diff options
author | David S. Miller <davem@davemloft.net> | 2018-08-02 20:55:32 +0300 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2018-08-02 20:55:32 +0300 |
commit | 89b1698c93a9dee043154f33d96bca9964e705f1 (patch) | |
tree | dd9dcb1965baae8edcf0b496aaa6a70609b6fc11 /drivers/usb/gadget/udc/aspeed-vhub/vhub.h | |
parent | ffd7ce3cd9c294f1ff49ec02cdbd1bc7cb913db6 (diff) | |
parent | e30cb13c5a09ff5f043a6570c32e49b063bea6a1 (diff) | |
download | linux-89b1698c93a9dee043154f33d96bca9964e705f1.tar.xz |
Merge ra.kernel.org:/pub/scm/linux/kernel/git/davem/net
The BTF conflicts were simple overlapping changes.
The virtio_net conflict was an overlap of a fix of statistics counter,
happening alongisde a move over to a bonafide statistics structure
rather than counting value on the stack.
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'drivers/usb/gadget/udc/aspeed-vhub/vhub.h')
-rw-r--r-- | drivers/usb/gadget/udc/aspeed-vhub/vhub.h | 33 |
1 files changed, 33 insertions, 0 deletions
diff --git a/drivers/usb/gadget/udc/aspeed-vhub/vhub.h b/drivers/usb/gadget/udc/aspeed-vhub/vhub.h index 2b040257bc1f..4ed03d33a5a9 100644 --- a/drivers/usb/gadget/udc/aspeed-vhub/vhub.h +++ b/drivers/usb/gadget/udc/aspeed-vhub/vhub.h @@ -462,6 +462,39 @@ enum std_req_rc { #define DDBG(d, fmt, ...) do { } while(0) #endif +static inline void vhub_dma_workaround(void *addr) +{ + /* + * This works around a confirmed HW issue with the Aspeed chip. + * + * The core uses a different bus to memory than the AHB going to + * the USB device controller. Due to the latter having a higher + * priority than the core for arbitration on that bus, it's + * possible for an MMIO to the device, followed by a DMA by the + * device from memory to all be performed and services before + * a previous store to memory gets completed. + * + * This the following scenario can happen: + * + * - Driver writes to a DMA descriptor (Mbus) + * - Driver writes to the MMIO register to start the DMA (AHB) + * - The gadget sees the second write and sends a read of the + * descriptor to the memory controller (Mbus) + * - The gadget hits memory before the descriptor write + * causing it to read an obsolete value. + * + * Thankfully the problem is limited to the USB gadget device, other + * masters in the SoC all have a lower priority than the core, thus + * ensuring that the store by the core arrives first. + * + * The workaround consists of using a dummy read of the memory before + * doing the MMIO writes. This will ensure that the previous writes + * have been "pushed out". + */ + mb(); + (void)__raw_readl((void __iomem *)addr); +} + /* core.c */ void ast_vhub_done(struct ast_vhub_ep *ep, struct ast_vhub_req *req, int status); |