mmc: mmci: reduce max_blk_count to avoid overflowing max_req_size

On a system with large pages (64k in my case), the following BUG is
triggered in MMC core:

[    2.338023] BUG: failure at drivers/mmc/core/core.c:221/mmc_start_request()!
[    2.338102] Kernel panic - not syncing: BUG!
[    2.338155] Call trace:
[    2.338228] [<ffffffc00008635c>] dump_backtrace+0x0/0x120
[    2.338317] [<ffffffc0003365ec>] dump_stack+0x14/0x1c
[    2.338403] [<ffffffc000336990>] panic+0xbc/0x1f0
[    2.338498] [<ffffffc00027a494>] mmc_start_request+0x154/0x184
[    2.338600] [<ffffffc00027abdc>] mmc_start_req+0x110/0x140
[    2.338701] [<ffffffc00028604c>] mmc_blk_issue_rw_rq+0x7c/0x39c
[    2.338804] [<ffffffc00028652c>] mmc_blk_issue_rq+0x1c0/0x468
[    2.338905] [<ffffffc000287564>] mmc_queue_thread+0x68/0x118
[    2.338995] [<ffffffc0000bc308>] kthread+0x84/0x8c

This is because of a 64k request with a max_req_size of 64k-1 bytes.

The following patch fixes the problem by limiting the max_blk_count
such that max_blk_count * max_blk_size == max_req_size. I couldn't
pursuade the compiler to emit a shift instead of a div without encoding
the shift explicitly.

Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Chris Ball <cjb@laptop.org>

1 files changed, 4 insertions, 3 deletions

diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c
index 0d955ffaf44e..11e589cd8233 100644
--- a/drivers/mmc/host/mmci.c
+++ b/drivers/mmc/host/mmci.c
@@ -1271,12 +1271,13 @@ static int __devinit mmci_probe(struct amba_device *dev,
         /*
          * Block size can be up to 2048 bytes, but must be a power of two.
          */
-        mmc->max_blk_size = 2048;
+        mmc->max_blk_size = 1 << 11;
 
         /*
-         * No limit on the number of blocks transferred.
+         * Limit the number of blocks transferred so that we don't overflow
+         * the maximum request size.
          */
-        mmc->max_blk_count = mmc->max_req_size;
+        mmc->max_blk_count = mmc->max_req_size >> 11;
 
         spin_lock_init(&host->lock);