On Wed, May 11, 2011 at 6:45 PM, Chris Ball cjb@laptop.org wrote:
Hi Per,
On Wed, May 11 2011, Per Forlin wrote:
From: Stefan Nilsson XK stefan.xk.nilsson@stericsson.com
If there is only 1 function interrupt registered it is possible to improve performance by directly calling the irq handler and avoiding the overhead of reading the CCCR registers.
Signed-off-by: Per Forlin per.forlin@linaro.org Acked-by: Ulf Hansson ulf.hansson@stericsson.com Reviewed-by: Nicolas Pitre nicolas.pitre@linaro.org
drivers/mmc/core/sdio_irq.c | 33 ++++++++++++++++++++++++++++++++- include/linux/mmc/card.h | 1 + 2 files changed, 33 insertions(+), 1 deletions(-)
diff --git a/drivers/mmc/core/sdio_irq.c b/drivers/mmc/core/sdio_irq.c index bb192f9..a0890ac 100644 --- a/drivers/mmc/core/sdio_irq.c +++ b/drivers/mmc/core/sdio_irq.c @@ -31,6 +31,17 @@ static int process_sdio_pending_irqs(struct mmc_card *card) { int i, ret, count; unsigned char pending;
- struct sdio_func *func;
- /*
- * Optimization, if there is only 1 function interrupt registered
- * call irq handler directly
- */
- func = card->sdio_single_irq;
- if (func) {
- func->irq_handler(func);
- return 1;
- }
ret = mmc_io_rw_direct(card, 0, 0, SDIO_CCCR_INTx, 0, &pending); if (ret) { @@ -42,7 +53,7 @@ static int process_sdio_pending_irqs(struct mmc_card *card) count = 0; for (i = 1; i <= 7; i++) { if (pending & (1 << i)) {
- struct sdio_func *func = card->sdio_func[i - 1];
- func = card->sdio_func[i - 1];
if (!func) { printk(KERN_WARNING "%s: pending IRQ for " "non-existant function\n", @@ -186,6 +197,24 @@ static int sdio_card_irq_put(struct mmc_card *card) return 0; }
+/* If there is only 1 function registered set sdio_single_irq */ +static void sdio_single_irq_set(struct mmc_card *card) +{
- struct sdio_func *func;
- int i;
- card->sdio_single_irq = NULL;
- if ((card->host->caps & MMC_CAP_SDIO_IRQ) &&
- card->host->sdio_irqs == 1)
- for (i = 0; i < card->sdio_funcs; i++) {
- func = card->sdio_func[i];
- if (func && func->irq_handler) {
- card->sdio_single_irq = func;
- break;
- }
- }
+}
/** * sdio_claim_irq - claim the IRQ for a SDIO function * @func: SDIO function @@ -227,6 +256,7 @@ int sdio_claim_irq(struct sdio_func *func, sdio_irq_handler_t *handler) ret = sdio_card_irq_get(func->card); if (ret) func->irq_handler = NULL;
- sdio_single_irq_set(func->card);
return ret; } @@ -251,6 +281,7 @@ int sdio_release_irq(struct sdio_func *func) if (func->irq_handler) { func->irq_handler = NULL; sdio_card_irq_put(func->card);
- sdio_single_irq_set(func->card);
}
ret = mmc_io_rw_direct(func->card, 0, 0, SDIO_CCCR_IENx, 0, ®); diff --git a/include/linux/mmc/card.h b/include/linux/mmc/card.h index d8dffc9..4910dec 100644 --- a/include/linux/mmc/card.h +++ b/include/linux/mmc/card.h @@ -191,6 +191,7 @@ struct mmc_card { struct sdio_cccr cccr; /* common card info */ struct sdio_cis cis; /* common tuple info */ struct sdio_func *sdio_func[SDIO_MAX_FUNCS]; /* SDIO functions (devices) */
- struct sdio_func *sdio_single_irq; /* SDIO function when only one IRQ active */
unsigned num_info; /* number of info strings */ const char **info; /* info strings */ struct sdio_func_tuple *tuples; /* unknown common tuples */
Thanks, looks good now -- pushed to mmc-next for .40.
This patch breaks libertas over SDIO, as the interrupt handler of that driver is now called even though the hardware didn't signal any interrupt condition. This is a problem for at least two reasons in this case:
a) the libertas code is currently structured in a way that it calls sdio_claim_irq() before everything is fully set up (which wasn't a problem so far as the hardware has an own register to mask interrupts). That results in a kernel Ooops as card->priv == NULL.
b) the libertas interrupt handler assumes that a received callback to its interrupt handler signals activity per se ("card->priv->activty_detected = 1").
I have a patch ready for libertas that works around the first problem, but I'm unsure if that's not fixing the wrong end. What the driver really should do is poll the card's CCCR register to find out if there was any interrupt at all and bail if that's not the case, but then we're back to the implementation without this patch.
The other option is to revert the patch again.
Any suggestions?
Daniel