commit a258f183ee4e48e9352b408e8448467eebdc6cd2

Happy new year, Jeff, Alan.

I'm curious why commit a258f183 got in? We already deal with this problem by executing SETXFER by polling. That's better IMHO for two reasons - no clutter in hot path && applies to drivers which use custom interrupt handlers.

Thanks.

tejun

Whats that commit in real code ?

The attached one.

tejun

From: Alan <[email protected]>
Date: Tue, 2 Oct 2007 13:53:05 -0700
Subject: [PATCH] libata: fix (hopefully) all the remaining problems with devices failing setup/identify

Two fixes in this test patch. One of them allows old CF cards to refuse pio mode setting, and one to wait for the drive to settle after a set features changes the speed settings, which is based upon the workarounds used by drivers/ide.

Please test and report back if you have an afflicted system. This patch isn't for merging just testing.

The CF card fix will still display errors when the card works (got fixes for that once I know the cure works) but should then be found/usable.

Not signed off by anyone, not for merging

---
 drivers/ata/libata-core.c |   15 +++++++++++++++
 1 files changed, 15 insertions(+), 0 deletions(-)

index 0c2f37e..2b24041 100644
@@ -6134,6 +6134,7 @@ inline unsigned int ata_host_intr(struct ata_port *ap,
 {
 	struct ata_eh_info *ehi = &ap->link.eh_info;
 	u8 status, host_stat = 0;
+	int i;
 
 	VPRINTK("ata%u: protocol %d task_state %d\n",
 		ap->print_id, qc->tf.protocol, ap->hsm_task_state);

@@ -6190,6 +6191,20 @@ inline unsigned int ata_host_intr(struct ata_port *ap,
 	if (unlikely(status & ATA_BUSY))
 		goto idle_irq;
 
+	if (unlikely(qc->tf.command == ATA_CMD_SET_FEATURES &&
+			qc->tf.feature == SETFEATURES_XFER)) {
+		/* Let the timings change settle and the drive catch up as
+		   some hardware needs up to 10uS to get its brain back in
+		   gear. Taken from the workarounds in drivers/ide done by
+		   Matthew Faupel/Niccolo Rigacci */
+		for (i = 0; i < 10; i++) {
+			if ((status & (ATA_BUSY | ATA_DRQ | ATA_ERR)) == 0)
+				break;
+			udelay(1);
+			status = ata_chk_status(ap);
+		}
+	}
+
 	/* ack bmdma irq events */
 	ap->ops->irq_clear(ap);
 

1.5.2.4

Ok I don't see how that is related to the switch to polling. The old IDE code found it neccessary and always did polling.

Alan

ata_host_intr() is just never traveled while executing SETXFER because SETXFERs are executed using polling which nicely works around quirky controllers which raises interrupt a tad bit too early during SETXFER.

tejun

Ok

Okay. Jeff, can you please revert the commit? Thanks.

tejun