hw/ppc: ppc440_uc: avoid multiply overflow in dcr_write_dma()
PATCH 1/1 hw/ppc: ppc440_uc: avoid multiply overflow in dcr_write_dma()
Coverity reports a OVERFLOW_BEFORE_WIDEN issue in dcr_write_dma(). When handling the DMA0_CR switch we're doing a multiplication between two integers (count and width), and the product is assigned to an uint64_t (xferlen). The int32 product can be overflow before widened.
Fix it by casting the first operand to uint64_t to force the product to be 64 bit.
--- hw/ppc/ppc440_uc.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
hw/ppc/ppc440_uc.c +1 -1
@@ -908,7 +908,7 @@ static void dcr_write_dma(void *opaque, int dcrn, uint32_t val) sidx = didx =0 ; width =1 << ((val &DMA0_CR_PW ) >>25 ); - xferlen = count * width; + xferlen = (uint64_t )count * width; wlen = rlen = xferlen; rptr =cpu_physical_memory_map (dma->ch[chnl].sa, &rlen, false);
Signature · 1 line
2.36.1
Quoted 4 lines · Daniel Henrique Barboza
Coverity reports a
OVERFLOW_BEFORE_WIDENissue indcr_write_dma(). When handling theDMA0_CRswitch we're doing a multiplication between two integers (count and width), and the product is assigned to anuint64_t(xferlen). The int32 product can be overflow before widened.
It can't in practice though as count is max 0xffff and width is max 8 so this sounds like a Coverity false positive. Maybe you could avoid it by changing the type of count or width to uint64_t instead of casting?
Quoted 5 lines
Fix it by casting the first operand to
uint64_tto force the product to be 64 bit.Fixes Coverity CID 1490852Fixes 3c409c1927ef ("ppc440_uc: Basic emulation of PPC440 DMA controller")
This line does not appear in 3c409c1927ef so this second Fixes line is bogus. This was added to fix other Coverity issues in eda3f17bcd7b96 but still did not make Coverity happy :-)
Regards,
BALATON Zoltan
Quoted 20 lines
Cc BALATON Zoltan <[email protected]>Signed-off-by Daniel Henrique Barboza <[email protected]>---
hw/ppc/ppc440_uc.c| 2 +-1 file changed, 1 insertion(+), 1 deletion(-)hw/ppc/ppc440_uc.c +1
-1@@ -908,7 +908,7 @@ static void dcr_write_dma(void *opaque, int dcrn, uint32_t val) sidx = didx =0 ; width =1 << ((val &DMA0_CR_PW ) >>25 ); - xferlen = count * width; + xferlen = (uint64_t )count * width; wlen = rlen = xferlen; rptr =cpu_physical_memory_map (dma->ch[chnl].sa, &rlen, false);
Quoted 7 lines · BALATON Zoltan
Quoted 4 lines · Daniel Henrique Barboza
Coverity reports a
OVERFLOW_BEFORE_WIDENissue indcr_write_dma(). When handling theDMA0_CRswitch we're doing a multiplication between two integers (count and width), and the product is assigned to anuint64_t(xferlen). The int32 product can be overflow before widened.It can't in practice though as count is max
0xffffand width is max 8 so this sounds like a Coverity false positive.
This time I believe it's worth fixing it. This code can be used as a base to do something similar somewhere else, where the variables don't have hard caps like we have here, and then we'll have a real overflow to deal with.
Quoted 1 line
Maybe you could avoid it by changing the type of count or width to
uint64_tinstead of casting?
That was my first idea but it would require more changes. I'll send a v2.
Quoted 8 lines
Quoted 6 lines
Fix it by casting the first operand to
uint64_tto force the product to be 64 bit.Fixes Coverity CID 1490852Fixes 3c409c1927ef ("ppc440_uc: Basic emulation of PPC440 DMA controller")This line does not appear in
3c409c1927efso this second Fixes line is bogus. This was added to fix other Coverity issues ineda3f17bcd7b96but still did not make Coverity happy :-)
My bad. I'll drop that "Fixes" line.
Daniel
Quoted 24 lines
Regards,
BALATON ZoltanQuoted 20 lines
Cc BALATON Zoltan <[email protected]>Signed-off-by Daniel Henrique Barboza <[email protected]>---
hw/ppc/ppc440_uc.c| 2 +-1 file changed, 1 insertion(+), 1 deletion(-)hw/ppc/ppc440_uc.c +1
-1@@ -908,7 +908,7 @@ static void dcr_write_dma(void *opaque, int dcrn, uint32_t val) sidx = didx = 0; width = 1 << ((val & DMA0_CR_PW) >> 25); - xferlen = count * width; + xferlen = (uint64_t)count * width; wlen = rlen = xferlen; rptr = cpu_physical_memory_map(dma->ch[chnl].sa, &rlen, false);
Balaton,
I had a change of heart. The code is too clear that it won't overflow. It felt overkill changing var types just for that.
Peter already marked it as Ignored - False Positive in Coverity as well. So this would be a code change to "look better". It didn't look particularly better after the changes I was about to send, so let's forget about it.
Daniel
Quoted 61 lines · Daniel Henrique Barboza
Quoted 7 lines · BALATON Zoltan
Quoted 4 lines · Daniel Henrique Barboza
Coverity reports a
OVERFLOW_BEFORE_WIDENissue indcr_write_dma(). When handling theDMA0_CRswitch we're doing a multiplication between two integers (count and width), and the product is assigned to anuint64_t(xferlen). The int32 product can be overflow before widened.It can't in practice though as count is max
0xffffand width is max 8 so this sounds like a Coverity false positive.This time I believe it's worth fixing it. This code can be used as a base to do something similar somewhere else, where the variables don't have hard caps like we have here, and then we'll have a real overflow to deal with.
Quoted 1 line
Maybe you could avoid it by changing the type of count or width to
uint64_tinstead of casting?That was my first idea but it would require more changes. I'll send a v2.
Quoted 8 lines
Quoted 6 lines
Fix it by casting the first operand to
uint64_tto force the product to be 64 bit.Fixes Coverity CID 1490852Fixes 3c409c1927ef ("ppc440_uc: Basic emulation of PPC440 DMA controller")This line does not appear in
3c409c1927efso this second Fixes line is bogus. This was added to fix other Coverity issues ineda3f17bcd7b96but still did not make Coverity happy :-)My bad. I'll drop that "Fixes" line.
Daniel
Quoted 24 lines
Regards,
BALATON ZoltanQuoted 20 lines
Cc BALATON Zoltan <[email protected]>Signed-off-by Daniel Henrique Barboza <[email protected]>---
hw/ppc/ppc440_uc.c| 2 +-1 file changed, 1 insertion(+), 1 deletion(-)hw/ppc/ppc440_uc.c +1
-1@@ -908,7 +908,7 @@ static void dcr_write_dma(void *opaque, int dcrn, uint32_t val) sidx = didx = 0; width = 1 << ((val & DMA0_CR_PW) >> 25); - xferlen = count * width; + xferlen = (uint64_t)count * width; wlen = rlen = xferlen; rptr = cpu_physical_memory_map(dma->ch[chnl].sa, &rlen, false);
Quoted 8 lines · Daniel Henrique Barboza
Balaton,
I had a change of heart. The code is too clear that it won't overflow. It felt overkill changing var types just for that.
Peter already marked it as Ignored - False Positive in Coverity as well. So this would be a code change to "look better". It didn't look particularly better after the changes I was about to send, so let's forget about it.
Fine with me. I may look at it later when Peter's second patch changing this code lands if there are any cleanups possible, I've noticed that now that we have xferlen = count * width it could be used elsewehere where that product appears for example. At that point I'll see if the type can be changed too. Until then it's fine as it is now.
Regards,
BALATON Zoltan
Quoted 9 lines · Daniel Henrique Barboza
Balaton,
I had a change of heart. The code is too clear that it won't overflow. It felt overkill changing var types just for that.
Peter already marked it as Ignored - False Positive in Coverity as well. So this would be a code change to "look better". It didn't look particularly better after the changes I was about to send, so let's forget about it.
Yeah, I marked it false positive because the limit on the two inputs to the multiply is very clear in the code. (Usually I put a note about why when I mark something an FP, but you have to expand the 'Triage' section in the right hand panel of the Coverity Scan web UI to show that text.)
-- PMM
Quoted 2 lines · BALATON Zoltan
I may look at it later when Peter's second patch changing this code lands if there are any cleanups possible
You mean the 2nd patch I sent in that RFC series? I'm not currently totally sure what I want to do with that. Looking at the code and at the docs that describe how the device works, there's definitely quite a bit of missing functionality. So I'm sort of undecided between "do a fair bit of overhaul to the device to fix up the more obviously missing pieces (but without much ability to test the changes)" and "just leave the code as it is in git at the moment, since it's sufficient for the only guest we know touches it to be able to boot". What would you prefer?
thanks
-- PMM
Quoted 5 lines · Peter Maydell
Quoted 2 lines · BALATON Zoltan
I may look at it later when Peter's second patch changing this code lands if there are any cleanups possible
You mean the 2nd patch I sent in that RFC series? I'm not
Yes I mean that patch.
Quoted 4 lines
currently totally sure what I want to do with that. Looking at the code and at the docs that describe how the device works, there's definitely quite a bit of missing functionality. So I'm sort of undecided between "do a fair
As the original commit message says I only aimed for the functionality AmigaOS seems to use and did not try to implement everything. What we have now seems to be enough for AmigaOS but I don't know if there are any other functionality it may want to use in rare cases that I don't know about or there are other guests that would do so.
Quoted 5 lines
bit of overhaul to the device to fix up the more obviously missing pieces (but without much ability to test the changes)" and "just leave the code as it is in git at the moment, since it's sufficient for the only guest we know touches it to be able to boot". What would you prefer?
I won't stop you if you want to improve this device but I also don't know how to test it apart from verifying AmigaOS still boots and works. It did with your second patch (I've sent a Tested-by) and I'll test any further patches. So just fixing the obviously wrong parts without adding any new functionality is probably enough if you don't want to spend too much time with it. I'll try to gather some traces on how AmigaOS uses it to help you to judge what's needed.
Thank you,
BALATON Zoltan