[PATCH] Check if an erase is needed to write

[Carl-Daniel Hailfinger]

For optimal partial reflashing, we have to find out which parts of the
chip can be written without erase. For that, the only criterion (except
a limit on the number of writes for very old chips) is whether the write
will only clear bits (set them to 0).
If (current&new==new) we can skip the erase.
If any bit would have to be set to 1, we need to erase.

Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006(a)gmx.net>

Index: flashrom-need_erase/flash.h
===================================================================
--- flashrom-need_erase/flash.h	(Revision 725)
+++ flashrom-need_erase/flash.h	(Arbeitskopie)
@@ -455,6 +455,7 @@
 int max(int a, int b);
 int check_erased_range(struct flashchip *flash, int start, int len);
 int verify_range(struct flashchip *flash, uint8_t *cmpbuf, int start, int len, char
*message);
+int need_erase(uint8_t *have, uint8_t *want, int len);
 char *strcat_realloc(char *dest, const char *src);
 
 #define OK 0
Index: flashrom-need_erase/flashrom.c
===================================================================
--- flashrom-need_erase/flashrom.c	(Revision 725)
+++ flashrom-need_erase/flashrom.c	(Arbeitskopie)
@@ -379,6 +379,26 @@
 	return ret;
 }
 
+/**
+ * Check if the buffer have can be programmed to the content of want without
+ * erasing. This is only possible if no bit has to be set to 1.
+ *
+ * @have        buffer with current content
+ * @want        buffer with desired content
+ * @len         length of the verified area
+ * @return      0 if no erase is needed, >0 otherwise
+ */
+int need_erase(uint8_t *have, uint8_t *want, int len)
+{
+	int failcount = 0;
+	int i;
+
+	for (i = 0; i < len; i++)
+		if ((have[i] & want[i]) != want[i])
+			failcount++;
+	return failcount;
+}
+
 struct flashchip *probe_flash(struct flashchip *first_flash, int force)
 {
 	struct flashchip *flash;


-- 
http://www.hailfinger.org/

[Stefan Reinauer]

Carl-Daniel Hailfinger wrote:
  For optimal partial reflashing, we have to find out
which parts of the
 chip can be written without erase. For that, the only criterion (except
 a limit on the number of writes for very old chips) is whether the write
 will only clear bits (set them to 0).
 If (current&new==new) we can skip the erase.
 If any bit would have to be set to 1, we need to erase.

Is that sufficient? Ie is it always ok to skip an erase if we're only
clearing bits?

[Carl-Daniel Hailfinger]

On 16.09.2009 19:07, Stefan Reinauer wrote:
  Carl-Daniel Hailfinger wrote:
   
  For optimal partial reflashing, we have to find
out which parts of the
 chip can be written without erase. For that, the only criterion (except
 a limit on the number of writes for very old chips) is whether the write
 will only clear bits (set them to 0).
 If (current&new==new) we can skip the erase.
 If any bit would have to be set to 1, we need to erase.

   
 Is that sufficient? Ie is it always ok to skip an erase if we're only
 clearing bits?
It depends on the chip, but given that the following appears in the ICH7
and all later datasheets as absolute flash requirement, I'm positive
most current flash chips support it.

-----------------
The system BIOS and Intel® Active Management Technology firmware usage
models
require that the serial flash device support multiple writes (minimum of
512 writes) to
a page (256 bytes) without requiring a preceding erase command.
-----------------

Flash vendors typically do not mention this in their datasheets, but
many BIOSes use flash for parameter storage (heck, even boot date/time
logging) since years.

We could restrict this feature to SPI chips only (which are almost
guaranteed to be recent enough), but since this patch only provides
infrastructure which is not used yet, I'd say the patch is harmless and
useful for future partial flashing tries.

Regards,
Carl-Daniel

-- 
http://www.hailfinger.org/

[Stefan Reinauer]

Carl-Daniel Hailfinger wrote:
  On 16.09.2009 19:07, Stefan Reinauer wrote:
   
  Carl-Daniel Hailfinger wrote:
   
     
  For optimal partial reflashing, we have to find
out which parts of the
 chip can be written without erase. For that, the only criterion (except
 a limit on the number of writes for very old chips) is whether the write
 will only clear bits (set them to 0).
 If (current&new==new) we can skip the erase.
 If any bit would have to be set to 1, we need to erase.

   
     
 Is that sufficient? Ie is it always ok to skip an erase if we're only
 clearing bits?
   
 It depends on the chip, but given that the following appears in the ICH7
 and all later datasheets as absolute flash requirement, I'm positive
 most current flash chips support it.

 -----------------
 The system BIOS and Intel® Active Management Technology firmware usage
 models
 require that the serial flash device support multiple writes (minimum of
 512 writes) to
 a page (256 bytes) without requiring a preceding erase command.
 -----------------
That means you can theoretically write it twice (2*256 byte writes on an
x8 device) without erasing it.
I don't think this is something we should rely on.

It also says serial flash device, which implies that LPC/FWH might
behave differently.

Stefan

[Carl-Daniel Hailfinger]

On 16.09.2009 23:40, Stefan Reinauer wrote:
  Carl-Daniel Hailfinger wrote:
   
  On 16.09.2009 19:07, Stefan Reinauer wrote:
   
     
  Carl-Daniel Hailfinger wrote:
   
     
       
  For optimal partial reflashing, we have to find
out which parts of the
 chip can be written without erase. For that, the only criterion (except
 a limit on the number of writes for very old chips) is whether the write
 will only clear bits (set them to 0).
 If (current&new==new) we can skip the erase.
 If any bit would have to be set to 1, we need to erase.

   
     
       
 Is that sufficient? Ie is it always ok to skip an erase if we're
only
 clearing bits?
   
     
 It depends on the chip, but given that the following appears in the ICH7
 and all later datasheets as absolute flash requirement, I'm positive
 most current flash chips support it.

 -----------------
 The system BIOS and Intel® Active Management Technology firmware usage
 models
 require that the serial flash device support multiple writes (minimum of
 512 writes) to
 a page (256 bytes) without requiring a preceding erase command.
 -----------------
   
 That means you can theoretically write it twice (2*256 byte writes on an
 x8 device) without erasing it.
 I don't think this is something we should rely on.

 It also says serial flash device, which implies that LPC/FWH might
 behave differently.
How should we handle the case where a chip is already erased? Do we
erase anyway? After all, we can't know if someone already wrote 0xff there.

Regards,
Carl-Daniel

-- 
http://www.hailfinger.org/

[Carl-Daniel Hailfinger]

On 17.09.2009 12:41, Carl-Daniel Hailfinger wrote:
  On 16.09.2009 23:40, Stefan Reinauer wrote:
   
  Carl-Daniel Hailfinger wrote:  
     
  On 16.09.2009 19:07, Stefan Reinauer wrote:
       
  Carl-Daniel Hailfinger wrote:
         
> For optimal partial reflashing, we have to find out which parts of the
> chip can be written without erase. For that, the only criterion (except
> a limit on the number of writes for very old chips) is whether the write
> will only clear bits (set them to 0).
> If (current&new==new) we can skip the erase.
> If any bit would have to be set to 1, we need to erase.
>
>           
 Is that sufficient? Ie is it always ok to skip an erase if we're only
 clearing bits?

 It depends on the chip, but given that the following appears in the ICH7
 and all later datasheets as absolute flash requirement, I'm positive
 most current flash chips support it.

 -----------------
 The system BIOS and Intel® Active Management Technology firmware usage
 models
 require that the serial flash device support multiple writes (minimum of
 512 writes) to
 a page (256 bytes) without requiring a preceding erase command.
 -----------------

 That means you can theoretically write it twice (2*256 byte writes on an
 x8 device) without erasing it.
 I don't think this is something we should rely on.

 It also says serial flash device, which implies that LPC/FWH might
 behave differently.

 How should we handle the case where a chip is already erased? Do we
 erase anyway? After all, we can't know if someone already wrote 0xff there.
To summarize: Write granularity is chip specific. The following write
granularities exist according to my datasheet survey:
- 1 bit. Each bit can be cleared individually.
- 1 byte. A byte can be written once. Further writes to an already
written byte cause the contents to be either undefined or to stay unchanged.
- 128 bytes. If less than 128 bytes are written, the rest will be
erased. Each write to a 128-byte region will trigger an automatic erase
before anything is written. Very uncommon behaviour.
- 256 bytes. If less than 256 bytes are written, the contents of the
unwritten bytes are undefined.

Note that chips with default 256-byte writes, which keep the original
contents for unwritten bytes, have a granularity of 1 byte.

This patch covers the 1-bit granularity variant, and since it is not
hooked up anywhere, I think it is a good starting point for implementing
a check for all variants. We will need an additional field in struct
flashchip for the minimum write granularity, though.

An ack would be appreciated. I'm open to adding the explanation above as
comment to the function and/or adding it to the commit message.

Regards,
Carl-Daniel

-- 
Developer quote of the month: 
"We are juggling too many chainsaws and flaming arrows and tigers."

[Carl-Daniel Hailfinger]

On 19.11.2009 17:51, Carl-Daniel Hailfinger wrote:
  On 17.09.2009 12:41, Carl-Daniel Hailfinger wrote:
   
  On 16.09.2009 23:40, Stefan Reinauer wrote:
   
     
  Carl-Daniel Hailfinger wrote:  
     
       
  On 16.09.2009 19:07, Stefan Reinauer wrote:
       
         
> Carl-Daniel Hailfinger wrote:
>         
>           
>> For optimal partial reflashing, we have to find out which parts of the
>> chip can be written without erase. For that, the only criterion (except
>> a limit on the number of writes for very old chips) is whether the write
>> will only clear bits (set them to 0).
>> If (current&new==new) we can skip the erase.
>> If any bit would have to be set to 1, we need to erase.
>>
>>           
>>             
> Is that sufficient? Ie is it always ok to skip an erase if we're only
> clearing bits?
>
>         
>           
 It depends on the chip, but given that the following appears in the ICH7
 and all later datasheets as absolute flash requirement, I'm positive
 most current flash chips support it.

 -----------------
 The system BIOS and Intel® Active Management Technology firmware usage
 models
 require that the serial flash device support multiple writes (minimum of
 512 writes) to
 a page (256 bytes) without requiring a preceding erase command.
 -----------------

       
 That means you can theoretically write it twice (2*256 byte writes on an
 x8 device) without erasing it.
 I don't think this is something we should rely on.

 It also says serial flash device, which implies that LPC/FWH might
 behave differently.

     
 How should we handle the case where a chip is already erased? Do we
 erase anyway? After all, we can't know if someone already wrote 0xff there.
   
 To summarize: Write granularity is chip specific. The following write
 granularities exist according to my datasheet survey:
 - 1 bit. Each bit can be cleared individually.
 - 1 byte. A byte can be written once. Further writes to an already
 written byte cause the contents to be either undefined or to stay unchanged.
 - 128 bytes. If less than 128 bytes are written, the rest will be
 erased. Each write to a 128-byte region will trigger an automatic erase
 before anything is written. Very uncommon behaviour.
 - 256 bytes. If less than 256 bytes are written, the contents of the
 unwritten bytes are undefined.

 Note that chips with default 256-byte writes, which keep the original
 contents for unwritten bytes, have a granularity of 1 byte.

 This patch covers the 1-bit granularity variant, and since it is not
 hooked up anywhere, I think it is a good starting point for implementing
 a check for all variants. We will need an additional field in struct
 flashchip for the minimum write granularity, though.
New patch. Handle 1-bit, 1-byte and 256-byte write granularity.

Stefan: I believe this addresses your concerns.

Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006(a)gmx.net>

Index: flashrom-need_erase/flash.h
===================================================================
--- flashrom-need_erase/flash.h	(Revision 770)
+++ flashrom-need_erase/flash.h	(Arbeitskopie)
@@ -485,6 +485,12 @@
 int bitbang_spi_write_256(struct flashchip *flash, uint8_t *buf);
 
 /* flashrom.c */
+enum write_granularity {
+	write_gran_1bit,
+	write_gran_1byte,
+	write_gran_256bytes,
+};
+
 extern char *programmer_param;
 extern int verbose;
 extern const char *flashrom_version;
@@ -496,6 +502,7 @@
 int max(int a, int b);
 int check_erased_range(struct flashchip *flash, int start, int len);
 int verify_range(struct flashchip *flash, uint8_t *cmpbuf, int start, int len, char
*message);
+int need_erase(uint8_t *have, uint8_t *want, int len, enum write_granularity gran);
 char *strcat_realloc(char *dest, const char *src);
 
 #define OK 0
Index: flashrom-need_erase/flashrom.c
===================================================================
--- flashrom-need_erase/flashrom.c	(Revision 770)
+++ flashrom-need_erase/flashrom.c	(Arbeitskopie)
@@ -407,6 +407,55 @@
 	return ret;
 }
 
+/**
+ * Check if the buffer have can be programmed to the content of want without
+ * erasing. This is only possible if no bit has to be set to 1.
+ *
+ * @have        buffer with current content
+ * @want        buffer with desired content
+ * @len         length of the verified area
+ * @return      0 if no erase is needed, 1 otherwise
+ */
+int need_erase(uint8_t *have, uint8_t *want, int len, enum write_granularity gran)
+{
+	int result = 0;
+	int i, j, limit;
+
+	switch (gran) {
+	case write_gran_1bit:
+		for (i = 0; i < len; i++)
+			if ((have[i] & want[i]) != want[i]) {
+				result = 1;
+				break;
+			}
+		break;
+	case write_gran_1byte:
+		for (i = 0; i < len; i++)
+			if ((have[i] != want[i]) && (have[i] != 0xff)) {
+				result = 1;
+				break;
+			}
+		break;
+	case write_gran_256bytes:
+		for (j = 0; j < len / 256; j++) {
+			limit = min (256, len - j * 256);
+			/* Are have and want identical? */
+			if (!memcmp(have + j * 256, want + j * 256, limit))
+				continue;
+			/* have needs to be in erased state. */
+			for (i = 0; i < limit; i++)
+				if (have[i] != 0xff) {
+					result = 1;
+					break;
+				}
+			if (result)
+				break;
+		}
+		break;
+	}
+	return result;
+}
+
 /* This function generates various test patterns useful for testing controller
  * and chip communication as well as chip behaviour.
  *


-- 
Developer quote of the month: 
"We are juggling too many chainsaws and flaming arrows and tigers."

[Carl-Daniel Hailfinger]

On 23.11.2009 15:33, Carl-Daniel Hailfinger wrote:
  On 19.11.2009 17:51, Carl-Daniel Hailfinger wrote:  
   
  To summarize: Write granularity is chip specific.
The following write
 granularities exist according to my datasheet survey:
 - 1 bit. Each bit can be cleared individually.
 - 1 byte. A byte can be written once. Further writes to an already
 written byte cause the contents to be either undefined or to stay unchanged.
 - 128 bytes. If less than 128 bytes are written, the rest will be
 erased. Each write to a 128-byte region will trigger an automatic erase
 before anything is written. Very uncommon behaviour.
 - 256 bytes. If less than 256 bytes are written, the contents of the
 unwritten bytes are undefined.
 New patch. Handle 1-bit, 1-byte and 256-byte write granularity.

 Stefan: I believe this addresses your concerns.

 Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006(a)gmx.net>
Ping?
This is http://patchwork.coreboot.org/patch/582/ in case you want to
look at the patch again.

Regards,
Carl-Daniel

-- 
Developer quote of the month: 
"We are juggling too many chainsaws and flaming arrows and tigers."

[Carl-Daniel Hailfinger]

This patch is 5 months old and it would be awesome if someone could
review it.
We need this if we ever want to support partial writes.

On 22.12.2009 02:38, Carl-Daniel Hailfinger wrote:
  On 23.11.2009 15:33, Carl-Daniel Hailfinger wrote:
   
  On 19.11.2009 17:51, Carl-Daniel Hailfinger
wrote:  
   
     
  To summarize: Write granularity is chip specific.
The following write
 granularities exist according to my datasheet survey:
 - 1 bit. Each bit can be cleared individually.
 - 1 byte. A byte can be written once. Further writes to an already
 written byte cause the contents to be either undefined or to stay unchanged.
 - 128 bytes. If less than 128 bytes are written, the rest will be
 erased. Each write to a 128-byte region will trigger an automatic erase
 before anything is written. Very uncommon behaviour.
 - 256 bytes. If less than 256 bytes are written, the contents of the
 unwritten bytes are undefined.
     
 New patch. Handle 1-bit, 1-byte and 256-byte write granularity.

 Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006(a)gmx.net>
   
 Ping?
 This is http://patchwork.coreboot.org/patch/582/ in case you want to
 look at the patch again.
Regards,
Carl-Daniel

-- 
Developer quote of the year:
"We are juggling too many chainsaws and flaming arrows and tigers."

[David Hendricks]

Looks good to me. Is there any reason not to ACK this? It seems like a
useful optimization, at least.

I think there may have been a small misunderstanding with the snippet
Carl-Daniel posted earlier. The note about usage model at the end of the
section might help alleviate fears of doing multiple writes without doing an
erase operation first. Here is the full section from the ICH7 datasheet from
April '07:

*5.25.4.3 Multiple Page Write Usage Model*
*         The system BIOS and Intel® Active Management Technology firmware
usage models*
*         require that the serial flash device support multiple writes
(minimum of 512 writes) to*
*         a page (256 bytes) without requiring a preceding erase command.
BIOS commonly*
*         uses capabilities such as counters that are typically implemented
by using byte writes*
*         to ‘increment’ the bits within a page that have been designated as
the counter. The*
*         Intel AMT firmware usage model requires the capability for
multiple data updates within*
*         any given page. These data updates occur via byte writes without
executing a*
*         preceding erase to the given page. Both the BIOS and Intel AMT
firmware multiple*
*         page write usage models apply to sequential and non-sequential
data writes.*
*Note:    This usage model requirement is based on any given bit only being
written once from a*
*         ‘1’ to a ‘0’ without requiring the preceding erase. An erase would
be required to change*
*         bits back to the ‘1’ state.*


On Thu, Feb 11, 2010 at 6:31 AM, Carl-Daniel Hailfinger <
c-d.hailfinger.devel.2006(a)gmx.net&gt; wrote:

  This patch is 5 months old and it would be awesome if
someone could
 review it.
 We need this if we ever want to support partial writes.

 On 22.12.2009 02:38, Carl-Daniel Hailfinger wrote:
  On 23.11.2009 15:33, Carl-Daniel Hailfinger
wrote:

> On 19.11.2009 17:51, Carl-Daniel Hailfinger wrote:
>
>
>> To summarize: Write granularity is chip specific. The following write
>> granularities exist according to my datasheet survey:
>> - 1 bit. Each bit can be cleared individually.
>> - 1 byte. A byte can be written once. Further writes to an already
>> written byte cause the contents to be either undefined or to stay
unchanged.
 >> - 128 bytes. If less than 128 bytes are
written, the rest will be
>> erased. Each write to a 128-byte region will trigger an automatic erase
>> before anything is written. Very uncommon behaviour.
>> - 256 bytes. If less than 256 bytes are written, the contents of the
>> unwritten bytes are undefined.
>>
>>
> New patch. Handle 1-bit, 1-byte and 256-byte write granularity.
>
> Signed-off-by: Carl-Daniel Hailfinger <
c-d.hailfinger.devel.2006(a)gmx.net&gt;
 
 Ping?
 This is http://patchwork.coreboot.org/patch/582/ in case you want to
 look at the patch again.
 Regards,
 Carl-Daniel

 --
 Developer quote of the year:
 "We are juggling too many chainsaws and flaming arrows and tigers."


 _______________________________________________
 flashrom mailing list
 flashrom(a)flashrom.org
 http://www.flashrom.org/mailman/listinfo/flashrom


-- 
David Hendricks (dhendrix)
Systems Software Engineer, Google Inc.

[Sean Nelson]

I see no problem with the patch, and it isn't bit-rotten or interferes 
with anything.
Acked-by: Sean Nelson <audiohacked(a)gmail.com>

[David Hendricks]

Stefan made a good point about it only applying to serial flash chips. But
as Carl-Daniel noted, this functionality is not used by anything at all
currently. Perhaps a comment or something can be added to note that it
should only be used for chips conforming to Intel's multiple page write
spec?

On Fri, Feb 26, 2010 at 7:39 PM, Sean Nelson <audiohacked(a)gmail.com> wrote:

  I see no problem with the patch, and it isn't
bit-rotten or interferes with
 anything.
 Acked-by: Sean Nelson <audiohacked(a)gmail.com>
Agreed. FWIW,

Acked-by: David Hendricks <dhendrix(a)google.com>

-- 
David Hendricks (dhendrix)
Systems Software Engineer, Google Inc.

[Carl-Daniel Hailfinger]

On 27.02.2010 08:00, David Hendricks wrote:
  Stefan made a good point about it only applying to
serial flash chips. But
 as Carl-Daniel noted, this functionality is not used by anything at all
 currently. Perhaps a comment or something can be added to note that it
 should only be used for chips conforming to Intel's multiple page write
 spec?
I added a lengthy comment about usage and about how the code applies to
chips. The good thing about the code is that with proper annotation in
struct flashchip, the code is usable for every chip out there regardless
of chip conformance with said Intel spec.

  On Fri, Feb 26, 2010 at 7:39 PM, Sean Nelson
<audiohacked(a)gmail.com> wrote:

   
  I see no problem with the patch, and it isn't
bit-rotten or interferes with
 anything.
 Acked-by: Sean Nelson <audiohacked(a)gmail.com>

 Agreed. FWIW,

 Acked-by: David Hendricks <dhendrix(a)google.com>
Thanks for the reviews!
Committed in r927.

Regards,
Carl-Daniel

-- 
"I do consider assignment statements and pointer variables to be among
computer science's most valuable treasures."
-- Donald E. Knuth