mmc: Add support to handle Urgent data transfer request
Urgent request notification stops currently running transaction
on bus.
In order to decrease a latency of a prioritized requests (aka Urgent
request), we might want to stop the transmission of a running "low
priority" request in order to handle the Urgent request. The urgency of
the request is decided by the block layer I/O scheduler. When the block
layer notifies the MMC layer of an urgent request and if the MMC layer is
blocked on current request to complete, it will be woken up.
The decision on whether to stop an ongoing transfer is taken according to
several parameters, one of them being the number of bytes already
transferred for ongoing transfer by host controller so far.
To calculate how many bytes were successfully programmed before stop,
CORRECTLY_PRG_SECTORS_NUM[245:242] parsed from EXT_CSD register. The
remainder of stopped request (and next prepared request in case it
exists) re-inserted back to the I/O scheduler to be handled after the
completion of the urgent request.
In case it is decided not to stop the ongoing transfer, MMC context will
wait for normal completion of the ongoing transfer, then already prepared
next request (if it exists) will be re-inserted back into block layer and
the urgent request fetched.
URGENT_REQUEST handling has following dependencies:
1. Host controller driver should support mmc_host op named "stop_request"
2. Block I/O scheduler should support re-insert API
3. eMMC card should support HPI (High Priority Interrupt) command
If any of the above dependencies are not met then urgent request mechanism
will not become operational.
Change-Id: Ic3fa1ca9463cc8991aefee940d8bfddf76c111d3
Signed-off-by: Konstantin Dorfman <kdorfman@codeaurora.org>
diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
index cb8e4f3..1a3c662 100644
--- a/include/linux/mmc/host.h
+++ b/include/linux/mmc/host.h
@@ -140,6 +140,8 @@
void (*hw_reset)(struct mmc_host *host);
unsigned long (*get_max_frequency)(struct mmc_host *host);
unsigned long (*get_min_frequency)(struct mmc_host *host);
+ int (*stop_request)(struct mmc_host *host);
+ unsigned int (*get_xfer_remain)(struct mmc_host *host);
};
struct mmc_card;
@@ -153,13 +155,25 @@
* Returns 0 if success otherwise non zero.
*/
int (*err_check) (struct mmc_card *, struct mmc_async_req *);
+ /* Reinserts request back to the block layer */
+ void (*reinsert_req) (struct mmc_async_req *);
+ /* update what part of request is not done (packed_fail_idx) */
+ int (*update_interrupted_req) (struct mmc_card *,
+ struct mmc_async_req *);
};
/**
* mmc_context_info - synchronization details for mmc context
* @is_done_rcv wake up reason was done request
* @is_new_req wake up reason was new request
- * @is_waiting_last_req mmc context waiting for single running request
+ * @is_waiting_last_req is true, when 1 request running on the bus and
+ * NULL fetched as second request. MMC_BLK_NEW_REQUEST
+ * notification will wake up mmc thread from waiting.
+ * @is_urgent wake up reason was urgent request
+ * @is_waiting is true, when first request is running on the bus,
+ * second request preparation started or mmc thread is
+ * waiting for the completion of the current request
+ * (latter case is like @is_waiting_last_req)
* @wait wait queue
* @lock lock to protect data fields
*/
@@ -167,6 +181,8 @@
bool is_done_rcv;
bool is_new_req;
bool is_waiting_last_req;
+ bool is_urgent;
+ bool is_waiting;
wait_queue_head_t wait;
spinlock_t lock;
};
@@ -269,6 +285,7 @@
#define MMC_CAP2_SANITIZE (1 << 15) /* Support Sanitize */
#define MMC_CAP2_INIT_BKOPS (1 << 16) /* Need to set BKOPS_EN */
#define MMC_CAP2_CLK_SCALE (1 << 17) /* Allow dynamic clk scaling */
+#define MMC_CAP2_STOP_REQUEST (1 << 18) /* Allow stop ongoing request */
mmc_pm_flag_t pm_caps; /* supported pm features */
int clk_requests; /* internal reference counter */