From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751600AbdKFCaJ (ORCPT ); Sun, 5 Nov 2017 21:30:09 -0500 Received: from mga09.intel.com ([134.134.136.24]:7168 "EHLO mga09.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751390AbdKFCaD (ORCPT ); Sun, 5 Nov 2017 21:30:03 -0500 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.44,350,1505804400"; d="scan'208";a="1033836772" From: Ricardo Neri To: Ingo Molnar , Thomas Gleixner , "H. Peter Anvin" , Andy Lutomirski , Borislav Petkov Cc: Peter Zijlstra , Andrew Morton , Brian Gerst , Chris Metcalf , Dave Hansen , Paolo Bonzini , Masami Hiramatsu , Huang Rui , Jiri Slaby , Jonathan Corbet , "Michael S. Tsirkin" , Paul Gortmaker , Vlastimil Babka , Chen Yucong , "Ravi V. Shankar" , Shuah Khan , linux-kernel@vger.kernel.org, x86@kernel.org, ricardo.neri@intel.com, Tony Luck , Ricardo Neri , Adam Buchbinder , Colin Ian King , Lorenzo Stoakes , Qiaowei Ren , Arnaldo Carvalho de Melo , Adrian Hunter , Kees Cook , Thomas Garnier , Dmitry Vyukov Subject: [PATCH v11 02/12] x86/insn-eval: Add support to resolve 32-bit address encodings Date: Sun, 5 Nov 2017 18:27:47 -0800 Message-Id: <1509935277-22138-3-git-send-email-ricardo.neri-calderon@linux.intel.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1509935277-22138-1-git-send-email-ricardo.neri-calderon@linux.intel.com> References: <1509935277-22138-1-git-send-email-ricardo.neri-calderon@linux.intel.com> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org 32-bit and 64-bit address encodings are identical. Thus, the same logic could be used to resolve the effective address. However, there are two key differences: address size and enforcement of segment limits. If running a 32-bit process on a 64-bit kernel, it is best to perform the address calculation using 32-bit data types. In this manner hardware is used for the arithmetic, including handling of signs and overflows. 32-bit addresses are generally used in protected mode; segment limits are enforced in this mode. This implementation obtains the limit of the segment associated with the instruction operands and prefixes. If the computed address is outside the segment limits, an error is returned. It is also possible to use 32-bit address in long mode and virtual-8086 mode by using an address override prefix. In such cases, segment limits are not enforced. Support to use 32-bit arithmetic is added to the utility functions that compute effective addresses. However, the end result is stored in a variable of type long (which has a width of 8 bytes in 64-bit builds). Hence, once a 32-bit effective address is computed, the 4 most significant bytes are masked out to avoid sign extension. The newly added function get_addr_ref_32() is almost identical to the existing function insn_get_addr_ref() (used for 64-bit addresses). The only difference is that it verifies that the effective address is within the limits of the segment. Cc: Dave Hansen Cc: Adam Buchbinder Cc: Colin Ian King Cc: Lorenzo Stoakes Cc: Qiaowei Ren Cc: Arnaldo Carvalho de Melo Cc: Masami Hiramatsu Cc: Adrian Hunter Cc: Kees Cook Cc: Thomas Garnier Cc: Peter Zijlstra Cc: Borislav Petkov Cc: Dmitry Vyukov Cc: Tony Luck Cc: Ravi V. Shankar Cc: x86@kernel.org Signed-off-by: Ricardo Neri --- arch/x86/lib/insn-eval.c | 112 ++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 106 insertions(+), 6 deletions(-) diff --git a/arch/x86/lib/insn-eval.c b/arch/x86/lib/insn-eval.c index a4427b4..e6cb68a 100644 --- a/arch/x86/lib/insn-eval.c +++ b/arch/x86/lib/insn-eval.c @@ -814,7 +814,11 @@ static int get_eff_addr_reg(struct insn *insn, struct pt_regs *regs, if (*regoff < 0) return -EINVAL; - *eff_addr = regs_get_register(regs, *regoff); + /* Ignore bytes that are outside the address size. */ + if (insn->addr_bytes == 4) + *eff_addr = regs_get_register(regs, *regoff) & 0xffffffff; + else /* 64-bit address */ + *eff_addr = regs_get_register(regs, *regoff); return 0; } @@ -846,7 +850,7 @@ static int get_eff_addr_modrm(struct insn *insn, struct pt_regs *regs, { long tmp; - if (insn->addr_bytes != 8) + if (insn->addr_bytes != 8 && insn->addr_bytes != 4) return -EINVAL; insn_get_modrm(insn); @@ -875,7 +879,13 @@ static int get_eff_addr_modrm(struct insn *insn, struct pt_regs *regs, tmp = regs_get_register(regs, *regoff); } - *eff_addr = tmp + insn->displacement.value; + if (insn->addr_bytes == 4) { + int addr32 = (int)(tmp & 0xffffffff) + insn->displacement.value; + + *eff_addr = addr32 & 0xffffffff; + } else { + *eff_addr = tmp + insn->displacement.value; + } return 0; } @@ -908,7 +918,7 @@ static int get_eff_addr_sib(struct insn *insn, struct pt_regs *regs, long base, indx; int indx_offset; - if (insn->addr_bytes != 8) + if (insn->addr_bytes != 8 && insn->addr_bytes != 4) return -EINVAL; insn_get_modrm(insn); @@ -946,12 +956,102 @@ static int get_eff_addr_sib(struct insn *insn, struct pt_regs *regs, else indx = regs_get_register(regs, indx_offset); - *eff_addr = base + indx * (1 << X86_SIB_SCALE(insn->sib.value)); + if (insn->addr_bytes == 4) { + int addr32, base32, idx32; + + base32 = base & 0xffffffff; + idx32 = indx & 0xffffffff; - *eff_addr += insn->displacement.value; + addr32 = base32 + idx32 * (1 << X86_SIB_SCALE(insn->sib.value)); + addr32 += insn->displacement.value; + + *eff_addr = addr32 & 0xffffffff; + } else { + *eff_addr = base + indx * (1 << X86_SIB_SCALE(insn->sib.value)); + *eff_addr += insn->displacement.value; + } return 0; } + +/** + * get_addr_ref_32() - Obtain a 32-bit linear address + * @insn: Instruction with ModRM, SIB bytes and displacement + * @regs: Register values as seen when entering kernel mode + * + * This function is to be used with 32-bit address encodings to obtain the + * linear memory address referred by the instruction's ModRM, SIB, + * displacement bytes and segment base address, as applicable. If in protected + * mode, segment limits are enforced. + * + * Returns: + * + * Linear address referenced by instruction and registers on success. + * + * -1L on error. + */ +static void __user *get_addr_ref_32(struct insn *insn, struct pt_regs *regs) +{ + unsigned long linear_addr = -1L, seg_base, seg_limit; + int eff_addr, regoff; + long tmp; + int ret; + + if (insn->addr_bytes != 4) + goto out; + + if (X86_MODRM_MOD(insn->modrm.value) == 3) { + ret = get_eff_addr_reg(insn, regs, ®off, &tmp); + if (ret) + goto out; + + eff_addr = tmp; + + } else { + if (insn->sib.nbytes) { + ret = get_eff_addr_sib(insn, regs, ®off, &tmp); + if (ret) + goto out; + + eff_addr = tmp; + } else { + ret = get_eff_addr_modrm(insn, regs, ®off, &tmp); + if (ret) + goto out; + + eff_addr = tmp; + } + } + + ret = get_seg_base_limit(insn, regs, regoff, &seg_base, &seg_limit); + if (ret) + goto out; + + /* + * In protected mode, before computing the linear address, make sure + * the effective address is within the limits of the segment. + * 32-bit addresses can be used in long and virtual-8086 modes if an + * address override prefix is used. In such cases, segment limits are + * not enforced. When in virtual-8086 mode, the segment limit is -1L + * to reflect this situation. + * + * After computed, the effective address is treated as an unsigned + * quantity. + */ + if (!user_64bit_mode(regs) && ((unsigned int)eff_addr > seg_limit)) + goto out; + + /* + * Data type long could be 64 bits in size. Ensure that our 32-bit + * effective address is not sign-extended when computing the linear + * address. + */ + linear_addr = (unsigned long)(eff_addr & 0xffffffff) + seg_base; + +out: + return (void __user *)linear_addr; +} + /* * return the address being referenced be instruction * for rm=3 returning the content of the rm reg -- 2.7.4