Bug 145146 - vfree() in interrupt context
Summary: vfree() in interrupt context
Keywords:
Status: CLOSED WONTFIX
Alias: None
Product: Red Hat Enterprise Linux 3
Classification: Red Hat
Component: kernel
Version: 3.0
Hardware: i386
OS: Linux
medium
medium
Target Milestone: ---
Assignee: Brian Maly
QA Contact: Brian Brock
URL:
Whiteboard:
Depends On:
Blocks: 170417 186960
TreeView+ depends on / blocked
 
Reported: 2005-01-14 19:42 UTC by Wendy Cheng
Modified: 2007-11-30 22:07 UTC (History)
8 users (show)

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Clone Of:
Environment:
Last Closed: 2006-04-25 21:37:36 UTC
Target Upstream Version:
Embargoed:


Attachments (Terms of Use)

Description Wendy Cheng 2005-01-14 19:42:59 UTC
Description of problem:

In vfree(), we have the following:

        if (in_interrupt()) {
                printk("vfree(): sleeping in interrupt!! \n");
#ifdef __i386__
                show_stack(NULL);
#endif
        }

I assume the intention here is to catch the faulty kernel module(s) and fix the
issue. So we got one here: 

Aug 24 15:40:42 elonxdl585 kernel: vfree(): sleeping in interrupt!!
Aug 24 15:40:42 elonxdl585 kernel: ca347e80 f8bc7d0c 00000000 ffffffff f8bcef9a
f8bd7000 00004000 00000000
Aug 24 15:40:42 elonxdl585 kernel:        f8bc7d0c 00000010 f8bc7a65 00000000
00000000 001c0e7d f8bc7da0 f8bd4080
Aug 24 15:40:42 elonxdl585 kernel:        c037e100 f8bc7dd2 00000000 ffffffef
00000000 c0134626 00000000 00000001
Aug 24 15:40:42 elonxdl585 kernel: Call Trace:   [<f8bc7d0c>] _lock_fdc [floppy]
0xac (0xca347e84)
Aug 24 15:40:42 elonxdl585 kernel: [<f8bcef9a>] floppy_release_irq_and_dma
[floppy] 0x21a (0xca347e90)
Aug 24 15:40:42 elonxdl585 kernel: [<f8bc7d0c>] _lock_fdc [floppy] 0xac (0xca347ea0)
Aug 24 15:40:42 elonxdl585 kernel: [<f8bc7a65>] set_dor [floppy] 0x165 (0xca347ea8)
Aug 24 15:40:42 elonxdl585 kernel: [<f8bc7da0>] motor_off_callback [floppy] 0x0
(0xca347eb8)
Aug 24 15:40:42 elonxdl585 kernel: [<f8bd4080>] motor_off_timer [floppy] 0x0
(0xca347ebc)
Aug 24 15:40:42 elonxdl585 kernel: [<f8bc7dd2>] motor_off_callback [floppy] 0x32
(0xca347ec4)
Aug 24 15:40:42 elonxdl585 kernel: [<c0134626>] __run_timers [kernel] 0xb6
(0xca347ed4)
Aug 24 15:40:42 elonxdl585 kernel: [<c0134342>] timer_bh [kernel] 0x62 (0xca347f00)
Aug 24 15:40:42 elonxdl585 kernel: [<c012efb5>] bh_action [kernel] 0x55 (0xca347f14)
Aug 24 15:40:42 elonxdl585 kernel: [<c012ee57>] tasklet_hi_action [kernel] 0x67
(0xca347f1c)
Aug 24 15:40:42 elonxdl585 kernel: [<c012ebe5>] do_softirq [kernel] 0x105
(0xca347f30)
Aug 24 15:40:42 elonxdl585 kernel: [<c010db48>] do_IRQ [kernel] 0x148 (0xca347f50)
Aug 24 15:40:42 elonxdl585 kernel: [<c010da00>] do_IRQ [kernel] 0x0 (0xca347f74)
Aug 24 15:40:42 elonxdl585 kernel: [<c0109100>] default_idle [kernel] 0x0
(0xca347f7c)
Aug 24 15:40:42 elonxdl585 kernel: [<c0109100>] default_idle [kernel] 0x0
(0xca347f90)
Aug 24 15:40:42 elonxdl585 kernel: [<c0109129>] default_idle [kernel] 0x29
(0xca347fa4)
Aug 24 15:40:42 elonxdl585 kernel: [<c01091c2>] cpu_idle [kernel] 0x42 (0xca347fb0)
Aug 24 15:40:42 elonxdl585 kernel: [<c01287a3>] printk [kernel] 0x143 (0xca347fcc)

Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.
  
Actual results:


Expected results:


Additional info:

Comment 1 Wendy Cheng 2005-01-14 19:55:04 UTC
Since this happens in the timer bottom half code, messing around with irq, waitq
etc seems to be messy. Why don't we just change the vmalloc() inside the
fd_dma_mem_alloc() into a kmalloc() ? The dma buffer size is either 32K or 64K. 

The current code path is fd_dma_mem_alloc() --> fd_routine[vdma_mem_alloc] -->
vmalloc().  

Comment 3 Wendy Cheng 2005-01-14 20:37:14 UTC
oops, I spoke too soon. After digging into the code more, it is messier than I
thought. It seems that the default mem allocation method is indeed the
kmalloc(), it only switches into vmalloc() when not enough memory available
(can_use_virtual_dma is 2):

static int can_use_virtual_dma=2;
/* =======
 * can use virtual DMA:
 * 0 = use of virtual DMA disallowed by config
 * 1 = use of virtual DMA prescribed by config
 * 2 = no virtual DMA preference configured.  By default try hard DMA,
 * but fall back on virtual DMA when not enough memory available
 */

static int use_virtual_dma;
/* =======
 * use virtual DMA
"drivers/block/floppy.c" 4522L, 116862C

Two options come to mind:

1. Fix the callback function. If it finds itself in interrupt context while
trying to floppy_release_irq_and_dma(), delay and schedule the vfree() until it
can be safely done (say on next floppy_open() ?).
2. Dis-allow vmalloc - this virtually dis-allows floppy device to get opened
when lowmem is tight.



Comment 4 Wendy Cheng 2005-01-14 21:11:42 UTC
Found a global variable called floppy_track_buffer - seems to be a natural fit
for the option #1 in above comment.  

Comment 6 Jason Orendorf 2005-09-08 00:02:17 UTC
Any updates on this bug?  I just saw the same problem:

Sep  7 10:01:57 cust kernel: vfree(): sleeping in interrupt!!
Sep  7 10:01:57 cust kernel: cd503e80 f905cd0c 00000000 ffffffff f9063f9a
f9049000 00004000 00000000
Sep  7 10:01:58 cust kernel:	    f905cd0c 00000010 f905ca65 00000000 00000000
001c1bc5 f905cda0 f9069060
Sep  7 10:01:58 cust kernel:	    c0391900 f905cdd2 00000000 ffffffef 00000000
c0135396 00000000 0000003f
Sep  7 10:01:58 cust kernel: Call Trace:   [<f905cd0c>] _lock_fdc [floppy] 0xac
(0xcd503e84)
Sep  7 10:01:58 cust kernel: [<f9063f9a>] floppy_release_irq_and_dma [floppy]
0x21a (0xcd503e90)
Sep  7 10:01:58 cust kernel: [<f905cd0c>] _lock_fdc [floppy] 0xac (0xcd503ea0)
Sep  7 10:01:58 cust kernel: [<f905ca65>] set_dor [floppy] 0x165 (0xcd503ea8)
Sep  7 10:01:58 cust kernel: [<f905cda0>] motor_off_callback [floppy] 0x0
(0xcd503eb8)
Sep  7 10:01:58 cust kernel: [<f9069060>] motor_off_timer [floppy] 0x0 (0xcd503ebc)
Sep  7 10:01:58 cust kernel: [<f905cdd2>] motor_off_callback [floppy] 0x32
(0xcd503ec4)
Sep  7 10:01:58 cust kernel: [<c0135396>] __run_timers [kernel] 0xb6 (0xcd503ed4)
Sep  7 10:01:58 cust kernel: [<c01350b2>] timer_bh [kernel] 0x62 (0xcd503f00)
Sep  7 10:01:58 cust kernel: [<c012fd25>] bh_action [kernel] 0x55 (0xcd503f14)
Sep  7 10:01:58 cust kernel: [<c012fbc7>] tasklet_hi_action [kernel] 0x67
(0xcd503f1c)
Sep  7 10:01:58 cust kernel: [<c012f955>] do_softirq [kernel] 0x105 (0xcd503f30)
Sep  7 10:01:58 cust kernel: [<c010df38>] do_IRQ [kernel] 0x148 (0xcd503f50)
Sep  7 10:01:58 cust kernel: [<c010ddf0>] do_IRQ [kernel] 0x0 (0xcd503f74)
Sep  7 10:01:58 cust kernel: [<c0109100>] default_idle [kernel] 0x0 (0xcd503f7c)
Sep  7 10:01:58 cust kernel: [<c0109100>] default_idle [kernel] 0x0 (0xcd503f90)
Sep  7 10:01:58 cust kernel: [<c0109129>] default_idle [kernel] 0x29 (0xcd503fa4)
Sep  7 10:01:58 cust kernel: [<c01091c2>] cpu_idle [kernel] 0x42 (0xcd503fb0)
Sep  7 10:01:58 cust kernel: [<c01294f3>] printk [kernel] 0x153 (0xcd503fcc)

Is there a fix or workaround for this problem?

Comment 12 Ernie Petrides 2006-04-25 21:37:36 UTC
There was not enough time to resolve technical issues with the patch,
and RHEL3 is now closed.



Note You need to log in before you can comment on or make changes to this bug.