Re: [PATCH 17/17] mmc: pass queue_limits to blk_mq_alloc_disk

[Christoph Hellwig <hch@lst.de>]

On Tue, Feb 20, 2024 at 11:01:05PM +0100, Geert Uytterhoeven wrote:
> Hi Christoph,
> 
> On Thu, Feb 15, 2024 at 9:16 AM Christoph Hellwig <hch@lst.de> wrote:
> > Pass the queue limit set at initialization time directly to
> > blk_mq_alloc_disk instead of updating it right after the allocation.
> >
> > This requires refactoring the code a bit so that what was mmc_setup_queue
> > before also allocates the gendisk now and actually sets all limits.
> >
> > Signed-off-by: Christoph Hellwig <hch@lst.de>
> 
> Thanks for your patch, which is now commit 616f876617927732 ("mmc: pass
> queue_limits to blk_mq_alloc_disk") in block/for-next.
> 
> I have bisected the following failure on White-Hawk (also seen on
> other R-Car Gen3/4 systems) to this commit:
> 
>     renesas_sdhi_internal_dmac ee140000.mmc: mmc0 base at
> 0x00000000ee140000, max clock rate 200 MHz
>     mmc0: new HS400 MMC card at address 0001
>     ------------[ cut here ]------------
>     WARNING: CPU: 1 PID: 20 at block/blk-settings.c:202
> blk_validate_limits+0x12c/0x1e0

This is:

	if (lim->virt_boundary_mask) {
		if (WARN_ON_ONCE(lim->max_segment_size &&
                                 lim->max_segment_size != UINT_MAX))
			return -EINVAL;

so we end up here with both a virt_boundary_mask and a
max_segment_size set, which is rather bogus.  I think the
problem is the order of check in the core blk_validate_limits
that artificially causes this.  Can you try this patch?

diff --git a/block/blk-settings.c b/block/blk-settings.c
index c4406aacc0efc6..2120b6f9fef8ea 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -182,16 +182,6 @@ static int blk_validate_limits(struct queue_limits *lim)
 	if (WARN_ON_ONCE(lim->seg_boundary_mask < PAGE_SIZE - 1))
 		return -EINVAL;
 
-	/*
-	 * The maximum segment size has an odd historic 64k default that
-	 * drivers probably should override.  Just like the I/O size we
-	 * require drivers to at least handle a full page per segment.
-	 */
-	if (!lim->max_segment_size)
-		lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
-	if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE))
-		return -EINVAL;
-
 	/*
 	 * Devices that require a virtual boundary do not support scatter/gather
 	 * I/O natively, but instead require a descriptor list entry for each
@@ -203,6 +193,16 @@ static int blk_validate_limits(struct queue_limits *lim)
 				 lim->max_segment_size != UINT_MAX))
 			return -EINVAL;
 		lim->max_segment_size = UINT_MAX;
+	} else {
+		/*
+		 * The maximum segment size has an odd historic 64k default that
+		 * drivers probably should override.  Just like the I/O size we
+		 * require drivers to at least handle a full page per segment.
+		 */
+		if (!lim->max_segment_size)
+			lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
+		if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE))
+			return -EINVAL;
 	}
 
 	/*