mtd: nftl: reduce stack usage in NFTL_movebuf()
PATCH 1/1 mtd: nftl: reduce stack usage in NFTL_movebuf()
From: Arnd Bergmann <[email protected]>
The code in the ntfl write function is rather complex, and it contains a 512 byte on-stack buffer. The combination of these two leads to using more than the per-function stack warning limit in some configurations, especially with KASAN enabled:
drivers/mtd/nftlcore.c:673:12: error: stack frame size (1328) exceeds limit (1280) in 'nftl_writeblock' [-Werror,-Wframe-larger-than]
Avoid this warning by moving the on-stack buffer into a separate function that only copies one part of the device to another.
This does not really help with the total maximum stack usage in the (non-KASAN) normal case, but it does two things:
- no single function has more than the warning limit
- the complexity goes down, so the parent function ends up spilling few local variables, and the total actually goes down slightly.
--- drivers/mtd/nftlcore.c | 43 +++++++++++++++++++++--------------------- 1 file changed, 21 insertions(+), 22 deletions(-)
drivers/mtd/nftlcore.c +21 -22
@@ -228,6 +228,25 @@ static u16 NFTL_findfreeblock(struct NFTLrecord *nftl, int desperate )return BLOCK_NIL ; } +static noinline_for_stack void NFTL_move_block(struct mtd_info *mtd, loff_t src, loff_t dst) +{ +unsigned char movebuf[512 ]; +struct nftl_oob oob; +size_t retlen; +int ret; + + ret =mtd_read (mtd, src,512 , &retlen, movebuf); +if (ret <0 && !mtd_is_bitflip (ret)) { + ret =mtd_read (mtd, src,512 , &retlen, movebuf); +if (ret != -EIO ) +printk ("Error went away on retry.\n" ); + } +memset (&oob,0xff ,sizeof (struct nftl_oob )); + oob.b.Status = oob.b.Status1 =SECTOR_USED ; + +nftl_write (mtd, dst,512 , &retlen, movebuf, (char *)&oob); +} +static u16 NFTL_foldchain (struct NFTLrecord *nftl,unsigned thisVUC,unsigned pendingblock ) {struct mtd_info *mtd = nftl->mbd.mtd;
@@ -389,9 +408,6 @@ static u16 NFTL_foldchain (struct NFTLrecord *nftl, unsigned thisVUC, unsigned p*/ pr_debug ("Folding chain %d into unit %d\n" , thisVUC, targetEUN);for (block =0 ; block < nftl->EraseSize /512 ; block++) { -unsigned char movebuf[512 ]; -int ret; -/* If it's in the target EUN already, or if it's pending write, do nothing */ if (BlockMap[block] == targetEUN || (pendingblock == (thisVUC * (nftl->EraseSize /512 ) + block))) {
@@ -403,25 +419,8 @@ static u16 NFTL_foldchain (struct NFTLrecord *nftl, unsigned thisVUC, unsigned pif (BlockMap[block] ==BLOCK_NIL )continue ; - ret =mtd_read (mtd, - (nftl->EraseSize * BlockMap[block]) + (block *512 ), -512 , - &retlen, - movebuf); -if (ret <0 && !mtd_is_bitflip (ret)) { - ret =mtd_read (mtd, - (nftl->EraseSize * BlockMap[block]) + (block *512 ), -512 , - &retlen, - movebuf); -if (ret != -EIO ) -printk ("Error went away on retry.\n" ); - } -memset (&oob,0xff ,sizeof (struct nftl_oob )); - oob.b.Status = oob.b.Status1 =SECTOR_USED ; - -nftl_write (nftl->mbd.mtd, (nftl->EraseSize * targetEUN) + - (block *512 ),512 , &retlen, movebuf, (char *)&oob); +NFTL_move_block (mtd, (nftl->EraseSize * BlockMap[block]) + (block *512 ), + (nftl->EraseSize * targetEUN) + (block *512 )); }/* add the header so that it is now a valid chain */
Signature · 1 line
2.39.5
Quoted 8 lines · Arnd Bergmann
The code in the ntfl write function is rather complex, and it contains a 512 byte on-stack buffer. The combination of these two leads to using more than the per-function stack warning limit in some configurations, especially with KASAN enabled:
drivers/mtd/nftlcore.c:673:12: error: stack frame size (1328) exceeds limit (1280) in 'nftl_writeblock' [-Werror,-Wframe-larger-than][...]
Applied to mtd/next, thanks!
[1/1] mtd: nftl: reduce stack usage in NFTL_movebuf()
commit: d22d5f47ecb2c4e140c0ee52204f4dd8fd0e85eb
Patche(s) should be available on mtd/linux.git and will be part of the next PR (provided that no robot complains by then).
Kind regards,
Miquèl