+ csr = viafb_mmio_read(VDMA_CSR0);
+ viafb_mmio_write(VDMA_CSR0, VDMA_C_ENABLE|VDMA_C_START);
+ spin_unlock_irqrestore(&global_dev.reg_lock, flags);
+ /*
+ * Now we just wait until the interrupt handler says
+ * we're done.
+ */
+ wait_for_completion_interruptible(&viafb_dma_completion);
+ viafb_mmio_write(VDMA_MR0, 0); /* Reset int enable */
+ mutex_unlock(&viafb_dma_lock);
+}
+EXPORT_SYMBOL_GPL(viafb_dma_copy_out);
+#endif
+
+/*
+ * Do a scatter/gather DMA copy from FB memory. You must have done
+ * a successful call to viafb_request_dma() first.
+ */
+int viafb_dma_copy_out_sg(unsigned int offset, struct scatterlist *sg, int nsg)
+{
+ struct viafb_vx855_dma_descr *descr;
+ void *descrpages;
+ dma_addr_t descr_handle;
+ unsigned long flags;
+ int i;
+ struct scatterlist *sgentry;
+ dma_addr_t nextdesc;
+ /*
+ * Get a place to put the descriptors.
+ */
+ descrpages = dma_alloc_coherent(&global_dev.pdev->dev,
+ nsg*sizeof(struct viafb_vx855_dma_descr),
+ &descr_handle, GFP_KERNEL);
+ if (descrpages == NULL) {
+ dev_err(&global_dev.pdev->dev, "Unable to get descr page.\n");
+ return -ENOMEM;
+ }
+ mutex_lock(&viafb_dma_lock);
+ /*
+ * Fill them in.
+ */
+ descr = descrpages;
+ nextdesc = descr_handle + sizeof(struct viafb_vx855_dma_descr);
+ for_each_sg(sg, sgentry, nsg, i) {
+ dma_addr_t paddr = sg_dma_address(sgentry);
+ descr->addr_low = paddr & 0xfffffff0;
+ descr->addr_high = ((u64) paddr >> 32) & 0x0fff;
+ descr->fb_offset = offset;
+ descr->seg_size = sg_dma_len(sgentry) >> 4;
+ descr->tile_mode = 0;
+ descr->next_desc_low = (nextdesc&0xfffffff0) | VIAFB_DMA_MAGIC;
+ descr->next_desc_high = ((u64) nextdesc >> 32) & 0x0fff;
+ descr->pad = 0xffffffff; /* VIA driver does this */
+ offset += sg_dma_len(sgentry);
+ nextdesc += sizeof(struct viafb_vx855_dma_descr);
+ descr++;
+ }
+ descr[-1].next_desc_low = VIAFB_DMA_FINAL_SEGMENT|VIAFB_DMA_MAGIC;
+ /*
+ * Program the engine.
+ */
+ spin_lock_irqsave(&global_dev.reg_lock, flags);
+ init_completion(&viafb_dma_completion);
+ viafb_mmio_write(VDMA_DQWCR0, 0);
+ viafb_mmio_write(VDMA_CSR0, VDMA_C_ENABLE|VDMA_C_DONE);
+ viafb_mmio_write(VDMA_MR0, VDMA_MR_TDIE | VDMA_MR_CHAIN);
+ viafb_mmio_write(VDMA_DPRL0, descr_handle | VIAFB_DMA_MAGIC);
+ viafb_mmio_write(VDMA_DPRH0,
+ (((u64)descr_handle >> 32) & 0x0fff) | 0xf0000);
+ (void) viafb_mmio_read(VDMA_CSR0);
+ viafb_mmio_write(VDMA_CSR0, VDMA_C_ENABLE|VDMA_C_START);
+ spin_unlock_irqrestore(&global_dev.reg_lock, flags);
+ /*
+ * Now we just wait until the interrupt handler says
+ * we're done. Except that, actually, we need to wait a little
+ * longer: the interrupts seem to jump the gun a little and we
+ * get corrupted frames sometimes.
+ */
+ wait_for_completion_timeout(&viafb_dma_completion, 1);
+ msleep(1);
+ if ((viafb_mmio_read(VDMA_CSR0)&VDMA_C_DONE) == 0)
+ printk(KERN_ERR "VIA DMA timeout!\n");
+ /*
+ * Clean up and we're done.
+ */
+ viafb_mmio_write(VDMA_CSR0, VDMA_C_DONE);
+ viafb_mmio_write(VDMA_MR0, 0); /* Reset int enable */
+ mutex_unlock(&viafb_dma_lock);
+ dma_free_coherent(&global_dev.pdev->dev,
+ nsg*sizeof(struct viafb_vx855_dma_descr), descrpages,
+ descr_handle);
+ return 0;
+}
+EXPORT_SYMBOL_GPL(viafb_dma_copy_out_sg);
+
+
+/* ---------------------------------------------------------------------- */
/*
* Figure out how big our framebuffer memory is. Kind of ugly,
* but evidently we can't trust the information found in the
diff --git a/drivers/video/via/via-core.h b/drivers/video/via/via-core.h
index ba64b36..3d03141 100644
--- a/drivers/video/via/via-core.h
+++ b/drivers/video/via/via-core.h
@@ -131,4 +131,33 @@ void viafb_irq_disable(u32 mask);
#define VDE_I_LVDSSIEN 0x40000000 /* LVDS Sense enable */
#define VDE_I_ENABLE 0x80000000 /* Global interrupt enable */
+/*
+ * DMA management.
+ */
+int viafb_request_dma(void);
+void viafb_release_dma(void);
+/* void viafb_dma_copy_out(unsigned int offset, dma_addr_t paddr, int len); */
+int viafb_dma_copy_out_sg(unsigned int offset, struct scatterlist *sg, int nsg);
+
+/*
+ * DMA Controller registers.
+ */
+#define VDMA_MR0 0xe00 /* Mod reg 0 */
+#define VDMA_MR_CHAIN 0x01 /* Chaining mode */
+#define VDMA_MR_TDIE 0x02 /* Transfer done int enable */
+#define VDMA_CSR0 0xe04 /* Control/status */
+#define VDMA_C_ENABLE 0x01 /* DMA Enable */
+#define VDMA_C_START 0x02 /* Start a transfer */
+#define VDMA_C_ABORT 0x04 /* Abort a transfer */
+#define VDMA_C_DONE 0x08 /* Transfer is done */
+#define VDMA_MARL0 0xe20 /* Mem addr low */
+#define VDMA_MARH0 0xe24 /* Mem addr high */
+#define VDMA_DAR0 0xe28 /* Device address */
+#define VDMA_DQWCR0 0xe2c /* Count (16-byte) */
+#define VDMA_TMR0 0xe30 /* Tile mode reg */
+#define VDMA_DPRL0 0xe34 /* Not sure */
+#define VDMA_DPR_IN 0x08 /* Inbound transfer to FB */
+#define VDMA_DPRH0 0xe38
+#define VDMA_PMR0 (0xe00 + 0x134) /* Pitch mode */
+
#endif /* __VIA_CORE_H__ */
--
1.7.0.1
--
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