On 04/01/2015 10:20 AM, Suravee Suthikulpanit wrote:
ACPI v5.1 introduced _CCA object for specifying cache coherency attribute
for devices. This patch implements a logic, which traverses device namespace
to parse the coherency information, and calling the corresponded
arch_setup_dma_ops().
It checks the _CCA during ACPI scan, and setup coherency during acpi_device
creation.
Then, this is propagted to the platform_device when it is created at later time.
CC: Mark Salter <msal...@redhat.com>
CC: Hanjun Guo <hanjun....@linaro.org>
CC: Al Stone <al.st...@linaro.org>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpa...@amd.com>
---
NOTE:
The original patch has been submitted by Mark Salter, and discussed here:
* http://comments.gmane.org/gmane.linux.acpi.devel/70424
One of the reveiw comment was dissusing whether we need the do/while loop for
travesing parents of the device. AFAIK, there are no APIs that would do
upsearch
from a particular node. The acpi_get_handle() is the closest one, but it is
using
ACPI_NS_NO_UPSEARCH.
This has been rebased and tested with:
* http://git.linaro.org/leg/acpi/acpi.git acpi-5.1-v11
* [V8 PATCH 0/3] Introduce ACPI support for ahci_platform driver
(https://lkml.org/lkml/2015/3/30/729)
drivers/acpi/acpi_platform.c | 5 +++-
drivers/acpi/scan.c | 56 ++++++++++++++++++++++++++++++++++++++++++++
2 files changed, 60 insertions(+), 1 deletion(-)
diff --git a/drivers/acpi/acpi_platform.c b/drivers/acpi/acpi_platform.c
index 1284138..47e37a8 100644
--- a/drivers/acpi/acpi_platform.c
+++ b/drivers/acpi/acpi_platform.c
@@ -108,9 +108,12 @@ struct platform_device *acpi_create_platform_device(struct
acpi_device *adev)
if (IS_ERR(pdev))
dev_err(&adev->dev, "platform device creation failed: %ld\n",
PTR_ERR(pdev));
- else
+ else {
+ arch_setup_dma_ops(&pdev->dev, 0, 0, NULL,
+ is_device_dma_coherent(&adev->dev));
dev_dbg(&adev->dev, "created platform device %s\n",
dev_name(&pdev->dev));
+ }
kfree(resources);
return pdev;
diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
index 230ec983..4d81c55 100644
--- a/drivers/acpi/scan.c
+++ b/drivers/acpi/scan.c
@@ -11,6 +11,7 @@
#include <linux/kthread.h>
#include <linux/dmi.h>
#include <linux/nls.h>
+#include <linux/dma-mapping.h>
#include <asm/pgtable.h>
@@ -57,6 +58,9 @@ struct acpi_device_bus_id{
struct list_head node;
};
+static int acpi_bus_type_and_status(acpi_handle handle, int *type,
+ unsigned long long *sta);
+
void acpi_scan_lock_acquire(void)
{
mutex_lock(&acpi_scan_lock);
@@ -1327,6 +1331,51 @@ void acpi_bus_put_acpi_device(struct acpi_device *adev)
put_device(&adev->dev);
}
+/**
+ * acpi_check_and_set_coherency - check and set cache coherency of a device
+ * @device: ACPI device
+ *
+ * Search a device and its parents for a _CCA method and calculate
+ * the effective coherency property of the device. If found, set up
+ * appropriate dma_ops for the device.
+ *
+ * Note From ACPIv5.1 spec:
+ * If used _CCA must be included under all bus-master-capable
+ * devices defined as children of \_SB. The value of _CCA is inherited
+ * by all descendants of these devices, so it need not be repeated for
+ * their children devices and will be ignored by OSPM if it is provided
+ * there.
+ */
+static int acpi_check_and_set_coherency(struct acpi_device *device)
+{
+ acpi_status status;
+ unsigned long long eff_cca = ~0ULL;
+ acpi_handle tmp, handle = device->handle;
+
+ /* Calculate effective _CCA from all parents of this device */
+ do {
+ unsigned long cca;
+
+ status = acpi_get_handle(handle, "_CCA", &tmp);
+ if (ACPI_SUCCESS(status)) {
+ status = acpi_evaluate_integer(tmp, "_CCA", NULL, &cca);
+ /* Other values besides 0 and 1 are reserved */
+ if (ACPI_FAILURE(status) || (cca != 0 && cca != 1))
+ return -EINVAL;
+ eff_cca = cca;
+ }
+
+ status = acpi_get_parent(handle, &handle);
+ if (ACPI_FAILURE(status))
+ break;
+ } while (ACPI_SUCCESS(status));
Having the above portion/loop be a separate function that could be
callable by others (such as drivers) would be good so that both the
ACPI code and the driver code look for the _CCA attribute the same
way - similar to the of_dma_is_coherent function.
Possibly even work this into a a generic call like the device property
functions (even though this doesn't fall into the _DSD area for ACPI) -
where for device tree the of_dma_is_coherent call is made and for ACPI
an acpi_dma_is_coherent call is made - device_dma_is_coherent?
Thanks,
Tom
+
+ if (eff_cca != ~0ULL)
+ arch_setup_dma_ops(&device->dev, 0, 0, NULL, eff_cca);
+
+ return 0;
+}
+
int acpi_device_add(struct acpi_device *device,
void (*release)(struct device *))
{
@@ -1398,6 +1447,13 @@ int acpi_device_add(struct acpi_device *device,
device->dev.parent = &device->parent->dev;
device->dev.bus = &acpi_bus_type;
device->dev.release = release;
+
+ result = acpi_check_and_set_coherency(device);
+ if (result) {
+ dev_err(&device->dev, "Error getting device _CCA
information\n");
+ goto err;
+ }
+
result = device_add(&device->dev);
if (result) {
dev_err(&device->dev, "Error registering device\n");
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
