Return-path: Received: from hrndva-omtalb.mail.rr.com ([71.74.56.124]:40892 "EHLO hrndva-omtalb.mail.rr.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750925Ab1EEUjx (ORCPT ); Thu, 5 May 2011 16:39:53 -0400 Date: Thu, 05 May 2011 15:39:46 -0500 From: Larry Finger To: linux-wireless@vger.kernel.org Cc: chaoming_li@realsil.com.cn, linux-kernel@vger.kernel.org Subject: [RFC] rtlwifi: Remove large arrays from stack - a question Message-ID: <4dc30b12.uDIivxjeQot+s05F%Larry.Finger@lwfinger.net> (sfid-20110505_224012_943899_65A23081) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: linux-wireless-owner@vger.kernel.org List-ID: In driver rtlwifi, one routine places two relatively large arrays on the stack - a 1D u8 array of size 128, and a 2D array of u16 with 128 * 4 elements. With the next driver to be used with rtlwifi, the sizes will be 256 and 256 * 4 respectively. As that will make the 2D array be 2048 bytes, I have changed the code to use kmalloc to allocate the space as shown in the patch below. This code has been tested. This method performs 6 kmalloc operations. My understanding is that each one allocates a page of memory, which makes the code allocate 6 KiB rather than the previous 5 * 128 or 5 * 256 bytes. This routine is called once for each load of an upper-level driver that uses rtlwifi. Given that the large arrays on the stack needed to be removed, I see three possible options: 1. Leave it this way as 6 allocations of O(0) are trivial. 2. Switch to a single allocation of up to 1280 B and set the pointers manually. 3. Add the arrays to a structure maintained by the upper-level driver, which would make the memory usage be permanent. My inclination is to use option 1. Are there strong arguments for one of the other alternatives, or any that I have not considered? Thanks, Larry Index: wireless-testing-new/drivers/net/wireless/rtlwifi/efuse.c =================================================================== --- wireless-testing-new.orig/drivers/net/wireless/rtlwifi/efuse.c +++ wireless-testing-new/drivers/net/wireless/rtlwifi/efuse.c @@ -235,7 +235,7 @@ void read_efuse(struct ieee80211_hw *hw, { struct rtl_priv *rtlpriv = rtl_priv(hw); struct rtl_efuse *rtlefuse = rtl_efuse(rtl_priv(hw)); - u8 efuse_tbl[rtlpriv->cfg->maps[EFUSE_HWSET_MAX_SIZE]]; + u8 *efuse_tbl; u8 rtemp8[1]; u16 efuse_addr = 0; u8 offset, wren; @@ -245,7 +245,7 @@ void read_efuse(struct ieee80211_hw *hw, rtlpriv->cfg->maps[EFUSE_MAX_SECTION_MAP]; const u32 efuse_len = rtlpriv->cfg->maps[EFUSE_REAL_CONTENT_SIZE]; - u16 efuse_word[efuse_max_section][EFUSE_MAX_WORD_UNIT]; + u16 **efuse_word; u16 efuse_utilized = 0; u8 efuse_usage; @@ -256,9 +256,24 @@ void read_efuse(struct ieee80211_hw *hw, return; } + /* allocate memory for efuse_tbl and efuse_word */ + efuse_tbl = kmalloc(rtlpriv->cfg->maps[EFUSE_HWSET_MAX_SIZE] * + sizeof(u8), GFP_KERNEL); + if (!efuse_tbl) + return; + efuse_word = kmalloc(EFUSE_MAX_WORD_UNIT * sizeof(u16 *), GFP_KERNEL); + if (!efuse_word) + goto done; + for (i = 0; i < EFUSE_MAX_WORD_UNIT; i++) { + efuse_word[i] = kmalloc(efuse_max_section * sizeof(u16), + GFP_KERNEL); + if (!efuse_word[i]) + goto done; + } + for (i = 0; i < efuse_max_section; i++) for (j = 0; j < EFUSE_MAX_WORD_UNIT; j++) - efuse_word[i][j] = 0xFFFF; + efuse_word[j][i] = 0xFFFF; read_efuse_byte(hw, efuse_addr, rtemp8); if (*rtemp8 != 0xFF) { @@ -285,7 +300,7 @@ void read_efuse(struct ieee80211_hw *hw, read_efuse_byte(hw, efuse_addr, rtemp8); efuse_addr++; efuse_utilized++; - efuse_word[offset][i] = (*rtemp8 & 0xff); + efuse_word[i][offset] = (*rtemp8 & 0xff); if (efuse_addr >= efuse_len) break; @@ -297,7 +312,7 @@ void read_efuse(struct ieee80211_hw *hw, read_efuse_byte(hw, efuse_addr, rtemp8); efuse_addr++; efuse_utilized++; - efuse_word[offset][i] |= + efuse_word[i][offset] |= (((u16)*rtemp8 << 8) & 0xff00); if (efuse_addr >= efuse_len) @@ -320,9 +335,9 @@ void read_efuse(struct ieee80211_hw *hw, for (i = 0; i < efuse_max_section; i++) { for (j = 0; j < EFUSE_MAX_WORD_UNIT; j++) { efuse_tbl[(i * 8) + (j * 2)] = - (efuse_word[i][j] & 0xff); + (efuse_word[j][i] & 0xff); efuse_tbl[(i * 8) + ((j * 2) + 1)] = - ((efuse_word[i][j] >> 8) & 0xff); + ((efuse_word[j][i] >> 8) & 0xff); } } @@ -336,6 +351,11 @@ void read_efuse(struct ieee80211_hw *hw, (u8 *)&efuse_utilized); rtlpriv->cfg->ops->set_hw_reg(hw, HW_VAR_EFUSE_USAGE, (u8 *)&efuse_usage); +done: + for (i = 0; i < EFUSE_MAX_WORD_UNIT; i++) + kfree(efuse_word[i]); + kfree(efuse_word); + kfree(efuse_tbl); } bool efuse_shadow_update_chk(struct ieee80211_hw *hw)