Queue bounce should start after the max page, not on it.
Signed-off-by: Chuck Ebbert <[email protected]>
---
Could this explain reported slowdown on x86_64 after limit was
changed in 2.6.16.7?
--- 2.6.17.9-64.orig/block/ll_rw_blk.c
+++ 2.6.17.9-64/block/ll_rw_blk.c
@@ -638,11 +638,11 @@ void blk_queue_bounce_limit(request_queu
/* Assume anything <= 4GB can be handled by IOMMU.
Actually some IOMMUs can handle everything, but I don't
know of a way to test this here. */
- if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
+ if (bounce_pfn <= (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
dma = 1;
q->bounce_pfn = max_low_pfn;
#else
- if (bounce_pfn < blk_max_low_pfn)
+ if (bounce_pfn <= blk_max_low_pfn)
dma = 1;
q->bounce_pfn = bounce_pfn;
#endif
--
Chuck
In-Reply-To: <[email protected]>
On Sat, 19 Aug 2006 06:11:34 -0400, Chuck Ebbert wrote:
> Could this explain reported slowdown on x86_64 after limit was
> changed in 2.6.16.7?
Oops, I meant 2.6.17.7:
| Date: Wed, 2 Aug 2006 13:20:02 -0700
| From: "Robin H. Johnson" <[email protected]>
| Subject: 2.6.17.7 leading to doubling system CPU usage?
| To: [email protected]
| Message-ID: <[email protected]>
And the patch only fixes the problem on my CDROM. The system only has 512 MB
of RAM, so DMA should always be enabled. I applied the below debugging patch
and got this (how do I find the dev for the unknown ones?):
# dmesg | fgrep -v debounce | fgrep -B 2 bounce
[ 16.971945] Probing IDE interface ide0...
[ 17.260734] hda: IC25N060ATMR04-0, ATA DISK drive
[ 17.931625] isa bounce pool size: 16 pages
[ 17.931672] q = ffff81001da606a0, dma = 1, bounce_pfn = 122608, max = 122608
[ 17.931745] q = ffff81001da606a0, dma = 0, bounce_pfn = 1048575, max = 122608
--
[ 17.932039] Probing IDE interface ide1...
[ 18.666664] hdc: MATSHITAUJ-840D, ATAPI CD/DVD-ROM drive
[ 19.002080] q = ffff81001da60050, dma = 1, bounce_pfn = 122608, max = 122608
[ 19.002142] q = ffff81001da60050, dma = 1, bounce_pfn = 122608, max = 122608
--
[ 21.726613] KBC0 MSE0 P2P AUDO
[ 21.726804] ACPI: (supports S0 S3 S4 S5)
[ 21.727335] q = ffff81001db23990, dma = 1, bounce_pfn = 122608, max = 122608 <= unknown dev
--
[ 25.001150] EXT3 FS on hda7, internal journal
[ 25.001156] EXT3-fs: mounted filesystem with ordered data mode.
[ 25.620562] q = ffff81001db23340, dma = 1, bounce_pfn = 122608, max = 122608 <== unknown dev
[ 25.620587] q = ffff81001db23340, dma = 0, bounce_pfn = 1048575, max = 122608
--
[ 25.676213] sd 0:0:0:0: Attached scsi disk sda
[ 25.692273] sd 0:0:0:0: Attached scsi generic sg0 type 0
[ 25.692372] q = ffff81001db22cf0, dma = 1, bounce_pfn = 122608, max = 122608 <== sda:
[ 25.692396] q = ffff81001db22cf0, dma = 0, bounce_pfn = 1048575, max = 122608 why 14 times???
[ 25.693374] q = ffff81001db22cf0, dma = 1, bounce_pfn = 122608, max = 122608
[ 25.693394] q = ffff81001db22cf0, dma = 0, bounce_pfn = 1048575, max = 122608
[ 25.693485] q = ffff81001db22cf0, dma = 1, bounce_pfn = 122608, max = 122608
[ 25.693503] q = ffff81001db22cf0, dma = 0, bounce_pfn = 1048575, max = 122608
[ 25.695692] q = ffff81001db22cf0, dma = 1, bounce_pfn = 122608, max = 122608
[ 25.695714] q = ffff81001db22cf0, dma = 0, bounce_pfn = 1048575, max = 122608
[ 25.695997] q = ffff81001db22cf0, dma = 1, bounce_pfn = 122608, max = 122608
[ 25.696015] q = ffff81001db22cf0, dma = 0, bounce_pfn = 1048575, max = 122608
[ 25.696316] q = ffff81001db22cf0, dma = 1, bounce_pfn = 122608, max = 122608
[ 25.696335] q = ffff81001db22cf0, dma = 0, bounce_pfn = 1048575, max = 122608
[ 25.696416] q = ffff81001db22cf0, dma = 1, bounce_pfn = 122608, max = 122608
[ 25.696434] q = ffff81001db22cf0, dma = 0, bounce_pfn = 1048575, max = 122608
[ 25.699613] usb-storage: device scan complete
[ 27.331383] eth0: link up, 100Mbps, full-duplex, lpa 0x45E1
[ 35.361254] q = ffff81001da606a0, dma = 0, bounce_pfn = 1048575, max = 122608 <== hda again ???
[ 35.361289] q = ffff81001da606a0, dma = 0, bounce_pfn = 1048575, max = 122608
[ 35.867099] q = ffff81001da60050, dma = 1, bounce_pfn = 122608, max = 122608 <== hdc again ???
[ 35.867136] q = ffff81001da60050, dma = 1, bounce_pfn = 122608, max = 122608
--- 2.6.17.9-32.orig/block/ll_rw_blk.c
+++ 2.6.17.9-32/block/ll_rw_blk.c
@@ -651,6 +651,15 @@ void blk_queue_bounce_limit(request_queu
q->bounce_gfp = GFP_NOIO | GFP_DMA;
q->bounce_pfn = bounce_pfn;
}
+
+ printk(KERN_ERR "q = %p, dma = %d, bounce_pfn = %lu, max = %lu\n",
+ q, dma, bounce_pfn,
+#if BITS_PER_LONG == 64
+ (unsigned long)(min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT)
+#else
+ blk_max_low_pfn
+#endif
+ );
}
EXPORT_SYMBOL(blk_queue_bounce_limit);
--
Chuck
On Saturday 19 August 2006 12:11, Chuck Ebbert wrote:
> Queue bounce should start after the max page, not on it.
The page with memory 0xffffffff...0xffffffff+4096 is already
beyond the limit. And 0xffffffff's PFN is 0xffffffff>>PAGE_SHIFT.
I don't see how the patch is correct.
-Andi