Re: [PATCH v2 05/20] kernel-shark: Add stream_id to kshark_entry

["Yordan Karadzhov (VMware)" <y.karadz@gmail.com>]

On 13.10.20 г. 3:05 ч., Steven Rostedt wrote:
> On Mon, 12 Oct 2020 16:35:08 +0300
> "Yordan Karadzhov (VMware)" <y.karadz@gmail.com> wrote:
> 
>> kshark_entry contains all information from one trace record, needed
>> in order to visualize the time-series of trace records. Here we
>> reorganize data fields to kshark_entry in order to make room for
>> the unique identifier of the Data stream this entry belongs to.
>>
>> Signed-off-by: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
>> ---
>>   src/libkshark.c |  2 +-
>>   src/libkshark.h | 12 +++++++++---
>>   2 files changed, 10 insertions(+), 4 deletions(-)
>>
>> diff --git a/src/libkshark.c b/src/libkshark.c
>> index 52aacd30..92e2450c 100644
>> --- a/src/libkshark.c
>> +++ b/src/libkshark.c
>> @@ -489,7 +489,7 @@ static inline void unset_event_filter_flag(struct kshark_context *kshark_ctx,
>>   	e->visible &= ~event_mask;
>>   }
>>   
>> -static void set_all_visible(uint16_t *v) {
>> +static void set_all_visible(uint8_t *v) {
>>   	/*  Keep the original value of the PLUGIN_UNTOUCHED bit flag. */
>>   	*v |= 0xFF & ~KS_PLUGIN_UNTOUCHED_MASK;
>>   }
>> diff --git a/src/libkshark.h b/src/libkshark.h
>> index 1165c512..fe0ba7f2 100644
>> --- a/src/libkshark.h
>> +++ b/src/libkshark.h
>> @@ -1,7 +1,7 @@
>>   /* SPDX-License-Identifier: LGPL-2.1 */
>>   
>>   /*
>> - * Copyright (C) 2017 VMware Inc, Yordan Karadzhov <y.karadz@gmail.com>
>> + * Copyright (C) 2017 VMware Inc, Yordan Karadzhov (VMware) <y.karadz@gmail.com>
>>    */
>>   
>>    /**
>> @@ -49,7 +49,10 @@ struct kshark_entry {
>>   	 * kshark_filter_masks to check the level of visibility/invisibility
>>   	 * of the entry.
>>   	 */
>> -	uint16_t	visible;
>> +	uint8_t		visible;
> 
> Why did the visible variable turn from 16bits to 8 bits?
> 
> That should be stated in the change log.
> 

After further discussing this on a call with Steven we decided to keep 
the "visible" field 16 bit and to define the "stream_id" to be a 16 bit 
integer as well. Those 16 bit will be taken from the event_id. This is 
OK because the Id of the trace event is coded with only 16 bit in the 
kernel anyway.

>> +
>> +	/** Data stream identifier. */
>> +	int8_t		stream_id;
>>   
>>   	/** The CPU core of the record. */
>>   	int16_t		cpu;
>> @@ -57,7 +60,7 @@ struct kshark_entry {
>>   	/** The PID of the task the record was generated. */
>>   	int32_t		pid;
>>   
>> -	/** Unique Id ot the trace event type. */
>> +	/** Unique Id of the trace event type. */
>>   	int32_t		event_id;
>>   
>>   	/** The offset into the trace file, used to find the record. */
>> @@ -319,6 +322,9 @@ struct kshark_data_stream {
>>   	struct kshark_data_stream_interface	interface;
>>   };
>>   
>> +/** Hard-coded maximum number of data stream. */
>> +#define KS_MAX_NUM_STREAMS	127
> 
> Will this become a variable?
> 
> Let's say we have a really powerful machine with a lot of memory. And
> we have a data file with 500 CPUs traced, and we want to look at all of
> them? Will this be a hard limit to change in the future without recompiling.
> 
> I've been hit by restrictions like this in code in the past, and know
> that things like this can stay in the code for a long time, to a point
> where they are no longer reasonable. I just want to make sure that we
> don't get stuck with that. I've learned to look at all hard limits with
> extra care, knowing that they are always a pain point in the future.


Now when we have 16 bit available for coding the stream Id, we no longer 
need to enforce such a small limit for the total number of streams loaded.

Thanks Steven!

Yordan

> 
> -- Steve
> 
> 
>> +
>>   /** Size of the task's hash table. */
>>   #define KS_TASK_HASH_SHIFT 16
>>   #define KS_TASK_HASH_SIZE (1 << KS_TASK_HASH_SHIFT)
>