[uml-devel] Two tiny changes to make DEBUG_SPINLOCKS 2 working

[Jan Hudec <bulb@uc...>]

Hello All,

I found DEBUG_SPINLOCKS=2 useful. But it's not working. There are two
small bugs:

1) cli() and others are not defined. Thus include/linux/spinlock.h
should pull in <asm/system.h> (or other header to get these).

2) there is a bug in __save_flags - argument name is different from what
is used in the definition.

Here is the respective patch (-p1 of course...):

--- umlinux/include/linux/spinlock.h.orig	2003-02-08 21:05:36.000000000 +0100
+++ umlinux/include/linux/spinlock.h	2003-02-08 21:03:30.000000000 +0100
@@ -110,6 +110,7 @@
 #define SPIN_LOCK_UNLOCKED (spinlock_t) { 0, 25, __BASE_FILE__ }
 
 #include <linux/kernel.h>
+#include <asm/system.h>
 
 #define spin_lock_init(x)	do { (x)->lock = 0; } while (0)
 #define spin_is_locked(lock)	(test_bit(0,(lock)))
--- umlinux/include/asm-um/system-generic.h.orig	2003-02-08 21:06:09.000000000 +0100
+++ umlinux/include/asm-um/system-generic.h	2003-02-08 21:00:54.000000000 +0100
@@ -41,7 +41,7 @@
 #define __cli() block_signals()
 #define cli() block_signals()
 
-#define __save_flags(x) do { (flags) = get_signals(); } while(0)
+#define __save_flags(x) do { (x) = get_signals(); } while(0)
 #define save_flags(x) __save_flags(x)
 
 #define __restore_flags(x) local_irq_restore(x)

-------------------------------------------------------------------------------
						 Jan 'Bulb' Hudec <bulb@...>

Re: [uml-devel] recursive fault

[Jon Smirl <jonsmirl@ya...>]

--- Jeff Dike <jdike@...> wrote:
> You need to update the kernel page tables with the
> new mapping and call
> flush_tlb_kernel_vm.  When the new process gets
> switched in, it will notice
> that its idea of the kernel mappings is out of date,
> and will remap everything.
> 

So I need a support function that does this:

void *vaddr = (void *) get_vm(size);
err = os_map_memory(vaddr, fd_dma, udma.dma_addr,
size, 1, 1, 1);
// update kernel page table
flush_tlb_kernel_vm();

So I'm building a mmap() routine for a dynamically
assigned  location that works with UML's address space
mechanism. Maybe this already exists and I don't need
to build it again?

I don't know very much about the internals of the VM
system. Is there an example of what I need to do to
update the kernel page tables?


=====
Jon Smirl
jonsmirl@...

__________________________________________________
Do you Yahoo!?
Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
http://mailplus.yahoo.com

Re: [uml-devel] recursive fault

[Jeff Dike <jdike@ka...>]

jonsmirl@... said:
> It was the first case I hit where I have done an mmap in one kernel
> process and then tried to use it from another that was forked before
> the mmap was made.

You need to update the kernel page tables with the new mapping and call
flush_tlb_kernel_vm.  When the new process gets switched in, it will notice
that its idea of the kernel mappings is out of date, and will remap everything.

				Jeff

Re: [uml-devel] Patch archive

[Christian Laursen <xi@bo...>]

David Coulson <david@...> writes:

> So, anyone who has patches which are non-trivial, please tell me where
> I can grab them so I can put them somewhere useful. If you ask nicely,
> I may also be able to offer some web space for distribution and such
> like.

I actually have a trivial patch, that I sent to the list a while ago,
but I never got a response about it. I apply it myself everytime I
build a new UML kernel.

It changes the contents of /proc/cpuinfo a bit to be more like on
a real linux. For one thing it makes top print the 'CPU states'
line, which will otherwise be missing.

--- /home/xi/projects/user-mode-linux/linux/arch/um/kernel/um_arch.c	Tue May 28 17:31:54 2002
+++ linux-2.4.18-uml/arch/um/kernel/um_arch.c	Thu May 30 19:44:11 2002
@@ -64,10 +64,11 @@
 		return 0;
 #endif
 
+	seq_printf(m, "processor\t: %d\n", index);
 	seq_printf(m, "bogomips\t: %lu.%02lu\n",
 		   loops_per_jiffy/(500000/HZ),
 		   (loops_per_jiffy/(5000/HZ)) % 100);
-	seq_printf(m, "host\t\t: %s\n", host_info);
+	seq_printf(m, "host\t\t: %s\n\n", host_info);
 
 	return(0);
 }


-- 
Med venlig hilsen
    Christian Laursen

Re: [uml-devel] [PATCH] Fix uml_utilities make install

[Petr Baudis <pasky@uc...>]

Dear diary, on Sat, Feb 08, 2003 at 05:18:50AM CET, I got a letter,
where Jeff Dike <jdike@...> told me, that...
> pasky@... said:
> > there is a weirdness with the uml_utilities --- you can set the
> > installation prefixes at the root Makefile, but the other Makefiles
> > just ignore it and override it by the default value.
> 
> It's a bug that the lower Makefiles just override the upper Makefile values
> for LIB_DIR and BIN_DIR, but it's a totally trivial one.
> 
> You seem not to have noticed DESTDIR, which is honored by all of the Makefiles.

Indeed I didn't, sorry. Then it should probably be like...

DESTDIR=/usr
BIN_DIR=$(DESTDIR)/bin
LIB_DIR=$(DESTDIR)/lib/uml

(Again I believe that appending $(DESTDIR) here rather than during install is
better since you append it on just one place instead a bunch of others and you
can override DESTDIR in some cases when you want some of the directories to
deviate from it; it's not probably that important, though.)

> > The attached patch (against uml_utilities_20030205) makes the
> > Makefiles in subdirs honour the user's choice of installation prefix
> > made in the root Makefile. Please apply.
> 
> Umm, no.  This is fixed now, but correctly.
> 
> Before proposing that stuff be blindly deleted, you might try to figure out
> what purpose it has, and whether you can fix whatever you think the problem
> while preserving that.

What purpose it has to be able to specifically set install directory for each
of the packages...? If you would want to be able to run make install from the
directories, it would miss DESTDIR anyway. This way you will have bunch of
duplicate stuff around all the makefiles, I don't know what purpose that would
serve.

Kind regards,

-- 
 
				Petr "Pasky" Baudis
.
"A computer is a state machine.
 Threads are for people who can't program state machines."
		-- Alan Cox
.
Crap: http://pasky.ji.cz/

Re: [uml-devel] Patch archive

[Jeff Dike <jdike@ka...>]

david@... said:
> So, anyone who has patches which are non-trivial, please tell me where
> I  can grab them so I can put them somewhere useful. If you ask
> nicely, I  may also be able to offer some web space for distribution
> and such like. 

Also, this might reduce the chance that I lose someone's useful patch in
my mail.  I try not to do that, but I make no guarantees.  I would personally
find it useful for there to be one place I can go to see what people want
to get in.

				Jeff

[uml-devel] recursive fault

[Jon Smirl <jonsmirl@ya...>]

I know now how I got into the recursive fault. It was
the first case I hit where I have done an mmap in one
kernel process and then tried to use it from another
that was forked before the mmap was made. I ran the
same code in skas mode and got much further. 

Now I guess I get to learn all about fault handling in
UML so that  I can catch this fault and build the
appropriate mapping. I have all of the needed data to
build the mmaping, I just need to associate the data
with the fault.

=====
Jon Smirl
jonsmirl@...

__________________________________________________
Do you Yahoo!?
Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
http://mailplus.yahoo.com

Re: [uml-devel] Debugging "Kernel panic: Segfault with no mm"

[Jeff Dike <jdike@ka...>]

david@... said:
> although gdb isn't being all that helpful. How should I go about
> debugging this thing? 

I don't know how unhelpful gdb is being, but the first step would be a stack
trace.  If 'bt' isn't working, then x/xa starting at the current stack pointer
might provide a clue as to what's on the stack.

				Jeff

Re: [uml-devel] System call 0=sys_ni_syscall???

[Jeff Dike <jdike@ka...>]

JEFFREY.R.GUY@... said:
>   [ 0 ] = sys_ni_syscall, which suggests to me that system call 0 is
> not implemented.  Why, then, is it being called so often? 

That's a bit strange.  Where are they being called from?  I've never seen
a syscall 0 executed.

				Jeff

Re: [uml-devel] [PATCH] Fix uml_utilities make install

[Jeff Dike <jdike@ka...>]

pasky@... said:
> there is a weirdness with the uml_utilities --- you can set the
> installation prefixes at the root Makefile, but the other Makefiles
> just ignore it and override it by the default value.

It's a bug that the lower Makefiles just override the upper Makefile values
for LIB_DIR and BIN_DIR, but it's a totally trivial one.

You seem not to have noticed DESTDIR, which is honored by all of the Makefiles.

> The attached patch (against uml_utilities_20030205) makes the
> Makefiles in subdirs honour the user's choice of installation prefix
> made in the root Makefile. Please apply.

Umm, no.  This is fixed now, but correctly.

Before proposing that stuff be blindly deleted, you might try to figure out
what purpose it has, and whether you can fix whatever you think the problem
while preserving that.

				Jeff

[uml-devel] Patch archive

[David Coulson <david@da...>]

Hi,

	I'd like to start compiling a public 'patch archive' for UML. I know 
Jeff has started to merge many patches in, but I'd like to keep track of 
patches people submit and stuff.

So, anyone who has patches which are non-trivial, please tell me where I 
can grab them so I can put them somewhere useful. If you ask nicely, I 
may also be able to offer some web space for distribution and such like.

David

-- 
David Coulson                                    email: d@...
Linux Developer /                          web: http://davidcoulson.net/
Network Engineer                                   phone: (216) 533-6967

Re: [uml-devel] get_vm()

[Jeff Dike <jdike@ka...>]

jonsmirl@... said:
> You told me I had to leave page size gaps between the blocks. Are
> these gaps needed or not?

Sigh.  There have to be gaps between the pieces that are actually used.
Extending the vm area one page beyond what's actually going to be used is
a clean way to close up the vm areas.

				Jeff

[uml-devel] Re: [uml-user] Using Event Notification

[Jeff Dike <jdike@ka...>]

lugzug@... said:
> mconsole_notify - sendto failed, errno = 111 

~ 1034: perror 111
Connection refused

> Do I need something listening on the socket BEFORE I launch the UML? 

You at least need something bound to the socket beforehand.  Your script can
bind the socket, launch UML, and then go back to listening on the socket.

				Jeff

Re: [uml-devel] UML kernel fsync() ??

[William Stearns <wstearns@po...>]

Good evening, Jeff,

On Fri, 7 Feb 2003, Jeff Dike wrote:

> wstearns@... said:
> > At the risk of sounding like an idiot, it would _seem_ to me that  if
> > an app performed an fsync() asking the uml kernel to commit the data
> > to  disk, that the uml kernel should also ask the host to commit its
> > data to  disk as well.
> 
> There's no sync operation in the block device interface.  The is only relevant
> operation is write.  When a block device sees writes, it does them, and has
> no way of knowing what the reason for them is.

	Aha!  That makes sense now.
	My mistake - thanks for clearing that up.
	Cheers,
	- Bill

---------------------------------------------------------------------------
	"Do what you love, and love what you do, and you will never work
another day in your life."
--------------------------------------------------------------------------
William Stearns (wstearns@...).  Mason, Buildkernel, freedups, p0f,
rsync-backup, ssh-keyinstall, dns-check, more at:   http://www.stearns.org
Linux articles at:                         http://www.opensourcedigest.com
--------------------------------------------------------------------------

Re: [uml-devel] get_vm()

[Jon Smirl <jonsmirl@ya...>]

--- Jeff Dike <jdike@...> wrote:
> Did you test that at all?  The assignment of
> this->end + PAGE_SIZE to start
> will leave the holes that you want to get rid of.

What is the right fix for the problem? You told me I
had to leave page size gaps between the blocks. Are
these gaps needed or not?  All I know is that if you
do get_vm(1) your current code will fail. I'm trying
to fix that problem.

I do need a free_vm() routine. DMA buffers get mapped
in and out of kernel address space in the network card
drivers.

This is the get_vm() diff from the last patch I sent.
The only change is to round up the length to page
size.

===== mem.c 1.9 vs edited =====
--- 1.9/arch/um/kernel/mem.c    Sat Jan 18 07:08:21
2003
+++ edited/mem.c        Fri Feb  7 19:58:42 2003
@@ -502,7 +502,7 @@
 {
        struct vm_reserved *this, *next;
        struct list_head *ele;
-       unsigned long start;
+       unsigned long start, stop;
        int err;

        down(&vm_reserved_sem);
@@ -518,10 +518,12 @@
  found:
        up(&vm_reserved_sem);
        start = (unsigned long)
UML_ROUND_UP(this->end) + PAGE_SIZE;
-       err = reserve_vm(start, start + len, NULL);
+       stop = start + (unsigned long)
UML_ROUND_UP(len);
+       err = reserve_vm(start, stop, NULL);
        if(err) return(0);
        return(start);
 }
+

Would something like this be clearer:

unsigned long get_vm(unsigned long len)
{
	struct vm_reserved *this, *next;
	struct list_head *ele;
	unsigned long start;
	int err;
	
	len = (unsigned long) UML_ROUND_UP(len);
	
	down(&vm_reserved_sem);
	list_for_each(ele, &vm_reserved){
		this = list_entry(ele, struct vm_reserved, list);
		next = list_entry(ele->next, struct vm_reserved,
list);
		if((this->start < next->start) && 
		   (this->end + len + PAGE_SIZE <= next->start))
			goto found;
	}
	up(&vm_reserved_sem);
	return(0);
 found:
	up(&vm_reserved_sem);
	start = this->end + PAGE_SIZE;
	err = reserve_vm(start, start + len, NULL);
	if(err) return(0);
	return(start);
}


=====
Jon Smirl
jonsmirl@...

__________________________________________________
Do you Yahoo!?
Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
http://mailplus.yahoo.com

[uml-devel] linux-2.4.21-pre4-djc1-1um released

[David Coulson <david@da...>]

I finally got myself up to date with the current UML release, so I've 
applied lots of useful patches:

Linux 2.4.20
Linux 2.4.21-pre4
UML Patch 2.4.19-50um
Freeswan 1.99
MPPE 2.2
Netfilter patch-o-matic 'base' dated 20030207
bridge-nf-0.0.7

The kernel binary, along with modules for PPP MPPE and a diff can be 
downloaded from:

http://kernels.usermodelinux.org/kernels/linux-2.4.21-pre4-djc1-1um/

David

-- 
David Coulson                                    email: d@...
Linux Developer /                          web: http://davidcoulson.net/
Network Engineer                                   phone: (216) 533-6967

Re: [uml-devel] UML kernel fsync() ??

[Jeff Dike <jdike@ka...>]

wstearns@... said:
> At the risk of sounding like an idiot, it would _seem_ to me that  if
> an app performed an fsync() asking the uml kernel to commit the data
> to  disk, that the uml kernel should also ask the host to commit its
> data to  disk as well.

There's no sync operation in the block device interface.  The is only relevant
operation is write.  When a block device sees writes, it does them, and has
no way of knowing what the reason for them is.

				Jeff

Re: [uml-devel] get_vm()

[Jeff Dike <jdike@ka...>]

jonsmirl@... said:
> After I get_vm() where is the routine to free it? 

There is none.  Up to now, all vm allocations done there have been permanent.

> If search_extable() doesn't find what it is looking for it returns
> NULL. 

Right, I missed the ->fixup attached to the call to search_exception_tables.

> The get_vm() problem is that it is possible for a small request to
> fall into the leftover space at the end of the previous allocated
> block. I attached another fix. 

Did you test that at all?  The assignment of this->end + PAGE_SIZE to start
will leave the holes that you want to get rid of.

				Jeff