lkml.org 
[lkml]   [2021]   [Jul]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH v2] coresight: tmc-etr: Speed up for bounce buffer in flat mode
Date
The AUX bounce buffer is allocated with API dma_alloc_coherent(), in the
low level's architecture code, e.g. for Arm64, it maps the memory with
the attribution "Normal non-cacheable"; this can be concluded from the
definition for pgprot_dmacoherent() in arch/arm64/include/asm/pgtable.h.

Later when access the AUX bounce buffer, since the memory mapping is
non-cacheable, it's low efficiency due to every load instruction must
reach out DRAM.

This patch changes to allocate pages with alloc_pages_node(), thus the
driver can access the memory with cacheable mapping in the kernel linear
virtual address; therefore, because load instructions can fetch data
from cache lines rather than always read data from DRAM, the driver can
boost memory coping performance. After using the cacheable mapping, the
driver uses dma_sync_single_for_cpu() to invalidate cacheline prior to
read bounce buffer so can avoid read stale trace data.

By measurement the duration for function tmc_update_etr_buffer() with
ftrace function_graph tracer, it shows the performance significant
improvement for copying 4MiB data from bounce buffer:

# echo tmc_etr_get_data_flat_buf > set_graph_notrace // avoid noise
# echo tmc_update_etr_buffer > set_graph_function
# echo function_graph > current_tracer

before:

# CPU DURATION FUNCTION CALLS
# | | | | | | |
2) | tmc_update_etr_buffer() {
...
2) # 8148.320 us | }

after:

# CPU DURATION FUNCTION CALLS
# | | | | | | |
2) | tmc_update_etr_buffer() {
...
2) # 2463.980 us | }

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---

Changes from v1:
Set "flat_buf->daddr" to 0 when fails to map DMA region; and dropped the
unexpected if condition change in tmc_etr_free_flat_buf().

.../hwtracing/coresight/coresight-tmc-etr.c | 56 ++++++++++++++++---
1 file changed, 49 insertions(+), 7 deletions(-)

diff --git a/drivers/hwtracing/coresight/coresight-tmc-etr.c b/drivers/hwtracing/coresight/coresight-tmc-etr.c
index acdb59e0e661..888b0f929d33 100644
--- a/drivers/hwtracing/coresight/coresight-tmc-etr.c
+++ b/drivers/hwtracing/coresight/coresight-tmc-etr.c
@@ -21,6 +21,7 @@

struct etr_flat_buf {
struct device *dev;
+ struct page *pages;
dma_addr_t daddr;
void *vaddr;
size_t size;
@@ -600,6 +601,7 @@ static int tmc_etr_alloc_flat_buf(struct tmc_drvdata *drvdata,
{
struct etr_flat_buf *flat_buf;
struct device *real_dev = drvdata->csdev->dev.parent;
+ ssize_t aligned_size;

/* We cannot reuse existing pages for flat buf */
if (pages)
@@ -609,11 +611,18 @@ static int tmc_etr_alloc_flat_buf(struct tmc_drvdata *drvdata,
if (!flat_buf)
return -ENOMEM;

- flat_buf->vaddr = dma_alloc_coherent(real_dev, etr_buf->size,
- &flat_buf->daddr, GFP_KERNEL);
- if (!flat_buf->vaddr) {
- kfree(flat_buf);
- return -ENOMEM;
+ aligned_size = PAGE_ALIGN(etr_buf->size);
+ flat_buf->pages = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO,
+ get_order(aligned_size));
+ if (!flat_buf->pages)
+ goto fail_alloc_pages;
+
+ flat_buf->vaddr = page_address(flat_buf->pages);
+ flat_buf->daddr = dma_map_page(real_dev, flat_buf->pages, 0,
+ aligned_size, DMA_FROM_DEVICE);
+ if (dma_mapping_error(real_dev, flat_buf->daddr)) {
+ flat_buf->daddr = 0;
+ goto fail_dma_map_page;
}

flat_buf->size = etr_buf->size;
@@ -622,6 +631,12 @@ static int tmc_etr_alloc_flat_buf(struct tmc_drvdata *drvdata,
etr_buf->mode = ETR_MODE_FLAT;
etr_buf->private = flat_buf;
return 0;
+
+fail_dma_map_page:
+ __free_pages(flat_buf->pages, get_order(aligned_size));
+fail_alloc_pages:
+ kfree(flat_buf);
+ return -ENOMEM;
}

static void tmc_etr_free_flat_buf(struct etr_buf *etr_buf)
@@ -630,15 +645,20 @@ static void tmc_etr_free_flat_buf(struct etr_buf *etr_buf)

if (flat_buf && flat_buf->daddr) {
struct device *real_dev = flat_buf->dev->parent;
+ ssize_t aligned_size = PAGE_ALIGN(etr_buf->size);

- dma_free_coherent(real_dev, flat_buf->size,
- flat_buf->vaddr, flat_buf->daddr);
+ dma_unmap_page(real_dev, flat_buf->daddr, aligned_size,
+ DMA_FROM_DEVICE);
+ __free_pages(flat_buf->pages, get_order(aligned_size));
}
kfree(flat_buf);
}

static void tmc_etr_sync_flat_buf(struct etr_buf *etr_buf, u64 rrp, u64 rwp)
{
+ struct etr_flat_buf *flat_buf = etr_buf->private;
+ struct device *real_dev = flat_buf->dev->parent;
+
/*
* Adjust the buffer to point to the beginning of the trace data
* and update the available trace data.
@@ -648,6 +668,28 @@ static void tmc_etr_sync_flat_buf(struct etr_buf *etr_buf, u64 rrp, u64 rwp)
etr_buf->len = etr_buf->size;
else
etr_buf->len = rwp - rrp;
+
+ if (etr_buf->offset + etr_buf->len > etr_buf->size) {
+ int len1, len2;
+
+ /*
+ * If trace data is wrapped around, sync AUX bounce buffer
+ * for two chunks: "len1" is for the trace date length at
+ * the tail of bounce buffer, and "len2" is the length from
+ * the start of the buffer after wrapping around.
+ */
+ len1 = etr_buf->size - etr_buf->offset;
+ len2 = etr_buf->len - len1;
+ dma_sync_single_for_cpu(real_dev,
+ flat_buf->daddr + etr_buf->offset,
+ len1, DMA_FROM_DEVICE);
+ dma_sync_single_for_cpu(real_dev, flat_buf->daddr,
+ len2, DMA_FROM_DEVICE);
+ } else {
+ dma_sync_single_for_cpu(real_dev,
+ flat_buf->daddr + etr_buf->offset,
+ etr_buf->len, DMA_FROM_DEVICE);
+ }
}

static ssize_t tmc_etr_get_data_flat_buf(struct etr_buf *etr_buf,
--
2.25.1
\
 
 \ /
  Last update: 2021-07-10 09:02    [W:0.117 / U:0.080 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site