On Oct 1, 2014, at 6:24 PM, Radim Krčmář <rkrcmar@xxxxxxxxxx> wrote: > 2014-09-30 20:49+0300, Nadav Amit: >> Intel SDM 17.2.4 (Debug Control Register (DR7)) says: "The processor clears the >> GD flag upon entering to the debug exception handler." This sentence may be >> misunderstood as if it happens only on #DB due to debug-register protection, >> but it happens regardless to the cause of the #DB. > > All real hardware behaves that way? I have no way of knowing. I know Intel’s phrasing is misleading, so I verified this behaviour in two ways: 1. I changed KVM not to trap #DB. I then changed kvm-unit-tests/debug.c to set DR7.GD prior to the watchpoint test, and printed once I entered the handler, before any DR was accessed by the handler. The result: we entered the handler once (afterwards I printed DR7 and saw GD is indeed clear). If #DB due to watchpoint did not clear GD, we would enter the handler twice. 2. I looked at bochs: https://github.com/larsr/bochs-svn/blob/master/cpu/exception.cc : if (vector == BX_DB_EXCEPTION) { // Commit debug events to DR6: preserve DR5.BS and DR6.BD values, // only software can clear them BX_CPU_THIS_PTR dr6.val32 = (BX_CPU_THIS_PTR dr6.val32 & 0xffff6ff0) | (BX_CPU_THIS_PTR debug_trap & 0x0000e00f); // clear GD flag in the DR7 prior entering debug exception handler BX_CPU_THIS_PTR dr7.set_GD(0); } > > Intel has another sentence after that > > [...], to allow the handler access to the debug registers. > > I suppose that the "the" is important, but I haven't verified it.[1] > Clearing GD on every #DB would also make the stated purpose[2] harder to > achieve without adding any benefit; it seems like a bug for Intel. The behaviour seems reasonable to me. Otherwise the CPU would re-enter the handler when the handler inspects DR6. > > > --- > 1: AMD [13.1.1.4 Debug-Control Register (DR7)] uses a similar wording > > General-Detect Enable (GD)—Bit 13. Software sets this bit to 1 to > cause a debug exception to occur when an attempt is made to execute > a MOV DRn instruction to any debug register (DR0–DR7). This bit is > cleared to 0 by the processor when the #DB handler is entered, > allowing the handler to read and write the DRn registers. The #DB > exception occurs before executing the instruction, and DR6[BD] is > set by the processor. Software debuggers can use this bit to > prevent the currently-executing program from interfering with the > debug operation. > > 2: Last sentence of [1] and also this from Intel > This condition is provided to support in-circuit emulators. > > When the emulator needs to access the debug registers, emulator > software can set the GD flag to prevent interference from the > program currently executing on the processor. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
