2015-08-02 18:26:57

by Hauke Mehrtens

Subject: [PATCH] bcma: fix access to host_pdev for PCIe devices

bus->host_pdev is part of a union so bus->host_pdev != NULL is probably
also true for PCIe devices, because there it accesses bus->host_pci. If
we access the dev member at the offset defined in struct
platform_device in struct pci_dev instead we probably get something
else.

This patch adds a new function which returns the host dev struct and
NULL if we do not have a host dev. When this gets registered on MIPS
brcm47xx we do not have a host dev in some situations.
This function could also be used in other places.

This problem was introduced in this commit:
commit cae761b5a6bdc597ba476a040fdcd5b4bc559b85
Author: Rafał Miłecki <[email protected]>
Date: Sun Jun 28 17:17:13 2015 +0200

bcma: populate bus DT subnodes as platform_device-s

Signed-off-by: Hauke Mehrtens <[email protected]>
---
drivers/bcma/bcma_private.h | 1 +
drivers/bcma/main.c | 30 +++++++++++++++++++++++++++---
2 files changed, 28 insertions(+), 3 deletions(-)

diff --git a/drivers/bcma/bcma_private.h b/drivers/bcma/bcma_private.h
index 15f2b2e..38f1567 100644
--- a/drivers/bcma/bcma_private.h
+++ b/drivers/bcma/bcma_private.h
@@ -34,6 +34,7 @@ int __init bcma_bus_early_register(struct bcma_bus *bus);
int bcma_bus_suspend(struct bcma_bus *bus);
int bcma_bus_resume(struct bcma_bus *bus);
#endif
+struct device *bcma_bus_get_host_dev(struct bcma_bus *bus);

/* scan.c */
void bcma_detect_chip(struct bcma_bus *bus);
diff --git a/drivers/bcma/main.c b/drivers/bcma/main.c
index 8d973c4..24882c1 100644
--- a/drivers/bcma/main.c
+++ b/drivers/bcma/main.c
@@ -7,7 +7,9 @@

#include "bcma_private.h"
#include <linux/module.h>
+#include <linux/mmc/sdio_func.h>
#include <linux/platform_device.h>
+#include <linux/pci.h>
#include <linux/bcma/bcma.h>
#include <linux/slab.h>
#include <linux/of_address.h>
@@ -269,6 +271,28 @@ void bcma_prepare_core(struct bcma_bus *bus, struct bcma_device *core)
}
}

+struct device *bcma_bus_get_host_dev(struct bcma_bus *bus)
+{
+ switch (bus->hosttype) {
+ case BCMA_HOSTTYPE_PCI:
+ if (bus->host_pci)
+ return &bus->host_pci->dev;
+ else
+ return NULL;
+ case BCMA_HOSTTYPE_SOC:
+ if (bus->host_pdev)
+ return &bus->host_pdev->dev;
+ else
+ return NULL;
+ case BCMA_HOSTTYPE_SDIO:
+ if (bus->host_sdio)
+ return &bus->host_sdio->dev;
+ else
+ return NULL;
+ }
+ return NULL;
+}
+
void bcma_init_bus(struct bcma_bus *bus)
{
mutex_lock(&bcma_buses_mutex);
@@ -388,6 +412,7 @@ int bcma_bus_register(struct bcma_bus *bus)
{
int err;
struct bcma_device *core;
+ struct device *dev;

/* Scan for devices (cores) */
err = bcma_bus_scan(bus);
@@ -410,13 +435,12 @@ int bcma_bus_register(struct bcma_bus *bus)
bcma_core_pci_early_init(&bus->drv_pci[0]);
}

+ dev = bcma_bus_get_host_dev(bus);
/* TODO: remove check for IS_BUILTIN(CONFIG_BCMA) check when
* of_default_bus_match_table is exported or in some other way
* accessible. This is just a temporary workaround.
*/
- if (IS_BUILTIN(CONFIG_BCMA) && bus->host_pdev) {
- struct device *dev = &bus->host_pdev->dev;
-
+ if (IS_BUILTIN(CONFIG_BCMA) && dev) {
of_platform_populate(dev->of_node, of_default_bus_match_table,
NULL, dev);
}
--
2.1.4

2015-08-10 19:21:07

by Kalle Valo

Subject: Re: bcma: fix access to host_pdev for PCIe devices

> bus->host_pdev is part of a union so bus->host_pdev != NULL is probably
> also true for PCIe devices, because there it accesses bus->host_pci. If
> we access the dev member at the offset defined in struct
> platform_device in struct pci_dev instead we probably get something
> else.
>
> This patch adds a new function which returns the host dev struct and
> NULL if we do not have a host dev. When this gets registered on MIPS
> brcm47xx we do not have a host dev in some situations.
> This function could also be used in other places.
>
> This problem was introduced in this commit:
> commit cae761b5a6bdc597ba476a040fdcd5b4bc559b85
> Author: Rafa? Mi?ecki <[email protected]>
> Date: Sun Jun 28 17:17:13 2015 +0200
>
> bcma: populate bus DT subnodes as platform_device-s
>
> Signed-off-by: Hauke Mehrtens <[email protected]>

Thanks, applied to wireless-drivers-next.git.

Kalle Valo