public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
* Re: 2.4.4-ac5 aic7xxx causes hang on my machine
@ 2001-05-09 12:59 Benedict Bridgwater
  2001-05-09 14:06 ` Alan Cox
  0 siblings, 1 reply; 15+ messages in thread
From: Benedict Bridgwater @ 2001-05-09 12:59 UTC (permalink / raw)
  To: Linux-Kernel

Justin Gibbs wrote:

> >I have a dual ppro 200MHZ W6LI motherboard.  I put 2.4.4-ac5 on last
> >night, and the machine hung at Freeing unused Kernel memory.  I
> >selectively backed off what I thought were relevant patches.  I got to
> >aic7xxx, and ac5 without it worked. I attached /proc/scsi/aic7xxx/0.
>
> This problem was fixed in rev. 6.1.13 of the aic7xxx driver.

Is this the same bug that would have caused SCSI timeouts/retries on an
Adaptec 2940 UW with a stock 2.4.3 kernel (Mandrake 8.0/Redhat 7.1)?
I've been trying to install Mandrake 8.0 from a  CD to a HD both
attached to a 2940 UW (Intel 440 FX motherboard), and the install took
hours of mostly zero CD/HD activity time, although it eventually
completed (the install fails to boot, though).

This has been reported to both Mandrake and Redhat:

http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=29555

I've been trying to find out if there's a fix (if it's aic7xxx 6.1.13
that's great!), but Redhat seem to believe it's a 2.4 kernel PCI bug:

Doug Ledford, from bugzilla:

> For clarification sake, it's not that the BIOS is mapping IRQ's wrong, and it's
> not that we have lost any functionality since the 2.2 kernel PCI code.  Quite
> the opposite, it's that PCI functionality has been added since the 2.2 kernel. 
> Specifically, the 2.4 kernel supports the notion of hot-plug PCI devices.  That
> requires that the kernel PCI layer know about/assign all the address space and
> IRQ resources to the various slots so that if sometime after booting up someone
> adds a new PCI card that has a PCI bridge chip, then the PCI layer has to be
> able to assign I/O and IRQ resources to the bridge chip itself, and those I/O
> and IRQ resources have to come from the pool of resources already allocated to
> the bus that the card was plugged into.  So, in order to insure that we will
> always be able to accept a hot plug card, the PCI layer in the 2.4 kernel now
> re-organizes those PCI resources at boot time to make sure that it is possible
> to plug a card with a PCI bridge chip into every PCI slot if the user so
> wishes.  In this process, the PCI code in the 2.4 kernel is evidently messing up
> the IRQ assignment on the Adaptec chipset.  This *only* happens on the L440GX
> motherboard from Intel as far as we can tell, and the root cause of the mistake
> hasn't been isolated any further than what I've told you here.  The reason that
> enabling SMP or UP-IOAPIC support solves the problem is that when IOAPIC support
> is enabled (which happens automatically with SMP support, or explicitly with UP
> kernels if you turn on UP-IOAPIC support), it runs *after* the PCI resource
> allocation code, and it re-routes the interrupts according the the computers MP
> table.  By doing that, it undoes whatever mistake the PCI code is making, and
> things work.  *THIS IS NOT A FIX*  The fact that it undoes the mistake the PCI
> code makes does not make the PCI code any better!  The PCI code still needs
> fixed.

Ben

Please CC any reply to bennyb@ntplx.net (I'm not a member of the list)

^ permalink raw reply	[flat|nested] 15+ messages in thread
* [Patch] Do not account shmem pages to the page cache
@ 2001-05-07 12:43 Christoph Rohland
  2001-05-07 13:04 ` 2.4.4-ac5 aic7xxx causes hang on my machine Andy Carlson
  0 siblings, 1 reply; 15+ messages in thread
From: Christoph Rohland @ 2001-05-07 12:43 UTC (permalink / raw)
  To: Linux Kernel Mailing List, MM mailing list

Hi,

The appended patch does it's own accounting of shmem pages and adjust
the page cache size to take these into account. So now again you will
see shmem pages as used in top/vmstat etc. This confused a lot of
people.

There is a uncertainty in the calculations since the vm may drop pages
behind shmem and the number of shmem pages is estimated too high. This
especially happens on truncate because first the page cache is reduced
and later the shmem readjusts it's count.

To prevent negative cache sizes the adjustment is only done if
shmem_nrpages > page_cache_size.

The latter part of the patch (all the init.c files) also exports the
shmem page number to the shared memory field in meminfo. This means a
change in semantics of this field but apparently a lot of people
interpret this field exactly this way and it was not used any more

The patches are on top of my encapsulation patch.

Greetings
		Christoph

diff -uNr 2.4.4-mSsu/fs/proc/proc_misc.c 2.4.4-mSsua/fs/proc/proc_misc.c
--- 2.4.4-mSsu/fs/proc/proc_misc.c	Sun Apr 29 20:32:52 2001
+++ 2.4.4-mSsua/fs/proc/proc_misc.c	Mon May  7 13:38:53 2001
@@ -140,6 +140,17 @@
 {
 	struct sysinfo i;
 	int len;
+	unsigned int cached, shmem;
+
+	/*
+	 * There may be some inconsistency because shmem_nrpages
+	 * update is delayed to page_cache_size
+	 * We make sure the cached value does not get below zero 
+	 */
+	cached = atomic_read(&page_cache_size);
+	shmem  = atomic_read(&shmem_nrpages);
+	if (shmem < cached)
+		cached -= shmem;
 
 /*
  * display in kilobytes.
@@ -153,8 +164,8 @@
                 "Swap: %8lu %8lu %8lu\n",
                 B(i.totalram), B(i.totalram-i.freeram), B(i.freeram),
                 B(i.sharedram), B(i.bufferram),
-                B(atomic_read(&page_cache_size)), B(i.totalswap),
-                B(i.totalswap-i.freeswap), B(i.freeswap));
+		B(cached), B(i.totalswap),
+		B(i.totalswap-i.freeswap), B(i.freeswap));
         /*
          * Tagged format, for easy grepping and expansion.
          * The above will go away eventually, once the tools
@@ -180,7 +191,7 @@
                 K(i.freeram),
                 K(i.sharedram),
                 K(i.bufferram),
-                K(atomic_read(&page_cache_size)),
+		K(cached),
 		K(nr_active_pages),
 		K(nr_inactive_dirty_pages),
 		K(nr_inactive_clean_pages()),
diff -uNr 2.4.4-mSsu/include/linux/shmem_fs.h 2.4.4-mSsua/include/linux/shmem_fs.h
--- 2.4.4-mSsu/include/linux/shmem_fs.h	Wed May  2 18:36:05 2001
+++ 2.4.4-mSsua/include/linux/shmem_fs.h	Mon May  7 12:52:00 2001
@@ -17,6 +17,8 @@
 	unsigned long val;
 } swp_entry_t;
 
+extern atomic_t shmem_nrpages;
+
 struct shmem_inode_info {
 	spinlock_t		lock;
 	struct semaphore 	sem;
diff -uNr 2.4.4-mSsu/mm/mmap.c 2.4.4-mSsua/mm/mmap.c
--- 2.4.4-mSsu/mm/mmap.c	Sun Apr 29 20:33:01 2001
+++ 2.4.4-mSsua/mm/mmap.c	Mon May  7 13:42:03 2001
@@ -55,13 +55,24 @@
 	 */
 
 	long free;
-	
+	unsigned long cached, shmem;
+
+	/*
+	 * There may be some inconsistency because shmem_nrpages
+	 * update is delayed to the page_cache_size
+	 * We make sure the cached value does not get below zero 
+	 */
+	cached = atomic_read(&page_cache_size);
+	shmem  = atomic_read(&shmem_nrpages);
+	if (cached > shmem)
+		cached -= shmem;
+
         /* Sometimes we want to use more memory than we have. */
 	if (sysctl_overcommit_memory)
 	    return 1;
 
 	free = atomic_read(&buffermem_pages);
-	free += atomic_read(&page_cache_size);
+	free += cached;
 	free += nr_free_pages();
 	free += nr_swap_pages;
 
diff -uNr 2.4.4-mSsu/mm/shmem.c 2.4.4-mSsua/mm/shmem.c
--- 2.4.4-mSsu/mm/shmem.c	Fri May  4 21:37:34 2001
+++ 2.4.4-mSsua/mm/shmem.c	Mon May  7 11:13:27 2001
@@ -3,7 +3,8 @@
  *
  * Copyright (C) 2000 Linus Torvalds.
  *		 2000 Transmeta Corp.
- *		 2000 Christoph Rohland
+ *		 2000-2001 Christoph Rohland
+ *		 2000-2001 SAP AG
  * 
  * This file is released under the GPL.
  */
@@ -45,6 +46,7 @@
 
 LIST_HEAD (shmem_inodes);
 static spinlock_t shmem_ilock = SPIN_LOCK_UNLOCKED;
+atomic_t shmem_nrpages = ATOMIC_INIT(0);
 
 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
 
@@ -52,6 +54,7 @@
  * shmem_recalc_inode - recalculate the size of an inode
  *
  * @inode: inode to recalc
+ * @swap:  additional swap pages freed externally
  *
  * We have to calculate the free blocks since the mm can drop pages
  * behind our back
@@ -62,12 +65,14 @@
  *
  * So the mm freed 
  * inodes->i_blocks/BLOCKS_PER_PAGE - 
- *			(inode->i_mapping->nrpages + info->swapped)
+ * 			(inode->i_mapping->nrpages + info->swapped)
  *
  * It has to be called with the spinlock held.
+ *
+ * The swap parameter is a performance hack for truncate.
  */
 
-static void shmem_recalc_inode(struct inode * inode)
+static void shmem_recalc_inode(struct inode * inode, unsigned long swap)
 {
 	unsigned long freed;
 
@@ -79,6 +84,7 @@
 		spin_lock (&info->stat_lock);
 		info->free_blocks += freed;
 		spin_unlock (&info->stat_lock);
+		atomic_sub(freed-swap, &shmem_nrpages);
 	}
 }
 
@@ -195,7 +201,7 @@
 out:
 	info->max_index = index;
 	info->swapped -= freed;
-	shmem_recalc_inode(inode);
+	shmem_recalc_inode(inode, freed);
 	spin_unlock (&info->lock);
 	up(&info->sem);
 }
@@ -250,14 +256,15 @@
 	entry = shmem_swp_entry(info, page->index);
 	if (IS_ERR(entry))	/* this had been allocted on page allocation */
 		BUG();
-	shmem_recalc_inode(page->mapping->host);
+	shmem_recalc_inode(page->mapping->host, 0);
 	error = -EAGAIN;
 	if (entry->val)
 		BUG();
 
 	*entry = swap;
 	error = 0;
-	/* Remove the from the page cache */
+	/* Remove the page from the page cache */
+	atomic_dec(&shmem_nrpages);
 	lru_cache_del(page);
 	remove_inode_page(page);
 
@@ -376,6 +383,7 @@
 	}
 
 	/* We have the page */
+	atomic_inc(&shmem_nrpages);
 	SetPageUptodate(page);
 	if (info->locked)
 		page_cache_get(page);
@@ -1275,6 +1283,7 @@
 	return 0;
 found:
 	add_to_page_cache(page, info->inode->i_mapping, offset + idx);
+	atomic_inc(&shmem_nrpages);
 	set_page_dirty(page);
 	SetPageUptodate(page);
 	UnlockPage(page);
diff -uNr 2.4.4-mSsu/arch/alpha/mm/init.c c/arch/alpha/mm/init.c
--- 2.4.4-mSsu/arch/alpha/mm/init.c	Sun Apr 29 20:31:56 2001
+++ c/arch/alpha/mm/init.c	Sun May  6 21:47:25 2001
@@ -402,7 +402,7 @@
 si_meminfo(struct sysinfo *val)
 {
 	val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = 0;
diff -uNr 2.4.4-mSsu/arch/arm/mm/init.c c/arch/arm/mm/init.c
--- 2.4.4-mSsu/arch/arm/mm/init.c	Sun Apr 29 20:31:56 2001
+++ c/arch/arm/mm/init.c	Sun May  6 21:47:01 2001
@@ -647,7 +647,7 @@
 void si_meminfo(struct sysinfo *val)
 {
 	val->totalram  = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram   = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = 0;
diff -uNr 2.4.4-mSsu/arch/cris/mm/init.c c/arch/cris/mm/init.c
--- 2.4.4-mSsu/arch/cris/mm/init.c	Sun Apr 29 20:31:57 2001
+++ c/arch/cris/mm/init.c	Sun May  6 21:47:03 2001
@@ -503,7 +503,7 @@
 
 	i = max_mapnr;
 	val->totalram = 0;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	while (i-- > 0)  {
diff -uNr 2.4.4-mSsu/arch/i386/mm/init.c c/arch/i386/mm/init.c
--- 2.4.4-mSsu/arch/i386/mm/init.c	Sun Apr 29 20:32:08 2001
+++ c/arch/i386/mm/init.c	Sun May  6 20:24:21 2001
@@ -570,7 +570,7 @@
 void si_meminfo(struct sysinfo *val)
 {
 	val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = totalhigh_pages;
diff -uNr 2.4.4-mSsu/arch/ia64/mm/init.c c/arch/ia64/mm/init.c
--- 2.4.4-mSsu/arch/ia64/mm/init.c	Sun Apr 29 20:32:11 2001
+++ c/arch/ia64/mm/init.c	Sun May  6 21:47:05 2001
@@ -151,7 +151,7 @@
 si_meminfo (struct sysinfo *val)
 {
 	val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = 0;
diff -uNr 2.4.4-mSsu/arch/m68k/mm/init.c c/arch/m68k/mm/init.c
--- 2.4.4-mSsu/arch/m68k/mm/init.c	Sat Nov  4 18:11:22 2000
+++ c/arch/m68k/mm/init.c	Sun May  6 21:47:45 2001
@@ -217,7 +217,7 @@
 
     i = max_mapnr;
     val->totalram = totalram_pages;
-    val->sharedram = 0;
+    val->sharedram = atomic_read(&shmem_nrpages);
     val->freeram = nr_free_pages();
     val->bufferram = atomic_read(&buffermem_pages);
     while (i-- > 0) {
diff -uNr 2.4.4-mSsu/arch/mips/mm/init.c c/arch/mips/mm/init.c
--- 2.4.4-mSsu/arch/mips/mm/init.c	Sat Nov  4 18:11:22 2000
+++ c/arch/mips/mm/init.c	Sun May  6 21:47:01 2001
@@ -343,7 +343,7 @@
 void si_meminfo(struct sysinfo *val)
 {
 	val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = 0;
diff -uNr 2.4.4-mSsu/arch/mips64/mm/init.c c/arch/mips64/mm/init.c
--- 2.4.4-mSsu/arch/mips64/mm/init.c	Sat Nov  4 18:11:22 2000
+++ c/arch/mips64/mm/init.c	Sun May  6 21:47:04 2001
@@ -411,7 +411,7 @@
 si_meminfo(struct sysinfo *val)
 {
 	val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = 0;
diff -uNr 2.4.4-mSsu/arch/parisc/mm/init.c c/arch/parisc/mm/init.c
--- 2.4.4-mSsu/arch/parisc/mm/init.c	Sun Dec 17 12:53:55 2000
+++ c/arch/parisc/mm/init.c	Sun May  6 21:47:02 2001
@@ -458,7 +458,7 @@
 
 	i = max_mapnr;
 	val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 #if 0
diff -uNr 2.4.4-mSsu/arch/ppc/mm/init.c c/arch/ppc/mm/init.c
--- 2.4.4-mSsu/arch/ppc/mm/init.c	Wed Apr 11 12:36:13 2001
+++ c/arch/ppc/mm/init.c	Sun May  6 21:47:05 2001
@@ -336,7 +336,7 @@
 
 	i = max_mapnr;
 	val->totalram = 0;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	while (i-- > 0)  {
diff -uNr 2.4.4-mSsu/arch/s390/mm/init.c c/arch/s390/mm/init.c
--- 2.4.4-mSsu/arch/s390/mm/init.c	Sun Apr 29 20:32:21 2001
+++ c/arch/s390/mm/init.c	Sun May  6 21:47:03 2001
@@ -271,7 +271,7 @@
 void si_meminfo(struct sysinfo *val)
 {
 	val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = 0;
diff -uNr 2.4.4-mSsu/arch/s390x/mm/init.c c/arch/s390x/mm/init.c
--- 2.4.4-mSsu/arch/s390x/mm/init.c	Sun Apr 29 20:32:22 2001
+++ c/arch/s390x/mm/init.c	Sun May  6 21:47:18 2001
@@ -284,7 +284,7 @@
 void si_meminfo(struct sysinfo *val)
 {
         val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = 0;
diff -uNr 2.4.4-mSsu/arch/sh/mm/init.c c/arch/sh/mm/init.c
--- 2.4.4-mSsu/arch/sh/mm/init.c	Sun Apr 29 20:32:23 2001
+++ c/arch/sh/mm/init.c	Sun May  6 21:47:26 2001
@@ -215,7 +215,7 @@
 void si_meminfo(struct sysinfo *val)
 {
 	val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = totalhigh_pages;
diff -uNr 2.4.4-mSsu/arch/sparc/mm/init.c c/arch/sparc/mm/init.c
--- 2.4.4-mSsu/arch/sparc/mm/init.c	Sun Apr 29 20:32:23 2001
+++ c/arch/sparc/mm/init.c	Sun May  6 21:47:04 2001
@@ -534,7 +534,7 @@
 void si_meminfo(struct sysinfo *val)
 {
 	val->totalram = totalram_pages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 	val->totalhigh = totalhigh_pages;
diff -uNr 2.4.4-mSsu/arch/sparc64/mm/init.c c/arch/sparc64/mm/init.c
--- 2.4.4-mSsu/arch/sparc64/mm/init.c	Sun Apr 29 20:32:25 2001
+++ c/arch/sparc64/mm/init.c	Sun May  6 21:47:02 2001
@@ -1512,7 +1512,7 @@
 void si_meminfo(struct sysinfo *val)
 {
 	val->totalram = num_physpages;
-	val->sharedram = 0;
+	val->sharedram = atomic_read(&shmem_nrpages);
 	val->freeram = nr_free_pages();
 	val->bufferram = atomic_read(&buffermem_pages);
 


^ permalink raw reply	[flat|nested] 15+ messages in thread
end of thread, other threads:[~2001-05-10 14:33 UTC | newest]

Thread overview: 15+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2001-05-09 12:59 2.4.4-ac5 aic7xxx causes hang on my machine Benedict Bridgwater
2001-05-09 14:06 ` Alan Cox
2001-05-09 15:25   ` Doug Ledford
2001-05-09 15:49     ` Alan Cox
2001-05-09 15:52       ` Doug Ledford
2001-05-09 16:10         ` Alan Cox
2001-05-09 16:21           ` Doug Ledford
2001-05-09 16:59             ` Alan Cox
2001-05-09 17:08               ` Doug Ledford
2001-05-09 17:16                 ` Alan Cox
     [not found]     ` <3AF96271.81C27F57@mandrakesoft.com>
     [not found]       ` <3AF9661B.1707F5E3@redhat.com>
     [not found]         ` <3AF96811.3C6C196D@mandrakesoft.com>
     [not found]           ` <3AF96B39.7C7888F4@redhat.com>
     [not found]             ` <3AF96D99.AB18CB4F@mandrakesoft.com>
     [not found]               ` <3AF97DE0.3C6455A6@ntplx.net>
2001-05-10 14:39                 ` Ben Bridgwater
2001-05-09 16:21   ` Andy Carlson
  -- strict thread matches above, loose matches on Subject: below --
2001-05-07 12:43 [Patch] Do not account shmem pages to the page cache Christoph Rohland
2001-05-07 13:04 ` 2.4.4-ac5 aic7xxx causes hang on my machine Andy Carlson
2001-05-07 13:58   ` Oyvind Jagtnes
2001-05-07 14:14   ` Justin T. Gibbs

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox