lkml.org 
[lkml]   [2011]   [Jul]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2 3/3] char drivers: ramoops debugfs entry
Il 01/07/2011 20:38, Sergiu Iordache ha scritto:
> On Fri, Jul 1, 2011 at 11:08 AM, Marco Stornelli
> <marco.stornelli@gmail.com> wrote:
>> Il 01/07/2011 03:28, Sergiu Iordache ha scritto:
>>>
>>> While ramoops writes to ram, accessing the dump requires using /dev/mem
>>> and knowing the memory location (or a similar solution). This patch
>>> provides a debugfs interface through which the respective memory
>>> area can be easily accessed. It also adds an entry to expose the record
>>> size which must be used to divide the memory area into individual dumps
>>> and a dump count entry.
>>>
>>
>> Good.
>>
>>> The entries added are:
>>> /sys/kernel/debug/ramoops/full - memory dump of the whole reserved area.
>>> /sys/kernel/debug/ramoops/count - number of dumps currently present
>>> (will be 0 after a restart).
>>
>> Is this count really needed?
>>
>>>
>>> Change-Id: Ifbab9af833be9ee0bc1c33bc9871f2fc0eb9d228
>>> Signed-off-by: Sergiu Iordache<sergiu@chromium.org>
>>> ---
>>> The patch was built on the 2.6.38 kernel and is based on the following
>>> patches which were applied from the mmotm tree:
>>> ramoops-add-new-line-to-each-print
>>> ramoops-use-module-parameters-instead-of-platform-data-if-not-available
>>>
>>> ramoops-use-module-parameters-instead-of-platform-data-if-not-available-checkpatch-fixes
>>>
>>> drivers/char/ramoops.c | 101
>>> ++++++++++++++++++++++++++++++++++++++++++++++++
>>> 1 files changed, 101 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/drivers/char/ramoops.c b/drivers/char/ramoops.c
>>> index f34077e..9c0e30e 100644
>>> --- a/drivers/char/ramoops.c
>>> +++ b/drivers/char/ramoops.c
>>> @@ -30,9 +30,15 @@
>>> #include<linux/platform_device.h>
>>> #include<linux/slab.h>
>>> #include<linux/ramoops.h>
>>> +#include<linux/uaccess.h>
>>> +#include<linux/debugfs.h>
>>>
>>> #define RAMOOPS_KERNMSG_HDR "===="
>>> #define MIN_MEM_SIZE 4096UL
>>> +#define RAMOOPS_DIR "ramoops"
>>> +#define RAMOOPS_FULL "full"
>>> +#define RAMOOPS_RS "record_size"
>>> +#define RAMOOPS_COUNT "count"
>>>
>>> static ulong record_size = 4096UL;
>>> module_param(record_size, ulong, 0400);
>>> @@ -67,6 +73,39 @@ static struct ramoops_context {
>>>
>>> static struct platform_device *dummy;
>>> static struct ramoops_platform_data *dummy_data;
>>> +static DEFINE_MUTEX(ramoops_mutex);
>>> +
>>> +/* Debugfs entries for ramoops */
>>> +static struct dentry *ramoops_dir, *ramoops_full_entry,
>>> *ramoops_rs_entry,
>>> + *ramoops_count_entry;
>>> +
>>> +/* Entry to have access to the whole memory area */
>>> +static ssize_t ramoops_read_full(struct file *file, char __user *buf,
>>> + size_t count, loff_t *ppos)
>>> +{
>>> + struct ramoops_context *cxt =&oops_cxt;
>>> +
>>> + mutex_lock(&ramoops_mutex);
>>> + if (*ppos + count> cxt->size)
>>> + count = cxt->size - *ppos;
>>> + if (*ppos> cxt->size) {
>>> + count = 0;
>>> + goto out;
>>> + }
>>> + if (copy_to_user(buf, cxt->virt_addr + *ppos, count)) {
>>> + count = -EFAULT;
>>> + goto out;
>>> + }
>>> + *ppos += count;
>>> +
>>> +out:
>>> + mutex_unlock(&ramoops_mutex);
>>> + return count;
>>> +}
>>> +
>>> +static const struct file_operations ramoops_full_fops = {
>>> + .read = ramoops_read_full,
>>> +};
>>>
>>> static void ramoops_do_dump(struct kmsg_dumper *dumper,
>>> enum kmsg_dump_reason reason, const char *s1, unsigned long
>>> l1,
>>> @@ -89,6 +128,7 @@ static void ramoops_do_dump(struct kmsg_dumper *dumper,
>>> if (reason == KMSG_DUMP_OOPS&& !cxt->dump_oops)
>>> return;
>>>
>>> + mutex_lock(&ramoops_mutex);
>>> buf = cxt->virt_addr + (cxt->count * cxt->record_size);
>>> buf_orig = buf;
>>>
>>> @@ -110,6 +150,7 @@ static void ramoops_do_dump(struct kmsg_dumper
>>> *dumper,
>>> memcpy(buf + l1_cpy, s2 + s2_start, l2_cpy);
>>>
>>> cxt->count = (cxt->count + 1) % cxt->max_count;
>>> + mutex_unlock(&ramoops_mutex);
>>> }
>>>
>>> static int __init ramoops_probe(struct platform_device *pdev)
>>> @@ -168,6 +209,51 @@ static int __init ramoops_probe(struct
>>> platform_device *pdev)
>>> goto fail1;
>>> }
>>>
>>> + /* Initialize debugfs entry so the memory can be easily accessed
>>> */
>>> + ramoops_dir = debugfs_create_dir(RAMOOPS_DIR, NULL);
>>> + if (ramoops_dir == NULL) {
>>> + err = -ENOMEM;
>>> + pr_err("debugfs directory entry creation failed\n");
>>> + goto out;
>>> + }
>>> +
>>> + ramoops_full_entry = debugfs_create_file(RAMOOPS_FULL, 0444,
>>> + ramoops_dir,
>>> NULL,&ramoops_full_fops);
>>> +
>>> + if (ramoops_full_entry == NULL) {
>>> + err = -ENOMEM;
>>> + pr_err("debugfs full entry creation failed\n");
>>> + goto no_ramoops_full;
>>> + }
>>> +
>>> + /*
>>> + * Since ramoops returns records of record_size it is useful to
>>> + * know the record size from userspace so we can parse the result
>>> + * Since the record size is usually small we don't mind converting
>>> + * it to a u32 from ulong.
>>> + */
>>> + ramoops_rs_entry = debugfs_create_u32(RAMOOPS_RS, 0444,
>>> + ramoops_dir, (u32
>>> *)&cxt->record_size);
>>> +
>>
>> Like above. The result can be parsed in an easy way due to
>> RAMOOPS_KERNMSG_HDR.
>
> I've added the count to make it easier to parse the records by getting
> record_size chunks out of the file. One of the problems I see is that
> you know the header (where it starts) but it's hard to find out where
> the last record in a series ends.

It was easy because the record size had a fixed length (4096), so maybe
at this point it can be sufficient the record size information. I see a
little problem however. I think we could use debugfs interface to dump
the log in an easy way but we should be able to dump it even with
/dev/mem. Specially on embedded systems, debugfs can be no mounted or
not available at all. So maybe, as you said below, with these new
patches we need a memset over all the memory area when the first dump is
taken. However, the original idea was to store even old dumps. In
addition, there's the problem to catch an oops after a start-up that
"clean" the area before we read it. At that point we lost the previous
dumps. To solve this we could use a "reset" paramater, but I think all
of this is a little overkilling. Maybe we can only bump up the record
size if needed. What do you think?

> By the way, is there any reason why the (whole) preserved ram area
> doesn't get cleared in ramoops when the first dump is taken? On one
> hand you could overwrite old dumps but they will get overwritten
> anyway and they are probably old and of no interest.
> Because right now after a dump you usually have:
>
> Useful data record (record_size size)
> Random data from ram (record_size size)
> Random data from ram (record_size size)
> Etc
>
> This makes it harder to parse (even if you know the record size, as
> there's a very small but possible chance that the header is in found
> in the random data without it being a valid dump).

The string "====" plus a timestamp? Maybe but really really difficult.

Marco


\
 
 \ /
  Last update: 2011-07-02 10:09    [W:0.741 / U:0.108 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site